Electricity load forecasting by artificial neural network model


Published on

  • Be the first to comment

No Downloads
Total views
On SlideShare
From Embeds
Number of Embeds
Embeds 0
No embeds

No notes for slide

Electricity load forecasting by artificial neural network model

  1. 1. INTERNATIONAL JOURNAL OF ELECTRICAL ENGINEERING International Journal of Electrical Engineering and Technology (IJEET), ISSN 0976 – 6545(Print), ISSN 0976 – 6553(Online) Volume 4, Issue 1, January- February (2013), © IAEME & TECHNOLOGY (IJEET)ISSN 0976 – 6545(Print)ISSN 0976 – 6553(Online)Volume 4, Issue 1, January- February (2013), pp. 91-99 IJEET© IAEME: www.iaeme.com/ijeet.aspJournal Impact Factor (2012): 3.2031 (Calculated by GISI) ©IAEMEwww.jifactor.com ELECTRICITY LOAD FORECASTING BY ARTIFICIAL NEURAL NETWORK MODEL USING WEATHER DATA Balwant singh Bisht1 and Rajesh M Holmukhe2 1 Post graduate student (M.E.Electrical Engineering), Electrical Engineering Department, Bharati Vidyapeeth Deemed University College of Engineering, Pune411043 (MS), India. Email: balwant_sb@rediffmail.com 2 Associate Professor in Electrical Engineering, Electrical Engineering Department, Bharati Vidyapeeth Deemed University, College of Engineering, Pune-411043(MS), India. Email: rajeshmholmukhe@hotmail.com ABSTRACT This paper discusses significant role of advanced technique in short-term load forecasting (STLF), that is, the forecast of the power system load over a period ranging from one hour to one week. An adaptive neuro - wavelet time series forecast model is adopted to perform STLF. The model is composed of several neural networks (NN) whose data are processed using a wavelet technique. The data to be used in the model are both the temperature and electricity load historical data. The temperature variable is included because temperature has a close relationship with electricity load. The calculation of mean average percentage error for a specific region under study in India is done and results obtained using MATLAB’S ANN toolbox. This study proposes a STLF model with a high forecasting accuracy. In this study absolute mean error (AME) value calculated is 1.24% which represents a reasonable degree of accuracy. Key words: short term load forecasting, artificial neural network, power system 1. INTRODUCTION Short term load forecasting (STLF) studies began at early 1960’s. In 1971, a load forecasting system was developed by researchers in United States which used statistical approach. Subsequent to 1990’s researchers started to implement different approaches for STLF other than statistical approach mainly due to their requirement for huge data sets to 91
  2. 2. International Journal of Electrical Engineering and Technology (IJEET), ISSN 0976 –6545(Print), ISSN 0976 – 6553(Online) Volume 4, Issue 1, January- February (2013), © IAEMEimplement STLF systems. In 2003 many STLF studies were carried using neural networkmodels.Load forecasting occupies a central role in the operation and planning of ElectricPower System. Load forecasting can be divided into three major categories: Long-term loadforecasting, Medium-term load forecasting and Short-term load forecasting. STLF precedesmany important roles carried in energy management systems (EMS), which continuoslymonitors the system and initiates the control actions in time critical Situations. STLF modelis critical important decision support tool for operating the electric power system securelyand economically. Load forecasting can be made by different methods like regression analysis, statisticalmethods, artificial neural networks, genetic algorithm, fuzzy logic, etc.In the recent years,many researchers have tried to use the modern techniques based on artificial intelligence. Ofall techniques, the artificial neural network (ANN) receives the most attention. ANN isregarded as an effective approach and is now commonly used for electricity load forecast.The reason for its popularity is its ease of use and its ability to learn complex input-outputrelationship. The ability to learn gives ANN a better performance in capturing nonlinearitiesfor a time series signal. Therefore, the study in this paper proposes a model comprisingneural networks as its forecasting tool. This paper explores an adaptive neuro-waveletmodel for Short Term Electricity Load Forecast (STLF). Both historical load and temperaturedata, which have important impacts on load level, are used in forecasting by the proposedmodel. To enhance the forecasting accuracy by neural networks, the non-decimated WaveletTransform (NWT) is introduced to pre-process these