Investigation of Two-Stage Load Forecasting to Minimize Error Residuals

Abstract

This paper discusses preliminary results obtained using a two-stage forecasting method for PJM Interconnection data sets. In stage 1, autoregressive integrated moving average (ARIMA) was deployed, and in stage 2, residuals from stage 1 was fed as an input to exponential smoothing (ES) method. The data contained demand values from 11 regions of PJM Interconnection. The datasets used to predict day-ahead demand values are both in 24-hour and 30-day format for 2016 calendar year. The accuracy of forecasting is evaluated using Mean Absolute Percentage Error (MAPE) and Mean Absolute Deviation (MAD) parameters. An economic dispatch was then carried using a linear programming formulation in Algebraic Mathematical Programming Language (AMPL) environment. The preliminary results indicate two stage process of ARIMA with ES and ARIMA with ARIMA outperforms one-stage application for this data set.