1 Implementation of a dynamic energy management system using real time

2 pricing and local renewable energy generation forecasts

3 Onur Elma, Akın Taşcıkaraoğlu, A. Tahir İnce, Uğur S. Selamoğulları1
4
5 Electrical Engineering Department, Yildiz Technical University, Esenler, Istanbul, Turkey
6
7 Abstract

8 The advanced communication and control technologies in smart grids enable end users to actively participate in balancing supply and demand

9 in response to electricity tariff changes by controlling their electricity consumption through demand response (DR) programs. In order to further

10 exploit the cost-saving potential in residential houses, home energy management (HEM) systems have gained increasing interest, particularly in

11 the last decade. HEM system basically focuses on the control of home appliances to reduce their electricity usage or to shift the operations of

12 predefined appliances to the periods with lower prices. However, the integration of local renewable generation units to the residential houses

13 considerably complicates the tasks of HEM systems. This study, therefore, proposes a novel dynamic HEM approach capable of integrating both

14 load and source side dynamics into decision-making process. In the new HEM approach, power consumption of appliances, electricity tariff and

15 power from renewable sources are dynamically taken into account with a 5-minute time step. A forecasting model is incorporated into the HEM

16 system for better matching of energy consumption to renewable energy generation. The simulation and experimental results show that the

17 proposed HEM system considerably improves cost savings for residential prosumers and can be implemented in real-world applications.

18 Keywords: Home energy management, renewable energy forecasting, dynamic control, real-time pricing, smart grid.
19

20 1. Introduction

21 A. Motivation and Background

22 Gradual increase in energy demand and energy prices, particularly in the developed and developing countries, raises the need

23 for further improvement at both electrical power generation side and consumption side. At the generation side, exploiting

24 renewable energy potential can be considered as a feasible and commercially-applicable method while various solutions such as

25 using grid connected storage systems, more efficient energy use in buildings, and implementing demand side management

26 programs, are generally adopted at the consumption side. In this regard, incorporating all these technologies in a common

27 environment has significant benefits from economic and technical points of view for both consumers and utilities. Smart homes

28 with residential power generating units, which have become widespread at the last decade with the emerging of smart grid

29 concepts, can be pointed out as one of the best examples of such advanced systems.

1
Corresponding author.
E-mail: selam@yildiz.edu.tr (Uğur Savaş Selamoğulları)
 2

30 A smart home can be defined as an intelligent home capable of remotely monitoring and controlling electrical appliances. In

31 these homes, a control system, called home energy management (HEM), is generally employed for the purpose of more efficient

32 energy use [1,2]. These homes might be equipped with various local energy generating units, such as wind turbines and solar

33 panels, supported by storage systems, for generating their own energy and even selling the excess energy to the grid if two-way

34 power flow is allowed. Two-way power flow also facilitates the use of different electric tariffs offered by utility companies, such

35 as time of use (ToU), real time pricing (RTP), critical peak pricing (CPP) in demand response (DR) applications [3–6].

36 The studies on DR applications generally focus on examining the effects of DR applications on power systems [7–14] or on the

37 power usage of large commercial/industrial buildings [15,16] due to several reasons such as: (i) higher contribution of the

38 aggregated power of many buildings or of large buildings on the power system operations, (ii) relatively predictable and

39 controllable characteristics of the aggregated power, and (iii) the lack of the availability of data at the residential level. However,

40 understanding the importance of the contribution of individual houses and appliances on the efficiency of DR applications,

41 particularly at the last decade, and also the increasing penetration of local power generation and storage units in houses have

42 directed the studies to DR implementations at residential dwelling and appliance levels. Furthermore, the development of high-

43 technology data collection tools (smart meters, smart plugs, etc.) has paved the way for these studies.

44 Residential DR applications aim to control the appliances with the objectives of reducing the overall power consumption and

45 transferring high-demands to off-peak hours. These two aims are of great importance for consumers to gain benefits using their

46 appliances based on the variations in electricity tariff. Residential DR applications also help utility companies to maintain short-

47 term balance between supply and demand. Reducing the overall demand by switching off certain appliances or shifting them to

48 off-peak periods, however, might disturb the end-user comfort to some extent. Minimization of end-user comfort violation in

49 residential DR applications while still accomplishing an improvement in power system operations and a reduction in consumer’s

50 electricity bills is an important topic from both the system operator and end-user points of view.

51 B. Literature Overview

52 A large number of studies dealing with DR applications has been presented in the literature. Among them, the efficiency of

53 heating, ventilation and cooling (HVAC) units in residential DR programs is investigated in different studies in terms of energy

54 savings [17], peak heating consumption reduction [18–20] and load balancing [21–23]. With similar objectives, the other

55 appliances, such as refrigerator [24], dishwasher [25] and electric water heater [26], are also considered in DR implementations.

56 A combination of various appliances is used in a number of DR applications for a higher controllable power potential. For instance,

57 wet appliances, namely, washing machine, dishwasher and tumble dryers, are considered in residential DR applications [27].

58 According to their preset priority, high-power appliances in a house, such as space cooling units, water heaters, clothes dryers and

59 electric vehicles (EV), are controlled with a model in [28]. Similarly, Chen et al. [29] and Fernandes et al. [30] investigate the
 3

60 dynamic and optimal operation of different appliances based on the user-defined preferences and the technical characteristics of

61 appliances. The load controlling operations are carried out to manage the power flow to the house in outage cases in [31]. Some

62 studies also use load forecasts for more effective scheduling in DR implementations [32,33]. Furthermore, several studies discuss

63 the contribution of customer surveys about the energy usage profiles on the appropriate scheduling tasks [34,35].

64 Domestic electrical appliances are generally classified into three different groups in the literature according to their

65 controllability: (i) interruptible loads that can be switched off for a short period of time, (ii) shiftable loads that can be programmed

66 to be used in a later period of time, and (iii) non-controllable loads that cause discomfort to the user when controlled. At the last

67 decade, the studies on matching the power generation of local generating and storage units located in or around the smart houses

68 with the consumption of interruptible loads, and on postponing the functioning of shiftable loads to the off-peak hours or to the

69 times in which the energy from renewable sources are high, have gained increased interest. For this purpose, Li and Hong present

70 a model that controls the loads regarding time-varying electricity tariff and stores the energy when the price is low and use the

71 stored energy when the price is high [36]. A mixed integer nonlinear programming (MINLP) model is proposed in [37] for optimal

72 energy usage in a grid-connected smart home, which has micro cogeneration system and underfloor heating/cooling unit as energy

73 sources, and a battery and a hot water tank as electric and thermal storage system, respectively.

74 Several studies also investigate the benefits of renewable energy sources on more efficient and cost-effective energy

75 management in smart homes. Considering the forecasting of weather characteristics, Wang et al. examine the DR operation of a

76 residential hybrid renewable energy system, including wind turbines, Photovoltaic (PV) panels and a battery group, to provide

77 energy to a single family house in a stand-alone mode [38]. Incorporating PV output power forecasts, several studies are conducted

78 for the grid-connected homes with PV panels and storage systems [39–41], and in addition to these components, a combined

79 cooling, heating and power (CCHP) system is used in [42] and [43].

80 Similar to the structure used in this study, excluding the smart home and experimental verification, various studies are realized

81 for more effective DR implementations combining PV panels with wind turbines in a grid-connected house [44–46]. Besides,

82 Cagigal et al. investigate DR operation in a real grid connected smart house with controllable appliances with PV forecasts and a

83 battery system [47]. In another smart home system, Lujano-Rojas et al. present a load management method using price forecasts,

84 energy consumption and renewable power generation, in which only one test day with hourly time step is discussed [48].

85 None of the literature studies given above considers an experimental DR implementation of a smart home with local renewable

86 energy sources. With this objective, only a few studies are available in the literature. Newsham et al. present a method for a smart

87 home with PV array [49], and Karfopoulos et al. employ a model in a smart home including wind turbine, PV panels and batteries

88 [50]. However, both studies consider a grid-independent smart house, which is a relatively easy task and considerably limits the

89 exploitation of renewable sources and batteries. Furthermore, these studies do not take a dynamic energy management into account
 4

90 within a short term time scale. To the best of our knowledge, the HEM algorithm proposed in this paper is the first study dealing

91 with the dynamic DR implementation of an experimental smart home having renewable energy sources and a battery.

92 Further explanations on the residential demand response implementations and a larger number of related literature studies can

93 be found in recent literature reviews, in which the demand response programs are investigated from the perspective of their

94 economic and operating benefits [51–54], from the perspective of behavioral factors in household demand response participation

95 [55,56], and from the perspective of their role in smart grids [57].

