Least Distance Predictor Model for Short Term Load Forecasting

Abstract

One of the foremost issues concerning the stability of power system around the world is regulation of frequency. A balance between supply and demand maintains the frequency constant or within a permissible range. In India, this balance is regulated by imposing Charges of Deviation on power utilities like distribution and generation companies that deviate from their scheduled transactions of energy. This charge is dependent on the system conditions and varies inversely with the system frequency. Imposition of these charges on the participants helps maintain grid discipline, increase grid efficiency, and make the participants more responsible and accountable. The main objective of this study is to identify a constructive approach to reduce the unscheduled energy transactions thereby reducing any deviations from the schedule of a distribution utility. It is proposed that this could be realized in real time through accurate short term forecasting of load demand. Two year (Jun 2011- May 2013) past data of daily load demand of Uttar Gujarat Vij Company Ltd. (UGVCL), a distribution utility in the State of Gujarat is used as a case example. In this thesis, Seasonal ARIMA model is taken as a base model for short term load forecasting. It is clearly seen from the results that the model is not able to capture the characteristics of particular group of days. To overcome this, a least distance predictor model is developed to forecast the daily load of the distribution utility. It results in better load characterization and improved forecast accuracy compared to the similar shape predictor model. This is achieved without using massive amounts of training data, thereby reducing time of execution.