An Intelligent Algorithm For Oil Pipeline Slow Leakage Detection

Pipeline leak detection is an important part to guarantee the safe operation of oil and gaspipelines. Since the detection accuracy and false alarm rate of common pipeline leak detectiontechnology are limited seriously by pipeline operation and leakage diversity, some new requires areproposed by the pipeline safety such as detecting small and slow leak, reducing the false alarm rateand so on. Thus, this paper studies the internal dynamic characteristics of pipe pressure signaldeeply, and the various types of operating conditions can be identified effectively, which is the trendto improve pipeline leak detection accuracy. The main research work is as follows:In this paper, the measured pipeline pressure signal as the study object, the original data arepre-processed to be de-noised by three wavelet packet. Based on the maximum Lyapunov exponent,the inherent complexity of pipeline fluid is explored. Six pipeline pressure time series in differentstates, for example, are reconstructed by using MATLAB software from the perspective ofnonlinear dynamics. Their nonlinear characteristic including chaotic characteristics andapproximate entropy values are extracted, which indicate the characteristic values distribution ofpressure signal in different states presents their internal complexity with different degrees.The feed forward neural networks are optimizes by swarm intelligence algorithms evolvedfrom bionics. Then the artificial bee colony algorithm(ABC) developing recently is analyzed andcompared with traditional particle swarm algorithm on the parts of basic theory and model of neuralnetwork optimization. By training international standards datasets, the BP neural network modeloptimized by ABC is validated to converge faster and satisfy the error requirements. Owing to itsmore effective classification accuracy, it can be applied to recognition of pipeline leakage state. Themeasured pipeline pressure signal of simulated leakage are collected on site. The nonlinearcharacteristics of pressure signal are extracted and the neural networks optimized by swamintelligent algorithms are established to detect pipeline leakage. The analysis results show that theweights and threshold parameters of network can be optimized by ABC more quickly andaccurately and much less steps are passed to achieve the training error requirement. The slow leaksclose to1%are detected by the optimizing network established, which the right recognition rate of91.67%greatly reduces the false alarm rate of pipeline detection system and improves the reliability.