| With the increasing demand for natural gas in my country year by year,underground gas storage as an important energy storage infrastructure has great practical significance for ensuring national energy security.Salt cavern gas storage has the advantages of low porosity and permeability,good creep characteristics,and strong self-damage recovery ability.As the core structure of the salt cavern gas storage,the injection and production string is susceptible to integrity damage due to its complex environment and high pressure.In order to explore the corrosion mechanism and failure cause of the injection and production string of the salt cavern gas storage.In this study,a corrosion prediction model based on machine learning algorithm was established,and then the corrosion rate of the injection and production string of the salt cavern gas storage was obtained by simulation.In order to provide quantitative data support for its safe operation and maintenance and risk early warning.The research is mainly divided into three parts.Firstly,the corrosion factor analysis of the injection and production string of the salt cavern gas storage is carried out.Since the corrosion data is usually high-dimensional and nonlinear,the kernel principal component analysis method is selected to extract the features of the corrosion data,and the concept of the wavelet kernel function is introduced to improve the feature extraction efficiency.The prediction model of the corrosion rate of the injection and production string of the subsequent salt cavern gas storage will be well prepared.Then,three machine learning methods were introduced into the corrosion prediction study of the injection and production string of the salt cavern gas storage.At the same time,the corrosion rate prediction models of the injection and production string of the salt cavern gas storage based on the back propagation network,the support vector machine and the extreme learning machine are established respectively.Then,the optimal prediction model was selected by comparing the results of three commonly used evaluation indexes.Finally,the salp swarm algorithm and the Improved gray wolf optimization algorithm are introduced to iteratively optimize the random parameters of the prediction model.Verified by an example,the model prediction effect optimized by the swarm intelligence algorithm is more accurate,and the established model provides a new method for predicting the corrosion rate of the injection and production string of the salt cavern gas storage.The research results show that the three established machine learning-based models for predicting the corrosion rate of the injection and production string of the salt cavern gas storage all show good prediction effects and model characteristics.Among them,the prediction effect of the pipe string corrosion rate prediction model based on ELM is the best highest.Its prediction results are better than those of pipe string corrosion rate prediction models based on back propagation network(BP)and support vector machines(SVM).In addition,the prediction effect of the model optimized by the salps swarm algorithm(SSA)and the improved gray wolf algorithm(IGWO)is better than that of the single prediction model,and the prediction accuracy of the IGWO-ELM model is better than that of the SSA-ELM model.This not only enriches the prediction method of the corrosion rate of the injection and production string of the salt cavern gas storage,but also provides an intelligent guarantee for the safe operation and maintenance and risk early warning of the injection and production system of the salt cavern gas storage. |
