A Generalized Approach for Short-Term Load Forecasting Using Artificial Neural Networks

Dr. Hafiz Abdur Rahman and his Research Team (K. M. Jawadur Rahman, Imran Uddin, , M Maksud Alam)

Abstract—Short-term electrical load forecasting is critical for smooth operation and cost effective management of power systems. Several works have been done to address the load forecasting problem using a variety of approaches and techniques, especially artificial neural networks (ANN). However, for ANN models, very little work has been done to generalize the model development process. A generalized approach for short-term load forecasting using ANN has been presented in this work. This generalized approach is applied to load dataset from five different countries and the statistical summaries of the performance of all five forecasting results are presented. The first stage of the workflow enhances the feature selection process. Socio-economic and meteorological factors affect the demand patterns of electrical load over a period of time. Since these factors vary from one country to another, the features of the forecasting model need to be customized according to the dataset that we are working with. To generalize this step, we have first trained a self-organizing map (SOM) using unsupervised learning. The responses of the map to different inputs are then used to select some of the features of the model. Our approach produces highly accurate load forecasting models across multiple datasets.