Electrical Power and Energy Systems 30 (2008) 569574

Contents lists available at ScienceDirect

Electrical Power and Energy Systems

journal homepage: www.elsevier.com/locate/ijepes

A hybrid society model for simulating residential electricity consumption

Minjie Xu a,b,*, Zhaoguang Hu b, Junyong Wu a, Yuhui Zhou a
a b

School of Electrical Engineering, Beijing Jiaotong University, Beijing, China State Power Economic Research Institute, Beijing, China

a r t i c l e

i n f o

a b s t r a c t
In this paper, a hybrid social model of econometric model and social inuence model is proposed for evaluating the inuence of pricing policy and public education policy on residential habit of electricity using in power resources management. And, a hybrid society simulation platform based on the proposed model, called residential electricity consumption multi-agent systems (RECMAS), is designed for simulating residential electricity consumption by multi-agent system. RECMAS is composed of consumer agent, power supplier agent, and policy maker agent. It provides the policy makers with a useful tool to evaluate power price policies and public education campaigns in different scenarios. According to an inuenced diffusion mechanism, RECMAS can simulate the residential electricity demandsupply chain and analyze impacts of the factors on residential electricity consumption. Finally, the proposed method is used to simulate urban residential electricity consumption in China. 2008 Elsevier Ltd. All rights reserved.

Article history: Received 26 July 2007 Received in revised form 15 May 2008 Accepted 15 August 2008

Keywords: Residential electricity consumption Multi-agent systems (MAS) Power price policy Electricity saving Social inuence model

1. Introduction With rapid development of economy and improvement of the living standard of people, residential power demand is growing rapidly, which creates a strong requirement to develop power resources management schemes. The traditional methods emphasize increasing power supplies, but they do not consider electricity saving, which results in resources waste and pollution problem. However, under the energy shortage condition, promoting the electricity saving awareness of people, changing the habits of electricity using, and improving electricity using efciency should be taken into account. Policy makers needs to combine structural and nonstructural methods to consider the potential of electricity saving of the consumers [1,2], the increasing population, and pollution. The structural and nonstructural methods include social education, power price policy, compensating method, and so on. However, it is difcult to evaluate the inuence of these methods on consumer behaviors. On one hand, the consumers have different beliefs, habits, skills and knowledge related to the environment, on the other hand, they interact with their friends, fellows, acquaintances, neighbors, and so on. According to the similarity between the consumers and the way agents, we adopt a multi-agent social simulation to evaluate the impact of policies in this paper. A hybrid social model for evaluat-

ing the impact of electricity saving policies is proposed, and a platform, called residential electric power consumption multi-agent systems (RECMAS), is designed to simulate the residential electricity demandsupply chain. The model benets from previous works: studies based on econometric model [3] and successful application of agent-based social simulation in water management [4]. RECMAS modies the traditional econometric models by designing a social simulation layer considering the social awareness on electricity saving policies and social education strategies. RECMAS enables the user to explore the inuences of the policies on total residential power consumption, and it facilitates the design, creation, modication, and simulation of different scenarios. This paper is organized as follows: the next section introduces some applications about multi-agent social simulation. Section 3 proposes the model of residential electricity demand, and puts more emphasis on consumers social environment for estimating social attribute variables. Section 4 describes the simulation in detail including its architecture, procedure and implementation. Section 5 performs the experiments in six scenarios. Finally, Section 6 concludes the results obtained in this paper. 2. Multi-agent social simulation Multi-agent systems establish a major research domain in articial intelligence [5], which was focused on the resolution of problems by society of agents. This eld is characterized by the study, design and implementation of articial agent society, which is the extensive use of computational modeling for real-world applications and social simulations [6]. The method of modeling

