Neural Net Time Series Of Roof Water Inrush Forecasting Based On Logistic Regressive

Cuimu coal mine is located in Yonglong mining area of Huanglong Jurassic coal field,which is rich in coal resources.However,in the mining face,there are many roof water inrush accidents with large amount of water,which cause great difficulties to the safe mining of coal and great threat to the safety of miners.In order to study and predict the roof water inrush rule,this paper realizes the study and prediction of the roof water inrush rule time series neural network based on logistic regression analysis by means of stress analysis of roof water inrush model,semi variogram regular interpolation of variable data,regression analysis and fitting of water inrush index,and study and prediction of water inrush rule,and obtains the following research results:(1)The numerical simulation shows that the development height of the water diversion fracture zone determines the accumulation and loss of the water in the separation zone,and the stress distribution of the overburden is the inducing factor of the development of the water diversion fracture zone,that is to say,the key to the study of roof water inrush is the stress analysis of the overburden.On the basis of numerical simulation of roof separation development,it is determined that there are three water inrush modes in the roof,and the water inrush can be attributed to the stress release of overburden fracture.On the basis of beam and plate fracture theory of the roof overburden,the gravity stress of the rock mass and the gravity stress of the fluid are discussed as the geological model of water inrush of the roof overburden.(2)Combined with the geological model and statistical law of overburden fracture,a regression analysis solution for selecting indexes by using the results is provided,and an initial index system of roof water inrush is designed,which takes the rock load and fluid load as the main components and the layered rock stratum as the calculation basis.(3)On the basis of geostatistical pan Kriging interpolation and the calculation principle of logistic regression,based on the 33 initial indexes of the initial index system of roof water inrush,the logistic regression model is established for the behavior of roof water inrush,and the best geometric parameter set with the accuracy of 0.78 is obtained.The regression model of roof water inrush is analyzed,and it is shown that in the overburden rock inrush In the process of water inrush,there is a certain degree of synchrony in the rock stratum,and the water inrush behavior has obvious plate breaking characteristics,and it is limited to four sides fixed support;three sides fixed support,one side simple support;two sides fixed support,two sides simple support,these three boundary conditions.The contribution of water inrush from different strata has certain gradient difference.(4)By using time series neural network and adding location description variables on the basis of the best fitting index system,the concept of time series accumulation is introduced for the learning of water inrush rules,and the neural network learning of water inrush rules of coal seam roof and the prediction of water inrush in the second panel are realized.