Research And Application Of Bayesian Optimized XGBoost In Minor Fault In Coalfields

Minor faults are a common geological hazard in coal mining,and studies have shown that their presence can lead to serious accidents such as roof collapse,water influx,and gas outbursts.Accurate prediction of minor faults is crucial for ensuring safety in coal mining,minimizing unexpected costs,and guiding on-site construction planning.However,due to the imprecise location and less prominent geological features,traditional techniques struggle to identify minor faults with high precision,and the accuracy of mainstream identification technologies still needs to be improved.In order to enhance the accuracy of seismic interpretation of minor faults,this study proposes using information value to reduce seismic attributes and optimizing XGBoost parameters with an improved Bayesian optimization algorithm for fault identification.First,the seismic attribute data in the mining area are preprocessed to remove abnormal and noisy samples.Then,for each processed feature,a chi-square binning method is applied to calculate the weight of evidence(WOE)in each bin,and the information value(IV)of each feature is obtained as an indicator of its importance.The features with low IV and high noise are then reduced.Meanwhile,a certain level of noise is added to the minor fault seismic data to enhance the model’s anti-noise ability.Next,an XGBoost model is constructed with a proposed method to balance the training weights of positive and negative samples due to the imbalanced distribution of samples.In addressing the issue of parameter selection,the Bayesian algorithm is employed to optimize the XGBoost model parameters.However,since the Bayesian optimization algorithm tends to be biased towards either "exploit" or "explore," it can result in low optimization efficiency and a tendency to converge to local optimums.To address this,an adaptive balancing factor change algorithm is proposed to dynamically balance the "exploit" and "explore" processes of the acquisition function PI,thereby improving the robustness of the parameter optimization process.Experimental results show that using the acquisition function-optimized Bayesian algorithm to optimize the XGBoost model with improved objective functions leads to a higher accuracy in identifying minor faults compared to other prediction methods,such as BP neural networks,support vector machines(SVM),K-nearest neighbors(KNN),Ada Boost,extreme learning machines(ELM),and random forests.The proposed XGBoost model framework(SAPI-Bay-Imp XGBoost)is expected to improve the accuracy of minor fault seismic identification.