Research On Algorithm For Fault Diagnosis Of Mine Ventilation System Windage Alteration For Imbalanced Data Set

The stable operation of mine ventilation system is an important factor to ensure the safety of mine production,orderly operation of equipment and disaster prevention and reduction.If the ventilation system fails,the underground air flow will change,which will lead to the accumulation of dust and gas,and reduce the stability of the mine ventilation system and the ability to resist disasters.It is of great significance to determine the location of fault branch timely and accurately to ensure the reliability and safety of mine ventilation system.In recent years,ventilation system windage alteration fault diagnosis method based on intelligent algorithm has gained wide attention.This kind of method can obtain good branch identification results under the condition of ideal data set.However,in actual working conditions,the fault sample data of mine ventilation system is imbalanced,and the traditional machine learning model has certain limitations in the case of imbalanced data fault diagnosis.Therefore,based on the characteristics of ventilation system monitoring data,the dissertation adopts statistical analysis,programming and other methods to study the fault diagnosis algorithm of ventilation system imbalance data.To improve the quality of wind speed monitoring data,a SDAE model for cleaning mine wind speed data is constructed by using the powerful feature extraction and reduction analysis capabilities of the SDAE model.The correlation between monitoring data time series is mined by using association rules,and the "dirty" data is classified and processed in collaboration with the SDAE model.The experimental results of wind speed data cleaning in Dongshan Coal Mine show that the SDAE model has the best performance when the learning rate is 0.05,the noise staining is 0.1,and the number of iterations is 1400.The MAE and RMSE values are 5.6% and1.87%,respectively.The effective cleaning of wind speed monitoring data is realized.The influences of two cases of imbalanced data on the performance by conventional machine learning algorithms are analyzed.The experimental results show that the imbalance data within the classes(there is a huge difference in the number of fault samples in each branch)will make the fault diagnosis classification model biased to the majority of fault samples and ignore the identification of other fault samples.Imbalanced data between classes(the number of normal samples is more than that of fault samples)makes it difficult for conventional machine learning algorithms to establish an effective model and fail to identify faults.The experimental results of different input characteristics show that under different classification models,the fault diagnosis results obtained by taking the air volume as the characteristic are better than the wind pressure characteristics.Aiming at the imbalance phenomenon within the classes,the Wasserstein distance generation Adversarial Network(WGAN-div)is built to carry out data enhancement processing on the imbalance data set.The residual blocks are added creatively in the WGAN-div network.Combined with random forest(RF)model in ensemble learning,fault branch diagnosis of ventilation system is realized.The experimental results show that compared with the traditional mine ventilation system resistance fault diagnosis method(SVM),the indexes of Re,Pr,G-mean and F1 of WGAN-div-RF model are increased by 4.7%,2.3%,10.1%,and 3.5%,respectively.Aiming at the imbalance phenomenon between the classes,the single classification is introduced.The multiple single classification support vector machines(OCSVM)are integrated to construct the MC-OCSVM ventilation system fault model.Scale factors are designed to uniformly process the output of each OCSVM model,and the problem of ventilation system fault diagnosis is transformed into the maximum decision distance.The modeling without fault samples participating in training is established.Experimental results show that the proposed algorithm can identify fault branches quickly and accurately,the accuracy is 93.2%,and the single fault diagnosis time is 1.2s,which has strong robustness and generalization.The dissertation has 59 pictures,47 tables,and 151 references.