* Corresponding author. Address: State Power Economic Research Institute, Integrated Building Room 518, No. 1 Ertiao, Baiguang Road, Xuanwu District, Beijing 100761, China. Tel.: +86 10 6341 6269; fax: +86 10 6341 6779. E-mail address: 05117299@bjtu.edu.cn (M. Xu). 0142-0615/$ - see front matter 2008 Elsevier Ltd. All rights reserved. doi:10.1016/j.ijepes.2008.08.005

570

M. Xu et al. / Electrical Power and Energy Systems 30 (2008) 569574

complex systems by multi-agent social simulations has been used in a wide range of scientic elds, for example, society economy [7], organization theory [8], residential water using management [4], power market simulation [9], and so on. Multi-agent systems can deal with very different models of individuals, ranging from simple entities to more complex ones. On the other hand, some actual systems may be too complex or too large to be solved by a single agent, and moreover in most cases, they are inuenced by the external environment. Hence, multi-agent social simulation is an available way to deal with complex systems. Generally, in a multi-agent social simulation, the society is composed of a number of agents that are able to interact with each other and the environment, and they differ from each other in their habits, skills and knowledge about the environment. 3. Residential electricity demand model 3.1. Inuence factors of residential electricity consumption Residential electricity consumption is inuenced by many factors which have been summarized in a number of literatures based on questionnaire surveys or statistical data. Ref. [10] obtained that household incomes, household appliances, area, power prices are related (directly or indirectly) to the total electricity consumption by statistical data, and presented a simplied bottom-up model for forecasting electricity demand. The model can be used to generate realistic domestic electricity consumption data on an hourly basis from a few up to thousands of households. According to the questionnaires, Ref. [11] analyzed residential energy consumption patterns, and presented relationships between electricity consumption and factors, which include power prices, residential incomes, and housing conditions. Ref. [1] examined relationships between the total energy consumption and value patterns or motivation to save energy, and found that the people with the most motivation are willing to save energy, even though the costs are higher and comfort is lower, and the people with the least motivation are not willing to save energy, even though the costs are less and the comfort is higher. As a result, the least motivated group requires more energy than the average and most motivated groups. Results of [2] showed that education of electricity saving plays an important role in implementing and propagating the ideas of energy-efcient behavior and awareness. Ref. [13] presented simulation results about the effects of various dynamic pricing schedules on the average monthly bills for typical single-family houses. The simulation results showed that price responsive load management is an efcient way of peak load management for the residential sector. The inuence factors of residential electricity consumption are listed as follows (Fig. 1):

Family incomes; Housing conditions; Household appliances; Area and weather; Electricity saving technologies; Pricing strategies for discouraging inefcient use of electricity; Public social education for educating the consumers to change the consumer behavior and to enhance electricity using efciency; and  Financial incentive programs for encouraging efcient use of electricity.        3.2. Residential electricity demand forecast model For estimating electricity demand, a variety of methods and econometric models have been used on the basis of the nature and availability of data. Refs. [13] and [14] summarized four types of parametric econometric models of electricity demand, whose mathematical representations include linear, log-linear, translog, and almost ideal demand systems (AIDS) functional forms. Accuracy and applicability of these models were analyzed by using Bootstrapping method [14] and the Bayesian approach [13]. Each of the parametric functional forms tested performs poorly, suggesting that they may be insufciently exible to provide valid results in certain applications. Ref. [3] presented a forecast model of urban residential electricity demand by using OLS method, the model reects the theory and quantity relationships between electricity consumption and factors, which include urban residential incomes and urban power prices. Electricity demand estimation is usually formulated as a generic model of form D f P; H which relates electricity demand D to power prices P and family attribute H. The residential electricity demand is estimated by using the following equation:

