SHORT TERM LOAD FORECASTING USING EMPIRICAL MODE DECOMPOSITION AND ARTIFICIAL NEURAL NETWORK: A CASE STUDY OF KATHMANU VALLEY, NEPAL

Abstract

Short-term Electric Load Forecasting is an important aspect of power system planning and operation for utility companies. Short Term Load Forecasting (STLF) has always been one of the most critical, sensitive and accuracy demanding factors of the power systems. An accurate STLF improves not only the system’s economic viability but also its safety, stability and reliability. The research presented in this work supports the argument of hybrid model based on Artificial Intelligence (AI) and Empirical Mode Decomposition (EMD) techniques in short-term load forecasting. In this research work, a hybrid short term load forecasting model based on EMD and Feed Forward Back Propagation (FFBP) algorithm of artificial neural network was developed. With the application of hybrid model the load demand for Sunday 22, 2076 B.S is forecasted and the result is compared with the actual data by calculating the Mean Absolute Deviation (MAD) and Mean Absolute Percentage Error (MAPE) between the actual and forecasted load. The MAD value of 8.53MW and MAPE of 4.82% is found between the original and forecasted load. The correlation factor analysis between the IMFs, residue and the original load data which is decomposed using EMD technique is carried out and the most informative IMFs are chosen to further forecast the load. The correlation between the original load and the residue is found to be 0.93 while the correlation between the IMFs and the input is very low which less than 0.2. The load is forecasted taking the residue as the input, in this case the MAD is found to be 3.27MW and the MAPE 1.79%.The model is compared with the basic FFBP algorithm of ANN. The output of both the model is compared by calculating the MAD and MAPE between the actual and the forecasted load. The EMD based hybrid model after Correlation Factor Analysis decreases the MAD by 75.95% and MAPE by 80.15% compared to basic FFBP model. For the validation of model it is used to forecast the load of New York Network for July 1, 2004 and the output is compared to other well known hybrid model such as WTNNEA, WGMIPSO and IEEMD-BPNN. The proposed EMD-FFBP model gives the MAPE of 4.34 % which is higher than the MAPE value obtained by other hybrid models, but it is within the limit of acceptance.