96 C. Contributions

97 Considering the drawbacks and advantages of the approaches presented in the literature and also the state-of-the-art on this

98 field, a novel HEM approach is presented in this study with four objectives: to minimize the energy cost in a residential smart

99 house particularly by reducing the household consumption when energy prices are high, to maximize the use of renewable energy

100 potential, to reduce the power profiles at the grid side by alleviating peak demand, and to run a dynamic home energy management

101 algorithm. In order for these objectives to be achieved, the balance between the power generation from the local renewable sources

102 and the energy consumption of the appliances in the house is maintained dynamically while avoiding energy purchases at high

103 costs as much as possible. For this purpose, at the generation side, the potential energy from the renewable sources is exploited

104 using a forecasting tool and at the consumption side, the operation of the appliances is proactively controlled in the cases of

105 insufficient generation in response to the existing energy price and taking the SOC of the battery and the end user’s preferences

106 into account.

107 The main contributions of the proposed study compared to the existing studies in this field are listed as follows:

108  Both load side and source side dynamics are taken into account in decision-making process with a 5-minute time step,

109 which considerably improves the efficiency of the real-time renewable energy resources scheduling and energy

110 management in small-scale systems. Such short-term management actions are also important for controlling of battery

111 storage systems as appropriate charge and discharge conditions for batteries must be satisfied at each time step.

112  Aiming to improve the profits on both source and load sides, a number of parameters is considered in the proposed

113 approach, namely, consumer load priorities, power demand of appliances, charge level of the battery and the time-varying

114 electricity price information.

115  Using a combined forecasting model, expected wind and solar generations from the residential renewable sources are

116 also considered in the scheduling process.

117  The effectiveness of the proposed approach is experimentally verified in a real smart home system comprising real

118 appliances, residential renewable energy sources (i.e., a wind turbine and PV panels) and battery storage.

119
 5

120 D. Organization of the paper

121 This paper is organized as follows: Section 2 describes the specifications of the smart home system. The proposed HEM system

122 structure is explained in detail in Section 3 and the renewable energy forecasting approach is summarized in Section 4. Section 5

123 presents the results with the objective of investigating the contribution of proposed HEM algorithm. The last section concludes

124 the study.

125 2. Description of the Smart Home System

126 The proposed HEM system is implemented in a real smart home, called Yildiz Technical University Smart Home Laboratory

127 (YTU-SHL). YTU-SHL has residential electricity generation units including a wind turbine and solar panels, as well as a storage

128 unit [59]. The smart home also includes various electrical appliances that can be monitored and controlled through internet-based

129 applications. A remotely-controllable inverter is used to connect to the grid when required. The local renewable sources,

130 controllable appliances and relevant control equipment make the real time control of both source side and load side possible.

131 2.1. Outline of the Smart Home System

132 A smart home including a living room with kitchen, a study room and bathroom has been built inside a research laboratory at

133 YTU Electrical Engineering Department, as shown in Fig. 1. The smart home is equipped with the household appliances that are

134 likely to be present in an average home in Turkey. The list of the appliances in the smart home and their rated power values are

135 given in Table 1. In order to monitor and control the appliances using a central PC, all the appliances are equipped with a

136 commercial smart plugs with ZigBee wireless communication.

137
138 Fig. 1 Smart Home Environment at YTU Electrical Engineering Department.

139 2.2 Renewable Energy and Power Conditioning System

140 The smart home generates its own energy using the renewable sources; a wind turbine with 2.3 kW power rating constructed

141 on 27 meter long pole at the vicinity of the smart home and a PV system with 3.3 kW output power rating located on the rooftop

142 of the building (Fig. 2a). The excess energy from renewable sources is stored in a battery storage system with 22 kWh energy

143 capacity (Fig. 2b). The required connections between these systems are provided by a smart inverter with an expert control option,

144 which allows the control of inverter by the HEM system (Fig. 3).
 6

145 In addition to the power and control equipment, a weather station is also installed to collect meteorological data such as solar

146 radiation, wind speed, ambient temperature and humidity. The general concept of the smart home is given in Fig. 4. Different

147 communication technologies are used in the smart home as purchased systems have their own communication protocols. The

148 communication between the control computer and the inverter is provided by RS232 communication, the communication with the

149 smart plugs is provided by ZigBee gateway, and Ethernet communication is used for weather station data transfer.

150 TABLE I
151 ELECTRICAL APPLIANCES IN THE SMART HOME
Electrical Measured rated Electrical Measured rated Electrical Measured rated
Appliances Power (kW) Appliances power (kW) Appliances Power (kW)
Refrigerator 0.160 Heater 1.790 TV 0.088
Iron 1.970 Hair straightener 0.140 AC 1.020
Toaster 0.700 Oven 2.150 PC desktop 0.055
Kettle 1.700 Cooker hood 0.190 PC monitor 0.017
Hair dryer 1.400 Dishwasher 1.770 LED Lighting 0.530
Vacuum cleaner 1.500 Microwave oven 1.150 Printer 0.009
Stove top 1.030 Wash machine 1.810

152
153 a) Wind turbine and PV panels b) Battery group
154 Fig. 2 Local renewable sources.

155
156 Fig. 3 Control and communication tools.

157 3. Home Energy Management System (HEMS)

158 The studies on HEMS in the literature are generally focused on the management of load side. In this study, a new HEMS is

159 proposed that considers both load side and source side to maximize renewable energy usage, to reduce electricity bill and to

160 maximize customer comfort. In the new HEMS, dynamic nature of renewable power generation, electricity tariff and load demand

161 are taken into account and a 5-minute time step is used to implement the HEM decisions.
 7

162 3.1. Residential Load Profiles

163
164 In this study, a typical residential load profile in Turkey is obtained using real household consumption measurements and a

165 survey on consumers’ habits [60]. Thus, representative load profiles are obtained for evaluating the proposed HEM algorithm.

166 The dynamic nature of the load demand can easily be seen in these load profiles. Then, the proposed HEM algorithm is tested

167 using weekday load profiles for winter and summer seasons, which are shown in Fig. 5.

168

169 Fig. 4 General concept of Smart Home Environment.

170
a)
7000 Lighting
TV
6000 Cordless Phone
Hair straightener
Printer
5000 Desktop PC
Monitor
Power (watt)

4000 Stove Top

Vacuum Cleaner
3000 Toaster
Kettle
Cooker Hood
2000 Microwave
Oven
1000 Dishwasher
Hair Dryer
0 Refrigerator
AC
00:00
00:30
01:00
01:30
02:00
02:30
03:00
03:30
04:00
04:30
05:00
05:30
06:00
06:30
07:00
07:30
08:00
08:30
09:00
09:30
10:00
10:30
11:00
11:30
12:00
12:30
13:00
13:30
14:00
14:30
15:00
15:30
16:00
16:30
17:00
17:30
18:00
18:30
19:00
19:30
20:00
20:30
21:00
21:30
22:00
22:30
23:00
23:30

Iron
Time (min.) Wash Machine

b)
171 Fig. 5 Load profiles for a typical household in Turkey:
172 a) Winter season b) Summer season
 8

173 3.2. Real Time Pricing

174
175 With the dissemination of smart meters in recent years, utility companies have started to benefit from different pricing policies

176 in order to better manage the balance between generation and consumption in power systems. Among others, RTP, ToU and CPP

177 methods are the most widely used pricing strategies both in literature and in real implementations of utility companies [61]. The

178 RTP method plays a key role particularly in DR applications [3,4]. Since a highly volatile information cannot be followed by the

179 consumers, HEM systems provide the most effective way to fully benefit from the RTP by continuously observing the tariff and

180 correspondingly controlling the power usage. Therefore, peak demand might be reduced in response to high electricity tariff in

181 these periods, providing more efficient use of existing power system capacity [62].

182 With the increased use of small-scale generation units at residential level, RTP implementations have started to become more

183 widespread since these generation units such as solar panels and wind turbines have very suitable structures for RTP due to their

184 dynamic generation characteristics. In order to exploit this dynamic nature, very short term scheduling applications ranging from

185 a few minutes to a few hours are of great importance. In this study, a scheduling application based on 5-minute real time pricing

186 is considered. For this purpose, a sample one-day RTP graph is generated based on the ToU tariff in Turkey and residential load

187 profiles (Figure 6). It should be noted that the proposed HEM algorithm also allows different RTP graphs to be adopted in the

188 system.
45

36
Price [cent/kWh]

27

18

0
00:00 04:10 08:20 12:30 16:40 20:50 23:59
Time [hour]
189
190
191 Fig. 6 Generated one-day RTP graph for a residential customer in Turkey.
192
193 3.3. Load Side Priority and Control
194
195 The power consumption of a residential customer is influenced by a large number of socio-economic and demographic factors

196 such as dwelling type, occupancy level (presence of the occupants at home), household income, education level and even

197 awareness of energy saving. It is generally not possible to control the resulting power consumption due to the mentioned factors

198 without affecting the consumer comfort. However, if the power consumed by a certain appliance can be controlled for some

199 periods based on an end-user determined priority, level of deterioration in consumers’ comfort can be limited. Furthermore, a

200 priority level for each appliance can be determined based on the preferences of the consumers.
 9

201 In this study, five appliances, air conditioner (AC), electric heater (EH), washing machine (WM), dish washer (DW) and oven

202 are defined as deferrable. The AC is considered to be used in summer while the EH is considered to be used in winter. These

203 deferrable loads are controlled depending on their priority and on the user-defined threshold power limit, both of which are

204 determined a priori by the consumer, in order to maximize customer comfort. In other words, when the power consumption exceeds

205 the user-defined threshold power limit, the use of appliances will be deferred depending on their priority. The defined priority

206 sequence is considered as oven, DW, WM and AC/EH, as shown in Fig. 7. This sequence is determined considering the technical

207 characteristics of these appliances, i.e., the suitability for frequent deferring operations, the effect of the delays of each appliance

208 on consumer comfort.

