A comparative analysis of PSO and LM based NN short term load forecast with exogenous variables for smart power generation

Abstract

Accurate short term load forecasting is essential for reliable operation and several decision making processes of the power system. However, forecast model selection, network training issues and improper input selection of forecast model may significantly decrease the prediction accuracy of forecast model. As a result operational cost and reliability of system affected dramatically. In this paper, particle swarm optimization (PSO) based neural network (NN) forecast model is presented and compared with Levenberg Marquardt (LM) based NN forecast model for 168 hours ahead load forecast case studies. The impact of day type, day of the week, time of day and holidays on load demand are also analyzed. The mean absolute percentage errors (MAPE) and regression analysis of NN training are used to measure the forecast model performance. Moreover, PSONN based forecast model produces higher forecast accuracy for all test case studies with confidence interval of 99%. In this research ISO-New England grid load and respective weather data is used to train and test the forecast model.