Modeling of De-Icing Heating System by Machine Learning Methods

The article proposes a methodology for using digital twins for modeling thermal power plants. Building a digital equivalent allows for virtual testing, monitoring technical condition, optimizing performance and reducing maintenance costs and resource consumption. The method is based on neural network predictive architectures. The input and output parameters of the system are considered as time series. The measured characteristics of installations are taken into account, including state characteristics and external factors. Research has been carried out on MLP and LSTM architectures. The best results were obtained for the LSTM network. The method allows you to predict about ten minutes of work from data about twenty minutes of work.