基于遗传优化BP神经网络的GPS高程异常拟合模型研究

GPS Elevation Abnormality Fitting Model Based on Genetic Optimization BP Neural Network

为解决标准BP神经网络模型存在的易陷入极小值、训练时间长、网络不稳定等问题,采用基于实数编码的遗传算法,优化网络初始权值和阈值,构建GPS高程异常拟合模型。通过实测数据进行计算分析,并将该模型的结果与平面拟合、二次曲面拟合及标准BP神经网络模型所得结果比较,得出如下结论:使用遗传优化BP神经网络进行高程异常拟合,模型误差和中误差均较小,故基于遗传优化BP神经网络模型具有较高的精度和较好的稳定性,可以应用于GPS高程异常拟合问题。

In order to solve the problems such as falling into minimum value easily, long training time, instability of net- work and so on, the genetic algorithm based on real-coding is adopted to optimize the network＇ s initial weights and threshold value, and build the GPS elevation abnormality fitting model. Using the measured data for calculation and anal- ysis, the result of the model is compared with the results of plane fitting model, quadratic curved-surface fitting model and standard BP neural network model, then the conclusions are dwawn as follows： when using the genetic optimization BP neural network for height abnormality fitting, the model error and mean-square error are all smaller, so the BP neural network model based on genetic optimization has higher precision and better stability, and can be applied to GPS height abnormality fitting.