Leakage diagnosis of heating pipe-network based on BP neural network

Abstract

Listen

Translate article

The leakage of heating pipe-network can lead to serious consequences: the heating quality unable to meet the needs of users, increasing the energy consumption of the heating system and so on. In order to improve the accuracy of leakage diagnosis (LD), this paper regards the LD of the heating pipe-network as a pattern recognition problem relying on the intelligent heating experimental pipe-network system of the hydraulic balance laboratory in Shandong Jianzhu University. Based on the experiment datasets, the model datasets and their cross datasets of the experimental pipe-network, we construct and train a BP (Back Propagation) heating pipe-network leakage diagnosis model (HPLDM) which is used to LD towards the single heat-source branch pipe-network and the double heat-sources with double loops pipe-network. For the single heat-source branch pipe-network, the prediction accuracy of the HPLDM built with the model datasets is 89.31%, 98.51% with the experiment datasets, and 99.70% with the cross datasets. For the double heat-sources with double loops pipe-network, the prediction accuracy of the HPLDM built with the model datasets is 100%, 97.03% with the experiment datasets, and 97.20% with the cross datasets. The experimental results show that the HPLDM based on BP neural network has a higher identification accuracy on the diagnosis of not only leakage location, but also leakage degree of the heating pipe-network. Furthermore, the prediction effectiveness of leakage location is better than that of leakage degree. Meanwhile, the HPLDM has strong generalization ability and some reference significance for the LD of other fluid pipe-network.