Prediction of Slope Deformation at the Dam Inlet Based on GA-BP Neural Network

The stability of dam intake slopes is critical to ensuring the safe and reliable operation of hydropower projects. Although backpropagation (BP) neural networks have been widely applied in slope deformation forecasting, they inherently exhibit limitations such as slow convergence and potential local minima issues. This study focuses on the left-bank intake slope of the Longtan Hydropower Station in Guangxi Province, where a hybrid genetic algorithm-BP (GA-BP) neural network model was developed to improve deformation prediction accuracy. Monitoring data collected from two distinct rock mass zones were meticulously selected and partitioned into training, validation, and test datasets for systematic analysis. Results demonstrate that the GA-BP model achieved high consistency with field-measured deformation trends. Specifically, the average prediction errors for measuring points M(4)1 and M(4)4 were 0.027 mm and 0.0070 mm respectively, representing significant improvements compared to the conventional BP model. In terms of computational efficiency, the GA-BP model converged to a precision of 2.46×10-6 within 12 iterations, whereas the BP model required 161 iterations to reach a lower precision of 2.3713×10-5. These findings highlight the GA-BP model's advantages in achieving faster convergence and higher prediction accuracy for dam slope deformation analysis.