Abstract：Considering that BP neural network model ignores the temporal correlation of slope monitoring data, and LSTM model falls into local optimality due to the subjectivity of hyperparameter selection, we propose a slope deformation prediction model based on the combination of genetic algorithm and long short-term memory network(GA-LSTM). We utilize the global search ability of genetic algorithm and the advantages of LSTM forecasting time series data. Taking the slope of Haiming mining open-pit as the research object, we adopt BP neural network model, LSTM network model and GA-LSTM network model, respectively, to predict and analyze the GNSS49 deformation of the slope monitoring point. We compare the times for each model to reach the convergence condition. The research results show that: The time difference between GA-LSTM model and other models to reach the same convergence condition is not large. The fitting accuracy of GA-LSTM model is between 0.1 mm and 0.2 mm, which is 5 to 7 times that of LSTM neural network model and 10 to 20 times that of BP neural network model, with high accuracy and stability. The predicted value is basically consistent with the actual monitoring data, which can provide scientific basis for the safe production, management and decision control of mine slope.
XIAO Haiping,WANG Shunhui,CHEN Lanlan et al. An Optimization Network Model for Slope Deformation Prediction Based on GA and LSTM Fusion and Its Application[J]. jgg, 2024, 44(5): 491-496.
