基于改进RBF神经网络的GNSS高程拟合

GNSS Height Fitting Based on Improved RBF Neural Network

针对传统的RBF神经网络模型在GNSS高程拟合中拟合精度较低、稳定性较差、相关因子需提前人为设置等问题,通过将改进的自适应权重粒子群优化算法与MATLAB RBF神经网络函数newrb相结合,实现RBF神经网络函数模型中隐含节点数和SPREAD值的自动优化选取,提高算法在GNSS高程拟合中的精度和稳定性。通过实例分析,该方法拟合精度高,可达到mm级精度,相对于传统的二次多项式模型精度提高17%,稳定性良好。

In the traditional RBF neural network model in the GNSS height fitting,the fitting accuracy is relatively low,the stability is relatively poor,and the correlation factors need to be set artificially in advance.This paper adopts the improved adaptive weight particle swarm optimization algorithm and MATLAB RBF newrb.The network function newrb combines to realize the automatic optimization of the number of hidden nodes and SPREAD in the RBF neural network function model,and improve the accuracy and stability of the algorithm in GNSS height fitting.Through the example analysis,the method has high fitting precision and can reach mm precision.Compared with the traditional quadratic polynomial model,the accuracy is improved by 17%and the stability is good.It has important reference value for accurately solving GNSS height anomaly.