基于CNN-BiLSTM的潮汐电站潮位预测  

Tide Level Prediction for Tidal Power Station Based on CNN-BiLSTM Network Model

作　　者：黄冬梅 王唱 胡安铎 孙锦中 孙园 李俊峰 HUANG Dongmei;WANG Chang;HU Anduo;SUN Jinzhong;SUN Yuan;LI Junfeng(College of Electronics and Information Engineering,Shanghai University of Electric Power,Shanghai 201306,China;College of Mathematics and Physics,Shanghai University of Electric Power,Shanghai 200090,China;State Power Investment Nanyang Thermal Power Co.,Ltd.,Nanyang 473000,Henan,China)

机构地区：[1]上海电力大学电子与信息工程学院,上海201306 [2]上海电力大学数理学院,上海200090 [3]国电投南阳热电有限责任公司,河南南阳473000

出　　处：《水力发电》2021年第10期80-84,共5页Water Power

基　　金：上海市科委地方院校能力建设项目(20020500700)。

摘　　要：潮汐电站的优化运行需要进行潮位预测。针对传统调和分析方法不能有效处理潮位非线性和非平稳的特性的问题,提出一种CNN-BiLSTM的预测模型,以滑动数据窗口构造潮位数据的特征图作为输入,利用1D CNN提取潮位数据深层特征,然后采用BiLSTM网络生成特征描述,最后输出预测结果。采用芝加哥港口的潮汐数据进行了实验,所提预测模型与调和分析及LSTM模型相比,均方根误差分别降低了66.26%和30.11%。算例结果表明CNN-BiLSTM模型可以实现高精度的短期潮位预测。The optimal operation of tidal power stations requires tide level prediction.Aiming at the problem that traditional harmonic analysis methods cannot effectively deal with the non-linear and non-stationary characteristics of tide level,a CNN-BiLSTM prediction model is proposed.The model uses a sliding data window to construct the feature map of tide level data as input and uses 1D CNN to extract the deep-layer feature of tide level data,then the BiLSTM network is exploited to generate description of feature and output the final prediction results.The experiments are carried out using the tide data from Chicago Port.Compared with the harmonic analysis and LSTM model,the proposed prediction model can reduce the RMSE by 66.26%and 30.11%respectively.The results of the calculation example show that the CNN-BiLSTM model can achieve high-precision short-term tide level prediction.

关 键 词：潮汐电站 潮位预测 调和分析 卷积神经网络 双向长短期记忆神经网络 

分 类 号：TV752[水利工程—水利水电工程]