A Heuristic Method for Short Term Load Forecasting Using Historical Data D.V.Rajan ? , C.Saravanan ? , S.S.Thakur ? A Abstract-Load forecasting plays an important role in power system planning and operation. In the present complex power system network under deregulated regime, power generating companies must be able to forecast their system demand and the corresponding price in order to make appropriate market decisions. Therefore, load forecasting, specially the short-term load forecasting (STLF) plays an important role for energy efficient and reliable operation of a power system. It provides input data for many operational functions of power systems such as unit commitment, economic dispatch, and optimal power flow and security assessment. This paper proposes a new and simple technique to calculate short term load forecasting using historical data and applied it to the Damodar Valley Corporation (DVC) grid operating under Eastern Grid (ERLDC-Eastern Regional Load Despatch Centre), India. This gives load forecasts half an hour in advance. The forecast error i.e. difference between calculated forecast load and real time load is a measure of the accuracy of the system, is found to be lower than other existing techniques like Holt's Method, Chow's Adaptive Control Method, Brown's One-Parameter Adaptive Method. n the present day, there are many issues and challenges in the deregulated electric power industry worldwide. The Indian power sector is undergoing structural metamorphosis and the various power generations, transmission and distribution companies are getting ready to take their rightful place in this sector to offer efficient service for which load forecasting is an effective tool. # Keywords Load forecasts with lead times from a few hours to seven days are essential in certain scheduling functions such as unit commitment and interchange evaluation. A wide variety of modeling techniques for STLF have been suggested in the literature in ref. 1-6. Conventional load forecasting methods like regression method in ref. 7 The short term load forecast plays an important role in economic operation and reliability of power systems. The main objective of the STLF is to advise the dispatcher in making a decision for economic dispatching. Therefore with an accurate model, it could also benefit dispatch systems to: supply load consistently estimate fuel allocation determine operational constraints determine equipment limitations The second objective of the STLF is security assessment and system updation. STLF system requires offline historical data to do predictions. The data helps to run the model in advance, therefore allows dispatcher to provide corrective counter measure to the system. In the proposed technique, historical load data obtained from DVC from the year 2010 to 2011 was used. The inputs used for the proposed method are, load at the particular time in the previous year, and two readings at half hour intervals of the same year along with the load of the half hour intervals in the present year. A mean absolute percentage error of 0.05% was achieved over the period of data which was tested on 1 week data. This represents on average a high degree of accuracy in the load forecast. Load forecasting is one of the most important inputs for prediction of electricity prices. The vital initiative behind prediction involves increasing number of models that estimate future values of an indicator based on its past values. Load forecasting can be done for different durations i.e. long term forecasts with lead time of more than one year, medium term forecasts with the lead time of one week to one year, short term forecasts with lead A historical data method has been used in this work to develop a model to make predictions of the load half an hour in advance, based on the relationship of processed data of previous year and data available for the current year. In this paper, the proposed short term load forecast using historical data (STLFHD) method has been tested on DVC load data in which the forecast has been made based on load data at a particular time of the previous year in steps of half hour and one hour and corresponding load data of the current year. An assumption has been made that the environment factor of power production system is same on the present day and the same day in the last year. Also, the two real time loads in thirty minutes difference is included in the calculation to make the forecast value more accurate. The equations devised for load forecasting using historical data are as given below: Mean Absolute Deviation (MAD) is the final accuracy measurement. This error measurement is the average of the absolute value of the error without regard to whether the error was an over estimate or underestimate. # Eqn (5) # Where = actual load at particular time instant and = forecast load at that time. The proposed technique was tested on historical data from the period 2010 to 2011. The data of first week February 2011 has been considered here for discussion and plotted graphs show for better understanding. Table 1 shows morning peak and Table 2 shows evening peak of Load data & load forecast data respectively. Where FL represents Forecast Load and HD represents Historical data Also, the results obtained clearly demonstrate that the proposed method is simple, fast, reliable, accurate, and effective for short term load forecasting and that this method can perform good prediction with least error. The results obtained in this work confirm the applicability as well as the efficiency of the proposed method in short-term load forecasting for the DVC grid load pattern located in eastern part of India. The method applied was able to determine the nonlinear relationship that exists between the historical load data supplied and on that basis, to make a prediction of what the load would be in the next half an hour. The forecasting reliability of the proposed method was evaluated by computing the mean absolute error between the real time load and forecasted load. The results have shown that the prediction is more accurate with least error. Finally, we concluded this technique is simple and fast and could be an important tool for short term load forecasting for inter connected grid systems. ![-HM-Holt's Method, CACM-Chow's Adaptive Control Method, BOPAM-Brown's One-Parameter Adaptive Method, RTL-Real time load Mean Absolute Percentage Error (MAPE), Short Term Load Forecasting (STLF).](image-2.png "") ![I time of 1 to 168 hours and very short-term load forecasting with lead time shorter than a day. STLF is a dynamic nonlinear input/output mapping function of many variables such as weather conditions, temperature etc. Auto regressive models and moving average mapping are well known examples that come under linear autoregressive models. The various tools available for load forecasting are Artificial Neural network.