Research On Prediction Method Of Coal And Gas Outburst Risk

Modern coal mining enterprises greatly improve the productivity,with the progress of industrial technology.However,this rapid development has brought a variety of industrial accidents caused by production safety problems.Safe production is an important condition to ensure the economic benefits of enterprises and the life and health of employees.Coal and gas outburst is an extremely complex dynamic disaster in the process of coal mine production.This kind of disaster will destroy the important underground mechanical equipment and disturb the ventilation system.In serious cases,it may even cause a large number of casualties.Although coal mining enterprises have established strict rules and regulations and monitoring system for this harm.But it is still difficult to achieve timely and accurate prediction,because coal and gas outburst is affected by many factors and there is a highly nonlinear relationship between them.If this problem can be solved,the corresponding protective measures can be made before the disaster.The life safety of underground workers will be guaranteed to the maximum extent.At present,some monitoring data for coal and gas outburst have been accumulated by most production enterprises.But the prominent disaster has the characteristics of many influencing factors and complex mechanism,and belongs to the small probability event.The outstanding data that can be used for model training is only a small part of the total.The following work is done to solve the prediction problem: First,eight factors that may lead to outburst are analyzed.The grey correlation analysis method is adopted,and five factors are selected as the next step of outstanding prediction indicators;Indicators combined with sample data are used to form a prediction indicator set.The training data set is imported into neural network model and support vector machine model for training and prediction;The oversampling algorithm theory is used to solve the sample class imbalance problem.The new samples are synthesized by the improved BSMOTE algorithm to balance the data set;Particle swarm optimization is used to solve the problem of complex parameter adjustment in training process.Finally,the improved particle swarm optimization algorithm and BSMOTE-SVM are combined to form PSO-BSMOTE-SVM model.The results show that the correct classification rate is 61.90% when using SVM model for prediction.The accuracy of BP neural network is only 47.62%.It can be seen that the performance of support vector machine is better than that of BP neural network under the condition of small samples.But the results of model training still can not meet the prediction accuracy requirements;BSMOTE algorithm is adopted when the failure reason of SVM model is analyzed.The shortcomings of the algorithm are improved accordingly.Firstly,the salient samples are divided into several clusters by cluster analysis.Secondly,the sample distance and near neighborhood density are considered comprehensively,and the "suspected noise points" in each cluster are identified and eliminated.The remaining samples are sorted according to the amount of information.After that,the new samples are synthesized by the improved formula.The simulation results show that the classification accuracy of the improved BSMOTE-SVM prediction model reaches 80.95%.It can be seen that the prediction accuracy is greatly improved by the oversampling algorithm;Accurate and appropriate parameters are difficult to determine quickly when BSMOTE-SVM model is used.The improved particle swarm optimization algorithm is used to optimize the parameters of the model.The simulation results show that the training time of PSO-BSMOTE-SVM prediction model is reduced to 6.241 seconds,and the classification accuracy reaches 92.86%.The accuracy requirement of coal and gas outburst prediction can be satisfied.