摘要
针对在钟差预报中小波神经网络预报结果的不稳定性问题,提出一种粒子群优化小波神经网络的钟差预报模型,以提高预研结果稳定性。该模型是将小波神经网络的各阈值和权值作为粒子的位置向量,通过粒子群算法寻求网络的阈值和权值的最优值,降低了网络出现局部极值的概率,从而改善了小波神经网络预报结果的稳定性。通过分析粒子群算法最优个体适应度值变化曲线图、比较粒子群算法优化前后WNN模型训练误差以及两种模型多次预报结果,验证了模型的有效性。并与二次多项式(QP)及灰色(GM(1,1))两种常用模型进行实验对比。实验结果表明:在24小时钟差预报中,所提方法相比于两种常用模型预报的精度分别提高了约49.7%和66%,并且预报得到钟差精度相比于超快速钟差产品提高了97.7%,质量明显优于超快速钟差产品。
Aiming at the instability of wavelet neural network prediction results in clock bias prediction,a clock bias prediction model based on particle swarm optimization wavelet neural network is proposed to improve the stability of prediction results.In the model,each threshold and weight of the wavelet neural network is taken as the position vector of the particle,and the optimal value of the threshold and weight of the network is sought by the particle swarm optimization algorithm,which reduces the probability of local extreme value of the network,and thus improves the stability of the prediction results of the wavelet neural network.The effectiveness of the model is verified by analyzing the variation curve of the fitness value of the optimal individual of PSO,comparing the training error of WNN model before and after PSO optimization,and multiple prediction results of the two models.Compared with two common models,quadratic polynomial(QP)and grey(GM(1,1)),the experimental results show that the accuracy of the proposed method is improved by 49.7%and 66%respectively,the accuracy of the predicted clock bias is improved by 97.7%compared with the ultra-fast clock bias product,and the quality of the predicted clock bias is obviously better than that of the ultra-fast clock bias product.
作者
王旭
张文
柴洪洲
WANG Xu;ZHANG Wen;CHAI Hongzhou(School of Resources and Civil Engineering,Liaoning Institute of Science and Technology,Benxi 117004,China;University of Science and Technology Liaoning,School of Civil Engineering,Anshan 114051,China;Institute of Surveying and Mapping,Zhengzhou 450001,China)
出处
《中国惯性技术学报》
EI
CSCD
北大核心
2023年第1期33-39,共7页
Journal of Chinese Inertial Technology
基金
国家自然科学基金项目(41604013)
2022年辽宁省教育厅基本科研项目资助(LJKMZ20221686)。
关键词
卫星钟差
粒子群算法
精密单点定位
小波神经网络
satellite clock bias(SCB)
particle swarm optimization
precise point positioning
wavelet neural network
