The main objective of this study is to develop a method for assessing the operating modes of central heating stations (CHS) based on forecasting the thermal power and analyzing the deviations of actual parameters from the reference values. To achieve this goal, the following tasks were solved: primary processing and cleaning of data on technological parameters of CHS and meteorological observations for the period 2022-2024 were performed; regression models (Random Forest, XGBoost, CatBoost and a fully connected neural network) were trained and compared for forecasting the thermal power; an algorithm for assessing and forecasting the operating modes of CHS was developed. The most important results are: the fully connected neural network model, which achieved the determination coefficient R² = 0.9997 and the mean absolute error MAE = 0.0099 Gcal/h; An algorithm developed based on the resulting model allows for differentiating deviations in central heating system operating modes by their source—from the heat supplier, from the consumer, and hidden abnormal conditions not detected by direct comparison with the temperature chart. The analysis revealed that a sharp increase in outside air temperature leads to an increase in the direct network water pressure due to the specifics of quantitative weather-dependent regulation, which can cause accelerated wear of heating networks. The significance of the obtained results lies in the fact that the proposed algorithm ensures the early detection of emergency and preemergency conditions, which helps reduce accident rates, reduce repair and maintenance costs, and improve the energy efficiency of district heating systems. The method can be adapted to various regions with centralized heating.