Di; t a bPi; t cHi; t ei; t;

where Di; t is the electricity demand for consumer agent i at time t; Pi; t is the vector of price variables; Hi; t is the vector of consumer attribute variables (i.e., incomes, household appliances, habits of consumer, and so on); ei; t is the error term; a; b, and c are coefcients to be estimated. Econometric model can reect the underlying relationships between the consumption of electricity and explanatory variables such as power prices, incomes, and other factors. However, the model does not reect function of public social educations, which are tools for educating and informing consumers on how to modify their habits of electricity using and enhance electricity using efciency [1]. It is usually difcult to evaluate the results of such public social educations. On one hand, public social educations have a direct impact on consumers who participate in them, on the other hand, there is an indirect impact on the other consumers realized by participates who propagate the ideas of electricity saving to their friends, fellows, acquaintances, neighbors, and so on. We can use multi-agent social simulation to obtain the inuence results of public social educations in this paper. In order to consider the inuence of public social educations, the hybrid social model is formulated as D f P; H; S, which describes the relationships between electricity demand D and power price P, family attribute H, and social attribute S. The electricity demand is dened as

Di; t a bPi; t cHi; t dSi; t ei; t;

where Si; t is the vector of social attribute variables, d is a coefcient to be estimated, and the others are dened as before. 3.3. Residential social environment
Fig. 1. Inuence factors of residential electricity consumption and their relationships.

In the power demandsupply chain, residents live in a society and interact with each other. To simulate their interactions, we

M. Xu et al. / Electrical Power and Energy Systems 30 (2008) 569574

571

Fig. 2. Social environment model. Fig. 3. Structure of RECMAS simulation platform.

begin with the formalization of the residential social environment. We use a square social grid, as shown in Fig. 2, to simulate the society of residents. The society is composed of a set of consumer agents (CA), whose communication represents the social relationship among them. Each CA is situated on a square social grid and its state is determined by its position on the grid. So a single CA is identied as CAx; y, where x and y are its coordinates on the grid. Social interaction between CAs is limited to a neighborhood, dened as its neighborhood environment. In Fig. 2, the 2-D grid of a side is equal to 6, the neighborhood scope is limited to 1, and the neighborhood area of CA(3, 4) is marked out by the rectangular frame. The social inuence model is realized in the neighborhood area, so CA(3, 4) electricity consumption is affected only by its three neighbors, CA(2, 3), CA(2, 5), and CA(4, 3). So the social attribute variables can be calculated by the following equation [4]:

Si; t f

ni X j0

swij ;

where Si; t is the social variables of CAi at time t, swij is the social weights that consumer agent CAi receives from its neighbor j; ni is the number of neighbors of CAi. f is a diffraction function for adjusting the sum of social weights, and it represents a consumers ability to comprehend signals of electricity saving. f is chosen as S function based on peoples cognitive competence, i.e.,

f x

1 ; 1 ex5

microcosmic agent role partition and function denition in agent simulation system from sociology and histology angle. It regards agent simulation system as an organization or society, and facilitates the design, creation, modication, and execution of the multi-agent system, and denes agent role, purview, responsibility, agreement, activity and security. Anyones role can be regarded as an entity function abstract description. By hybrid society model, we think over urban residential electricity demand problem in one area. There is only one PSA in the area, whose main mission is to offer power according to residential demand and advise power price to PMA according to their benet, and role schema is shown in Table 1. CA has two roles: one is electricity consumer that indicates its function; the other is consumers neighbor that indicates its society character. Its roles are shown in Tables 2 and 3, respectively. Each CA uses the hybrid society model to estimate its personal demand and reports its demands to PSA; and at the same time, the CA communicates with its neighbors to propagandize electricity saving policy and power price policy, its inuence is reected by society weight. The inuence mechanism has been given in Section 3.3. PMA enacts the power price according to the total power supply, demand, and society environment, whose role schema is shown in Table 4. 4.3. The simulation procedure of hybrid society model In RECMAS, all agents interact with each other and implement the overall simulation procedure, which is shown in Fig. 4. Step 1: Scenario input and initialization. User species the scenario to be simulated, including the size of the group, CAs initial states and neighborhood limit, iteration interval time, and maximal iteration. Step 2: PMA enacts the price and informs PSA and CAs the price. Step 3: CA receives price information, and communicates with its neighbors according to the social inuence mechanism. Step 4: CA estimates personal electricity demand by the hybrid social model. Step 5: CA reports its demand to PSA. Step 6: PSA collects all residents demands, calculates the total demand and the total costs, and reports the results to PMA. Step 7: PMA adjusts the price if needed, turn to step 2. Step 8: When the iteration is over, the simulation results are obtained.

