LSTM-модель потребления тепловой энергии в многоэтажном жилом здании

Системная инженерия и инфокоммуникации. 2025. № 4. С. 11–14.

Ершов И. А.

Translator: Д. В. Кутузов

The heat consumption of residential buildings is a stochastic series. It is necessary for the design of thermal energy regulators the creation of a neural network model. In the paper, the model is carried out based on Long Short-Term Memory (LSTM). The high accuracy of reproducing the series was achieved by training the model on a 2013-2023 dataset from the city of Tomsk. The characteristics of buildings and the outside air temperature were took into account for modeling. A comparison of model dependencies with real commercial accounting data is carried out. The outcomes demonstrate the possibility of design a weather-dependent thermal energy regulator based on machine learning methods. Examples of specific series modeling are given that characterized by a sharp data change and a complete lack of trend, which is typical for modern weather conditions.

Keywords: машинное обучение, linear regulator of a stochastic system