Wavelet neural networks(WNNs)possess inherent drawbacks in data prediction,such as slow convergence and susceptibility to local optima.Conversely,artificial bee colony(ABC)algorithms exhibit strong global search capability and fast convergence,albeit with a tendency to slow down when approaching optimal solutions and weaker exploration ability in later stages.This study employed the ABC algorithm to optimize a WNN,and constructed an ABC-WNN analytical model.With the excavation of foundation pit as the case for study,the model was utilized to predict deformation induced by deep excavation in an existing shield tunnel.Predictions were then compared against those from standalone BP neural network and WNN models in terms of mean squared error and average absolute error.The results reveal that:(1)The ABC-WNN model exhibits high goodness-of-fit with actual engineering data,with a maximum relative error of merely 2.41×10-5,indicating reliable predictive capability.(2)Statistical characteristics of the ABC-WNN model are the lowest,specifically with values of 0.557 and 0.563,signifying that the ABC-optimized wavelet neural network model provides higher prediction accuracy,improved computational stability,and faster convergence for tunnel deformation quantification.These findings offer a novel approach for predicting shield tunnel deformation in analogous engineering contexts.
万方数据
维普
