Energy Consumption Forecasting Based on Big Data Technologies

Forecasting energy consumption is one of the most important tasks for many companies. Accurate prediction allows to reduce costs and it influences the technological processes of an organization. In this paper, the essential approaches to model building for predicting future energy consumption were explored. The historical data about electricity consumption for 2016, 2017 in the metallurgical company were used for a model creation. The following algorithms for prediction were investigated: ARIMA, Xgboost, Random Forest, LSTM, Regularized linear regression. A comparative analysis of these methods showed that Xgboost, LSTM, Regularized linear regression demonstrated a satisfactory accuracy. The ensemble method based on these models was proposed. The results showed improvement in the generalization ability as well as alleviation of several unstable-prediction problems. The key idea of the described approach is to take the geometric mean of the methods involved into the ensemble. The proposed model allowed to improve accuracy and achieve a reduction of the lost profits by five million rubles annually.