209
210 Fig. 7 Appliances priority mode.

211 With respect to the second parameter used for the deferring decisions, a power consumption threshold, which has been found

212 effective in the literature [19,28,58] due to its favorable effects on DR programs, is adopted in the HEM model. It should be noted

213 that several other approaches could be also used as a benchmark to evaluate the performance of the proposed HEM algorithm.

214 The threshold power limit is a parameter and can be defined by end users depending on the desired comfort level and also on

215 various socio-economic factors given above. It can be indicated that there is a trade-off between end user comfort and cost saving;

216 the higher the threshold limit and resulting end user comfort, the lower the cost saving is. However, the threshold power limit

217 value should be defined carefully in order to avoid frequent interventions on the energy consumption. For instance, if this limit is

218 set to zero, the HEM algorithm will then act based on real time pricing and renewable energy generation, and more frequent HEM

219 decision in terms of appliance control will be made. Therefore, the resulting cost saving can be improved; however, in this case,

220 consumer comfort violation will be probably increased significantly with the more frequent interventions and load recovery

221 problem is likely to be faced in the system operation. In this study, a power limit of 3 kW is used since our target is to control

222 power-intensive appliances instead of low power loads so that the customer comfort is not affected significantly.

223 It should be noted that the oven is considered as a deferrable load in this study contrary to a large number of similar studies in

224 the literature. Since the HEM uses 5-min time steps compared to commonly-used hourly time step in the literature, the dynamic
 10

225 control of oven does not affect its overall operation. AC and EH are controlled based on the internal temperature (𝑇𝑖 ). 𝑇𝑖 is set to

226 21o 𝐶 for summer and is set to 22o 𝐶 for winter.

227 3.4. Real Time Home Energy Management Algorithm

228
229 The proposed HEM algorithm has the capability of controlling both the load and source components together dynamically. In

230 the control process of the appliances, short-term power generation potential of renewable sources are taken into account in order

231 to calculate the expected energy excess or deficiency in the following time step. Furthermore, the RTP is continuously checked

232 by the proposed algorithm with two main objectives: (i) to make load shifting decisions and (ii) to maximize the profit by taking

233 power from grid when the price is low and by selling power to the grid when the price is high. The state of charge (SOC) of the

234 batteries is controlled continuously and they are charged at specified conditions so that the SOC is kept within an appropriate

235 interval. With respect to the grid side, load shifting operations decrease the magnitude of peak demand, which provides more

236 effective power system operation.

237 The proposed HEM algorithm uses a high amount of data collected by various sources in the decision-making process, as shown

238 in Fig. 8. The source of main inputs used in the algorithm can be gathered into four groups: (i) the data collected from the smart

239 home consisting of measured values such as temperature and power consumption of appliances, and of a user-defined threshold

240 power value, (ii) the electricity price values from the grid, (iii) the data collected from the weather station to be used in the

241 forecasting process within the HEM algorithm, and (iv) the data coming from renewable energy sources and battery system.

242

243 Fig. 8 Inputs and outputs of the proposed HEM algorithm.

244 It should be noted that the proposed HEM algorithm also makes use of the parameters belonging to the PV panels and wind

245 turbine as input to calculate the required power outputs. All the inputs used in the HEM algorithm, and in which order and how

246 they are used in the decision-making process are shown in detail in the general flow chart given in Fig. 9.

247 The main steps of the proposed HEM algorithm can be stated as follows: First, the system components in HEMS are initialized.

248 Then, the weather data, i.e., solar radiation, temperature and wind speed, are forecasted for the next 5-minute with a forecasting
 11

249 method to be explained in the following section. The power generation of local energy sources (𝑃𝑔𝑒𝑛 ) are calculated using these

250 forecasts in the wind turbine model and PV model, which are also given in detail in the following section. After one 5-minute

251 forecasting horizon, the real measurements of the weather characteristics collected during this period are included in the available

252 input data set and the new forecasts are carried out, which are in turn used in the wind turbine and PV models as inputs to calculate

253 the total expected power value from renewable sources (𝑃𝑔𝑒𝑛 ) for the next 5-minute period. Also, a consumer-defined power

254 threshold limit (𝑃𝑇ℎ ) is used in the proposed algorithm in order to check if the current residential power consumption is above or

255 below the defined power limit. When load demand is above the 𝑃𝑇ℎ value, the shiftable loads are controlled by following the

256 priority order provided that other conditions are also satisfied.

257
258 Fig. 9 General flowchart of the proposed HEM algorithm.

259 Comparing the estimated renewable generation power with the power consumed at home (𝑃𝑙𝑜𝑎𝑑 ), the net power (𝑃𝑛𝑒𝑡 ) is

260 calculated, as shown in Eq. 1.

 12

𝑃𝑛𝑒𝑡 = 𝑃𝑔𝑒𝑛 − 𝑃𝑙𝑜𝑎𝑑 (1)

261 The net power, together with the SOC value read through the inverter at each decision step and RTP value received from the

262 grid, are used to calculate the power that is given to/taken from the grid. The decision about power flow, which are decided by the

263 energy management system, is implemented through the power inverter at YTU-SHL. The inverter circuit can run both on-grid

264 and off-grid, and it can also charge the battery when necessary. In the case of transferring power to the grid, the current set value

265 (𝐼𝑔𝑓 ) is used to control inverter circuit. This current value is calculated as net power divided by grid voltage (𝑉𝑔 ), as given in Eq.

266 2. Only active power is considered for calculating 𝐼𝑔𝑓 value since the reactive power is negligible at residential electrical systems.

267 When SOC is high (over 95%), the maximum grid feeding current value (𝐼𝑔𝑓𝑚𝑎𝑥 ) that the experimental system can securely

268 provide, which is 30 A, is used instead of the 𝐼𝑔𝑓 value calculated in Eq. 2.

𝐼𝑔𝑓 = 𝑃𝑛𝑒𝑡 /𝑉𝑔 (2)

269 In the case of reasonable energy prices and/or low power generation from renewable energy sources, battery charging current

270 drawn from the grid (𝐼𝑐𝑐 ) is calculated as given in Eq. 3. The 𝐼𝑐𝑐 value is determined as the half of maximum current that the

271 battery can be charged and obtained by dividing battery power (𝑃𝑏𝑎𝑡𝑡𝑒𝑟𝑦 ) by battery voltage (𝑉𝐷𝐶 ).

𝐼𝑐𝑐 = (𝑃𝑏𝑎𝑡𝑡𝑒𝑟𝑦 × 0.05)/𝑉𝐷𝐶 (3)

272 If SOC decreases below a threshold that would reduce the lifetime of batteries (which is considered as 30% in this study), the

273 SOC level must be raised over this critical level immediately [63]. The current value in this case, called critical charging current

274 (𝐼𝑐𝑐𝑚𝑎𝑥 ), is calculated as given in Eq. 4. 𝐼𝑐𝑐𝑚𝑎𝑥 value must not exceed the 10% of battery current capacity, which is determined

275 considering the economic lifetime of battery [64,65].

𝐼𝑐𝑐𝑚𝑎𝑥 = (𝑃𝑏𝑎𝑡𝑡𝑒𝑟𝑦 × 0.10)/𝑉𝐷𝐶 (4)

276 Another advantage of the proposed HEM algorithm is the consideration of RTP value in HEM decisions. The suitability of a

277 price value (i.e., whether it is expensive or not) at each time step is determined by comparing dynamically changing RTP value

278 with dynamic threshold price (DTP). In this study, DTP values are defined using the three ToU tariff system in Turkey.

279 4. Renewable Power Forecasting Model

280 The power forecasts of the residential renewable energy sources are incorporated in the proposed HEM algorithm for the

281 purpose of exploiting maximum power from these sources. With this objective, first, wind speed, solar radiation and temperature

282 are forecasted using a combined approach. The wind speed forecasts are then converted to the wind power forecasts using a

283 commercial wind turbine power curve while the PV system model given in Ref. [63] is used to calculate solar power forecasts

284 using the forecasted solar radiation and temperature values.

