Study On The Evolution Law Of Coal Mine Gas Distribution Field And Its Spatio-temporal Modeling

Gas disaster is the worst disaster form in the coal mine which often causes heavycasualties and significant economic losses. A lot of coal mine gas accident occurred inthe working face roadway area, it is the key to gas control. So far, prediction of gasdata is mainly composed of time series methods while construction of the gasdistribution field use space information as the primary means. But transport anddistribution of gas is closely related to time and space, ignoring the relationshipbetween time and space to handle problems independently will result in the loss ofimportant prior knowledge. This dissertation takes the gas migration regularity as thestarting point of gas safely problems solution, analysises the gas diffusion andmigration in theoretically, proposes solving gas problem by spatio-temporal modeling,and takes the data-driven method as research mode. It studies on the following aspects:A new spatio-temporal extreme learning model (STELM) is proposed to predict gasconcentration by carrying on the space extension of time predict method extremelearning model (ELM) which was selected as the basic learning model. In order tosimplify the complexity of spatio-temporal modeling, it chooses spatial correlation asinput weighting and time panel data of the adjacent site as the input of spatio-temporalneurons. The algorithm only requires two input parameters: space delay operatorboundary value and time delay operator boundary value. The academical simulationdata and the field monitoring data show that the proposed method has largeimprovement on the generalization ability compared with the forecasting methodwhich only depends on the time dimension information; On this basis, it proposesselective ensemble learning methods of STELM which based on L1regularization bycombining with selective ensemble learning ideas, which further enhances theprediction accuracy and generalization performance; In order to solve the gas fieldreconstruction problem, it learns from kriging model and extends time dimension,uses the product-sum model by fitting the spatial semi-variogram and timesemi-variation function to construct spatio-temporal semi-variogram function. Thus, itcan be easy to achieve the internal relationship between time and space. The proposednew reconstruction method of gas distribution field based on the spatio-temporalkriging improves interpolation effect significantly; In order to make the proposedspatio-temporal modeling methods get better results in the actual scene in the case ofa fixed gas sensor number, multi-objective particle swarm optimization algorithm is adopted to model and improved, incremental rate dominant is used as the fitnesswinning strategy, elitist and passive update mechanism is added, which is moreapplicable to the layout optimization scene of coalmine gas sensors.