Machine Learning in Predicting Heat Loss in Heat Supply Systems

Background. The study presents the application of machine learning methods to predict heat loss in the heat supply system of the city of Kazan based on data for the period 2020-2023. Forecasting heat loss allows you to increase the efficiency of the heat supply system, reduce the cost of heat production and transportation. The paper examines various machine learning models, such as linear regression, ensemble methods (random forest, gradient boosting) and neural networks, and evaluates their accuracy and applicability in the context of urban infrastructure.

Purpose. Improving the accuracy of forecasting heat loss in the Kazan city heat supply system in order to optimize operation and reduce operating costs.

Materials and methods. The study is based on operational data of the heat supply system, weather data, as well as infrastructure characteristics of pipelines. The main research methods include linear regression, regularization methods (Lasso, ridge regression), ensemble learning (random forest and gradient boosting) and multilayer perceptron (MLP). The MSE, MAPE and R2 metrics, as well as cross-validation, were used to evaluate the models.

Results. The analysis showed that machine learning methods, especially gradient boosting and neural networks, make it possible to achieve high accuracy in predicting heat loss (R² = 0,89). The use of these methods helps to increase energy efficiency and reduce operating costs in heat supply systems.