data. The objective of this study is conduct out short-term load forecasting using MATLAB’SANN Toolboxes. Artificial Neural Network (ANN) Method is applied to forecast the short-term load for a large power system in one state of India. A nonlinear load model is proposedand several structures of ANN for short term forecasting are tested. The data used in the model, both the weather and electricity load historical data wereobtained from metrological office of Government of India (GOI), Pune(India) and state loaddispatch center,Mumbai(India).Field visits to state load dispatchcenter,Mumbai(India) weredone.2. LITERATURE SURVEY One of the first published studies was done by Heinemann et al. in 1966 which dealtwith the relationship between temperature and load. Lijesen and Rosing (1971) developedload orecastig systemwhich used statistical approach. In this study, estimated average rootmean square error value was 2.1%. Hagan and Behr (1987) forecasted load using a timeseries model. With this model, the nonlinear relationship between load and temperature dataduring winter months was clearly observed. Park et al (1991) claimed that the statisticalmethods like regression and interpolation did not provide reliable prediction performancesthat of artificial neural network (ANN). The average absolute errors of the one-hour and 24-hour ahead forecasts were calculated as 1.40% and 2.06%, respectively. This method wasfound successful when compared with the regression method for 24 hour ahead forecasts withan error of 4.22%. Xu(2003) considered market forecasting by using various new techniques,such as wavelet, neural network and support vector machine.The author explored howdifferent models for electricity Load and price forecast have been developed, which are 92
  3. 3. International Journal of Electrical Engineering and Technology (IJEET), ISSN 0976 –6545(Print), ISSN 0976 – 6553(Online) Volume 4, Issue 1, January- February (2013), © IAEMEable to forecast at one or more time steps ahead. Dong et al (2003) presented an adaptiveneuro-wavelet model for Short Term Electricity Load Forecast (STLF). Both historicalload and temperature data, which have important impacts on load level, are used inforecasting by the proposed model. To enhance the forecasting accuracy by neuralnetworks, the Non-decimated Wavelet Transform (NWT) is introduced to pre-processthese data. Benaouda et al (2005) looked at wavelet multiscale decomposition basedautoregressive approach for the prediction of one-hour ahead load based on historicalelectricity load data.Ching-Lai Hor et al (2005) made use of forecasting model includingboth climate-related and socioeconomic factors that can be used very simply by utilityplanners to assess long-term monthly electricity patterns using long-term estimates ofclimate parameters, gross domestic product (GDP), and population growt. Myint et al(2008) proposed a novel model for short term loadforecast (STLF) in the electricitymarket.In this study the prior electricity demand data are treated as time series. The modelis composed of several neural networks whose data are processed using a wavelettechnique. The model is created in the form of a simulation program written withMATLAB.3. KEY FEATURES OF ARTIFICIAL NEURAL NETWORK BASED SHORTTERM LOAD FORECASTING (ANNSTLF) Advantages of ANNSTLF1. Adaptive learning: An ability to learn how to do tasks based on the data given fortraining or initial experience.2. Self-Organization: An ANN can create its own organization or representation of theinformation it receives during learning time.3. Real Time Operation: ANN computations may be carried out in parallel, and specialhardware devices are being designed and manufactured which take advantage of thiscapability. Limitations in the ANNSTLFSeveral difficulties exist in short-term load forecasting such as precise hypothesis of theinput-output relationship, generalization of experts’ experience, the forecasting ofanomalous days, inaccurate or incomplete forecasted weather data.4. MATHEMATICAL MODEL OF A NEURON A neuron is an information processing unit that is fundamental to the operation ofa neural network. The three basic elements of the neuron model are. A set of weights, anadder for summing the input signals and activation function for limiting the amplitude ofthe output of a neuron. 93
