Electrical & Computer Engineering Technical Reports

Document Type

Technical Report

Publication Date

2024

Abstract

In Electric load prediction (ELP), we predict the electricity demand at aggregated levels which is vital for the proper functioning of smart grid (SG) and keeping a balance between load and supply demand. ELP can be categorized as very short-term, short-term, medium-term, and long-term ELP. In this report, we propose support vector regression for hourly day-ahead short-term ELP. In this report, we explain the concepts of actual risk, empirical risk, structural risk, complexity, and over-fitting. We present the concept of Vapnik Chervonenkis (VC) dimension and interpret the VC theorem that describes the bound on the actual risk. We present the concept of support vector machine (SVM) and its mathematical background and discuss the support vector machine criteria and develop the analysis that leads to the dual solution of the SVM and its main results. Then, we discuss about support vector regression (SVR). We present different classification performance metrics, regression performance metrics, and normalization techniques with comprehensive details. Mean absolute percentage error is used as a prediction accuracy metric. We implement SVR to predict 24 electric load values of the next day based on load, temperature, and dew point values of previous days. We evaluated the performance of SVR and compared it with the persistence and multiple linear regression methods. The results show that SVR outperforms these methods.

Download

Share

COinS