A Time-Frequency Collaborative Linear Model Heat Load Prediction Algorithm

The smart heat system is one of the important infrastructures of the city. It is of great significance for energy efficiency, environmental protection, and residents' comfort. Heat load prediction becomes crucial as the core task of smart heating system. However, current heat load prediction models only focus on the time domain information ignoring the information from the frequency domain. Therefore, this study proposes the time-frequency cooperative linear model (TFCoLinear) for heat load prediction. This model can better capture the complex features of the data by combining the time-domain and frequency-domain analyses to improve the accuracy of the model for long-term prediction, and the use of a simple linear model as the basis of the model has relatively few parameters, which reduces the computational cost. In this paper, detailed comparative analysis with SOTA models such as DLinear, RNN, CNN-LSTM, and Seq2seq is carried out by using the actual operational datasets of three heat exchange stations. The results show that the heat load prediction models based on time-frequency hybrid decomposition linear models proposed in this paper exhibit higher accuracy and generalization ability, and their mean absolute percentage errors (MAPE) are all less than 3.60%.