(ANN) Fuzzy logic (FL) Autoregressive model Similar day approach Time series Expert system Support vector machine Out of these methods, ANN and FL are the popular and commonly used mathematical tools in ref. 12-15 for load forecasting applications. The traditional Approaches in ref. 16 like Holt's Method, Chow's Adaptive Control Method, Brown's One-Parameter Adaptive Method have been applied on the historical load data of DVC for month of February 2011.](image-3.png "") ![load of current year at required time. =load of the previous year at the same time at which forecast is being done in current year. =load of the previous year half hour before the current forecast time. =load of the previous year one hour before the current forecast time. =load of the current year half hour before the current forecast time. =load of the current year one hour before the current forecast time. =Absolute average value of difference of half hour values. =Absolute average value of difference of hour values. The mean absolute percentage error (MAPE) which indicates the efficiency of the devised model for predicting the load in advance and studying the performance of the system and it does not accentuate large error. Equation (4) illustrates the MAPE formula. MAPE ?...??.eqn (4) Where = Forecast load of current year at required time. =load of the previous year at the same time and N represents the total number of data (time samples=48).](image-4.png "") 1![Figure 1: Comparison of proposed STLFHD method of forecasting with other methods for 3 rd Feb.2011.](image-5.png "Figure 1 :") 2![Figure 2 : Comparison of proposed STLFHD method of forecasting with other methods for 6 th Feb.2011.Journal of Researches in Engineering](image-6.png "Figure 2 :") 23![Figure 2](image-7.png "Figure 2 Fig. 3 :") ![](image-8.png "") 1TimeYear01.02.1105.02.1107.02.11Load (MW)Load (MW)2011Load (MW)600F FL148718372049.5 2TimeYear02.02.1104.02.1106.02.11Load (MW)Load (MW)2011Load (MW)FL1504174419611800HD150517461964FL1515.517541968.51830HD151617591967FL15031750.51977.51900HD150117431976FL1506174919781930HD150817481978FL1504.5176119862000HD150317641988FL1489.51805.519822030HD148918061981FL1511.5186419942100HD151318691997 3MethodsDateHMCACMBOPAMSTLFHD(ProposedMethod)01.02.110.8537710.0851330.6498260.08875802.02.111.8234550.099010.7214140.095503.02.112.1813190.3815171.2651120.05427904.02.111.6352350.1353380.8243620.20870405.02.111.6870370.6618411.6440670.13402706.02.110.6823340.2240320.6640180.1601207.02.111.1433430.34510.9067550.18286 411Volume XI Issue v v v v VII Version IDateHMMethods CACMBOPAMProposed MethodResearches in Engineering01.02.11 02.02.11 03.02.11 04.02.11 05.02.110.673998 1.122309 1.588924 1.522361 1.184650.120297 0.178835 0.847906 0.260466 1.6338520.404542 0.516869 1.126689 1.014949 1.571948STLFHD 0.08681 0.09902 0.045264 0.284057 0.171404Global Journal of06.02.110.5434460.449890.714490.14484107.02.111.3066220.7283711.1892240.215153 ( J ) 2011 December ( J ) 2011 December December The authors acknowledge Damodar Valley Corporation for their constant support and cooperation in providing real time data to bring this work in this form. ## Global ( J ) * Onehour-ahead load forecasting using neural network, IEEE transaction of power system TSenjyu HTakara KUezato TFunabasi IEEE Trans. on Power Systems 2. T. Haida, S. Muto 17 1 2002. 1994 Regression based peak load forecasting using a transformation technique * Fuzzy logic for short-term load forecasting DKRanaweera NFHubele GGKarady Elect. Power Energy Syst 18 4 1996 * Fuzz-ified neural network approach for load for load fore casting problems DKChaturvedi PSSatsangi PKKalra International Journal on Engineering Intelli-gent Systems 9 1 March 2001 CRL Publishing * Hybrid adaptive techniques for electic-load forecasting using ANN and ARIMA AAEidesouky MMEikateb IEE Proceedings-Generation, Transmission and Distribution 147 2000 * Short term load forecasting using fuzzy neural networks AGBakirtzis JBTheocharis SJKiartzis KJSatsios IEEE Transactions on Power Systems 10 1995. 1995 IEEE Transactions on Power Systems * A generalized knowledge-based short-term load forecasting techniques SRahman OHazim Computing-A Fusion of Foundations, Methodologies and Applications Heidelberg Springer-Verlag 1993. April 2004 8 * Short term load forecasting GGross FDGaliana Proceedings of the IEEE 75 1212 December 1987 * Wavelet transfom of meteorological parameters and gravity waves ZCan ZAslan OOguz AHSiddiqi Annals Geophysics 23 2005 * Identification of ARMAX model for short term load forecasting: an evolutionary programming approach HTYang CMHuang CLHuang IEEE Trans. on Power Systems 11 1 1996 * Analysis and evaluation of five short-term load forecasting techniques IMoghram SRahman IEEE Transactions on Power Systems 4 October 1989 * A novel approach to short-term load forecasing using fuzzy neuralnetworks SEPapadaki JBTheocharis SJKiartzis AGBakirtzis IEEE Transactions on Power Systems 13 1998 * A Neural Network Short Term Load Forecasting Model for Greek Power Systems AGBakirtzis VPetridis SJKiartzis MCAlexiadis AHMaissis IEEE Transactions on Power Systems 11 1996 * LJKrajewski LRRitzman 1993 * Operations Management, Strategy and Analysis Addision-Weslev Publishing Company, Inc 436 * SMakridakis SCWheelwright VEMcgee Forecasting:Methods and Applications NewYork John Wiley & Sons 1983 86 * Improved generalized neuron model for short term load forecasting DKChaturvedi MMohan RKSingh PKKalra International Journal on Soft