Analysis of Several Optimization Strategies of BP Neural Network in GNSS Height Anomaly Fitting

Several optimization algorithms based on the Back Propagation (BP) neural network are often used to improve the fitting accuracy of GNSS height anomaly. However, the performance and applicability of these algorithms have not been systematically studied. This study takes two kinds of GNSS/leveling points with different spatial distributions as examples, evaluates the performance and applicability of commonly used genetic algorithm (GA), simulated annealing algorithm (SA), particle swarm optimization (PSO), and the latest sparrow search algorithm (SSA) methods in improving the fitting accuracy of BP neural network. Additionally, this study also attempts to add the input data related to terrain for the first time to analyze whether it can improve the fitting accuracy under uneven point distribution. Results show that these optimization algorithms can significantly improve the accuracy and stability of the BP neural network fitting results, and PSO and SSA methods indicate the best improvement. Additionally, when the distribution of fitting points is uneven, adding appropriate input data can also effectively improve the reliability of the BP neural network.