Machine Learning for Forecasting Electricity Consumption at Water-Supply Pumping Stations

This article examines potential opportunities for reducing electricity purchase costs for a water supplier’s pumping stations. The importance of achieving the required accuracy in forecasting electricity consumption when changing electricity price categories is assessed. A classification of days of the week for forecasting electricity consumption is proposed, and an example of a machine learning model is presented. The feasibility of improving forecasting accuracy using the CatBoostRegressor ensemble model is demonstrated. The cost reduction calculation results obtained using the selected model are described and analyzed. A method for achieving maximum reduction in electricity purchase costs achieved with the achieved forecasting quality is proposed.