Forecasting Electricity Prices Using an Improved Time Series Ensemble Model: A Case Study of the Indian Electricity Market

Abstract

Forecasting of electricity prices is a need to the actors in the electricity markets which happen to be deregulated. This paper illustrates a new scheme by providing an upgraded time series ensemble forecasting algorithm on Indian electricity market. The manner in which it is executed has the pre-processing of the data to address the missing values, the stationarization of the variance, season and non-stationarization. We considered six baseline time series models, namely the Autoregressive (AR), Autoregressive Integrated Moving Average (ARIMA), Exponential Smoothing Model (ESM), Theta, Nonparametric Autoregressive (NAR), and Neural Network Autoregressive (NNA) models. Equal weighting techniques used to develop three ensemble models are in-sample error weighting scheme and out-of-sample error weighting technique. The Ensemble of three samples out of the sample produces the best outcome that is measured with the use of MAE, MAPE, RMSE and others. The given framework is highly accurate and it can be applied to all other energy markets.