PREDICTIVE CONTROL OF THE CITY HEAT SUPPLY SYSTEM USING LINEAR REGRESSION AND GRADIENT BOOSTING MODEL

Abstract

The paper analyzes the evaluation of the quality of performance of control models of the city heat network. The results of the analysis are a recommendation for choosing the optimal control model in terms of accuracy and resources required for its training. This recommendation will allow to realize an intellectual module for the decision support system. The intellectual module will be used in the realization of the automated control system of the heat network of the city and will allow more economically, in terms of resource consumption to ensure the maintenance of the required temperature regime in the apartment buildings of consumers. Purpose of the research. Selection of a model that will allow to calculate with greater accuracy the value of losses in the heat supply network of the city. Application of such a model will allow to predict the behavior of the heating network and, in accordance with this, to choose the control action. Materials and methods. Linear model, as the simplest in training, and showed high accuracy in predicting the network in established modes, as well as a model based on decision trees, built using the method of gradient bousting. Results. The capabilities of a decision tree-based model trained on the basis of statistical data to predict the value of heat losses in the network taking into account the thermal inertia of the system and predicted values of air temperature have been investigated. The application of such a model, which showed good results in the study, is justified. Performance estimates for the linear model as well as for the model based on decision trees are given. Conclusion. The proposed methods and models are tested on real data, which confirms the possibility of their use in the development of intelligent information system of heat supply management.