Short Term Load Forecasting Using PSO-DWT-MLR at System and End-User Levels

Abstract

The electric loads of both power system and power consumers have non-stationary and uncertain characteristics that lead to difficulties in constructing an adequate model to accurately predict the load variations. This paper proposes a novel prediction model of short term load forecasting (STLF) for both system load and aggregated load of power consumers customers. The prediction method uses a particle swarm optimization based discrete wavelet transformation in multiple linear regression model (PSO-DWT-MLR) to capture the non-linear relationship between the load demand and the exogenous inputs. PSO was used to select the optimal combination of details and approximations data from DWT to construct an MLR model. Associated with actual weather information, validation of the proposed model is conducted in both system-side data set and end-user data set in Independent System Operator-New England (ISO-NE) and aggregated load data respectively. The results demonstrate that PSO-DWT can boost the performance of MLR for prediction of nonstationary load conditions and can provide more accurate prediction than existing methods.