x 2 1; 10:

4. Simulating residential electricity consumption based on the hybrid social model 4.1. RECMAS structure The structure of the simulation platform RECMAS based on multi-agent, as shown in Fig. 3, includes three kinds of agents: policy maker agent (PMA), electricity consumer agent (CA), and power supplier agent (PSA). PMA is responsible for enacting the power price in each period. CA plays the major role in the simulation, using the hybrid social model to estimate personal electricity demand. PSA is responsible for collecting CAs demands and calculating the total demand, and it can give some suggestions to PMA according to its own account. 4.2. Agent role This paper designs every agent function and state in simulation platform by Gaia method [15]. It describes macroscopical and

572 Table 1 The power supplier role schema Role Description Activities Protocols Permissions Power supplier agent

M. Xu et al. / Electrical Power and Energy Systems 30 (2008) 569574

Simulating power suppliers behavior. Offering power according to residential demand and advising power price to PMA according to their benet Wait start step, query customer, calculate step total demand Submit total demand, send price advice, get customer demand Read: power price, personal demand, Write: step total demand, Send: price advice PSA wait start step get customer demand calculate step total demand submit total demand send price advie a True

Responsibilities Liveness Safety

a

x y notes x followed by y; x notes x occurs 1 or more times.

Table 2 The electricity consumer role schema Role Description Activities Protocols Permissions Responsibilities Liveness Safety
a

Electricity consumer agent Simulating electricity consumers behavior. Calculating electricity demand according to incomes, price, society environment, and so on Wait start step, consum electricity Send personal demand, contact neighbour, receive price Read: power price, neighbours list, personal parameters, last step demand, step ID Write: step personal consumption PSA wait start step receive price contact neighbour consum electricity send personal demand a Personal electricity demand > 0

x y notes x followed by y; x notes x occurs 1 or more times, x notes x occurs 0 or more times.

Table 3 The neighbor role of schema Role Description Neighbour Simulating consumers neighbor behavior. Propagandizing electricity saving policy and power price policy according to their society property and common education degree Calculate weights Reply neighbour Read: neighbours list, social weight parameters, step ID Write: social weights

Activities Protocols Permissions

Responsibilities Liveness PSA calculate weights reply neighbour a Safety Neighbour [neighbour list]
a x y notes x followed by y; x notes x occurs 0 or more times, x notes that x is optional.

Table 4 The policy maker role schema Role Description Policy maker agent Simulating policy maker behavior. Enacting the power price according to total power supply, demand relationship, and society environment Decide policy Send policy Read: total electricity demand Change: policy Fig. 4. Simulation steps of hybrid social model.

Activities Protocols Permissions

Responsibilities Liveness PSA decide policy send policy a Safety True

a

x y notes x followed by y; x notes x occurs 1 or more times.

In China, this phenomenon is very outstanding. Because electrication rate in urban area is higher than that in rural area, electricity consumption personal per year in urban area was two times more than that in rural area in 10th ve-year plan period. In rural area, biomass energy is about 50% in total energy consumption, such as straw and stalk. As an example, urban residential electricity demand is analyzed in this paper. 5.1. Calibration and validation of the model

5. Experiment analysis It is generally agreed that there is a lager difference in energy consumption, especially electricity, between urban and rural areas.

