Short-term Load Forecasting using Support Vector Machines Augmented by an Improved Bacterial Foraging Algorithm

In this paper, the short-term load forecasting problem is addressed using a Support Vector Machine (SVM) Model. In order to optimize the parameters of the model, an improved Dynamic Bacterial Foraging Algorithm (DBFA) is utilized. Load forecasting is a very significant task for the power system operation and planning. An accurate prediction of loading has a great impact on the system’s stability, reliability, economic dispatch and other operational aspects. Tuning the parameters of the SVM model is a key factor for the algorithm to converge since it is based on gradient searches. Specifically, Kernel function and penalty factor of the basic SVM model are highly dependent on prior experience. In this perspective, the DBFA was applied to optimize the model’s parameter selection procedure using historical data as training sets for the SVM model. The proposed model was validated using applicable historical load data of a total of 30 days as the training and testing samples. A comparison of the results with those obtained by the basic SVM model was performed to administrate the effectiveness and superiority of the model.