Medium term electric load forecasting using Lancsoz Bidiagonalization with singular value decomposition

Abstract

The term forecast stands for predictions of future events and conditions. The process of making such predictions is called forecasting. The main purpose of forecasting is to meet future requirements, reduce unexpected cost and provide a potential input to decision making regarding electrical power production and dispatch. In operating a power system, the mission of the utility/company, from the forecasting point of view, is to match demand for electric energy with available supply. This leads to the fact that a major objective of any power company is accurately predicting future loads. In this research, medium term electrical load forecasting for Peshawar region is studied using Lanczos Bidiagonalization with Singular Value Decomposition. Here, electrical hourly loads are processed in three steps. A polynomial fit is performed to access the non-linear trend of the hourly loads of each year. This is followed by applying the SVD method to the difference between the hourly loads and their trend. SVD serves to extract both the cyclic and the random components of the numerical data. Finally, prediction is done using matrix completion via Lanczos Bidiagonalization.