基于BP神经网络的霍普菲尔德模型改进研究

摘要

为了分析对流层延迟的时空变化规律、提高对流层延迟的改正精度,利用BP神经网络处理非线性问题的优势,改进传统的霍普菲尔德模型得到一种新的融合模型(Hop+BP模型)。分别对比 Hop+BP模型与传统的霍普菲尔德模型、多元线性回归模型、BP 神经网络等模型的计算结果,得到如下结论：霍普菲尔德模型存在一个明显的系统误差,精度较低；多元线性回归的预测精度有所提高,但是其本质是将数据强制拟合,缺少物理解释,难以推广使用；传统的BP神经网络的计算精度较之霍普菲尔德模型有80%的提高,但存在明显的不稳定性；Hop+BP 模型具有预测精度高、稳定性好等优点,预测中误差为1.1 cm,明显优于传统方法。%In order to analyze the law of temporal and spatial variation of troposphere wet delay and promote the amendatory precision of troposphere delay,this paper improves the traditional Hopfield troposphere model to form a new fusion model(Hop+BP model)by the advantage of Back Propagation neural network for solving nonlinear problems.Through comparing the fusion model with traditional Hopfield troposphere model,multiple linear regression model,and Back Propagation neural network model respectively,this paper draws the following conclusion:Hopflied troposphere model has a clear systematic error with a lower precision;multiple linear regression model has a higher precision,but with the nature of mandatory fitting, which lack of physical interpretation thus being difficult for popularization and application;the computational precision of Back Propagation neural network model is 80%,which is higher than Hopfield troposphere model,but has obvious instability;the fusion model embraces the advantages of high precision and stability,and with a deviation of 1.1 centimeter,which is evidently better than traditional methods.