Reasonable prediction of foundation settlement is of great significance to saving construction cost, ensuring construction quality and ensuring the safety of construction process.In neural network to reduce the error of gradient descent, easy to fall into local optimum and higher requirements for the number of original data and shortcomings of simulated annealing optimization model is put forward, realize the prediction of finding the global optimal and focus on the time delay neural network are optimized by the coordination factor (FTDNN) and nonlinear regression neural network (NARNN) the accuracy of the two models, Improve the prediction error caused by insufficient original data.The results show that the accuracy of the simulated annealing optimized prediction model is significantly higher than that of the two neural networks alone, which has certain engineering significance.