  4. 4. International Journal of Electrical Engineering and Technology (IJEET), ISSN 0976 –6545(Print), ISSN 0976 – 6553(Online) Volume 4, Issue 1, January- February (2013), © IAEME Figure 1 Model of an artificial neural network (ANN)5. NEURAL NETWORK (NN) MODEL WITH WAVELET ENHANCEMENTFOR TIME SERIES FORECAST To improve the quality of the raw input signal for time series forecast, theneural network model is enhanced with multi-scale wavelet transform. Figure belowshows an illustration of the wavelet enhanced neural network model for time seriesforecast.The inputs given are: Hourly load demand for the full day, day of the week,min/max/ average daily temperature and min/max daily humidity. Figure 2 Input-output schematic for load forecasting6. MEAN AVERAGE PERCENTAGE ERROR (MAPE) ; PA = Actual load demand, PF = Forecasted load demand, N= Number of time sections. 94
  5. 5. International Journal of Electrical Engineering and Technology (IJEET), ISSN 0976 –6545(Print), ISSN 0976 – 6553(Online) Volume 4, Issue 1, January- February (2013), © IAEMEFollowing table shows the different methods of the electric load forecasting and meanabsolute percentage error based on literature survey. The figure shows the neural networkmethod is more accurate than any other methods. Therefore method used in this study i.eartificial neural network based load forecasting method is justified. Table 1 Various STLF methods & MAPE Forecasting MAPE Methods Regression 3.74% Time Series 3.13% Expert System 2.74% Fuzzy Logic 2-3% Super Vector 2.14% Machine Neural Network 1.81%Figure 3 Actual v/s forecasted load for Sunday7. RESULTS The results obtained from testing the trained neural network on new data for 24 hoursof a day over a one-week period are presented below in graphical form. Each graph shows aplot of both the predicted and actual electrical load in MW values against the hour of theday. The absolute mean error % (AME %) between the predicted and actual loads for eachday has been calculated and presented in the table. Overall, these error values translate to anabsolute mean error of 1.24% for the network. This represents a high degree of accuracy inthe ability of neural networks to forecast electric load. Figure 4 Neural network training Figure 5 Mean squared error at first little iteration. at first little iteration 95
  6. 6. International Journal of Electrical Engineering and Technology (IJEET), ISSN 0976 –6545(Print), ISSN 0976 – 6553(Online) Volume 4, Issue 1, January- February (2013), © IAEME Figure 6 Neural network training after Figure 7 Mean squared error after 10000 10000 iterations iteration Figure 8 Comparison between actual targets and predictions Table 2 Mean average percentage error Day Marc Augu Januar h 3rd st 2nd y 1st week week week Sunday 0.74 2.28 1.47 Monday 0.69 0.72 1.09 Tuesday 0.18 0.23 2.38 Wednesday 1.41 1.81 0.21 Thursday 1.50 1.07 0.62 Friday 0.62 1.27 3.08 Saturday 3.24 0.94 0.39 Average 1.14 1.18 1.32 96
  7. 7. International Journal of Electrical Engineering and Technology (IJEET), ISSN 0976 –6545(Print), ISSN 0976 – 6553(Online) Volume 4, Issue 1, January- February (2013), © IAEME 8. Characteristics of the power system loadVarious factors that influence the system load behavior, can be classified into the majorcategories as weather, time, economy, random disturbance etc Figure 9 Hour load profile of grid in this study for a week of March 2010From the above daigram it is seen that ,typically load is low and stable from 0:00 to 6:00; itrises from around 6:00 to 9:00 and then becomes flat again until around 12:00; then itdescends gradually until 17:00; thereafter it rises again until 19:00; it descends again until theend of the day.CONCLUSION The result of adaptive neuro-wavelet time series forecast model used for one dayahead short term load forecast for the considered area under study in India has a goodperformance and reasonable prediction accuracy. Its forecasting reliabilities were evaluatedby computing the mean absolute error between the exact and predicted electrcity loadvalues.We were able to obtain an Absolute Mean Error (AME) of 1.24% which represents ahigh degree of accuracy. The results suggest that ANN model with the developed structurecan perform good prediction with least error and finally this neural network could be animportant tool for short term load forecasting. The accuracy of the electricity load forecast iscrucial in better power system planing and reliability.ACKNOWELDGEMENT Bharati Vidyapeeth Deemed University College of Engineering, Pune (India) forproviding MATLAB software and all necessary lab & library facilities.Metrologicaldepartment of Government of India for weather data. State load dispatch center, Mumbai(India) for load data. 97
