Research On Mine Gas Monitoring Technology Based On Data Fusion

The thesis topic comes from a school-enterprise cooperation project.With the development of sensor technology,more and more raw information on gas concentration is collected during the mining process.How to process the huge amount of data,mine the useful information and eliminate conflicting data becomes the focus of the next research.Multi-sensor data fusion technology collects data from different types of sensors distributed down the mine for comprehensive analysis,from which the correct data can be obtained to make accurate decisions.Traditional D-S evidence theory is used as a classical data fusion algorithm,but it needs improvement in solving problems such as high data conflict and basic probability assignment.In this paper,we propose a mine gas monitoring system containing a sparrow search algorithm optimised BP neural network（SSA-BP）gas detection system and a belief Hellinger distance（BHD）optimised D-S evidence fusion theory from a practical situation.The SSA-BP algorithm is mainly applied to signal acquisition and prediction of data received from sensors,and the BHD optimised D-S algorithm is applied to underground data processing and data fusion to make decisions.The main research elements of the thesis are as follows:（1）To study the theory related to multi-sensor data fusion,analyse the relationship between beliefs and rationality of D-S evidence theory,and for the conflict measurement of D-S evidence theory use the conflict coefficient k to balance the conflict degree of BPA function and ensure the normalization of data fusion results.（2）To improve the accuracy of mine gas detection,the SSA-BP neural network regression prediction model for mine gas concentration is proposed.The iterative operation rate of SSA-BP neural network is significantly improved compared with other regression methods,which proves that the algorithm is more capable of finding the optimal operation.This indicates that the optimization of SSA significantly improves the inversion capability and detection accuracy of the BP neural network model.（3）In order to solve the impact of the high conflict problem on the data fusion results,a data fusion algorithm with belief in Hellinger distance-optimized D-S evidence theory is proposed.The algorithm uses the Hellinger distance to accurately measure the distance difference between different evidence sources,effectively solving the confidence problem of the traditional D-S evidence theory0.The belief entropy measure is used to measure the information weight of each evidence body,which effectively solves the problem of evidence conflict transfer to the recognition framework and reduces the error of fusion results from multi-sensor high conflict data in mine gas detection.The experimental results show that the confidence value of belief Hellinger distance-optimised D-S evidence theory for the correct evidence body when disturbed by conflicting data is0.9982,with an accuracy rate of 97.94%.（4）The data collected from the mine were pre-processed and the confidence function was calculated based on Bayesian theory to provide the accuracy of the evidence fusion results in the next step,with the basic probability assigned according to the hazard level of each evidence source.A two-stage data fusion structure is designed.Firstly,the first fusion performs an initial sieving of the calculated BPA to obtain data results for each gas environmental factor;next the second stage fusion uses the belief Hellinger distance optimisation D-S evidence theory to perform decision-level fusion of multi-sensor data samples.The method is applied to actual mine gas detection and the results show that the concentration errors for CH₄,CO,O₂and CO₂are 0.0074,0.0084,0.0070 and 0.0083respectively,demonstrating the effectiveness of the method in resolving high conflict data errors in mine gas detection.