 13

285 4.1 Forecasting of Meteorological Parameters

286 In order to determine the power outputs of wind turbine and PV panels, a meteorological parameter forecasting model is

287 developed. The forecast model consists of two modules; a Wavelet transform (WT) model for decomposition of data into more

288 regular components that are easier to model, and an artificial neural network (ANN) model for estimating each component before

289 an aggregation process. The selection of such a model combination is mainly due to the favorable effects of data decomposition

290 on capturing the daily cycles in weather conditions and the proven efficiency of these models in forecasting applications in the

291 literature [66]. The first model in the proposed combined approach, the WT model, decomposes the time series of each weather

292 parameter into a set of components with different frequency levels. Therefore, these components, especially those in a lower

293 frequency band, can be forecasted with a higher accuracy [67]. In this study, three-level wavelet decomposition is used, resulted

294 in one approximation, A3, and three detail components, D1, D2 and D3. Regarding the ANN model, a multilayer feed forward

295 back propagation (FFBP) network is considered with different number of layers and neurons for each decomposed level, which

296 are chosen with a preliminary analysis on effectiveness of each parameter. It should also be noted that a recursive approach, which

297 is based on updating the input set with recent measurements at the end of each prediction horizon, is adopted in the proposed

298 model in order to follow the latest trend in considered weather parameters.

299 4.2 Wind Power Prediction

300 The forecasted wind speed time series are first applied to wind profile power law to calculate the speed at the hub height of the

301 wind turbine used in the smart home system and then obtained wind speed forecasts at hub height are converted to wind power

302 forecasts through the power curve of the wind turbine from the manufacturer, as shown in Fig. 10.
3

2.5

2
Power [kW]

1.5

0.5

0
0 5 10 15 20 25
Wind Speed [m/s]
303

304 Fig. 10 Power curve of the wind turbine.

305 4.3 PV Power Prediction

306 A cell-level model, which enables both series and parallel configurations for the purpose of modeling a certain commercial PV

307 unit, is used in the study [63]. The model calculates the power available from the solar panels using temperature and solar radiation

308 forecasts. In this model, short circuit current, 𝐼𝑠𝑐 , and open circuit voltage, 𝑉𝑜𝑐 , values are calculated as in Eqs. 5 and 6:
 14

I sc*  dI 
I sc  *
G 1  (Tc  Tc* ) sc  (5)
G  dTc 

dVoc G
Voc  Voc*  (Tc  Tc* )  Vt .ln( * ) (6)
dTc G

∗
309 with 𝐼𝑠𝑐 and 𝑉𝑜𝑐∗ being the reference values provided by panel manufacturers, 𝐺 being measured solar radiation and 𝑇𝑐 being the

310 PV cell temperature. The other reference values, 𝐺 ∗ and 𝑇𝑐∗ , are taken as 1000 W/m2 and 25 oC, respectively. The value 𝑉𝑡 is

311 constructed by the combination of constants and 𝑇𝑐 is given as in Eq. 7

Tc  Ta  Ct .G (7)

312 where 𝑇𝑎 is ambient temperature and 𝐶𝑡 is the equation constant calculated as in Eq. 8. The nominal operating cell temperature

313 (NOCT) varies between 42o 𝐶 and 46o 𝐶 resulting in a change in 𝐶𝑡 value between 0.0272 and 0.0321oC/(W/m2) [63].

NOCT (o C )  20
Ct  (8)
800

314 In real PV systems, an MMPT circuit is generally used to maximize the amount of power obtained from available solar radiation.

315 The maximum power point, 𝑃𝑚 = 𝑉𝑚 × 𝐼𝑚 , is calculated using open circuit voltage and short circuit current, as given in Eqs. 9

316 and 10 [63].

 b 
Vm  Voc . 1  ln a  rs (1  a b )  (9)
 voc 

I m  I sc .(1  a  b ) (10)

317 The coefficients 𝑎 and 𝑏 are given by Eqs. 11 and 12, as follows:

a  voc  1  2voc rs (11)

a (12)
b
1 a

318 where 𝑣𝑜𝑐 and 𝑟𝑠 values can be found using 𝑣𝑜𝑐 = 𝑉𝑜𝑐 / 𝑉𝑡 and 𝑟𝑠 = 𝑅𝑠 /(𝑉𝑜𝑐 /𝐼𝑠𝑐 ) [63]. Further explanations about the structure

319 and performance of the cell-level model can be found in Ref. [63].

320 5. Case Studies

321 The proposed HEM algorithm implemented in the smart home system is investigated for two representative time periods: a

322 daily load profile for winter and a daily load profile for summer. In the testing process, the first task is to carry out the forecasts

323 of the residential renewable energy sources for the next five-minute-ahead time step using the past measurements taken from the

324 weather station. It can be indicated that the forecasting accuracy directly affects the efficiency of the proposed HEM algorithm. A

325 combined forecasting approach which uses the unique advantage of two different methods is therefore considered in this study
 15

326 and a comparison with benchmark models is provided to evaluate its performance. The accuracy of the proposed forecasting model

327 based on WT and ANN is testified against two of the most widely-used forecasting methods in the literature, namely, persistence

328 and ANN methods, and the results are provided in Table II. The persistence method simply assumes that the weather variables

329 will remain same during the considered prediction horizon. This assumption gives quite high-accuracy forecasts for very short

330 terms ranging from a few seconds to a few hours depending the characteristics of the weather variable [68]. The latter model,

331 which has dominated the forecasting applications in the last a few decades, is particularly appropriate for the modeling of volatile

332 weather patterns due to its capability of nonlinear input-output mapping.

333 TABLE II
334 COMPARISON OF VARIOUS STATISTICAL ERROR METRICS FOR A SET OF FORECASTING METHODS INCLUDING THE PROPOSED APPROACH
Wind Speed Solar irradiance Temperature
Forecasting
Season MAE RMSE NRMSE MAE RMSE NRMSE MAE RMSE NRMSE
Models
(m/s) (m/s) (%) (W/m2) (W/m2) (%) x100(°C) x100(°C) (%)
Persistence 1.14 1.48 14.45 8.37 13.15 6.03 2.14 4.63 1.01
Winter ANN 1.05 1.35 13.20 13.46 13.06 5.99 2.99 5.14 1.12
WT-ANN 0.87 1.14 11.18 7.72 12.39 5.68 2.51 4.28 0.93
Persistence 0.93 1.22 11.86 26.08 87.15 10.12 5.25 8.85 1.55
Summer ANN 0.99 1.16 11.27 36.07 85.73 9.96 5.68 9.07 1.59
WT-ANN 0.72 0.94 9.11 24.79 68.50 7.96 5.29 8.77 1.54

335 In order to evaluate the effectiveness of the proposed model as well as two benchmark methods, the statistical error metrics

336 Mean Absolute Error (MAE), Root Mean Squared Error (RMSE) and Normalized RMSE (NRMSE) are used in Table II. Among

337 a variety of candidate performance criteria, these three negatively-oriented metrics are considered in the study since they allow

338 comparisons between the accuracy of the models given in Table II and those of similar forecasting approaches presented in the

339 literature. Regarding the wind speed forecasts, it is seen from Table II that ANN model outperforms the persistence model for

340 both season in terms of all three error metrics and the proposed model gives the best results. The only exception here is the MAE

341 value of ANN model in summer season, which is higher than that of persistence model. A higher MAE value obtained despite

342 lower RMSE and NRMSE values indicates a relatively higher mean difference between the wind speed measurements and

343 forecasts during the prediction horizon, on the contrary, a lower peak error since RMSE is calculated based on the square of

344 differences. With respect to solar irradiance forecasts, the MAE values of ANN model are quite higher for both cases than those

345 of persistence model, which is the result of relatively stable characteristics of solar irradiance compared to wind speed. The

346 proposed model provides the highest accuracy again for the predictions of this variable. Lastly, regarding the temperature forecasts,

347 it is seen that persistence model achieves lower error metrics compared to ANN model due to the “quasi-stationary” profile of

348 temperature for very short periods. The best forecasts are accomplished with the proposed model again. It is noted that the first

349 two error metrics given in Table II for temperature forecasts are multiplied by one hundred in order to better point out the small

350 differences. The NRMSE values are shown as they are, since this is an already scale-independent (i.e., normalized) metric. In

351 conclusion, it can be indicated that persistence and ANN methods show a good performance depending on the volatility of the
 16

352 related variable; however, the proposed WT-ANN method significantly increases the forecasting accuracy for all circumstances

353 due to two main contributions of data decomposition: (i) facilitation of forecasting task by dividing time series into more regular

354 subseries, and (ii) enabling to forecast each subseries with different ANN model parameters depending on the data pattern.

355 Compared to the NRMSE values of the persistence and ANN models, the proposed model improves the forecasts by 11.69% and

356 11.38% respectively, which is averaged for three weather variables and both test periods, and can be considered as significant for

357 very short term forecasts.

358 When the high-accuracy forecasts of weather parameters are accomplished, the output powers of wind turbine and PV panels

359 are calculated for the next 5-minute period using the models given in Subsections 4.2 and 4.3. The proposed HEM algorithm then

360 manages the power flow using other available information including the current load demand and RTP signal, apart from the

361 renewable power forecasts. The RTP signal here is assumed to be received from the grid and transferred to the HEM system.

