| Timely detection and location of pipelines leakage can avoid serious consequences.In the pipelines leakage detection by infrasonic method,the weak leakage signals are submerged in the background noise,and the useful leakage characteristics cannot be extracted from the signals.The traditional filtering noise reduction methods are easy to lose useful signals in the process of signals processing,and cannot guarantee the stable and effective application of small leakage signals of long-distance pipelines in practice.The paper tries new ideas and methods to construct a variety of leakage characteristics identification models to process signals and detect pipelines leakage.The work done is as follows:1.Combining the idea of information fusion and multivariate feature extraction technology,the characteristics of infrasonic signals and negative pressure wave signals are extracted and fused in all aspects.Combining the random forest model and the idea of recursive feature elimination to get the optimal characteristics subset,and train the random forest model,which realizes the classification of normal samples and leakage samples,and provides guidance for the follow-up research.2.Applying chaos theory to pipelines leakage detection,the chaotic critical state detection method for multi-frequency signals is proposed,which improves the selection criteria for the desired frequency of the multi-frequency signals to be measured.The adaptive stochastic resonance signals enhancement model based on the desired frequency is proposed.A new index is used to replace the signal-to-noise ratio to evaluate the signals enhancement effect.Based on this,the structural parameters are adjusted adaptively,which realizes the enhancement of useful signals and obtains obvious leakage characteristics.3.Based on the adaptive stochastic resonance algorithm based on the desired frequency,the dynamic recursive normalization algorithm and the accumulative fast difference algorithm are proposed,and to construct two real-time continuous pipelines leak detection models to achieve online continuous detection of the signals,which solves the problem that most methods cannot detect the signals continuously.All of the above methods have verified the effectiveness and practicality with actual signals. |