Gradient boosting machine for predicting return temperature of district heating system: A case study for residential buildings in Tianjin

Abstract

Accurate prediction of the return temperature is critical to energy efficiency of the district heating system (DHS). The support vector machines (SVMs) and artificial neural networks (ANNs) have been widely applied to forecast energy consumption of the DHS recently. However, the parameters of SVM and ANN are difficult to be optimized due to their specific request for inputs and time-consuming training process. To explore the performance of the decision tree-based ensemble algorithms in the return temperature prediction task of the DHS, four return temperature prediction models based on the operational data of a heating system in Tianjin are established, namely Support Vector Regression (SVR), Multilayer perceptron (MLP), Random Forest (RF) and light gradient boosting machine (LGBM). The RF and LGBM are two typical decision tree-based ensemble algorithms. The historical supply temperature, outdoor temperature, relative humidity, wind speed and air quality index (AQI) are used as original inputs of the models. For computational models (SVR, MLP), the input features should be further transformed. The experimental results demonstrate that the LGBM model outperforms others in all standard evaluation measures. It shows that the tree-based ensemble models without complicated feature transformation achieves considerable results. Moreover, a week-based time series data splitting strategy is developed and compared with the traditional method. The experimental results show that the novel method can improve the performance of models except MLP. Overall, the performance of tree-based ensemble algorithms (RF and LGBM), SVR, and MLP are compared based on a case study in this article, illustrating the potential of the tree-based ensemble algorithms in thermal load prediction.