In order to testify the rationality of hybrid social model, we need to validate and calibrate the model. Because it is difcult to evaluate public social education, validating and calibrating of model is not an easy task. In the model, education scenarios were set

M. Xu et al. / Electrical Power and Energy Systems 30 (2008) 569574

573

according to annual Campaign of Energy Saving Propaganda Week. In China, propaganda and education of energy saving is highly emphasized. From 2001 to 2005, annual Campaign of Energy Saving Propaganda Week had been successfully held. The topic of every year was Energy saving, sustainable development, Energy saving by law, sustainable development, Energy saving, all-round construction of the well-off society, Electricity saving for alleviating the bottleneck restrictions in the economic development, Mobilization and participation, construct energy saving society, respectively. Other parameters of the model were set by using data in [3]. In simulation environment, we set 10 10 society spaces and 80 CAs, and the CAs are categorized into four classes and randomly put on the 100 units. Urban residential electricity demand was forecasted in the period of 20012005 in China. The results are shown in Fig. 5. In Fig. 5, results of hybrid social model, results of econometric model [3] and actual electricity consumption personal per year (Statistical data) are compared. 5.2. Simulation experiment scenarios The platform was used to simulate six scenarios, which are as follows:  Scenario 1: Power price is increased by 5%, without considering the public social educations.  Scenario 2: Power price is increased by 7.5%, without considering the public social educations.  Scenario 3: Power price is adjusted to the real price, without considering the public social educations.  Scenario 4: Power price is increased by 5%, considering the public social educations, with the implementation of education or other information policy.  Scenario 5: Power price is increased by 7.5%, considering the public social educations, with the implementation of education or other information policy.  Scenario 6: Consumers income levels remain unchanged every year.

Table 5 Types of consumer Consumer type A. Opinion leaders B. Socially apathetic C. Opinion seekers D. Opinion receivers Proportion 10 20 40 30 Ability to promote High None Low Low Consumption reduction Low None High Low

 Type A: Opinion leaders, 10%, are supposed to be environmental aware; their ability to further lower their electricity consumption is generally limited.  Type B: Socially apathetic, 20%, neither promote nor comprehend electricity saving signals and have a negative attitude about conservation.  Type C: Opinion seekers, 40%, are supposed to be socially sensitive and act as opinion followers. They can be easily inuenced by families through their social relations with opinion leaders.  Type D: Opinion receivers, 30%, are the average consumers. Their attitude is passive, as they need considerable pressure by their contacts to start changing their habits of electricity use.

5.4. Simulation results Simulation platform was fullled by JAVA based on MAGE [17] which is developed by Institute of Computing Technology of Chinese Academy of Sciences. Each simulation period was supposed to be one year. The quantitative estimations obtained for the period from 2006 to 2010 are shown in Fig. 6 in the form of personal demand (kWh/year). A simulation for Scenario 6 was run under the hypothesis of the consumers income level to be constant every year, with slow increase in electricity demand. Compared with other scenarios, the experiment results show that the growth of incomes is the main reason that leads to the electricity demand growth. The comparisons between Scenario 1 and Scenario 4, Scenario 2 and Scenario 5 indicate that public social education and information policy can promote consumers awareness of electricity saving and make more efcient electricity using. Thus, through public social education, it can save about 2% electricity per capita every year. For example, in China, 562,120,000 urban residents can save 33.72 millions kWh electricity and reduce 3,170,400 tons carbon dioxide emission (carbon dioxide emission coefcient with unit electricity: 9.4 tons per ten thousand kWh) in 2005. The comparison between Scenario 1 and Scenario 3 shows that the power price is increased by 5%, which can save about from 2% to 3% electricity per capita every year. The comparison between Scenario 2 and Scenario 3 shows that the power price is increased by 7.5%, which can save about from 3% to 4% electricity per capita every year. 6. Conclusions In this paper, we proposed a hybrid society model and RECMAS for analyzing residential electricity demand. The hybrid social model extends traditional models by integrating social inuence model. RECMAS modies the traditional econometric models by designing a social simulation layer considering the social awareness on electricity saving policies and social education strategies. RECMAS is designed based on multi-agent, which can simulate different scenarios to obtain the inuence of power prices, residential incomes and public social educations on residential electricity demand. The advantage of RECMAS is that it supports communication and inuence between consumers in society. Every consumer

5.3. Consumer types Residential energy consumption value mode was analyzed in [16] by society questionnaire, and eight kinds of consumption conceptions were summarized; four types of consumers were analyzed in [4]. This paper combines the results in [4] and [16], consumer agents are clustered in four types, with respect to their abilities to promote and comprehend electricity saving signals. They are listed in Table 5.

