基于CEEMDAN-LSTM的陶岔渠首水深预测  

Research on Water Depth Prediction of Taocha Canal Headwork Based on CEEMDAN-LSTM Model

作　　者：陈伟 吕学斌 梁雪春 CHEN Wei;LYU Xuebin;LIANG Xuechun(College of Mathematics and Science,Nanjing Tech University,Nanjing 211816,China;College of Electrical Engineering and Control Science,Nanjing Tech University,Nanjing 211816,China)

机构地区：[1]南京工业大学数理科学学院,江苏南京211816 [2]南京工业大学电气工程与控制科学学院,江苏南京211816

出　　处：《人民黄河》2022年第5期89-94,共6页Yellow River

基　　金：国家自然科学基金青年基金资助项目(11801267);江苏省研究生科研与实践创新计划项目(SJCX21_0427)。

摘　　要：采用基于自适应噪声的完全经验模态分解-长短期记忆神经网络模型(CEEMDAN-LSTM)对水深进行预测。首先利用中位值平均滤波法对数据进行预处理,再采用CEEMDAN方法对历史水深序列进行分解获得历史水深的高、低频以及残差序列,然后对得到的各个分量使用LSTM神经网络预测,最后叠加各分量预测值重构水深预测结果。以陶岔渠首为研究对象,CEEMDAN-LSTM模型测试结果表明:该模型相比支持向量机回归、BP神经网络、长短期记忆神经网络、经验模态分解-长短期记忆神经网络模型有更强的预测性能。Taking the unique high complexity and non-linear characteristics of water level data into consideration, this paper adopted the complete ensemble empirical mode decomposition with adaptive noise and long short-term memory network model(CEEMDAN-LSTM)to predict the water level. Firstly, it applied the median average filtering method for data preprocessing and CEEMDAN method for historical water level sequence. Then, it used the LSTM neural network to predict the obtained components. After that, it superimposed the predicted values of each IMF to reconstruct the water level prediction results. Taking the monitoring Taocha canal headwork as an example, the results of CEEMDAN-LSTM in the test set show that the model has stronger predictive performance, compared with SVR, BP neural network, LSTM and EMD-LSTM models.

关 键 词：水深预测 中位值平均滤波法 CEEMDAN分解 LSTM神经网络模型 陶岔渠首 

分 类 号：TV213.4[水利工程—水文学及水资源]