以霜冰优化算法优化CNN-BiLSTM-Attention的参考蒸散量估算

doi:10.7522/j.issn.1000-694X.2025.00167

摘要/Abstract

摘要：

有限气象参数条件下借助于深度学习实现蒸散量的准确估算对干旱区有限水资源的高效利用和管理具有重要意义。当前基于混合深度学习模型CNN-BiLSTM-Attention的蒸散发估算忽视了参数优化，导致估算精度难以契合实际应用需求。本文提出了一种新的霜冰优化算法（RIME）优化CNN-BiLSTM-Attention的超参数的混合模型RIME-CNN-BiLSTM-Attention，实现了有限气象参数条件下临泽县参考蒸散量（ET₀）的准确预测。与CNN-BiLSTM-Attention相比，混合模型RIME-CNN-BiLSTM-Attention的平均绝对百分比误差（MAPE）从14.56%下降到14.09%，可决系数从0.8654上升到0.8930。此外，数值结果表明混合模型RIME-CNN-BiLSTM-Attention的模型性能优于分别采用哈里斯鹰优化算法（HHO）、鱼鹰优化算法（OOA）、北方苍鹰算法（NGO）对CNN-BiLSTM-Attention进行优化的混合模型HHO-CNN-BiLSTM-Attention、OOA-CNN-BiLSTM-Attention、NGO-CNN-BiLSTM-Attention，意味着所构建混合模型RIME-CNN-BiLSTM-Attention具有更加稳健的模型性能和更高的计算精度，能够实现研究区域ET₀的准确估算。

关键词: 参考蒸散量, 霜冰优化算法, 卷积神经网络, 双向长短期记忆网络, 注意力机制

Abstract:

Accurately estimating evapotranspiration using deep learning models under limited meteorological parameter conditions is of great significance for the efficient utilization and management of limited water resources in arid regions. However， current deep-learning-based hybrid model CNN-BiLSTM-Attention neglect the parameter optimization， making it difficult to meet the requirements of practical application. In this study， a novel hybrid model RIME-CNN-BiLSTM-Attention was proposed to achieve accurate prediction of daily reference evapotranspiration （ET₀） in Linze County under limited meteorological parameter conditions by using the RIME Optimization Algorithm to optimize the hyper-parameter of CNN-BiLSTM-Attention. Compared with CNN-BiLSTM-Attention， the mean absolute percentage error （MAPE） of RIME-CNN-BiLSTM-Attention decreases from 14. 56% to 14. 09%， and the coefficient of determination increases from 0. 8654 to 0. 8930； Furthermore， the numerical results show that the hybrid model RIME-CNN-BiLSTM-Attention outperformed than that of HHO-CNN-BiLSTM-Attention， OOA-CNN-BiLSTM-Attention， and NGO-CNN-BiLSTM-Attention， where CNN-BiLSTM-Attention was separately optimized by the Harris Hawks Optimization （HHO）， Osprey Optimization Algorithm （OOA）， and Northern Goshawk Optimization （NGO）， suggesting that the proposed hybrid model RIME-CNN-BiLSTM-Attention has more robust model performance and higher calculation accuracy， and can achieve accurate estimation of ET₀ in the study area.

Key words: reference evapotranspiration, RIME optimization algorithm, convolutional neural network, bidirectional long short-term memory network, attention mechanism

中图分类号:

P246.2

付桐林, 金晶. 以霜冰优化算法优化CNN-BiLSTM-Attention的参考蒸散量估算[J]. 中国沙漠, 2025, 45(3): 302-312.

Tonglin Fu, Jing Jin. Estimating reference evapotranspiration using CNN-BiLSTM-Attention enhanced by RIME optimization algorithm[J]. Journal of Desert Research, 2025, 45(3): 302-312.

图/表 7

图1 BiLSTM的结构和LSTM的细胞

Fig.1 Structure of BiLSTM and cells of LSTM

图2 RIME-CNN-BiLSTM-Attention的流程图

Fig.2 The flow chart of RIME-CNN-BiLSTM-Attention

图3 RIME-CNN-BiLSTM-Attention、HHO-CNN-BiLSTM-Attention 、OOA-CNN-BiLSTM-Attention、NGO-CNN-BiLSTM-Attention及CNN-BiLSTM-Attention的估算结果

Fig.3 The forecasting results of RIME-CNN-BiLSTM-Attention，HHO-CNN-BiLSTM-Attention，OOA-CNN-BiLSTM-Attention，NGO-CNN-BiLSTM-Attention，and CNN-BiLSTM-Attention

表1 启发式算法优化CNN-BiLSTM-Attention所得的最优参数

Table 1 The optimal parameters of CNN-BiLSTM-Attention

模型	学习率	神经元个数	键值	正则化参数
RIME-CNN-BiLSTM-Attention	0.0096	18	13	0.0005
HHO-CNN-BiLSTM-Attention	0.0063	16	8	0.0003
OOA-CNN-BiLSTM-Attention	0.0096	18	13	0.0005
NGO-CNN-BiLSTM-Attention	0.0078	20	23	0.0006

表2 各模型在测试阶段的性能评价指标

Table 2 The performance metrics of each model in testing stage

模型	NSCE	MSE	RMSE	MAE	MAPE/%	R²
RIME- CNN-BiLSTM-Attention	0.8830	0.2879	0.5366	0.4013	14.09	0.8930
OOA- CNN-BiLSTM-Attention	0.8816	0.3249	0.5700	0.4349	14.34	0.8816
NGO- CNN-BiLSTM-Attention	0.8796	0.3303	0.5747	0.4352	14.28	0.8796
HHO- CNN-BiLSTM-Attention	0.8769	0.3378	0.5812	0.4482	14.38	0.8769
CNN-BiLSTM-Attention	0.8754	0.3548	0.5957	0.4761	14.56	0.8654

