Application Research On Height Prediction Model Of Fractured Water-Conducting Zone Based On Sample Optimization And Support Vector Regression

The long-term exuberant high consumption demand of coal resources not only promotes the sustainable development of China's national economy and social progress,but also brings a series of prominent problems such as surface subsidence,mine water inrush and low recovery rate of coal resources.It is found that it is of great significance to scientifically predict the development height of fractured water-conducting zone for solving the above problems.At present,the commonly used methods for the height predicting have their own advantages and disadvantages in practical application,but none of them has established a general prediction model.In this paper,literature review,theoretical analysis,field investigation and test are used to apply collaborative mining technology to design a high quality sample integration idea,which can effectively integrate height prediction parameters,automatically expand sample database and optimize sample data structure.On this basis,the establishment and evaluation of multiple nonlinear regression models based on empirical formula correction and support vector regression models using different parameter optimization algorithms are studied.This paper mainly studies the following aspects:(1)Based on the theoretical analysis of multiple factors controlling the development height of fractured water-conducting zone,the height prediction parameter system has been established.The height prediction database is also constructed,which can co-invoke a large amount of real-time or historical original production data reflecting the current situation of coal seam mining and revealing the occurrence status of real coal strata in the basic database of mining engineering collaboration platform.With the increase of the number of the platform stationed in the mine or working face,the corresponding sample data will be updated automatically and the sample database will be expanded.(2)The gray correlation analysis is carried out on the part of complete samples in height sample database,and the main control parameters for height prediction were selected.In addition,this paper proposes a new comprehensive evaluation index of roof strata to form high-quality sample data structure with five attribute dimensions including development height of fractured water-conducting zone,buried depth of coal seam,lithology proportion coefficient of hard rock,mining height and goaf inclined length,etc.(3)With the help of prediction database,69 sets of dimensionally optimized high-quality sample data sets are efficiently integrated to complete the multiple nonlinear regression model modified by empirical formula and the support vector regression model optimized by cross validation and genetic algorithm.The results show that the accuracy of MNR model is generally lower than that of SVR model,and the GA-SVR model optimized by genetic algorithm has higher accuracy,stronger generalization ability and better ability to meet the requirements of field application of height prediction.(4)The first mining face of 22101 coal mine was selected to complete the high-efficiency integration of prediction parameters,selection of main control input parameters and the height prediction test process of model result output,the prediction results are applied to the feasibility analysis of coal seam mining under gulch.The field survey results of borehole peep show that the prediction accuracy of GA-SVR model accords with the actual engineering requirement,and its guidance in precise mining of coal seam under gulch has certain reliability.