Development of a forecasting model for optimizing energy consumption at coal enterprises

Source abstract

The study object is daily data on electricity consumption of one of the coal mines in the Karaganda basin for 2024. This article solves the problem of the lack of accurate tools that can predict complex and variable modes of energy consumption in a coal mine and thereby ensure more efficient management of energy-intensive installations. This article presents a comparative analysis of three electricity demand forecasting models using data from a coal mine in the Karaganda basin for 2024. The study explores the effectiveness of both classical approaches (seasonal ARIMA model and simple exponential smoothing) and an LSTM neural network model. To handle non-stationary data, the first difference method was applied, allowing the time series to be stationary. The forecast was generated for 7 days in advance. A comparative analysis of the models’ accuracy was conducted using the MAPE metric on both the training and test sets. The study found that the LSTM model demonstrated the best results with a MAPE of 5.37% on the test set demonstrating its superior ability to capture complex data dynamics compared to ARIMA and simple exponential smoothing. The developed predictive LSTM model can be effectively used in automated energy monitoring and management systems, providing accurate short-term load forecasts for coal mines and other mining and metallurgical enterprises with complex and volatile energy structures, provided the initial data is highly reliable and complete