Electrical load forecasting using support vector machines: A case study

In this study, an application with electrical load forecasting is made by a machine learning method that has recently become popular: Support Vector Machines (SVM). Load forecasting with SVM can model the nonlinear relation with the factors that affect the load in addition to the accurate modelling of the load curve at the weekends and on important calendar days. The data gathered from the Istanbul European Side are used as a sample for the application. In addition to the past load data, daily average temperature, calendar days, holidays and electricity price are considered as an attribute in forecasting. The program Libsvm is used for modelling the system. The results are compared with the Artificial Neural Network (ANN) and real values. In addition to that, another data set is constructed with the same values but without average daily temperatures to observe the effect of weather conditions in such an electrical load forecasting application. It is concluded that the Support Vector Machine algorithm is superior in both data sets to Artificial Neural Networks and is suitable for electrical load forecasting applications.