Study On Slope Stability Prediction Method Based On GRNN And ANFIS

Among the slope stability prediction and evaluation methods,the traditional numerical analysis method has a large amount of calculation and is strongly dependent on experience,and cannot reflect the dynamic open and nonlinear characteristics of the slope.Based on the generalized Hoek-Brown criterion reflecting the nonlinear failure of rock masses,this paper aims at the evaluation of the stability of rock slopes,and comprehensively considers the factors affecting the stability of slopes such as strength parameters and geometric parameters,and establishes a method based on the improved bat algorithm.The Generalized Regression Neural Network(GRNN)model predicts the safety factor and state of the slope;Based on comprehensive consideration of external natural factors,geological factors and engineering factors,an adaptive neuro-fuzzy inference system based on subtractive clustering is established to grade and evaluate the stability of the slope.Specific research contents are as follows:(1)Taking the actual slope in the engineering as a reference,by changing the strength parameters in the Hoek-Brown criterion and the geometric parameters of the slope geometry,the slope body with changed parameters is modeled and simulated by the strength reduction method(FLAC3D software)and the simplified Bishop method in the limit equilibrium method(Slide software),The safety factors of each slope are obtained as the sample source of the artificial neural network.(2)In the slope stability analysis,the safety factor and the slope state are important indicators to judge the slope stability.The generalized regression neural network has a simple structure and a good prediction effect.In practical engineering,the generalized regression neural network is chosen as the prediction model.Since the smoothing factor of the generalized regression neural network model layer has a greater impact on the network,the bat algorithm(BA)is used to determine the value of the smoothing factor;crossover and mutation operators are used to increase the diversity of the bat population so as to solves the problem that bat algorithm is trapped in local optimum in the iterative process.The comparative experiments show that the improved network has higher accuracy in predicting the safety factor and state of the slope.Using the network to predict the stability of slopes with different strength parameters and geometric parameters,the variation trend of safety factors obtained shows has good consistency with the actual situation,which further verifies the applicability and accuracy of the network.(3)In the actual slope engineering,in addition to considering the influence of the slope's own strength parameters and geometric parameters,it is also necessary to fully consider the random and uncertain factors such as natural factors,geological factors and engineering construction factors on the stability of the slope.In view of the complex fuzzy and non-linear relationship between various influencing factors and slope stability,the prediction model based on neuro-fuzzy inference system(ANFIS)is used to predict its stability level.Firstly,the subtraction clustering algorithm(SCM)was used to determine the number of membership functions of the adaptive neuro-fuzzy system,and the improved BA algorithm is used to determine the two neighborhood radii in the subtractive clustering.Finally,an improved BA-SCM-ANFIS slope stability rating model was established.When the network parameters were updated,the adaptive momentum stochastic optimization algorithm(Adam)was used to replace the traditional gradient descent method to accelerate the parameter updating speed.The evaluation results of the output grade of the network predicted samples and the actual grade were consistent with the actual situation,which confirmed the effectiveness of the network.In summary,in view of the nonlinear problem in the slope failure process,the improved GRNN network model and the ANFIS-based prediction model was used to predict and evaluate the slope stability.