Medium-Term Electricity Consumption Forecasting in Algeria Based on Clustering, Deep Learning and Bayesian Optimization Methods

Abstract

Listen

Translate article

Forecasting electricity consumption of aggregate or individual consumers is a challenge of production and distribution electricity enterprise to manage their electricity demand and reducing electricity loss. In this context, we investigate the problem of improving the accuracy of forecasting electricity consumption of the economic sector in Algeria. This paper presents a medium-term electricity consumption forecasting method based on a combination of deep learning and clustering techniques. The proposed approach aims to effectively extract the similarity of consumers’ consumption; and performing forecasting accurately at the reduced level. In the first step, two clustering methods, namely K-means and K-Shape, are used to extract similarities among consumers. Then, a deep learning model, based on Gated Recurrent Units, for each cluster with a Bayesian optimizer is employed to extract patterns of the consumers’ electricity consumption. To validate the proposed method, we compared our results to two enterprise classifications: Activity Sector and Maximum Power Demand. Several experiments have been conducted with almost 2000 clients and 14 years of monthly electricity consumption from Bejaia, Algeria. The results show that K-Shape reaches much higher prediction accuracy. The best results in global prediction obtained a MAPE equals to 6.57% and the grouping customer on Activity Sector a MAPE equals to 15.52% in individual.