Abstract
This paper presents a stochastic load model that uses a regression equation coupled with a time series model. It is developed as a simple model without compromising accuracy. In this model, the fast Fourier transform (FFT) is used to identify multi-seasonality. Based on the FFT results, a 24-hour set of regression equations has been developed which incorporates the hourly temperature variations. Weekly seasonality is handled by providing weekday and weekend nonlinear regression equations. Annual seasonality is incorporated by treating the four seasons separately. The Levenberg-Marquard method is used to estimate the parameters of nonlinear functions. This method is used because of its superiority over the widely used Gauss-Newton and steepest descent methods for estimating model parameters and to avoid "slow down" in the search process. A residual discrete time series is determined by using the autoregressive integrated moving average (ARIMA) model. Test results using PJM-market load data indicate the effectiveness of the proposed model to predict the daily electricity load.