图4 误差评价指标的柱形图

Fig.4 Column chart of error evaluation indicators

图5 各模型评价指标的雷达图

Fig.5 Radar diagrams of each model evaluation index

参考文献 25

1	任珩，赵文智，杨荣，等.河西走廊绿洲农业水生产力提升的途径与对策［J］.中国沙漠，2024，44（5）：217-224.
2	邢铭强，马可，陈彩亮，等.河西地区农业绿色发展水平测度及耦合协调提升路径［J］.中国沙漠，2024，44（6）：207-219.
3	Fu T L， Li X R， Jia R L，et al.A novel integrated method based on a machine learning model for estimating evapotranspiration in dryland［J］.Journal of Hydrology，2021，603：126881.
4	Fu T L， Li X R.Hybrid the long short-term memory with whale optimization algorithm and variational mode decomposition for monthly evapotranspiration estimation［J］.Scientifc Reports，2022，12：20717.
5	Chen S， He C， Huang Z，et al.Using support vector machine to deal with the missing of solar radiation data in daily reference evapotranspiration estimation in China［J］.Agricultural and Forest Meteorology，2022，316：108864.
6	Bilali E A， Hadri A， Taleb A，et al.A novel hybrid modeling approach based on empirical methods，PSO，XGBoost，and multiple GCMs for forecasting long-term reference evapotranspiration in a data scarce-area［J］.Computers and Electronics in Agriculture，2025，232：110106-110116.
7	Dong J， Xing L W， Cui N B，et al.Estimating reference crop evapotranspiration using optimized empirical methods with a novel improved Grey Wolf Algorithm in four climatic regions of China［J］.Agricultural Water Management，2024，108620.
8	Wu L F， Huang G M， Fan J L，et al.Hybrid extreme learning machine with meta-heuristic algorithms for monthly pan evaporation prediction［J］.Computers and Electronics in Agriculture，2020，168：105-115.
9	Farshad A， Saeid M， Babak M，et al.Application of an artificial intelligence technique enhanced with intelligent water drops for monthly reference evapotranspiration estimation［J］.Agricultural Water Management，2021，244：106622.
10	Roy D K， Sarkar T K， Kamar S S A，et al.Daily prediction and multi-step forward forecasting of reference evapotranspiration using LSTM and Bi-LSTM models［J］.Agronomy，2022，12：594.
11	Fu T L， Li X R.Estimating the monthly pan evaporation with limited climatic data in dryland based on the extended long short-term memory model enhanced with meta-heuristic algorithms［J］.Scientifc Reports，2023，13（1）：5960-5969.
12	Farzad R， Ahmadi F， Sharafati A，et al.Investigating hybrid deep learning models and meta-heuristic algorithms in predicting evaporation from a reservoir：a case study of Dez Dam［J］.Earth Science Informatics，2023，16：3597-3618.
13	Elzain E H， Abdalla A O， Abdallah M，et al.Innovative approach for predicting daily reference evapotranspiration using improved shallow and deep learning models in a coastal region：a comparative study［J］.Journal of Environmental Management，2024，354：120246.
14	Ferreira L B， Cunha F.Multi-step ahead forecasting of daily reference evapotranspiration using deep learning［J］.Computers and Electronics in Agriculture，2020，178：105728.
15	Dong J， Xing L， Cui N，et al.Estimating reference crop evapotranspiration using improved convolutional bidirectional long short-term memory network by multi-head attention mechanism in the four climatic zones of China［J］.Agricultural Water Management，2024，292：108665.
16	Su H， Zhao D， Heidari A A，et al.RIME：a physics-based optimization［J］.Neurocomputing，2023：183-214.
17	Heidari A A， Mirjalili S， Faris H，et al.Harris hawks optimization：algorithm and applications［J］.Future Generation Computer Systems，2019，97：849-872.
18	Dehghani M， Trojovský P.Osprey optimization algorithm：a new bio-inspired metaheuristic algorithm for solving engineering optimization problems［J］.Frontiers in Mechanical Engineering，2023.
19	Dehghani M， Hubálovský Š， Trojovský P.Northern goshawk optimization：a new swarm-based algorithm for solving optimization problems［J］.Ieee Access，2021，9：162059-162080.
20	Vaswani A， Shazeer N， Parmar N，et al.Attention is all you need［Z］.10.48550/arXiv.1706.03762，2017.
21	Ji X B， Zhao W Z， Jin B W，et al.Seasonal variations in energy exchange and evapotranspiration of an oasis-desert ecotone in an arid region［J］.Hydrological Processes，2021，35（9）：1-16.
22	许澳康，吉天琪，石晶.甘肃黑河流域生态网络构建与优化［J］.中国沙漠，2025，45（2）：61-70.
23	雷军，程新平，薛春，等.黑河流域中游北部荒漠区植物群落特征与稳定性［J］.中国沙漠，2024，44（6）：187-194.
24	Wang S X， Shi J R， Yang W Y，et al.High and low frequency wind power prediction based on Transformer and BiGRU-Attention［J］.Energy，2024，288：129753.
25	Wang W H， Xi C P.Short-term electricity load forecasting based on TPE-optimized CNN-BIGRU-attention［J］.Journal of Physics：Conference Series，2023，2496：1.

[1]	高博钰, 杨波, 张德国. U-Net深度卷积神经网络在沙脊线提取中的应用[J]. 中国沙漠, 2021, 41(5): 21-32.
[2]	张亦然, 刘廷玺, 童新, 段利民, 王昕. 基于U型神经网络的沙丘-草甸相间地区无人机影像植被覆盖度提取及其影响因素[J]. 中国沙漠, 2021, 41(3): 16-24.