Forecasting the Short-Term Electricity In Steel Manufacturing For Purchase Accuracy on Day-Ahead Market

Abstract

Forecasting electricity consumption in the most accurate way is crucial for purchase on the day-ahead market in steel manufacturing. This study is aimed to predict short-term electricity consumption regarding the day-ahead market purchase by employing important features of electricity consumption time-series data. We utilize Random Forest (RF), Gradient-Boosted Trees (GBT), and Generalized Linear Models (GLM), as they are appropriate for the given problem and widely used regression algorithms for prediction purposes. This study leverages the regression algorithms in the Apache Spark Machine Learning library. The performance of the prediction models is evaluated based on the standard deviation of the residuals (RMSE) and the proportion of variance explained (R-squared). We additionally discuss the distribution of prediction errors of the models. Experiments show that the RF model outperforms the GBT and GLM. It is considered that the results can contribute to accurate forecasting of short-term electricity demand for purchasing on the day-ahead.