362 According to the direction and magnitude of the power flow as well as to the user-defined threshold power limit, the appliances

363 are switched ON or OFF starting from the high-priority or low-priority appliances, respectively. As mentioned previously, the

364 case studies are carried out separately for summer and winter seasons.

365 The energy consumption of the smart home, the energy generated by the renewable units, the power received from the grid, and

366 variation of the battery SOC are shown in Fig. 11 for the load profiles and weather conditions in summer season. As seen, the

367 power consumption (Fig. 11a) and generation (Fig. 11b) are generally not consistent during the selected day, which clearly points

368 out the necessity for a storage system and a control system. Using these components in dealing with the power demand-supply

369 interactions, the power taken from the grid is substantially decreased, resulted in a high profit for the prosumers. It should be noted

370 that there are both positive and negative power values in Fig. 10b, which indicate the power drawn from grid and supplied to the

371 grid, respectively. In terms of the energy transfer, the total energy received from the grid is only 41.5 kWh (Fig. 11c) during the

372 whole day, while the energy given to the grid is 13.8 kWh for a dynamic pricing scheme. Compared to the case in which no control

373 method is implemented, the proposed HEM method provides a reduction of $4.5 (excluding the taxes and other prices) in the

374 customer’s bill for the selected day. This benefit can be further increased if the monetary value of residential load consumption is

375 included and the incentives for renewable energy sources are considered. It is also important to highlight that the battery charge

376 level is continuously observed in energy transfers to prevent any overcharge condition.

377 From the utility side, the most important advantage of the proposed HEM system can be indicated as the reduction in the peak

378 power around the 7:00 AM (between the minutes 400 and 450). As seen in Figs. 11a and 11b, the peak power drawn from the grid

379 is decreased by over 1.0 kW. Furthermore, it can ben indicated that local power generation substantially contributes to the

380 decrement of transmission and distribution losses in the power systems.

 17

381 The data for the power generation and consumption are also given for the winter load profile and weather conditions, as shown

382 in Fig. 12. Regarding the energy values, an amount of 27.6 kWh is generated by the renewable sources, and 15.13 kWh is supplied

383 by the grid. As a result, a saving of $1.36 is achieved in the customer’s bill in this winter day by the HEM algorithm. Besides, a

384 higher reduction in peak demand compared to the summer case is obtained, which is over 3 kW.

385 It is clear in Fig. 12 that the power generation from PV panels are decreased aligned with the decrease in solar radiation in

386 winter season. In this case, therefore, a higher amount of energy compared to the case in summer is used from the batteries, which

387 leads to more reduction in battery SOC. In both cases, the battery SOC remains between 89% and 95%. One of the main reasons

388 for keeping this value high as much as possible is to enable the house to run for a reasonable time in the case of a possible grid

389 fault.

7500
Normal load profile Load profile with HEM
6000
Power [W]

4500
3000
1500
0
0 36 72 108 144 180 216 252 288
a)
6000
Prenew Pgrid
3000
Power [W]

-3000

-6000

-9000
0 36 72 108 144 180 216 252 288
b)
SOC
0.95
SOC

0.9
0.85
0.8
0 36 72 108 144 180 216 252 288
c)
Time [5-min intervals]
390
391 Fig. 11 Simulation results of the source side powers for summer season: a) Power consumption in the smart home with and without HEM b)
392 Power generation of the renewable sources and power drawn from grid with the proposed HEM algorithm c) Battery SOC.

7500 Normal load profile Load profile with HEM

Power [W]

6000
4500
3000
1500
0
0 36 72 108 144 180 216 252 288
a)
6000
Prenew Pgrid
3000
Power [W]

0
-3000
-6000
-9000
0 36 72 108 144 180 216 252 288
b)
0.99
SOC
0.95
SOC

0.9

0.85

0.8
0 36 72 108 144 180 216 252 288
c)
393 Time [5-min intervals]

394 Fig. 12 Test results of the source side powers for winter season. a) Power consumption in the smart home with and without HEM b) Power
395 generation of the renewable sources and power drawn from grid with the proposed HEM algorithm c) Battery SOC.
 18

396 Furthermore, in order to show the effectiveness of RTP on energy management, the cases with RTP and without RTP are

397 examined for a weekday load profile in winter. The power exchanges with the grid and variations of SOC for two cases is shown

398 in Fig. 13. In order to observe the price dependent variations in power flow in detail, a period of 30 minute is also shown in Fig.

399 14. As seen, energy is purchased from the grid in some periods despite the high prices and sold to the grid in periods with low

400 prices when RTP is not taken into account. However, the algorithm including RTP information generally provide a power

401 exchange with higher economic benefits.

9000
Normal load profile Pgrid with RTP Pgrid without RTP
6000
Power [W]

3000
0
-3000
-6000
-9000
0 36 72 108 144 180 216 252 288
a)
Battery SOC with RTP Battery SOC without RTP

0.95
SOC [%]

0.9

0.85

0.8
0 36 72 108 144 180 216 252 288
b)
402 Time [5-min intervals]

403 Fig. 13 Effect of RTP on energy management a) Power exchange with grid b) Battery SOC.

8
Pgrid-without RTP Pgrid-with RTP
6
in these periods, although the price expensive in y-axis positive values indicate that the power is drawn from the grid
Power [kW]

power drawn from the grid instead of feeding while negative values indicate that the power to the grid
3

-3

-6

-9
100 105 110 115 120 125 130
a)
25
RTP DTP
Price[cent/kWh]

22

19

16

13

10
100 105 110 115 120 125 130
b)
404 Time [5-min intervals]

405 Fig. 14 A zoomed representation of power exchange with and without RTP a) Power drawn from the grid with/without the proposed HEM
406 algorithm b) RTP and DTP.

407 In the study, the contribution of including renewable power forecasts in the HEM system is also investigated. The comparisons

408 of the power exchange with grid and the variation of SOC level for two cases, i.e., with and without forecasts, are illustrated in

409 Fig. 15. As seen, the renewable power forecasts affect the power flow positively, particularly for morning and evening periods,

410 enabling the better control of power exchange with the grid. Using forecasts in the HEM algorithm decreases the amount of power

411 drawn from the grid during the peak hours.

412
 19

413 5.1 Experimental Verification of the Proposed HEM Algorithm

414 The effectiveness of the proposed HEM algorithm has been investigated in various simulation studies as given above and its

415 performance has been evaluated in terms of power and cost savings. As this paper focuses on the dynamic energy management

416 for a structure with a high number of components, such as a smart home with different appliances, local renewable energy sources

417 and a storage system, it is not possible to compare its saving potential to those presented in the literature studies based mainly on

418 hourly time scale. Instead, the applicability and effectiveness of the proposed HEM algorithm is experimentally tested at YTU-

419 SHL.

8000
Normal load profile Pgrid-with Prenew forecasting Pgrid-without Prenew forecasting
6000

3000
Power [W]

-3000

-6000

-9000
0 36 72 108 144 180 216 252 288
a)
0.99
Battery SOC with RTP Battery SOC without RTP

0.95
SOC [%]

0.9

0.85
0 36 72 108 144 180 216 252 288
b)
420 Time [5-min intervals]

421 Fig. 15 Effect of power forecasts on energy management a) Power exchange with the grid b) Battery SOC

422 Since our aim is to prove that the proposed HEM algorithm can be implemented in real time, a 75-minute sample load profile

423 consisting of major home appliances shown in Figure 16 is used in the experimental study. In the sample load profile, only the

424 heater (Level 1-0.820 kW) is considered as shiftable load for experimental validation of appliance shifting. 𝑃𝑇ℎ value is taken as

425 3 kW. A MATLAB program is developed to combine different communication protocols into single program as different

426 communication protocols are used by smart plugs, weather station and power inverter at YTU-SHL. For example, ZigBee is used

427 in smart plugs while RS232 is used in power inverter.

428 After all collected data from smart plugs, weather station and the power inverter are processed; a HEM decision is made for the

429 next 5-minute period. The data collection and process require time to generate the HEM decision for the next 5 minute. It takes

430 around 1 minute to generate the HEM decision. Once a decision is made the HEM algorithm does not intervene the system for the

431 next 5-minute period. Thus, there is around (5+1)-minute delay in the proposed HEM algorithm in following real life.

432 The experimental study has been completed between 11:00am-12:15pm. Obtained results with the implementation of the HEM

433 algorithm, according to the flowchart in Fig. 9, are shown in Fig. 17. In this figure, the load demand with and without HEM

434 algorithm, the heater control, battery SOC, power supplied through the inverter (𝑃𝑓𝑒𝑒𝑑𝑖𝑛𝑔 ), total renewable power, PV and wind
 20

435 powers, the power difference between total renewable power and load demand (𝑃𝑛𝑒𝑡 ), DTP and RTP values are shown. RTP values

436 are taken from Fig. 6 between 11:00am-12:15pm.

