A Novel PCA_ERF Method for Leakage Detection of District Heating System

Abstract

The leakage of district heating system can lead to serious consequences. Therefore, the leakage detection of the district heating system has always been the focus of research in various industries. Relying on the intelligent heating experimental pipe-network system in Shandong Jianzhu University, this paper takes 4 topological heating experimental pipe-networks as the research objects, constructs the real-time operation datasets, simulation datasets and the cross datasets of the above two, creatively proposes a PCA-ERF (Principal Component Analysis-Extremely Random Forest) based method for the leakage detection task. The method adopts PCA to map the original pressure and flow data of the heating network into the vector space, which has a stronger feature expression ability firstly; then the decision trees for classification are trained by ERF with stronger randomness; finally, the final classification results are obtained by integrating the judgment of all the decision trees. The experimental results show that the PCA_ERF method shows excellent performance under different cross-data ratios, especially when the cross-data ratio is 2:1, the accuracy of the proposed PCA-ERF method in the leakage prediction for 4 different topologies is 98.08%, 97.1%, 98.92% and 97.64% respectively, which can complete the leakage detection task of complex heating network with multiple topologies.