Pipeline Leakage Detection Method Based on Convolutional Neural Network

In the early years, the fiber optic sensing detection method was used for pipeline leakage detection. It uses the principle of fiber optic interference to lay fiber optic sensors along the pipeline, transmit the collected vibration signals back to the computer for relevant processing and detect whether there is any leakage in the pipeline. In recent years, the rapid development of artificial intelligence has led researchers to associate it with pipeline detection. Experts have conducted research on adaptive filtering and Kalman filtering of pipeline leakage signals under strong environmental noise. Then they used methods such as empirical mode decomposition and wavelet transform to extract features from the filtered signal. Finally, they used support vector machines to detect pipeline leaks. Experts can also use the pressure, flow rate, and valve data collected from the pipeline network as inputs to the deep belief network to detect pipeline leakage. This paper studies the current status of leak detection methods and proposes a leak detection algorithm based on one-dimensional convolutional neural networks that can automatically extract features. Due to the great dependence of deep learning on the amount of data, the larger the dataset, the better the network learning effect. The actual number of signals collected is far from meeting the needs of the network, so this paper proposes a data set enhancement technique for one-dimensional time series signals to expand the data set. Experiments have compared and analyzed the recognition accuracy of different parameters of the model, and the final results show that the model in this paper has good performance.