基于深度信念网络的钱塘江潮位预测方法

A Method for Tide Prediction of Qiantang River Based on Deep Belief Network

浅层网络不能很好地挖掘钱塘江各站点间潮位数据及其他数据的原始特征关系,其浅层网络模型学习后获得的是没有层次结构的单层特征,影响预测数据重构精度,为此,提出了一种基于深度学习的钱塘江潮位预测方法。首先利用多个受限玻尔兹曼机RBM提取钱塘江各水文站点的潮位数据以及农历"日"的数据特征值,然后在网络的最后一层设置BP神经网络实体分类器,将RBM提取到的特征数据作为BP神经网络的输入特征向量,利用BP神经网络对实体关系分类器进行有监督地训练,最后将误差反向传播至每一层RBM来微调整个网络参数,最终建立深度信念网络DBN预测模型。实验结果表明,深度信念网络预测模型比神经网络预测模型预测钱塘江潮位更具稳定性,为钱塘江潮位的预测提供了一种方法。

In view of the shallow network can not dig the Qiantang River tidal data and other data original characteristics,and monolayer feature without hierarchical structure obtained via shallow network model learning,which affect the reconstruction accuracy of the predicted data.A method for predicting the tidal level of the Qiantang River based on deep learning was presented in this paper.Firstly the multiple limited Boltzmann machine(RBM)was used to extract the Qiantang River hydrological site of the tide data and the lunar calendar“day”data eigenvalues.Secondly BP neural network entity classifier was set in the last layer of the network.Thirdly the feature data extracted by RBM was input the BP neural network to supervise the training entity relations classifier,and calculation error was backed to each layer RBM to fine tune the entire network parameters.Finally the depth belief network(DBN)prediction model was established.The experiment results show that the DBN prediction model was more stable to predict tide of Qiantang River,which provides a method for the prediction of Qiantang River tide.