Grade leakage detection in actual heating networks based on CUSUM method and BP neural network

To address the current difficulties and low efficiency in detecting leakage faults in heating networks, this paper proposes a grade leakage fault detection system for pipe networks that integrates CUSUM (cumulative sum) and BP neural network (BPNN). Firstly, the system employs the CUSUM method (first level) to detect the make-up water flow in the heating network and determine whether a leakage exists. If a leakage is detected, the BPNN (second level) is then utilized to precisely locate the leakage position. Taking the actual heating network in a mining area as the research object, and combining its operational data during the heating period with simulation data, data processing is carried out using PCA (principal component analysis) and data normalization methods to construct and train a BPNN model for detecting leakage positions in the actual heating network. Finally, a CUSUM-BPNN grade leakage fault detection system for the heating network in this mining area is developed. By using on-site blowdown valves of the supply and return water pipes to simulate leakages, the system is tested on three heat exchange stations and their corresponding heating networks. The results show that the system can accurately identify leakage faults and rapidly locate the pipe sections where the leakages occur, with leakage alarm delays within 2 minutes. There are few cases of undetected faults or false alarms, verifying the reliability and efficiency of the system developed in this paper.