Research On Remaining Life Prediction And Maintenance Strategy Of Buried Pipelines Under External Corrosion

During the service of buried oil and gas pipelines,accidents occur due to various factors such as soil environment,third-party damage,microorganisms,etc.Therefore,it is especially important to reasonably predict the change of maximum corrosion depth during the service life of buried oil and gas pipelines,to improve the accuracy of the remaining life prediction of pipelines,and to give reasonable maintenance strategies according to the actual situation for the safe operation of pipelines.To this end,this paper has done the following work:(1)Maximum corrosion depth prediction of buried pipeline based on GSWOA-GRNN: Firstly,the data are processed by outlier processing and Z-score normalization,and then the factors affecting the maximum corrosion depth of buried pipeline are correlated by using Spearman’s coefficient to get 9 indicators that are highly correlated with them,and the high efficiency of GSWOA search is used to find the optimal value of the smoothing parameter in GRNN.GSWOA search for the optimal value of the parameters to improve the accuracy of the prediction of the maximum corrosion depth of buried pipes.The results show that GSWOA-GRNN has excellent performance in terms of goodness of fit,model reliability and prediction accuracy compared with the traditional machine learning model and the initial GRNN model.The model can predict the maximum corrosion depth in a section of buried pipeline according to the field environment index,so as to determine the corrosion susceptibility point in the section.(2)Dynamic remaining life prediction of buried pipelines based on LSTM-Gamma process: The excellent mathematical properties of the Gamma process are combined with the excellent data fitting capability of LSTM to predict the remaining life of buried pipelines by taking the vulnerable point in a certain pipe section of buried pipelines as the research object.Specifically,after smoothing the data with Savitzky-Golay wavelets,the scale parameters of the Gamma process and the shape parameters at different degradation moments are estimated using Markov chain Monte Carlo(MCMC),and then the estimated shape parameters are used as the input of the LSTM for single-input multiple-output training,and the parameters of the LSTM are updated at every time interval,and finally The remaining life of the pipe at a certain service time is obtained by deriving the probability density function of the remaining life.(3)Research on buried pipeline maintenance strategy based on deep reinforcement learning: Based on the theory of incomplete maintenance,the LSTM-Gamma process degradation model is used as the degradation path of the pipeline,and after determining the preventive maintenance actions and their repair rates,and designing the reward function for evaluating each maintenance action,an updated learning mechanism based on the reward function is established to realize the buried oil and gas pipeline maintenance strategy based on deep reinforcement learning Optimal selection.Deep reinforcement learning(DDQN)is used to solve the model: the greedy factor ε with continuous decay enables the intelligent body to choose more repair options in the early stage to avoid falling into local optimum.The Q-network is then used for action selection and the target network is updated asynchronously to solve the problem of DQN overestimation.The cost-efficiency ratio of the optimal repair strategy is analyzed for different initial states after the completion of the iteration to obtain the optimal strategy for pipe repair in a given computational period.