基于改进PSO-BP模型的钟差预报研究  

Research on Clock Difference Prediction Based on Improved PSO-BP Model

作　　者：张颖博 刘音华[1,2] 刘娅 ZHANG Yingbo;LIU Yinhua;LIU Ya(National Timing Center,Chinese Academy of Sciences,Xi'an 710600,China;University of Chinese Academy of Sciences,Beijing 101408,China)

机构地区：[1]中国科学院国家授时中心,西安710600 [2]中国科学院大学,北京101408

出　　处：《宇航计测技术》2024年第1期41-47,共7页Journal of Astronautic Metrology and Measurement

摘　　要：针对BP神经网络训练时陷入局部最优解导致预报钟差不稳定的问题,采用改进粒子群优化神经网络的钟差预报模型。首先改进粒子群优化算法中几个重要参数生成的方法,再将BP神经网络的初始权值和阈值作为粒子的位置,通过改进的粒子群优化算法迭代,寻找网络的最优初始权值和阈值,提高BP神经网络钟差预报的稳定性和准确性。从理论上分析改进后的PSO算法原理,利用改进后的模型预测钟差,经过分析全局最优适应度曲线、粒子群优化前后BP模型多次预报钟差的试验,证明该算法优化的有效性。与ARMA和GM(1,1)等传统的预报模型相比,基于改进的粒子群优化神经网络模型的钟差预报精度分别提高了86.5%和79%。In order to solve the problem that the clock error prediction caused by falling into the local optimal solution during the training process of BP neural networks,the improved particle swarm optimization BP neural network is used for the clock difference forecast model.Firstly,the method of generating several essential parameters in the particle swarm optimization algorithm is improved.Then the initial weights and thresholds of the BP neural network are used as the positions of the particles.The improved particle swarm optimization algorithm iteratively searches the optimal initial weights and thresholds of the network to improve the clock error prediction of BP neural networks,stability and accuracy.This paper analyzes the improvement principle and uses this model to predict clock error,which proves the effectiveness of the optimization of the algorithm after analyzing the global optimal fitness curve and the experiments of the BP model before and after the particle swarm optimization many times,forecasting the clock difference.Compared with the traditional forecasting models such as the ARMA model and CM(1,1)mode,the accuracy of the clock difference forecast based on the improved particle swarm optimization neural network model is improved by 86.5%and 79%,respectively.

关 键 词：钟差预报 粒子群优化算法 BP神经网络 

分 类 号：P127.1[天文地球—天体测量]