Short term load forecasting using artificial neural networks

Abstract

Short term load forecasting of load demand is an essential step in planning and operating an electric power system. The purpose of this project is to get familiar with load forecasting and with Artificial Neural Network (ANN) technique used for short term load forecasting, and build our own forecasting model using ANN. This project presents an overview of the types of load forecasting and its uses, the factors that affect a load forecast and the framework of short term load forecast. Next, ANN is taken into a more detailed fashion. Finally, a model is built using MATLAB and tested for the city of Houston, Texas. Historical load data and historical weather data were obtained from ERCOT website and wunderground website, respectively. The forecasts were made for the first week of March 2015. The inputs used were Hourly temperature, Minimum temperature of the day, Maximum temperature of the day, Average temperature of the day, Minimum humidity of the day, Maximum humidity of the day, Average humidity of the day, Average wind speed of the day, Hours, Is working day, Last week same day hourly load, Previous day hourly load, Previous day 24 hour average. The number of hidden layer neurons was chosen to be 17. The output layer has 1 neuron. The data were trained sufficiently. A mean absolute percentage error of 0.04% was detected for the whole model, showing a high level of accuracy.