基于静态与动态神经网络的运河水位预报

发布时间：2018-05-02 19:02

本文选题：城市运河 + 人工神经网络　；参考：《天津大学学报(自然科学与工程技术版)》2017年03期

【摘要】：以杭州市上塘河运河为例,分别建立自回归、静态时延神经网络及动态反馈神经网络模型预报受闸门及泵站影响的城市河道未来1~3,h的水位变化,并对模型预报误差、不同水位区间下的误差-频数关系及训练数据的不确定性进行分析.研究结果表明:在预见期1~3,h内,时延神经网络预报效率系数均达到0.9以上,比自回归模型分别提高1.34%,、5.57%,和6.86%,,比动态反馈网络分别提高0.21%,、1.97%,和1.98%,;动态网络模型在人为调控的影响下仍能通过训练模拟出降雨径流关系,对数据具有更好的自学习与调适能力;时延神经网络模型随数据减少预测精度也减少,在仅保留1个洪水场次下效率系数最大降低8.05%,,而动态反馈网络效率系数随训练数据量变化基本在0.11%,内波动,因此在数据量较少的情况下宜建构动态模型.
[Abstract]:Taking the Shangtang River Canal in Hangzhou as an example, the autoregressive, static time-delay neural network and dynamic feedback neural network are established to predict the water level change of the urban river affected by the sluice gate and pumping station in the future, and the prediction error of the model is given. The error-frequency relation and the uncertainty of training data are analyzed in different water levels. The results show that the prediction efficiency coefficients of the delayed neural networks are more than 0.9 in the predicted period of 1 ~ 3 h. Compared with the autoregressive model, the dynamic network model increases 5.57 and 6.86, 0.21 and 1.97, and 1.98, respectively, compared with the dynamic feedback network, and the dynamic network model can still simulate the rainfall runoff relationship through training under the influence of artificial regulation and control, and has better self-learning and adjusting ability to the data. The prediction accuracy of time-delay neural network model decreases with the decrease of data, and the efficiency coefficient decreases by 8.05 when only one flood event is retained, while the dynamic feedback network efficiency coefficient changes basically at 0.11 with the training data, and fluctuates internally. Therefore, the dynamic model should be constructed in the case of less data.
【作者单位】：浙江大学建筑工程学院;杭州市河道管理总站;
【基金】：浙江省水利科技计划资助项目(H20151104) 教育部博士点新教师资助项目(J20131413)~~
【分类号】：U612.23

【相似文献】