Study and application of deeply optimized neural network in roof stability evaluation

Deep coal seam mining causes instability and collapse of coal seam roof frequently, which seriously affects the safety production and threatens the personal safety of underground personnel. In order to evaluate the stability of coal roof accurately, this paper select 6th coal seam in Kongduigou Coalfield of Jungar Coalfieldas research object, analyzes the geological and hydrogeological data, and study the lithology, rock combination, sandstone thickness, fault, fold, seam inclination, rock quality index, and rock compressive strength on the influence of the roof stability, drawing the main control factor 3D mapping projection surface maps. Select 58 borehole data points as the input samples (50 training sets and 8 test sets), use genetic algorithm (GA) to optimize the network random initial weights and threshold initial and sparrow search algorithm (SSA) for secondary optimization for the BP neural network training and learning, establishing GA-BP neural network based on SSA optimization (SSA-GA-BP neural network) coal roof stability evaluation model, which is used to predict and evaluate the 6th coal roof stability of the research area after the training error accuracy reached the requirements. The fuzzy comprehensive evaluation method, BP neural network, GA-BP neural network and SSA-GA-BP neural network are also used to predict and evaluate the 6th coal roof stability. Compare the evaluation results of each model with the actual value. The results show that the error of coal seam roof stability evaluation of SSA-GA-BP neural network is smallest, with the accuracy 88%, and the model is successfully applied to predict the roof stability of the 6th coal seam in Kongduigou Coalfield, which provides a scientific evaluation method and theoretical basis for the evaluation of coal seam roof stability.