摘要
受降水量、径流等因素的影响,水库的长期水位预测面临巨大挑战。提出了一种新的基于长短期记忆(Long Short Term Memory, LSTM)网络的时间序列模型,对沂沭泗流域中的石梁河水库水位进行了预测和性能评价。该模型整合了降雨、水流和土壤含水量等历史信息,并通过实验获取最优预测步长,从而提高了模型的预测准确度,并且稳定性更好,避免出现较大的误差。实验使用Nash-Sutcliffe效率(NSE)、Pearson相关系数平方(R;)和绝对均方根误差(Root Mean Square Errors, RMSE)等评价指标,与基本的多层感知机模型和卷积神经网络(Convolutional Neural Networks, CNN)比较,得出如下结论:① LSTM模型的预测值不存在明显较小的波峰或波谷;②模型的预测精度不会随着预测时间步长的增加而急剧下降;③在真实的洪水事件预测中,雨量较小时不会引起预报线的波动,且预测洪峰时偏离度较小。当然,如何在大规模流域中应用该模型,以及对流域中的多个水库水位同时预测等问题,将在未来的工作中进行进一步的研究和分析。
Due to the influence of precipitation, runoff and other factors, the long-term prediction of reservoir water level is faced with great challenges.A novel time series model based on Long Short Term Memory(LSTM) network is proposed to predict the water level and evaluate the performance of Shiliang River reservoir in Yishusi Basin.This model integrates historical information such as rainfall, water flow and water content in soil, and obtains the optimal prediction step through experiments, so as to improve the prediction accuracy of the model with better stability and avoid large errors.In the experiments, Nash-Sutcliffe Efficiency(NSE),Squared Pearson Correlation Coefficient(R;),Root Mean Square Errors(RMSE) and other evaluation criteria are used to make comparison with the basic Multilayer Perceptrons(MLP) model and the Convolutional Neural Network(CNN) model.It is concluded that:① The predicted values of LSTM model do not have significantly smaller peaks or troughs;② The prediction accuracy of the model does not decrease sharply with the increase of the prediction time step;③ In the prediction of real flood events, small rainfall does not cause the fluctuation of forecast line, and the deviation degree of flood peak prediction is small.Of course, how to apply the model in large-scale basins and how to predict the water level of multiple reservoirs simultaneously will be further studied and analyzed in the future.
作者
刘威
尹飞
LIU Wei;YIN Fei(School of Networking and Communication Engineering,Jinling Institute of Technology,Nanjing 211169,China;Water Conservancy Bureau of Lianyungang City,Lianyungang 222006,China)
出处
《无线电工程》
北大核心
2022年第1期83-87,共5页
Radio Engineering
基金
国家自然科学基金(41801303)。
关键词
水位预测
深度学习
长短期记忆模型
洪水预测
water level prediction
deep learning
long short term memory
flood prediction