7
Oven, WM, DW and heater

6
Kettle
Wash machine Oven TV and PC
Vacuum cleaner
5 Hair dryer
Power [kW]

Dish Washer
2
Hair dryer

Stove top and cooker hood

1

0
5 10 15 20 25 30 35 40 45 50 55 60 65 70 75
437 Time [min]
438 Fig. 16 Sample load profile used for experimental validation of the proposed HEM algorithm.
8
Normal load profile Load profile with HEM Heater
Power [kW]

4
2
2 4 6 8 10 12
0
11:00 11:15 11:30 11:45 12:00 12:15
1
Wind power
Power [kW]

0.75
0.5
0.25
0
11:00 11:15 11:30 11:45 12:00 12:15
4
Power [kW]

PV power
3
2
1
0
11:00 11:15 11:30 11:45 12:00 12:15
4
Total renewable power
Power [kW]

3
2
1
0
11:00 11:15 11:30 11:45 12:00 12:15
439 Time [hour]

6
Power [kW]

P feeding
3
0
-2
-5
-8
11:00 11:15 11:30 11:45 12:00 12:15
2
Power [kW]

Pnet
0

-2

-4
11:00 11:15 11:30 11:45 12:00 12:15
98
SOC [%]

SOC
96
94
92
90
88
11:00 11:15 11:30 11:45 12:00 12:15
20
Price [cent/kWh]

RTP DTP
18
16
14
12
11:00 11:15 11:30 11:45 12:00 12:15
440 Time [hour]

441 Fig. 17 Experimental results of the proposed HEM system.

 21

442 The initial approximately 6 minutes of the 75-minute total time is used to generate the first HEM decision for the next 5-minute

443 period. The HEM decisions are taken based on SOC, 𝑃𝑛𝑒𝑡 , 𝑃𝑙𝑜𝑎𝑑 , 𝑃𝑇ℎ , RTP and DTP values. There are 13 HEM decisions within

444 the 75-minute experimental study and explanations for these decisions are given below:

445 1st decision: 𝑆𝑂𝐶 > 0.95, so no other parameter is checked. The heater continues its operation and 𝑃𝑓𝑒𝑒𝑑𝑖𝑛𝑔 value is

446 maximized using the 𝐼𝑔𝑓𝑚𝑎𝑥 value. It should be noted that the power to the grid is found by subtracting 𝑃𝑙𝑜𝑎𝑑 from the

447 𝑃𝑓𝑒𝑒𝑑𝑖𝑛𝑔 .

448 2nd decision: 0.3 < 𝑆𝑂𝐶 < 0.95, 𝑃𝑙𝑜𝑎𝑑 > 3kW, 𝑃𝑛𝑒𝑡 < 0 and RTP < DTP. Since 𝑃𝑙𝑜𝑎𝑑 > 3kW, the heater is turned off.

449 3rd decision: 0.3 < 𝑆𝑂𝐶 < 0.95, 𝑃𝑙𝑜𝑎𝑑 > 3kW, 𝑃𝑛𝑒𝑡 < 0 and RTP < DTP. Since 𝑃𝑙𝑜𝑎𝑑 > 3kW, the heater is kept off.

450 Calculated 𝐼𝑔𝑓 current is supplied by the inverter.

451 4th decision: 0.3 < 𝑆𝑂𝐶 < 0.95, 𝑃𝑙𝑜𝑎𝑑 <3kW, 𝑃𝑛𝑒𝑡 < 0 and RTP < DTP. Since 𝑃𝑙𝑜𝑎𝑑 < 3kW, the heater turned on.

452 Calculated 𝐼𝑔𝑓 current is supplied by the inverter.

453 5th decision: 𝑆𝑂𝐶 > 0.95, so no other parameter is checked and 𝑃𝑓𝑒𝑒𝑑𝑖𝑛𝑔 value is maximized using the 𝐼𝑔𝑓𝑚𝑎𝑥 value. The

454 heater is kept on.

455 6th decision: 0.3 < 𝑆𝑂𝐶 < 0.95, 𝑃𝑙𝑜𝑎𝑑 > 3kW, 𝑃𝑛𝑒𝑡 < 0 and RTP < DTP. Since 𝑃𝑙𝑜𝑎𝑑 > 3kW, the heater is turned off.

456 Calculated 𝐼𝑔𝑓 current is supplied by the inverter.

457 7th decision: 0.3 < 𝑆𝑂𝐶 < 0.95, 𝑃𝑙𝑜𝑎𝑑 < 3kW, 𝑃𝑛𝑒𝑡 < 0 and RTP < DTP. Since 𝑃𝑙𝑜𝑎𝑑 < 3kW, the heater is turned on.

458 Calculated 𝐼𝑔𝑓 current is supplied by the inverter.

459 8th decision: 𝑆𝑂𝐶 > 0.95, so no other parameter is checked and 𝑃𝑓𝑒𝑒𝑑𝑖𝑛𝑔 value is maximized using the 𝐼𝑔𝑓𝑚𝑎𝑥 value. The

460 heater is kept on. Since there is a sudden decrease in renewable power generation, the use of battery is increased and the

461 SOC value gets lower during this time frame.

462 9th decision: 0.3 < 𝑆𝑂𝐶 < 0.95, 𝑃𝑙𝑜𝑎𝑑 > 3kW, 𝑃𝑛𝑒𝑡 < 0 and RTP > DTP. Since 𝑃𝑙𝑜𝑎𝑑 > 3kW the heater is turned off.

463 Since RTP > DTP, 𝑃𝑓𝑒𝑒𝑑𝑖𝑛𝑔 value is maximized using the 𝐼𝑔𝑓𝑚𝑎𝑥 value. The renewable power generation is still low and

464 the SOC gets lower during this time frame.

465 10th decision: 0.3 < 𝑆𝑂𝐶 < 0.95, 𝑃𝑙𝑜𝑎𝑑 > 3kW, 𝑃𝑛𝑒𝑡 < 0 and RTP < DTP. Since 𝑃𝑙𝑜𝑎𝑑 > 3kW the heater is kept off.

466 Calculated 𝐼𝑔𝑓 current is supplied by the inverter.

467 11th decision: 0.3 < 𝑆𝑂𝐶 < 0.95, 𝑃𝑙𝑜𝑎𝑑 < 3kW, 𝑃𝑛𝑒𝑡 > 0 and RTP < DTP. Since 𝑃𝑛𝑒𝑡 > 0 and RTP < DTP the heater is

468 turned on. Calculated 𝐼𝑔𝑓 current is supplied by the inverter.

469 12th decision: 0.3 < 𝑆𝑂𝐶 < 0.95, 𝑃𝑙𝑜𝑎𝑑 > 3kW, 𝑃𝑛𝑒𝑡 < 0 and RTP > DTP. Since 𝑃𝑙𝑜𝑎𝑑 > 3kW the heater is turned off.

470 Since RTP > DTP, 𝑃𝑓𝑒𝑒𝑑𝑖𝑛𝑔 value is maximized using the 𝐼𝑔𝑓𝑚𝑎𝑥 value.
 22

471 13th decision: 0.3 < 𝑆𝑂𝐶 < 0.95, 𝑃𝑙𝑜𝑎𝑑 < 3kW, 𝑃𝑛𝑒𝑡 < 0 and RTP < DTP. Since 𝑃𝑙𝑜𝑎𝑑 < 3kW, the heater turned on.

472 Calculated 𝐼𝑔𝑓 current is supplied by the inverter.

473 In summary, the proposed HEM algorithm first collects all the available information from the system and the price information

474 from the grid. Forecasting algorithm is then run to predict the renewable power generation, and hence the net power, which is the

475 difference between the renewable power generation and the load demand, is calculated. By evaluating the net power and also the

476 other influencing factors mentioned above, i.e., threshold power limit, values related to energy price (RTP and DTP) and SOC, at

477 every 5-minute time step, energy management decisions are made. If the battery SOC is within its allowed operating range (i.e.,

478 between 0.30 and 0.95), the decisions are mainly based on the RTP and household load demand. Otherwise, if the SOC value is

479 higher than its maximum allowed value, power feeding mode becomes active with the predefined maximum current value for the

480 next 5-minute period or if it is lower than its minimum allowed value, the battery charging is activated with the critical charging

481 current for the next 5-minute period to protect the battery. In order to increase the customer saving for the SOC values within the

482 bounds, the power from the battery is used as much as possible to supply the load demand depending on the price values and net

483 power. It should be also noted that power feeding mode with current set value is activated when RTP is lower than DTP as well

484 as net load is different from zero, and that power feeding with predefined maximum current value is activated when RTP is higher

485 than DTP as well as net load is zero or positive. Besides, battery charging is activated to charge the battery with minimal cost if

486 RTP is lower than DTP. Lastly, device shifting is activated when all conditions are satisfied and load demand value is higher than

487 the threshold power value.

