СУЧАСНІ МЕТОДИ ПРОГНОЗУВАННЯ ЕНЕРГОСПОЖИВАННЯ БУДІВЕЛЬ

The relevance of the study is driven by the growing need to enhance the energy efficiency of buildings through thermomodernization and the demand for accurate forecasting tools to ensure effective energy management. In the context of climate change and rising energy prices, the selection of optimal forecasting models has become a critical task for housing and communal services as well as urban infrastructure. The aim of the study is to substantiate contemporary approaches to forecasting building energy consumption considering thermomodernization measures, based on the analysis of regression, neural, and hybrid models, in order to identify their advantages, limitations, and practical effectiveness. Methodology. A comparative analysis of recent studies was conducted, covering statistical, neural network, and hybrid forecasting models. The accuracy, scalability, flexibility, and adaptability of models to post-retrofit conditions were assessed. Particular attention was paid to deep learning architectures (LSTM, GRU), hybrid combinations (ARIMA+LSTM, CNN+ELM), and digital twin technologies. Results. It was established that the highest forecasting accuracy is achieved by neural network models, particularly deep architectures and ensembles, with average errors not exceeding 3–5%. Although regression methods are less accurate, they remain useful for baselineestimates and evaluating the impact of climatic variables. The effectiveness of hybrid approaches that combine trend modeling with neural network-based residual learning was demonstrated. The potential of digital twins as a tool for predictive and adaptive energy management was identified. Scientific novelty. For the first time, forecasting models were systematized in the context of building thermomodernization, with the identification of optimal approaches for different types of tasks. The relevance of applying intelligent forecasting systems integrated with digital twins as a new paradigm in building energy management was substantiated. Conclusions. It was proven that accurate forecasting of energy consumption is only possible when accounting for changes caused by thermomodernization and applying flexible models capable of adapting to new building operation conditions. It was determined that the integration of artificial intelligence methods with physical modeling and digital twinsenhances the accuracy and applicability of forecasts for practical energy management. Prospects for further research. It is advisable to develop combined modeling methods that consider behavioral factors in consumption, and to expand the empirical base for assessing the effectiveness of digital twins in retrofittedbuildings of various types.