Professional Documents
Culture Documents
743
Prediction of daily peak loads Prediction of daily total en-
ergy usage indicates that house category and usage of stand-alone
Total number of occupants Total number of occupants freezer devices are key factors for the daily total energy
Number of occupants > 15 Number of occupants > 15 consumption, but are weakly related to the daily peak
years old years old load. Such results follow common sense: the type of
Whether there is a dishwasher Whether there is a dishwasher
Whether there is a tumble dryer Number of bedrooms the house is related to the thermal performance and
Number of occupants < 15 Whether there is a tumble dryer inertia of the house, which crucially affects the total
years old amount of energy – required to maintain a desirable
Whether there is an electric Number of occupants < 15
cooker years old temperature in the house. Next, we show that a data-
Number of bedrooms Whether there is a stand-by driven model based on socio-economic factors can
freezer indeed find associations between user profiles and their
Type of the cooking appliances Average age of occupants
Average age of occupants Type of the house consumption behavior.
Employment status Employment status
744
Model Peak Load: R2 Total Load: R2 Peak loads prediction Daily total consumption prediction
5
RF(100) 0.5240 0.7261
4
RF(200) 0.5127 0.7236
3
2
RF(300) 0.5208 0.7100
1
RF(400) 0.5157 0.7070
5
0 5 10 15 20 25 30 0 5 10 15 20 25 30
SVM 0.4816 0.6572 Day Day
Table 2: Generalization error of peak load and total (a) Prediction of daily peak (b) Prediction of daily to-
electricity consumption prediction, with the Random load of one user during one tal consumption usage of one
Forest Model (RF) and SVM. Results for the RF model month. user during one month.
include values for the tree parameter (in parenthesis)
we used. Figure 1: An example of energy consumption indicator
forecasting.
745
Prediction of daily peak loads Prediction of daily total en-
ergy usage lack of accurate temperature data that affects our study,
Total number of occupants Total number of occupants since (due to privacy reasons) it was not possible to
Number of occupants > 15 Number of occupants > 15 join our temperature dataset to specific geographical
years old years old coordinates corresponding to the households in our
Type of the cooking appliances Average daily heating degree
days study. Overall, the ranking produced by the random
Whether there is an electric Number of bedrooms forest method is in line to that obtained through a
cooker manual inspection of the data by domain experts.
Average daily heating degree Whether there is a dishwasher
days
Whether there is a tumble dryer Month of year
Whether there is a dishwasher Whether there is a tumble dryer
Type of the appliances to heat Whether there is an electric
water in the house heater 5. Conclusion
Whether there is an instant elec- Whether there is a stand-by
tric shower facility freezer
Whether there is an electric Type of the appliances to heat
heater water in the house
In this paper, the issue of forecasting daily elec-
Table 3: Top 10 factors in prediction tasks. tricity usage patterns and evaluating the influence of
clients’ socio-economic factors on energy consumption
has been addressed.
peak loads and total energy consumption. The ranking A forecasting method based on random forest that
obtained is consistent to that obtained using the mutual seamlessly incorporates both clients’ socio-economic
information criterion in Sec. 3. The number of occu- factors and environmental factors has been proposed,
pants is ranked as the most important feature, affecting resulting in an ensemble of split-branch decision rule
both daily peak load level and the total amount of chains, whereby a voting mechanism is used to achieve
energy consumption, which follows the intuition of a stable prediction. The rule-chain based structure
a direct relation between household size and energy enabled the explaination of how each input features
consumption. Main appliances, such as cooking and contributes to energy consumption forecasting, thus
heating devices, are common features for the two unveiling the underlying physical association between
forecasting tasks. Indeed, they contribute to a large predictors and the daily energy usage pattern. These
extent to instantaneous peaks in energy consumption, learned associations can either be used to discover
and are generally responsible for aggregate energy con- unknown factors of energy consumption behaviors,
sumption as well. Appliances that have a daily cycle or they can be applied as a complementary decision
(e.g. freezers) and average daily heating degree days support to human experts.
are specially useful for estimating daily consumption,
rather than representative of peak demand. Experimental results based on a large-scale energy
consumption dataset showed that our methodology
We remark that, for peak load forecasting, the dramatically outperforms state-of-the-art approaches,
random forest method boosts the importance of the both in terms of prediction accuracy and in terms
water heating appliances, while it completely neglects of scalability. Additionally, our work evaluated the
indirect factors, such as the number of bedrooms, importance of socio-economic factors in the fore-
social class of the chief income and employment status. casting tasks. Our result can guide the construction
For total daily energy consumption, the random forests of socio-economic surveys without requiring human
method assigns higher importance scores to factors intervention, leading to a decreased intrusiveness of
related to heating usage – including environmental measurement campaigns.
heating degree index, month of year, the year when
the house is built, the number of bedrooms – while As shown in the experiments conducted in this
it decreases the importance of social class, age and work, daily peak load forecasting is a difficult task,
employment status of the owner. Interestingly, environ- compared to aggregate energy consumption. As such,
mental temperature seems not play an important role an extension of this work will build detailed usage
in both forecasting tasks, which is counter intuitive. patterns of large power appliances that play a key role
A possible reason for this result is related to high in determining peak loads and will integrate historical
heating loads at all times due to Irish weather status. peak load information to extend the input set of the
Furthermore, such results can also be explained by the forecasting model.
