APPLICATION OF ECHO STATE NETWORKS FOR ENERGY CONSUMPTION FORECASTING IN MINING ENTERPRISES

Litvinov Stepan Nikolaevich

This article addresses the problem of forecasting electricity consumption in mining enterprises, where the high energy intensity of processes and complex data dynamics require more accurate analytical methods. Traditional approaches, such as regression analysis and ARIMA time series, often prove insufficient for accounting for nonlinear dependencies and seasonal fluctuations. We propose the use of Echo State Networks (ESN) — a machine learning method based on recurrent neural networks that can better handle these complexities. The article discusses key characteristics of mining enterprises, including high energy intensity, seasonality, nonlinear consumption dynamics, and the presence of noise in the data. The primary goal of the work is to improve the energy efficiency of the mining industry by enhancing the accuracy of electricity consumption forecasting using ESN. The results of the study showed an increase in representativeness metrics of 0.1 (from 0.72 to 0.82) and a decrease in mean absolute error from 5.2% to 3.8% compared to ARIMA.

Keywords:echo state network, forecasting, energy consumption, mining enterprises, recurrent neural networks