Study On Location And Intensity Inversion Of Mine Gas Explosion Source

Due to the extremely complicated causes of gas explosion accidents,many acci-dent scenes were seriously damaged and there were few survivors.The determination of the location and intensity of gas explosion sources had brought great difficulties to the emergency rescue and accident investigation of accidents.The empirical method of de-termining the location and intensity of gas explosion sources cannot be able to adapt to current needs.This thesis conducts an study on location and strength inversion of mine gas explosion source,in order to let the emergency rescue headquarters develop disaster relief plans based on scientific analysis and argumentation,provide reference for accident investigation at the gas explosion site,prevent the occurrence of similar accidents and en-sure the safe production of coal mines.This thesis was funded by the National Key R&D Program of China(Project No.2016YFC0801800).In this thesis,a mathematical model and a black box model have been established for the location and intensity inversion of gas explosion sources by using theoretical anal-ysis and data-driven ideology.The well-posed condition and ill-posed condition of the inversion problem are studied.Least squares estimation based on improved genetic algo-rithm optimization,BP neural network based on improved particle swarm optimization algorithm and improved generalized regression neural network are introduced into the inversion process.The inversion is based on the position data and overpressure data of different measuring points obtained by gas explosion experiments and gas explosion nu-merical simulations.The gas explosion inversion database is established based on the inversion data,and the inversion method has been applied in the gas explosion accident of Tunlan mine.Research indicates that:(1)The required parameters for solving the inversion problem of gas explosion source position and intensity are the position data and the overpressure data of different measur-ing points in the inversion coordinate system;The inversion problem under well-posed condition requires two sets of position data and overpressure data.Three or more sets of position data and overpressure data are required under ill-posed condition.(2)Under the well-posed condition,the inversion based on the mathematical model and two sets of position and overpressure data and using least squares estimation is less accurate,and the mean square error is 3539.58.This is because the measurement data has observation errors,and the derivation of the inversion mathematical model is based on the simplification of the model During the inversion process,various system errors are coupled to each other and amplified,resulting in low reliability of the inversion result.(3)Under the ill-posed condition,the inversion based on the mathematical model and eight sets of position and overpressure data and using the least squares estimation opti-mized by the improved genetic algorithm,uses seven sets of point locations and intensities data.The inversion results show that the improvement measures have avoided the "pre-mature phenomenon" and "evolutionary stagnation" of genetic algorithm;The change of population size of genetic algorithm has little effect on the fitness function;With the in-crease of population size of genetic algorithm,the inversion accuracy of the explosion source position is improved,and the inversion accuracy of the explosion source intensity is reduced,but the total mean square error decreases with the increase of the population size;The mean square error of the inversion result is 445.53,which indicates that the in-version accuracy is greatly improved under the well-posed condition,and the inversion reliability is enhanced.(4)Based on black box model of the location and intensity inversion of gas explo-sion source and eight sets of position and overpressure data,the BP neural network method based on improved particle swarm optimization is used to study the inversion under ill-posed condition.The inversion results show that the improvement measures improve the global search ability of the particle swarm optimization algorithm and avoid the "prema-ture phenomenon";The population size of the particle swarm optimization algorithm is not as large as possible,and smaller population size shortens the convergence time of par-ticle swarm optimization algorithm;The mean square error of the inversion result is 6.11,and the inversion precision is greatly improved compared with the least squares estima-tion based on the improved genetic algorithm optimization,and the inversion reliability is further enhanced.(5)Based on black box model of the location and intensity inversion of gas explosion source and eight sets of position and overpressure data,the generalized regression neural network method is used to study the inversion under ill-posed condition.The inversion results show that the improvement measures have obtained the optimal training sample input mode and the optimal expansion speed;The inversion result has a mean square error of 4.47,and the inversion accuracy is better than the inversion method of BP neural network based on improved particle swarm optimization.(6)A gas explosion inversion database has been established based on the inversion data.Multiple sets of position and overpressure data has been obtained according to the damage situation of the gas explosion in the Tunlan mine and the numerical simulation.The inversion of the gas explosion accident in the Tunlan mine has used three methods,which are improved least squares estimation based on genetic algorithm optimization,BP neural network based on improved particle swarm optimization algorithm and improved generalized regression neural network.The inversion results show that the most widely used and accurate one is the BP neural network optimized by particle swarm optimization based on the black box model;when the inversion data is small,it is suitable to use the least squares estimation optimized by the improved genetic algorithm based on mathemat-ical model optimization;when using the generalized regression neural network inversion method based on the black box model,the inversion data samples of the same size as the inversion problem need to be provided as the training samples of the black box model.The main academic contribution of the research work in this thesis is that it uses both the traditional theoretical analysis modeling method and the nonlinear scientific model-ing method.These modeling methods involve the integration of multidisciplinary and multidisciplinary fields.In the process of inversion,a variety of optimization algorithms based on group intelligence algorithm and artificial neural network are introduced.The inversion results can provide decision support for emergency rescue and post-accident investigation at the scene of gas explosion accidents,which has strong practical signifi-cance.