Abstract: [Purposes] To address the deficiencies of traditional geological subsidence prediction methods in nonlinear data processing and the issue of relying on experience in selecting parameters for RBF neural networks,and to achieve accurate prediction of subsidence along transmission lines. [Methods] An intelligent prediction model based on an improved particle swarm optimization algorithm to optimize the RBF neural network is proposed. [Findings] Comparative experiments using subsidence data from the Kunming transmission corridor show that the PSO-RBF model outperforms comparative models such as BP,RBF,PSO-BP,SVR,and LSTM in multiple indicators including MAE,RMSE,MAPE,and R²,with highly consistent prediction results and good generalization ability. [Conclusions] The superiority and engineering applicability of the PSO-RBF model in geological subsidence prediction are verified,providing an effective method for related intelligent early warning.

Key words: particle swarm optimization, RBF neural network, settlement prediction, adaptive parameter optimization, intelligent warning