Аннотация
This paper presents a comprehensive solution for predictive maintenance of urban heating systems based on modern machine learning algorithms. The primary objective of the study is the early detection of potential equipment failures, thereby increasing network reliability and reducing unplanned downtime. To simulate operating conditions, a synthetic dataset was created, including temperature, pressure, flow, and vibration parameters collected at regular intervals, simulating real-world monitoring conditions. Data preprocessing included removing outliers, normalizing parameter ranges, and repairing gaps, ensuring sample consistency and quality. To enhance model interpretability, feature significance assessment was performed using the SHapley Additive exPlanations (SHAP) method, enabling the identification of key factors influencing system condition. The study considered several machine learning algorithms, including logistic regression, support vector machines, random forests, artificial neural networks, and gradient boosting. Their effectiveness was assessed using cross-validation. Gradient boosting demonstrated the most stable and reliable results, confirming its ability to model complex nonlinear relationships and detect anomalies in real time. Logistic regression and random forest also demonstrated high classification performance, while support vector machines and neural networks were less effective on this task. These results confirm that ensemble machine learning methods supported by big data processing technologies are a promising tool for building predictive maintenance systems in heat supply networks. The proposed methodology can be integrated into modern monitoring platforms, enabling proactive maintenance strategies, more efficient resource planning, and reduced operational risks. Future research should focus on validating the proposed methodology using real-world production data and exploring hybrid frameworks that combine ensemble learning methods with neural networks to further enhance the robustness, adaptability, and practical applicability of predictive systems.