488 Since the SOC value is generally between 0.30 and 0.95 during the experiment, HEM decisions are mainly based on the RTP

489 value and 𝑃𝑙𝑜𝑎𝑑 value at the decision point. The main objective of the proposed HEM algorithm is to minimize grid use and to

490 maximize the exported power to the grid. The results of the experimental study verify the applicability of the proposed HEM

491 algorithm in real life.

492 6. Conclusions

493 In this study, a new dynamic HEM algorithm is proposed for decreasing the total cost of the energy from the grid while

494 maximizing the user comfort in residential houses. The effectiveness of the proposed algorithm is first simulated for two sample

495 daily load profiles of winter and summer and then examined in a real smart home consisting of various appliances and local

496 renewable energy sources. Incorporating real-time demand and price information as well as short-term forecasts of residential

497 renewable power generations from wind turbine and PV panels in the HEM model, it is aimed to increase the capability of the

498 model to schedule the five deferrable loads more effectively and to maximize the benefits from the local renewable generation. A

499 dynamic (5-minute) HEM algorithm is developed and the applicability of the proposed HEM algorithm is experimentally verified

500 using a 75-minute sample load profile consisting of major home appliances at YTU-SHL. Results of both simulation study and
 23

501 experimental study show that the proposed HEM algorithm has the ability of controlling the sources and loads dynamically under

502 time-varying renewable generation and electricity price. It should be noted that the proposed HEM algorithm is a dedicated tool

503 for short-term energy management; however, it might be effectively used for longer-term managements by modifying the

504 algorithm and forecasting method. This is also considered as a near-term target of the authors.

505 Acknowledgement
506 This work was supported in part by Yildiz Technical University Research Projects Fund under Grant 2014-04-02-DOP01 and

507 Istanbul Development Agency Fund under Grant KCE-27.

508 References
509
510 [1] Beaudin M, Zareipour H. Home energy management systems: A review of modelling and complexity. Renew Sustain Energy Rev 2015;45:318–35.
511 doi:10.1016/j.rser.2015.01.046.
512 [2] Vega a. M, Santamaria F, Rivas E. Modeling for home electric energy management: A review. Renew Sustain Energy Rev 2015;52:948–59.
513 doi:10.1016/j.rser.2015.07.023.
514 [3] Khan AR, Mahmood A, Safdar A, Khan ZA, Khan NA. Load forecasting, dynamic pricing and DSM in smart grid: A review. Renew Sustain Energy
515 Rev 2016;54:1311–22. doi:10.1016/j.rser.2015.10.117.
516 [4] Chakraborty S, Ito T, Senjyu T. Smart pricing scheme: A multi-layered scoring rule application. Expert Syst Appl 2014;41:3726–35.
517 doi:10.1016/j.eswa.2013.12.002.
518 [5] Amir-Hamed Mohsenian-Rad, and Leon-garcia A. Price Prediction in Real-Time Electricity. IEEE Journals Mag 2010;1:120–33.
519 doi:10.1109/TSG.2010.2055903.
520 [6] Vardakas JS, Zorba N, Verikoukis C V. A Survey on demand response programs in smart grids: pricing methods and optimization algorithms. IEEE
521 Commun Surv Tutorials 2014;17:152–78. doi:10.1109/COMST.2014.2341586.
522 [7] Siano P, Sarno D. Assessing the benefits of residential demand response in a real time distribution energy market. Appl Energy 2016;161:533–51.
523 doi:10.1016/j.apenergy.2015.10.017.
524 [8] Sarker MR, Ortega-Vazquez MA, Kirschen DS. Optimal Coordination and Scheduling of Demand Response via Monetary Incentives. IEEE Trans Smart
525 Grid 2015;6:1341–52. doi:10.1109/TSG.2014.2375067.
526 [9] Koutitas G, Tassiulas L. Periodic flexible demand: Optimization and phase management in the smart grid. IEEE Trans Smart Grid 2013;4:1305–13.
527 doi:10.1109/TSG.2013.2246874.
528 [10] Kilkki O, Seilonen I. Optimized Control of Price-Based Demand Response With Electric Storage Space Heating 2015;11:281–8.
529 doi:10.1109/TII.2014.2342032.
530 [11] Jiang B, Fei Y. Smart home in smart microgrid: A cost-effective energy ecosystem with intelligent hierarchical agents. IEEE Trans Smart Grid 2015;6:3–
531 13. doi:10.1109/TSG.2014.2347043.
532 [12] Harsha P, Sharma M, Natarajan R, Ghosh S. A framework for the analysis of probabilistic demand response schemes. IEEE Trans Smart Grid
533 2013;4:2274–84. doi:10.1109/TSG.2013.2278238.
534 [13] Hakimi SM, Moghaddas-Tafreshi SM. Optimal planning of a smart microgrid including demand response and intermittent renewable energy resources.
535 IEEE Trans Smart Grid 2014;5:2889–900. doi:10.1109/TSG.2014.2320962.
536 [14] Douglass PJ, Garcia-Valle R, Nyeng P, Ostergaard J, Togeby M. Smart demand for frequency regulation: Experimental results. IEEE Trans Smart Grid
537 2013;4:1713–20. doi:10.1109/TSG.2013.2259510.
538 [15] Schibuola L, Scarpa M, Tambani C. Demand response management by means of heat pumps controlled via real time pricing. Energy Build 2015;90:15–
539 28. doi:10.1016/j.enbuild.2014.12.047.
540 [16] Ma K, Hu G, Spanos CJ. Punishment Mechanism and Application to Industrial Refrigerated Warehouses 2015;11:1520–31.
541 [17] Qela B, Mouftah HT. Observe , Learn , and Adapt ( OLA )— An Algorithm for Energy Management in Smart Homes Using Wireless Sensors and Arti
542 fi cial Intelligence 2012;3:2262–72.
543 [18] Yoon JH, Baldick R, Novoselac A. Dynamic demand response controller based on real-time retail price for residential buildings. IEEE Trans Smart Grid
544 2014;5:121–9. doi:10.1109/TSG.2013.2264970.
545 [19] Corradi O, Ochsenfeld H, Madsen H, Pinson P. Controlling electricity consumption by forecasting its response to varying prices. IEEE Trans Power
546 Syst 2013;28:421–9. doi:10.1109/TPWRS.2012.2197027.
547 [20] Al-Mulla A, Elsherbini A. Demand management through centralized control system using power line communication for existing buildings. Energy
548 Convers Manag 2014;79:477–86. doi:10.1016/j.enconman.2013.12.011.
549 [21] Zhang Y, Lu N. Parameter selection for a centralized thermostatically controlled appliances load controller used for intra-hour load balancing. IEEE
550 Trans Smart Grid 2013;4:2100–8. doi:10.1109/TSG.2013.2258950.
551 [22] Mehta N, Sinitsyn NA, Backhaus S, Lesieutre BC. Safe control of thermostatically controlled loads with installed timers for demand side management.
552 Energy Convers Manag 2014;86:784–91. doi:10.1016/j.enconman.2014.06.049.
553 [23] Lu N. An evaluation of the HVAC load potential for providing load balancing service. IEEE Trans Smart Grid 2012;3:1263–70.
554 doi:10.1109/TSG.2012.2183649.
555 [24] Zehir MA, Bagriyanik M. Demand Side Management by controlling refrigerators and its effects on consumers. Energy Convers Manag 2012;64:238–
556 44. doi:10.1016/j.enconman.2012.05.012.
557 [25] Finn P, O’Connell M, Fitzpatrick C. Demand side management of a domestic dishwasher: Wind energy gains, financial savings and peak-time load
558 reduction. Appl Energy 2013;101:678–85. doi:10.1016/j.apenergy.2012.07.004.
559 [26] Pourmousavi SA, Patrick SN, Nehrir MH. Real-time demand response through aggregate electric water heaters for load shifting and balancing wind
560 generation. IEEE Trans Smart Grid 2014;5:769–78. doi:10.1109/TSG.2013.2290084.
561 [27] Labeeuw W, Stragier J, Deconinck G. Potential of active demand reduction with residential wet appliances: A case study for Belgium. IEEE Trans Smart
562 Grid 2015;6:315–23. doi:10.1109/TSG.2014.2357343.
563 [28] Pipattanasomporn M, Kuzlu M, Rahman S. An Algorithm for Intelligent Home Energy Management and Demand Response Analysis. IEEE Trans Smart
564 Grid 2012;3:2166–73. doi:10.1109/TSG.2012.2201182.
 24

