Research on data-driven identification and prediction of heat response time of urban centralized heating system

Abstract Heat response time (HRT) is one of the key dynamic response characteristics of urban centralized heating system (UCHS). HRT is also critical to the operation dispatch of large-scale UCHS. This study proposes a data-driven approach to identify and predict HRT. Due to the diversity of operational data, this study applies correlation analysis and feature fusion (both linear and nonlinear) to generate feature sets. This study further develops the prediction models of HRT made up of three different feature sets and four machine learning models: Linear Regression (LR), Support Vector Machine Regression (SVR), Random Forest (RF), and Extreme Gradient Boosting (XGBoost). A UCHS in Zhengzhou is selected as the demo site to show the effectiveness of this approach. HRT of investigated substations range from 1000 to 3000 seconds. The correlation analysis indicates that the heating area’s square root is most relevant to HRT (correlation coefficient close to 0.68) compared to other features. Feature fusion is critical in HRT analysis. The performances of all four prediction models are improved with fused features added to sets. XGBoost model outperforms other models in terms of model accuracy. Quantification of HRT could be useful in the optimized operation control of large-scale UCHS.