In order to accurately predict the settlement of CFG pile composite subgrade, the observation time, cumulative fill thickness, soft soil thickness, soft soil compression modulus and pile length are taken as input variables, and a BP prediction model with a network structure of 5-5-1 is constructed based on the MATLAB platform. The BP prediction model is optimized by particle swarm optimization and genetic algorithm, and then the measured data of the new construction project of Qiaotou village in Zhaoqing city is simulated, finally the prediction performances of two kinds of optimization models are compared with the ordinary BP model. The results show that it is feasible to use PSO-BP and GA-BP prediction models to predict the settlement of CFG pile composite subgrade, and the prediction accuracy is high, and the prediction results are significantly better than the ordinary BP settlement prediction models. The results can provide some references for the settlement prediction of composite subgrade.