Comparative analysis of methods of short-term electric load forecasting one step forward

Abstract

ANFIS ARIMA «Naive» forecast Short-term TLF Forecasting methodsShort-term load forecasting (LF) for industrial enterprises is important and complex theme in the electricity, and so, it is of growing scientific interest. Precision short-term LF is important for proper planning and operation of power systems.This article investigates statistical methods and models of artificial intelligence for forecasting industrial enterprises one step forward. Literature review showed that the most popular methods of short-term LF one step forward are: autoregressive integrated moving average model (ARIMA), naive forecast and adaptive neuro-fuzzy inference system (ANFIS), which includes the following models: Grid Partitioning, Subtractive Clustering, FCM Clustering. Therefore, these methods were chosen for further modeling of LF. For ANFIS into account external factors: day factor (if a weekday, then 1, if a holiday, then 0), time of day, day of the week.An industrial enterprise with the manufacture of plastic products was the object of study where measurements vere provided every day for half an hour (measurements every 48 hours), from January 11, 2015 to June 11, 2015, taking into account holidays and weekends. This study used mean square error (RMSE) as standard measurements to verify the accuracy of forecasting. Software used for research was MATLAB 2020b, with toolbox: Fuzzy Logic Toolbox and Econometrics Toolbox.Step by step construction of the PEN models one step forward for these methods and models was developed. As a result, modeling ARIMA(2,1,2) was superior to other models with the least errors RMSE, training and test 0.0317 and 0.0354 respectively. The results showed that one-step prediction was the most effective model ARIMA. The prospect of further research is to develop models of these methods for multi-step forecasting and comparison with each other.