Research On Fault Prediction For Gas Recovery Of Single Coal Bed Methane Well Based On SVM

Along with the advancement of industrialization and the growing demand for new energy in our country, the coalbed methane as a new environmental protection energy gradually brought to the attention of the countries all over the world. However, the exploration of coalbed methane is a large-scale and complicated system, once accident occurs, heavy economic loss and severe casualty would result. In order to ensure reliable operation of coal-bed methane mining system, it not only requires the system fast accurate diagnosis as failure happens,but it’s also hope pre-diagnosis before the failure occurs, so that the worker can have enough time to take measures to prevent the occurrence of the failure. CBM single well gas recovery system is the basic node of the CBM gathering system. When a fault occurs in the single well gas recovery system, it is likely to cause paralysis of the entire CBM gas recovery system.Therefore, the main content of this study is aimed at single well gas recovery system fault prediction of coalbed methane.This paper is discussed the basic principle and the technological process of single well gas extraction system in the coalbed methane.Based on the above,this paper makes a in-depth analysis of the key parameters and several faults occurred in the process of single well’s gas production.CBM single well gas recovery system has the characteristic of running stability and gas extraction parameters changing quite gently. For these features, the research has made a forecasting method based on the time series parameters of Support Vector Machine (SVM). The method resolves the data prediction problem of fault prediction of coalbed methane gas recovery system. Aiming at the influence of the parameters of SVM to prediction performance, this paper introduces several kinds of optimization methods which are frequently used, it contains of grid search method, cross validation method, genetic algorithm and particle swarm optimization. the research has explored the process of selection and crossover operator of genetic algorithm which has been introduced into the traditional Particle Swarm Optimization (PSO) algorithm and has established a support vector machine model based on modified particle swarm optimization algorithm. The Final Prediction Error (FPE) principle is used to optimize the embedding dimension of the model, the parameters of the SVM model was optimized by using improved PSO algorithm, root mean square error is used to estimate the accuracy of the model. The model was applied in the trend forecasting of gas extraction parameters of coalbed methane gas recovery system. Comparison with the model optimized by the grid search method, PSO algorithm and the BP neural network shows that both one step forecasting accuracy and multi-step forecasting accuracy by the advanced PSO-SVM model are higher than those models.Then, for a whole fault forecast system, in addition to predict whether it has the fault, but it also needs to further determine the type of fault. For locates the types of the fault, on the basis of traditional SVM,this paper further researches the Fuzzy Support Vector Machine (FSVM) so that it can eliminate the effects on the modeling results by the noise. In the modeling processes, this paper uses the fuzzy C mean clustering algorithm to confirm the fuzzy membership degree of the training sample. Based on the above, it establishes the FSVM fault diagnosis model based on improved PSO. The simulation results show that comparison with the traditional SVM model, the FSVM model has a higher diagnostic accuracy, it also verifies the validity of this model.