  8. 8. International Journal of Electrical Engineering and Technology (IJEET), ISSN 0976 –6545(Print), ISSN 0976 – 6553(Online) Volume 4, Issue 1, January- February (2013), © IAEMEREFERENCES[1] H. K. Alfares and M. Nazeeruddin, “Electric load forecasting: literature survey and classification,” International Journal of Systems Science, vol. 33, no. 1, 2002.[2] D. Benaouda, F. Murtagh, J. L. Starck, and O. Renaud, “Wavelet-Based Nonlinear Multiscale Decomposition Model for Electricity Load Forecasting,” Sciences-New York, no. 1, pp. 1-47, 2005.[3] Ly Fie Sugianto Xue-Bing Lu School of Business Systems. “Demand forecasting in the deregulated market: a bibliography survey”[4] M. Buhari and S. S. Adamu, “Short-Term Load Forecasting Using Artificial Neural Network,” Computer, vol. I, 2012.[5] Z. Y. Dong, X. Li, Z. Xu, K. L. Teo, and S. Lucia, “weather depenent electricity market forecasting with neural networks , wavelet and data mining techniques” ,School of Information Technology and Electrical Engineering,1998.[6] M. T. Hagan and S. M. Behr, “3, August 1987,” Power, vol. 0, no. 3, pp. 785-791, 1987.[7] C.-lai Hor, S. J. Watson, and S. Majithia, “Analyzing the Impact of Weather Variables on Monthly Electricity Demand,” IEEE transactions on power systems, vol. 20, no. 4, pp. 2078- 2085, 2005.[8] F. Mosalman, A. Mosalman, H. M. Yazdi, and M. M. Yazdi, “One day-ahead load forecasting by artificial neural network,” Power, vol. 6, no. 13, pp. 2795-2799, 2011.[9] P. Murto,“Neural network models for short-term load forecasting”, Department of Engineering Physics and Mathematics Pauli Murto,” 1998.[10] D. C. Park, R. J. Marks, L. E. Atlas, and M. J. Damborg, “Electric load forecasting using an artificial neural network - Power Systems, IEEE Transactions on,” Power, vol. 6, no. 2, pp. 442- 449, 1991.[11] M. Park, “Adaptive forecasting,” Power, pp. 1757-1767.[12] L. Wang, “Short-term Electricity Load Forecasting Based on Particle Swarm Algorithm and SVM,” no. Vc.[13] C. Xia, J. Wang, and K. Mcmenemy, “Electrical Power and Energy Systems Short , medium and long term load forecasting model and virtual load forecaster based on radial basis function neural networks,” International Journal of Electrical Power and Energy Systems, vol. 32, no. 7, pp. 743-750, 2010.[14] Z. Xu, Z. Y. Dong, W. Q. Liu, and S. Lucia,“Neural network models for electricity,” 1987.[15] M. M. Yi, K. S. Linn, and M. Kyaw,“Implementation of Neural Network Based Electricity Load Forecasting,” Engineering and Technology, pp. 381-386, 2008.[16] G. Zhang, B. E. Patuwo, and M. Y. Hu, “Forecasting with artificial neural networks : The state of the art,” International Journal of Forecasting, vol. 14, pp. 35-62, 1998.[17] E. Banda K. A. Folly, Short Term Load Forecasting Using Artificial Neural Network, IEEE, POER Tech, 2007[18] Z. Xu , Z. Y. Dong , W. Q. Liu,Neural Network Models For Electricity Market Forecasting[19] Z.Y. Dong X. Li Z. Xu K. L. Teo, Weather Dependent Electricity Market Forecasting with Neural Networks, wavelet and Data Mining Techniques[20] A.padmaja, V.S.vakula, T.Padmavathi and S.V.Padmavathi, “Small Signal Stability Analysis Using Fuzzy Controller And Artificial Neural Network Stabilizer” International Journal of Electrical Engineering & Technology (IJEET), Volume 1, Issue 1, 2010, pp. 47 - 70, Published by IAEME[21] Soumyadip Jana, Sudipta Nath and Aritra Dasgupta, “Transmission Line Fault Classification Based On Wavelet Entropy And Neural Network” International Journal of Electrical Engineering & Technology (IJEET), Volume 3, Issue 2, 2012, pp. 94 - 102, Published by IAEME 98
  9. 9. International Journal of Electrical Engineering and Technology (IJEET), ISSN 0976 –6545(Print), ISSN 0976 – 6553(Online) Volume 4, Issue 1, January- February (2013), © IAEMEWebsites 1. http://mahasldc.in/ Accessed in year 2012 2. http://www.getcogujarat.com/ Accessed in year 2012 3. http://www.imdpune.gov.in/ Accessed in year 2010 4. http://www.kalkitech.com/ Accessed in year 2011 5. http://www.mathworks.com /Accessed in year 2011 6. http://www.sldcguj.com / Accessed in year 2012 7. http://www.wunderground.com/ Accessed in year 2010 8. www.mahasldc.in / Accessed in year 2012Field Visits 1. Metrological department,Pune(India) 2. State load dispatch center,Mumbai(India) 3. State Electrcity Board office,Pune(India)Acronymns 1. AME :Absolute mean error 2. ANN:Artificial neural network 3. ANNSTLF: Artificial neural network based Short term load forecasting 4. EMS : Energy management systems 5. GDP : Gross domestic product 6. GOI : Government of India 7. MAPE: Mean average percentage error 8. NN : Neural Networks 9. NWT: Non-decimated Wavelet Transform 10. STLF : Short-term load forecasting 99