A Leakage Detection Method for Heating-Network Based on Improved Random Forest

Accurately detecting of heating-network leakage was of vital importance. In this paper, a leakage detection method based on improved random forest was proposed. In this method, to address the lack problem of actual leakage data, a hydraulic simulation model was constructed. Then a comprehensive leakage expert database was constructed with simulation data. Based on the above expert database, four algorithms (including normalization, principal component analysis, standardization and regularization) were adopted for data pre-processing. Finally, an improved random forest model, was trained and tested with the processed expert data-base. One urban heating-network was selected as a study case. The results demonstrated that the improved random forest achieved detection accuracies exceeding 90% across 9 data sample proportions (training set: testing set). Moreover, when detecting leakage locations and degrees, the precision, recall, and F1 score consistently exceeded 90%. In addition, the leakage detection method proposed in this paper exhibited higher accuracy in detecting heating-network leakage compared to traditional detection methods based on general random forest and support vector machine.