565 [29] Chen Z, Wu L, Fu Y. Real-time price-based demand response management for residential appliances via stochastic optimization and robust optimization.
566 IEEE Trans Smart Grid 2012;3:1822–31. doi:10.1109/TSG.2012.2212729.
567 [30] Fernandes F, Morais H, Vale Z, Ramos C. Dynamic load management in a smart home to participate in demand response events. Energy Build
568 2014;82:592–606. doi:10.1016/j.enbuild.2014.07.067.
569 [31] Rastegar M, Fotuhi-firuzabad M. Response Programs 2015;6:1453–62. doi:10.1109/TSG.2014.2338794.
570 [32] Song M, Alvehag K, Widén J, Parisio A. Estimating the impacts of demand response by simulating household behaviours under price and CO2 signals.
571 Electr Power Syst Res 2014;111:103–14. doi:10.1016/j.epsr.2014.02.016.
572 [33] Costanzo GT, Member S, Zhu G, Anjos MF, Savard G. A System Architecture for Autonomous Demand Side Load Management in Smart Buildings
573 2012;3:2157–65.
574 [34] Vivekananthan C, Mishra Y, Ledwich G, Li F. Demand response for residential appliances via customer reward scheme. IEEE Trans Smart Grid
575 2014;5:809–20. doi:10.1109/TSG.2014.2298514.
576 [35] Li SH, Zhang D, Roget AB, O’Neill Z. Integrating Home Energy Simulation and Dynamic Electricity Price for Demand Response Study. IEEE Trans
577 Smart Grid 2014;5:779–88. doi:10.1109/TSG.2013.2279110.
578 [36] Li XH, Hong SH. User-expected price-based demand response algorithm for a home-to-grid system. Energy 2014;64:437–49.
579 doi:10.1016/j.energy.2013.11.049.
580 [37] Anvari-moghaddam A, Monsef H, Rahimi-kian A. A. A. Moghaddam, H. Monsef, and A. R. Kian, “Optimal smart home energy management considering
581 energy saving and a comfortable lifestyle,” IEEE Trans. Smart Grid, vol. 6, no. 1, pp. 324–332, Jan. 2015. 2015;6:324–32.
582 [38] Wang X, Palazoglu A, El-Farra NH. Operational optimization and demand response of hybrid renewable energy systems. Appl Energy 2015;143:324–
583 35. doi:10.1016/j.apenergy.2015.01.004.
584 [39] Yu Z, Jia L, Murphy-Hoye MC, Pratt A, Tong L. Modeling and stochastic control for home energy management. IEEE Trans Smart Grid 2013;4:2244–
585 55. doi:10.1109/TSG.2013.2279171.
586 [40] Shahgoshtasbi D, Jamshidi MM. A new intelligent neuro-fuzzy paradigm for energy-efficient homes. IEEE Syst J 2014;8:664–73.
587 doi:10.1109/JSYST.2013.2291943.
588 [41] Adika CO, Wang L. Energy Management 2014;5:673–82.
589 [42] Brahman F, Honarmand M, Jadid S. Optimal electrical and thermal energy management of a residential energy hub, integrating demand response and
590 energy storage system. Energy Build 2015;90:65–75. doi:10.1016/j.enbuild.2014.12.039.
591 [43] Lee S, Kwon B, Lee S. Joint energy management system of electric supply and demand in houses and buildings. IEEE Trans Power Syst 2014;29:2804–
592 12. doi:10.1109/TPWRS.2014.2311827.
593 [44] Zong Y, Kullmann D, Thavlov A, Gehrke O, Bindner HW. Application of model predictive control for active load management in a distributed power
594 system with high wind penetration. IEEE Trans Smart Grid 2012;3:1055–62. doi:10.1109/TSG.2011.2177282.
595 [45] Arif A, Javed F, Arshad N. Integrating renewables economic dispatch with demand side management in micro-grids: A genetic algorithm-based
596 approach. Energy Effic 2014;7:271–84. doi:10.1007/s12053-013-9223-9.
597 [46] Anvari-Moghaddam A, Monsef H, Rahimi-Kian A. Cost-effective and comfort-aware residential energy management under different pricing schemes
598 and weather conditions. Energy Build 2015;86:782–93. doi:10.1016/j.enbuild.2014.10.017.
599 [47] Castillo-Cagigal M, Gutierrez A, Monasterio-Huelin F, Caamano-Martin E, Masa D, Jimenez-Leube J. A semi-distributed electric demand-side
600 management system with PV generation for self-consumption enhancement. Energy Convers Manag 2011;52:2659–66.
601 doi:10.1016/j.enconman.2011.01.017.
602 [48] Lujano-Rojas JM, Monteiro C, Dufo-López R, Bernal-Agustín JL. Optimum residential load management strategy for real time pricing (RTP) demand
603 response programs. Energy Policy 2012;45:671–9. doi:10.1016/j.enpol.2012.03.019.
604 [49] Newsham GR, Galasiu AD, Armstrong MM, Beausoleil-Morrison I, Szadkowski F, Sager JM, et al. The zero-peak house: Full-scale experiments and
605 demonstration. Energy Build 2013;64:483–92. doi:10.1016/j.enbuild.2013.05.004.
606 [50] Karfopoulos E, Tena L, Torres A, Salas P, Jorda JG, Dimeas A, et al. A multi-agent system providing demand response services from residential
607 consumers. Electr Power Syst Res 2015;120:163–76. doi:10.1016/j.epsr.2014.06.001.
608 [51] Prüggler N. Economic potential of demand response at household level-Are Central-European market conditions sufficient? Energy Policy 2013;60:487–
609 98. doi:10.1016/j.enpol.2013.04.044.
610 [52] Oconnell N, Pinson P, Madsen H, Omalley M. Benefits and challenges of electrical demand response: A critical review. Renew Sustain Energy Rev
611 2014;39:686–99. doi:10.1016/j.rser.2014.07.098.
612 [53] Newsham GR, Bowker BG. The effect of utility time-varying pricing and load control strategies on residential summer peak electricity use: A review.
613 Energy Policy 2010;38:3289–96. doi:10.1016/j.enpol.2010.01.027.
614 [54] Martínez Ceseña EA, Good N, Mancarella P. Electrical network capacity support from demand side response: Techno-economic assessment of potential
615 business cases for small commercial and residential end-users. Energy Policy 2015;82:222–32. doi:10.1016/j.enpol.2015.03.012.
616 [55] Gyamfi S, Krumdieck S, Urmee T. Residential peak electricity demand response - Highlights of some behavioural issues. Renew Sustain Energy Rev
617 2013;25:71–7. doi:10.1016/j.rser.2013.04.006.
618 [56] Gyamfi S, Krumdieck S. Price, environment and security: Exploring multi-modal motivation in voluntary residential peak demand response. Energy
619 Policy 2011;39:2993–3004. doi:10.1016/j.enpol.2011.03.012.
620 [57] Siano P. Demand response and smart grids - A survey. Renew Sustain Energy Rev 2014;30:461–78. doi:10.1016/j.rser.2013.10.022.
621 [58] Hassan NU, Khalid YI, Yuen C, Huang S, Pasha MA, Wood KL, et al. Framework for minimum user participation rate determination to achieve specific
622 demand response management objectives in residential smart grids. Int J Electr Power Energy Syst 2016;74:91–103. doi:10.1016/j.ijepes.2015.07.005.
623 [59] Tascikaraoglu A, Uzunoglu M, Tanrioven M, Boynuegri AR, Elma O. Smart grid-ready concept of a smart home prototype: A demonstration project in
624 YTU. 4th Int. Conf. Power Eng. Energy Electr. Drives, Istanbul: 2013.
625 [60] Elma O, Selamogullari US. A new home energy management algorithm with voltage control in a smart home environment. Energy 2015;91:720–31.
626 doi:10.1016/j.energy.2015.08.094.
627 [61] Comed Hourly Pricing 2016. https://hourlypricing.comed.com/live-prices/.
628 [62] Mohsenian-Rad A-H, Leon-Garcia A. Optimal Residential Load Control With Price Prediction in Real-Time Electricity Pricing Environments. IEEE
629 Trans Smart Grid 2010;1:120–33. doi:10.1109/TSG.2010.2055903.
630 [63] Elma O, Selamogullari US. A comparative sizing analysis of a renewable energy supplied stand-alone house considering both demand side and source
631 side dynamics. Appl Energy 2012;96:400–8.
632 [64] Victron Energy, Gel and AGM Batteries n.d. https://www.victronenergy.com/batteries/gel-and-agm-batteries (accessed April 15, 2016).
633 [65] Batteries n.d. http://enerjienstitusu.com/2013/04/22/akuler-hakkinda-bilmeniz-gerekenler/ (accessed April 15, 2016).
634 [66] Tascikaraoglu A, Uzunoglu M. A review of combined approaches for prediction of short-term wind speed and power. Renew Sustain Energy Rev
635 2014;34:243–54. doi:10.1016/j.rser.2014.03.033.
636 [67] Tascikaraoglu A, Sanandaji BM, Poolla K, Varaiya P. Exploiting sparsity of interconnections in spatio-temporal wind speed forecasting using Wavelet
637 Transform. Appl Energy 2016;165:735–47. doi:10.1016/j.apenergy.2015.12.082.
638 [68] Tascikaraoglu A, Sanandaji BM. Short-term residential electric load forecasting: A compressive spatio-temporal approach. Energy and Buildings
639 2016;111:380-92. doi:10.1016/j.enbuild.2015.11.068.