基于深层神经网络的原子钟钟差预测算法

Prediction Algorithm of Atomic Clock Difference Based on Deep Neural Network

原子钟的钟差预测在原子钟时间和频率控制中发挥着重要作用,它关系着时标计算的准确性和守时系统的稳定性,良好的钟差预测有助于实现高精度时间和频率控制。为进一步提升钟组时间的长期稳定度,提高原子钟的钟差预测精度,构建了一个适用于钟差预测的深度学习网络预测模型,分析了该网络模型超参数在钟差数据上的影响,并给出了最佳参数指标,即隐藏层层数为3层,隐藏层神经元个数为256个。基于UTC-clock(i)进行数据实验对比,深度学习网络预测模型相比于最小二乘法的均方根误差更小,结果表明,在长期预测上,优化后的深层网络模型比最小二乘法模型的预测效果提升了86.93%,该深层网络模型更适于高精度时间、频率控制,在守时系统钟差数据处理方面具有应用前景。

s The clock error prediction of atomic clock plays an important role in the time and frequency control of atomic clock,which is related to the accuracy of time scale calculation and the stability of the punctual system.In order to further improve the long-term stability of clock group time and improve the accuracy of clock error prediction of atomic clocks,a deep learning network prediction model suitable for clock error prediction is constructed.The influence of the network model's hyperparameters on clock error data is analyzed.And the best parameter index is given.That is,the number of hidden layers is 3,and the number of hidden layer neurons is 256.The results show that in terms of long-term prediction,the prediction effect of the optimized deep network model is 86.93% higher than that of the least squares model,and the deep network model is more suitable for high-precision time and frequency control,and has application prospects in the processing of clock error data in punctual systems.