Parametric Model Based Approach for Consumer Load Prediction

Abstract

Various load prediction techniques have been proposed to predict consumer load which represents the activities of the consumer on the distribution network. Usually, these techniques use cumulative energy consumption data of the consumers connected to the power network to predict consumer load. However, this data fails to reveal and monitor the activities of individual consumers represented by consumer load consumption pattern. A new approach of predicting individual consumer load based on autoregressive moving average model (ARMA) is proposed in this study. Sub- optimal technique of parameter estimation based on Prony method was used to determine the model order of the ARMA models ARMA (10, 8), ARMA (8, 6) and ARMA (6, 4).  ARMA (6, 4) was found to be appropriate for consumer load prediction with an average mean square error of 0.00006986 and 0.0000685 for weekday and weekend loads respectively. The energy consumption data acquired from consumer load prototype for one week, with 288 data points per day used in our previous work, was used and 5-minute step ahead load prediction is achieved. Furthermore, a comparison between autoregressive AR (20) and ARMA (6, 4) was carried out and ARMA (6, 4) was found to be appropriate for consumer load prediction. This facilitates the monitoring of individual consumer activities connected on the power network.