Research On Comprehensive Prediction Methodology Of Rockburst Hazard In Coal Mine Based On Machine Learning

Because rockburst is controlled by the coupling of various factors such as coal mine geology and mining technology,the mechanism of rockburst occurrence is complicated,which makes it difficult to monitor,predict and analyze rockburst disasters.There is still a lack of comprehensive multiple-factors prediction method of rock burst.This paper adopts a machine learning method based on multidisciplinary data of coal mines,and uses the geological data,monitoring data,coal seam mining data and high-energy mine earthquake records of a working face in shandong province of a thousand-meter deep well coal mine to influence the factors and the occurrence of mine earthquake and rock burst to summarize and analyze the rules of characteristics,research on comprehensive prediction of rockburst hazard based on machine learning methods.By establishing mine earthquake intensity,rockburst hazard levels,the intelligent analysis model of related prediction problems such as types and related analysis methods are proposed to provide new methods for scientific monitoring,prediction and prevention of rockburst disasters during coal mining.The main research is as follows:1.In view of the outburst of rock burst and the difficulties in monitoring and analyzing the precursor information,the main factors influencing the occurrence of rock burst in coal mines are analyzed by using the data obtained from microseismic monitoring,ground sound monitoring,coal powder monitoring,roadway stress monitoring,and support resistance monitoring of working face.These factors are divided into geological factors and production factors,and the mining geological factors and monitoring data are divided into two aspects According to the data types of discrete and time continuous,the prediction methods are divided into static prediction and dynamic prediction.2.Based on the mining geological factors of the working face,a prediction model of rockburst danger level based on Bayes decision theory is established.Bayes discriminant analysis and Bayes network inference are used to predict the risk level of rockburst,realize static prediction,and the weights of different influencing factors are discussed to provide a new quantitative method and analysis method for the study of rockburst prediction.3.In order to accurately predict the intensity of mine shocks when rockburst occurs,the monitoring data related to rockburst such as coal powder monitoring,roadway stress monitoring and working surface support resistance monitoring are used as the discriminating factors for prediction.The magnitude of mine earthquake monitoring is used as the actual result to establish a model for predicting the magnitude of mine earthquakes.Considering the difference of influencing factors and applicable conditions,the radial basis neural network and the multi-layer perceptron neural network are used to predict the magnitude of mine shock,respectively,and more accurate results are obtained.4.To resolve the problem of dynamic prediction of the type of rockburst using the time-frequency distribution characteristics of microseismic monitoring signals,a deep learning model based on gated recurrent unit recurrent neural network(GRU-RNN)is proposed.The inputs of the GRU-RNN model are continuous multi-channel microseismic monitoring signals.The mechanism can implicitly extract the morphological distribution,amplitude and spectrum characteri stics of the microseismic signal to implement the discriminant analysis of the inducing factors of rockburst.For the dynamic prediction of rockburst hardard level,a restricted Bolzmann machine process neural network(RBM-PNN)model based on continuous multi-channel microseismic monitoring signals and ground sound monitoring signals is preresented.The two deep learning models can improve the existing methods to distinguish the comprehensive characteristics of different types of signals,and have a good adaptability to the dynamic prediction problem in mechanism.5.To address the issue of dynamic prediction of rockburst hazard fusion of multi-source time series data in mine geology,mining production and safety monitoring,a dynamic fuzzy inference neural network is proposed.The model combines fuzzy logic inference and neural network learning mechanism for signal characteristics,expresses domain knowledge based on fuzzy sets and membership functions,and adaptively establishes inference logic and fuzzy discriminant rules based on multi-source process signal sample sets,which can be effectively fused multi-source process information and prior knowledge,and it is suitable for modeling and predictive analysis in the case of small sample sets.Based on the above,combined with the practical application of the comprehensive prediction of rock burst in the mining of 1412 working face in shandong province and the prediction results are more accurate.