746
References [11] McLoughlin, F., Duffy, A., and Conlon,M., “Char-
acterising domestic electricity consumption patterns
[1] Aman, S., Simmhan, Y., and Prasanna, by dwelling and occupant socio-economic variables:
V.K.,“Improving Energy Use Forecast for Campus An Irish case study”, Energy and Buildings, vol.48,
Micro-grids using Indirect Indicators”, in Proceedings pp.240-248, 2012.
of the International Workshop on Domain Driven [12] Beckel, C., Sadamori, L., and Santini,S.,“Towards
Data Mining (DDDM), pp.1-9, 2011. automatic classification of private households using
[2] Tennesee Valley Authority, “Energy Vision 2020: In- electricity consumption data”, in Proceedings of the
tegrated Resource Plan/Environmental Impact State- Fourth ACM Workshop on Embedded Sensing Systems
ment”, Chapter 6, December, 1995. for Energy-Efficiency in Buildings, pp.169-176, 2012.
[3] Kolter, Z. and Ferreira, J., “A large-scale study on [13] Beckel, C., Sadamori, L., and Santini, S., “Auto-
predicting and contextualizing building energy usage”, matic socio-economic classification of households us-
in Proceedings of the Twenty-Fifth AAAI Conference ing electricity consumption data”, in Proceedings of
on Artificial Intelligence, pp.330-338, 2011. the fourth international conference on Future energy
systems, pp.75-86, 2013.
[4] Centra, M., “Hourly Electricity Load Forecasting: An
Empirical Application to the Italian Railways”, World [14] Taylor, J.W., and Buizza,R., “Neural Network Load
Academy of Science, Engineering and Technology, Forecasting With Weather Ensemble Prediction”,
vol.5, pp.888-895, 2011. IEEE Transactions on Power Systems, vol.17, pp.626-
632, 2002.
[5] Hesham K.A. and Nazeeruddin, M., “Electric load
forecasting: literature survey and classification of [15] Mohandes, M., “Support vector machines for short-
methods”, International Journal of Systems Science, term electrical load forecasting”, International Journal
vol.33, pp.23-34, 2002. of Energy Research, vol.26, pp.335-345, 2002.
[6] Humeau, S.F.R.J., Wijaya, T. K., Vasirani, M. and
[16] El Desouky, A., and Elkateb, M., “Hybrid adaptive
Aberer, K.,“Electricity Load Forecasting for Residen-
techniques for electric-load forecast using ANN and
tial Customers: Exploiting Aggregation and Correla-
ARIMA”, in Proceedings of Generation, Transmission
tion between households”, Sustainable Internet and
and Distribution, vol.147, pp.213-217, 2000.
ICT for Sustainability (SustainIT), pp.1-6, 2013.
[7] Aung, Z., Touky, M., Williams, J.R., and Herrero, [17] Abreu, J., and Pereira, F., “Household Electricity
S., “Towards Accurate Electricity Load Forecasting Consumption Routines and Tailored Feedback”, in
in Smart Grids”, in Proceedings of The Fourth In- Proceedings of ACEEE Summer Study on Energy
ternational Conference on Advances in Databases, Efficiency, pp.193-206, 2012.
Knowledge, and Data Applications, pp.52-57, 2012.
[18] “Eletricity Customer Behaviour Trial”, Commission
[8] Miranda, V., and Monteiro, C., “Fuzzy Inference Ap- for Energy Regulation, Ireland,2011.
plied to Spatial Load Forecasting”, in Proceedings of
International Conference on Electric Power Engineer- [19] Bishop, C.M., “Pattern Recognition and Machine
ing, pp.35-40, 1999. Learning”, Spring Verlag, 2006.
[9] Fu, C.W., and Nguyen, T.T, “Models for long-term [20] Fan, R.E., Chang, K.W., Hsieh, C.J., Wang, X.R.,
energy forecasting”, in Proceedings of IEEE Power and Lin, C.J., “LIBLINEAR: A library for large
Engineering Society General Meeting, pp.235-239, linear classification”, Journal of Machine Learning
2003. Research, vol.9, pp.1871-1874, 2008.
[10] Kermanshahi B. S., and Iwamiya H., “Up to year 2020
load forecasting using neural nets”, Electric Power
System Research, Elsevier, vol.24, pp.787-797, 2002.
747