摘要
为解决径流预测中高维非线性特征捕捉困难与模型参数优化复杂的问题,提出了融合Circle混沌映射与高斯变异策略的改进蛇鹭优化算法(ISBOA),并以此对构建的CNN-BiGRU-Attention深度学习耦合模型进行超参数智能寻优。利用黄河上游唐乃亥水文站径流数据,在月、半月、旬尺度及1~3步长预见期下进行模拟验证,并与CNN、GRU、CNN-BiGRU等多种基准模型进行对比。结果表明:经ISBOA优化后的CNN-BiGRU-Attention模型预测精度最高、稳定性最好,显著优于优化前的模型及其他对比模型,充分验证了所提ISBOA算法在多时间尺度径流预测模型优化中的有效性与优越性。
To address the challenges in capturing high-dimensional nonlinear features and optimizing complex model parameters for runoff prediction,this study proposes an Improved Snake-Eagle Optimization Algorithm(ISBOA)that integrates a Circle chaotic map and a Gaussian mutation strategy.The proposed ISBOA is then employed to intelligently optimize the hyperparameters of a constructed deep learning coupled model,CNN-BiGRU-Attention.Using runoff data from the Tangnaihai Hydrological Station in the upper reaches of Yellow River,simulation validations are conducted across multiple temporal scales(monthly,half-monthly,and ten-day)and with forecast lead times of 1-3 steps.The performance is also benchmarked against various baseline models,including CNN,GRU and CNN-BiGRU.The results indicate that the CNN-BiGRU-Attention model optimized by ISBOA achieves the highest prediction accuracy and stability,significantly outperforming its unoptimized counterpart and other comparative models,which thoroughly validates the effectiveness and superiority of the proposed ISBOA algorithm in optimizing runoff prediction models across multiple time scales.
作者
杨小鹏
周千涵
侯添甜
王心怡
胡可意
纪徐洋
朱非林
YANG Xiaopeng;ZHOU Qianhan;HOU Tiantian;WANG Xinyi;HU Keyi;JI Xuyang;ZHU Feilin(College of Hydrology and Water Resources,Hohai University,Nanjing 210024,Jiangsu,China)
出处
《水力发电》
2026年第4期9-18,68,共11页
Water Power
基金
国家自然科学基金资助项目(52009029)
中央高校基本科研业务费专项资金资助(B240201123)
中国长江电力股份有限公司科研项目资助(Z242302021)。
关键词
径流预测
卷积神经网络
双向门控循环单元
注意力机制
改进的蛇鹭优化算法
runoff prediction
Convolutional Neural Network
Bidirectional Gated Recurrent Unit
Attention mechanism
Improved Snake-Eagle Optimization Algorithm
