基于门控循环网络的海浪波倾角预测研究

Research on prediction of slope of wave based on GRU network

海浪的波倾角是一种非线性随机时间序列,对于舰载机的起降和舰载稳定平台的控制具有重要的指导意义。传统的时间序列预测方法诸如自回归滑动平均预测、神经网络预测法等,有时无法提供较高的精度。因此,提出一种基于门控循环网络的海浪模型时间序列预测法。结果表明,在平均绝对百分比误差上分别比BP神经网络、循环神经网络(RNN)、长短期记忆网络(LSTM)下降了85.0%、80.3%、34.4%;在均方根误差上分别下降了27.1%、37.7%和7.5%。与传统方法相比,门控循环网络在处理长依赖时间序列上更有优势,预测精度更高。

Slope of wave is a kind of non-linear random time series, which has important guiding significance for the takeoff and landing of carrier-based aircraft and the control of ship-based stabilized platform. Traditional time series prediction methods, such as Autoregressive and Moving average prediction and neural network prediction, sometimes fail to provide high accuracy. Therefore, a time series prediction method based on gated recurrent network (GRU) is proposed in this paper. The results show that the average absolute percentage error ( MAPE) is 85? 0%, 80. 3% and 34. 4% lower than BP, RNN and LSTM, and the root mean square error (RMSE) is 27. 1%, 37. 7% and 7. 5% lower than BP, RNN and LSTM. Compared with traditional methods, GRU network has more advantages in dealing with longdependent time series and get higher prediction accuracy.