Intelligent Leak Detection And Localization Of Multi-conditions Pipeline Based On Signal Processing

As China’s pipeline construction continues to develop,the current state of the pipeline network is becoming increasingly complex.Many pipelines directly or indirectly enter periods of frequent leaks,posing hidden risks to pipeline operation safety.Therefore,it is of great significance to combine signal processing and pattern recognition technology to carry out technical research on intelligent leak detection and location of pipelines,which can effectively improve the efficiency of pipeline leakage detection and location and ensure the safe operation of pipelines.To address the problems of insufficient anti-noise ability of existing leak signal singularity detection algorithms and excessive dependence of detection effect on signal noise reduction processing technology,a double deviation-based pipeline leak singularity detection algorithm is proposed.The reliability and efficiency of the double deviation value in detecting singularities under high noise are demonstrated,and the reliability and accuracy of the algorithm in detecting leak singularities under different degrees of noise reduction are verified by comparing the noise reduction algorithm and its parameter range on the simulation data and experimental data.The experimental results show that,compared with the wavelet mode maximum first-order derivative algorithm,the proposed algorithm improves the detection rate and accuracy by 153.1%and 519.4%,respectively,on five sets of high-noise simulation data,and improves the detection rate and accuracy by 175.9% and 415.0%,respectively,on four sets of high-noise leakage experimental data.The algorithm ensures the accuracy of the leak signal singularity detection while effectively reducing the false alarm rate and leakage rate of the leak detection results,and the improvement of the leak detection localization effect is remarkable.To address the problems of massive data scale,high cost of utilization and unintuitive visualization during pipeline operation,a time series segmentation algorithm based on double deviation is proposed and combined with the leak singularity detection and localization algorithm.The reliability and efficiency of extracting time series features by double deviation values are demonstrated,and the reliability,noise immunity,adaptability,compression accuracy and segmentation effectiveness of the proposed algorithm are compared and verified using simulated data containing different trends and different noises,different kinds of public data sets and oilfield field pipeline leakage data,respectively.The experimental results show that compared with four representative methods in selected time series segmentation and feature extraction,the proposed algorithm has an average reduction of fitting error up to 46.8%～79.8%on simulated data,shows stable and efficient adaptability and compression ability on public data sets,and has obvious segmentation effectiveness and high data compression ability on pipeline leakage data.While improving the segmentation fitting efficiency,the algorithm can effectively distinguish the trend changes before and after the leak time node,further improving the analysis capability of the leak signal data.To address the problem of false leakage detection caused by non-leakage conditions during pipeline operation,a "signal detection + pattern recognition" method for pipeline leak detection is proposed.The classification accuracy of SVM,LSSVM and TWSVM algorithms with different kernel functions and construction method settings are compared and verified on the experimental data of circular pipeline.The experimental results show that the TWSVM with RBF kernel function and "one-to-one" multi-classifier construction method has a high accuracy(99.32%)in classifying different working conditions of pipelines under the conditions of preferable parameters σ and C,which solves the influence of working conditions with similar data characteristics on the leak detection results to a certain extent and further improve the efficiency of pipeline leak detection.Based on the above research results,the visualization page is designed for the purpose of simplicity and efficiency,and an oil pipeline leak detection and location system based on multiple methods such as pipeline leak signal data analysis,leak singularity identification and location,signal data PLA fitting and compression,and leak working condition identification is realized.