An improved load forecast model using factor analysis: An Australian case study

This paper presents a mathematical model to predict the half-hourly load in one year based on solar energy index, holiday index and weather index. The weather index and the holiday index are taken as the eigenvalues of the Support Vector Machine (SVM), Elman neural network (Elman), and BP artificial neural network (BP neural network), and then the final optimal model is selected through a comparison with the actual data. In the past studies in the literature, weather indicators were often defined as temperature and humidity, and the relationship among various kinds of weather data and weather influence factor on the load were rarely studied. By contrast, in this paper, the model will reduce the forecast error caused by the uncertainty of the weather through the factor analysis of eight weather factors, considering the relationship between each weather data and their influence on the load demand. In addition, the holiday and non-holiday and peak hour and off-peak hour power usage will be dealt with through 0-1 variable definition to get the holiday index. Furthermore, this paper also applies the above-mentioned existing models, and critically evaluates the experimental results of these models to select the optimal one. The historical data used for the experiment are selected from three regions in Australia. © 2017 IEEE.