Research On The Windage Alteration Fault Diagnosis Of Mine Ventilation System Based On Deep Reinforcement Learning

The mine ventilation system windage alteration faults diagnosis model based on traditional machine learning algorithms achieves fault diagnosis by passively learning the nonlinear mapping relationship between input and output.It has poor intelligence,weak generalization ability,and poor performance when data distribution is unbalanced.Therefore,it is urgent to develop a new fault diagnosis technology.The deep reinforcement learning algorithm does not require manual feature extraction,and can achieve end-to-end learning from raw data to target output through interaction between agents and the environment.It is an effective method for achieving intelligent decision-making.Therefore,in this dissertation,a deep reinforcement learning algorithm is used to intelligently diagnose windage alteration faults in the mine ventilation system.In order to verify the effectiveness,intelligence,and strong generalization ability of the deep reinforcement learning algorithm in the field of mine fault diagnosis,the following research has been carried out:(1)Deep reinforcement learning is the combination of deep learning and reinforcement learning.In order to determine the appropriate deep learning network structure and hyperparameters,a deep learning model for windage alteration faults diagnosis of mine ventilation system based on one-dimensional convolutional neural network(1DCNN)was constructed.The influence of network structure,hyperparameter and sample attributes on the fault diagnosis performance of the model is explored,and it is found that the model performance is significantly affected by sample attributes,network structure and hyperparameters.The hyperparameters of the model are optimized based on Bayesian optimization algorithm(BO),and the optimized model is used to build a deep reinforcement learning Q network.(2)A fault diagnosis model for mine ventilation system based on deep reinforcement learning is constructed,and end-to-end intelligent autonomous decision-making from mine ventilation system state parameters to fault categories is realized.This method takes the air volume data of the mine after faults as the input of the model,and takes the fault roadway number or fault volume as the optional action for the current input.The fault feature is extracted through a deep learning network,and the Q value of the current state-action pair is fitted.The reinforcement learning model is used to identify the fault location and volume.Simulation experiments and on-site empirical applications have proved the intelligence,generalization,and feasibility of the method in the field of windage alteration fault diagnosis of mine ventilation systems.Compared with the results of traditional machine learning and deep learning based fault diagnosis models,the advantages of deep reinforcement learning models in terms of intelligence and generalization ability are proved.(3)Aiming at the problem of unbalanced fault samples,a method for constructing reinforcement learning reward functions based on KMeans++ algorithm and spatial distance is proposed.This method uses KMeans++ algorithm to cluster the samples,designs a reward function for the centroid based on the imbalance ratio,biases the reinforcement learning reward value with the Euclidean distance between the samples and the centroid in the same cluster,and constructs a personalized reward function for each sample.The diagnostic accuracy of the DQN model for the fault location of the sample sets UB1,UB2,and UB3 with different unbalance ratios is 98.75%,97.50%,and 95.00%,respectively.This effectively avoids the "bias" phenomenon of the intelligent model for most types of samples in the training sample,and achieves accurate prediction of the fault location under unbalanced samples.The mine ventilation system windage alteration faults diagnosis model constructed based on deep reinforcement learning algorithm has strong representation ability,generalization ability,and independent decision-making ability.Based on this method,it is expected to establish a universal architecture for mine ventilation system windage alteration faults diagnosis,providing theoretical guidance and technical support for achieving intelligent ventilation in mines.The dissertation has 87 figures,40 tables and 122 references.