基于递归神经网络的风暴潮增水预测

发布时间：2018-02-17 01:38

本文关键词： 风暴潮增水预测数值预报机器学习静态数据时序特性 BP神经网络递归神经网络　出处：《智能系统学报》2017年05期 　论文类型：期刊论文

【摘要】：风暴潮增水的准确预测能极大地减少人员伤害和经济损失,具有重要的实用价值。传统的风暴潮预报方法主要包括经验和数值预报,很难建立起相对准确的模型。现有的基于机器学习风暴潮预报方法大都只提取出静态数据间的关系,并没有充分挖掘出风暴潮数据背后的时序关联特性。文中提出了一种基于递归神经网络的风暴潮增水预测方法。本文对风暴潮时序数据进行特定的处理,并设计合适结构的递归神经网络,从而完成时序数据的预测。相较于传统的BP神经网络,递归神经网络能更好地应对时序数据的预测问题。将该方法用于潍坊水站的增水预测中,结果表明,相对于BP神经网络,递归神经网络能得到更好的预测结果,误差更小。
[Abstract]:The accurate prediction of storm surge increase water can greatly reduce human injury and economic loss, which is of great practical value. The traditional storm surge forecasting methods mainly include experience and numerical forecast. It is difficult to build a relatively accurate model. Most of the existing methods based on machine learning for storm surge prediction only extract the relationship between static data. The time series correlation characteristics behind storm surge data are not fully excavated. In this paper, a prediction method of storm surge increase water based on recurrent neural network is proposed. In this paper, storm surge time series data are specially processed. A recursive neural network with suitable structure is designed to predict the time series data. Compared with the traditional BP neural network, The recursive neural network can better deal with the prediction of time series data. The results show that the recurrent neural network can get better prediction results and the error is smaller than BP neural network.
【作者单位】：北京航空航天大学宇航学院图像处理中心;国家海洋局北海预报中心;
【基金】：国家自然科学基金项目(61671037)
【分类号】：P731.34;TP183

【相似文献】