Abstract: Aiming at the problem that the accuracy of constructing GNSS height anomaly fitting model in a complex terrain area is limited, a method based on the reptile search algorithm(RSA) is proposed to optimize the BP neural network. The RSA is used to solve the problems of local extremes and gradient descent of BP neural networks by global optimization of the weights and thresholds of neurons between the layers of traditional BP neural networks. At the same time, the height data of encrypted points above the third class level survey accuracy were selected as the sample set and learned and trained using RSA-BP neural network. Compared with the least squares support vector machine and the multi-surface function fitting performance, the RSA-BP neural network model has the highest fitting accuracy, the best stability and the best fit with the actual height anomalies.

Key words: reptile search algorithm, BP neural network, fitting model of heigth anomaly, geodetic height, normal height