An intelligent water supply pipeline leakage detection method based on SV-WTBSVM

Water supply pipeline leakage not only wastes resources but also causes dangerous accidents. Therefore, detecting the state of pipelines is a critical task. With the expansion of the scale of the water supply pipeline, the amount of data collected by the leak detection system is gradually increasing. Moreover, there is an imbalance of sample in the data. This makes the detection performance of traditional leakage detection methods deteriorate. To solve the above issues, this paper proposes a pipeline leakage intelligent detection method based on a support vector weighted twin-bound support vector machine (SV-WTBSVM). Noise in the data negatively affects the performance of the classifier. To eliminate the effect of noise, a hybrid denoising algorithm based on improved complete ensemble empirical mode decomposition with adaptive noise (ICEEMDAN) is used for denoising to filter out the noise in the data. Twin bound support vector machine (TBSVM) is a classical classification algorithm that has been widely used in the detection of pipeline leakage. To solve the decrease in classification accuracy caused by sample imbalance, the SV-WTBSVM algorithm oversamples the minority class samples based on the distance density and integrally undersamples the majority class samples to obtain a balanced sample. Since pipelines often have multiple working conditions, the SV-WTBSVM used for binary classification cannot meet this requirement, and this paper combines the SV-WTBSVM with the 'one-to-one' strategy to address the multi-classification problem. Finally, experiments have verified that the SV-WTBSVM algorithm not only retains the advantages of fast training speed and simple operation of the TBSVM but also improves the classification accuracy and generalization ability of the algorithm when dealing with imbalanced data.