Fig. 5. Comparison of econometric model and hybrid social model.

574

M. Xu et al. / Electrical Power and Energy Systems 30 (2008) 569574

References
[1] Vringer K, Aalbers T, Blok K. Household energy requirement and value patterns. Energy Policy 2007;35(1):55366. [2] Dintchev OD, Calmeyer JE, Delport GJ. Efcient and sustainable usage of electricity in South Africa: the role of the tertiary education institutions. Eng Sci Educ J 2000;9(2):539. [3] Song W, Gu AL, Wu ZX. Urban residential electricity consumption forecast model. Electr Power 2006;39(9):6670. [4] Athanasiadis IN, Mentes AK, Mitkas PA, Mylopoulos YA. A hybrid agent-based model for residential water demand. Simulation 2005;81(3):17587. [5] Weiss G. Multi-agent systems: a modern approach to distributed articial intelligence. MIT Press; 1999. [6] Jenning NR, Panzarasa P. The organization of sociality: a manifesto for a new science of multi-agent systems. In: Proceedings of the 10th European workshop on multi-agent systems (MAAMAW-01), Annecy, France; 2001. [7] Lopez A, Hernndez C, Pajares J. Towards a new experimental socio-economics: complex behavior in bargaining. J Socio-Econ 2002;31(4):4239. [8] Rivkin J, Siggelkow N. Balancing search and stability: interdependencies among elements of organizational design. Manage Sci 2003;49(3):290311. [9] Yuan JH, Ding W, Hu ZG. Acritical study of agent-based computational economics and its application in research of electricity market theory. Power Syst Technol 2005;29(7):4751. [10] Paatero JV, Lund PD. A model for generating household electricity load proles. Int J Energy Res 2005;30(5):27390. [11] Houri A, Ibrahim-Korfali S. Residential energy consumption patterns: the case of Lebanon. Int J Energy Res 2005;29(8):75566. [13] Moholkar A, Klinkhachorn P, Feliachi A. Effects of dynamic pricing on residential electricity bill. In: Power systems conference and exposition, IEEE PES, vol. 2; 2004. p. 10305. [13] Xiao N, Zarnikau J, Damien P. Testing functional forms in energy modeling: An application of the Bayesian approach to US electricity demand. Energy Econ 2007;29(2):15866. [14] Zarnikau J. Functional forms in energy demand modeling. Energy Econ 2003;25(6):60313. [15] Wooldridge M, Jenning NR, Kinny D. The Gaia methodology for agent-oriented analysis and design. Auton Agent Multi-Agent Syst 2000;3(3):285312. [16] Hessing-Couvret E, Reuling A. The WIN-Model, Value Systems in the Netherlands. TNS-NIPO, Amsterdam; 2002. [17] Shi ZZ, Zhang HJ, Dong MK. MAGE: multi-agent environment, ICCNMC-03; 2003.

Fig. 6. Simulation results.

communicates with its neighbors to promote electricity saving policies, changes the consumer behavior and improves electricity using efciency. RECMAS is a residential society simulation platform and can provide a policy simulation environment for decision-maker to reduce the cost of actualizing policy. The hybrid society model and RECMAS were used to simulate six scenarios and obtain quantitative estimations of future demand. They provide the decision-maker with a tool to evaluate the effect of power price policies and social education strategies on residential electricity consumption. Appendix A. Supplementary data Supplementary data associated with this article can be found, in the online version, at doi:10.1016/j.ijepes.2008.08.005.