| 引用本文: | 崔 嵩,吴 锦,张乃丰,等.基于SCSSA-CNN-BiLSTM模型的冬小麦生育期
日土壤水分预测研究[J].灌溉排水学报,2025,44(8):1-8. |
| CUI Song,WU Jin,ZHANG Naifeng,et al.基于SCSSA-CNN-BiLSTM模型的冬小麦生育期
日土壤水分预测研究[J].灌溉排水学报,2025,44(8):1-8. |
|
| 摘要: |
| 【目的】为准确预测大田土壤含水率，便于指导农业生产和节水灌溉。【方法】本文结合正余弦和柯西变异的麻雀优化算法（SCSSA）与卷积神经网络（CNN）、双向长短期记忆神经网络（BiLSTM）进行组合，构建了SCSSA-CNN-BiLSTM土壤水分混合预测模型。利用五道沟实验站2022年10月—2023年6月冬小麦全生育期的气象因子和3个土层（10、30、50 cm）土壤含水率，对冬小麦全生育期根系密集层20 cm土层土壤含水率进行预测。【结果】①SCSSA-CNN-BiLSTM模型能有效捕捉土壤水分时空演变规律，经SCSSA优化后的模型预测精度显著提升，RMSE从1.394降至0.774，降幅达44.5%；②SCSSA-CNN-BiLSTM模型展现出较好的预测能力，模型的土壤含水率预测值与实测值基本吻合。③与GA（遗传算法）、PSO（粒子群算法）、SSA（麻雀搜索算法）相比，SCSSA（麻雀优化算法）通过融合多重优化策略显著提升了模型超参数优化性能，模型可用于冬小麦生育期土壤水分预测。【结论】本文提出的SCSSA-CNN-BiLSTM模型对冬小麦20 cm土层土壤含水率的预测精度较高，可为农业实时适量灌溉提供科学依据。 |
| 关键词: 神经网络；麻雀优化算法；土壤水分预测；冬小麦 |
| DOI：10.13522/j.cnki.ggps.2025183 |
| 分类号: |
| 基金项目: |
|
| Daily soil moisture prediction during winter wheat growth season using an SCSSA-CNN-BiLSTM model |
|
CUI Song, WU Jin, ZHANG Naifeng, LIU Meng, HU Yongsheng,
HE Yanan, GU Yue, LONG Xinya, WANG Zhenlong
|
|
1. School of Water Conservancy and Civil Engineering, Northeast Agricultural University, Harbin 150030, China; 2. College of Hydrology and Water Resources, Hohai University, Nanjing 210098, China; 3. Wudaogou Hydrological Experimental Station, Anhui Academy of Water Conservancy Sciences, Bengbu 233000, China; 4. Nanjing Institute of Water Conservancy Science, Nanjing 210029, China
|
| Abstract: |
| 【Objective】Accurate prediction of field soil moisture is crucial for managing agricultural production and water-saving irrigation. This paper proposes a new method to predict soil moisture changes.【Method】The hybrid deep learning model, SCSSA-CNN-BiLSTM, was integrated with Sine Cosine Cauchy Sparrow Search Algorithm (SCSSA) for hyperparameter optimization. It was then combined with Convolutional Neural Networks (CNN) for spatial feature extraction and Bidirectional Long Short-Term Memory (BiLSTM) networks for temporal sequence learning. The model was trained using meteorological data and soil moisture measured at three depths - 10, 30 and 50?cm - at the Wudaogou Experimental Station between October 2022 and June 2023. It was then used to predict soil moisture in the 0-20 cm root zone during the winter wheat growing season.【Result】① The optimized model accurately captured the spatiotemporal variation in soil moisture, with the SCSSA enhancement reducing RMSE by 44.5% from 1.394 to 0.774. ② The proposed model was superior to other methods; its statistical metrics are R2=0.960, RMSE=0.774, MSE=0.599, MAE=0.528 and MAPE=1.84%. The predicted results agreed well with observed data. ③ Comparative analysis showed that SCSSA optimization significantly outperformed GA, PSO and SSA in hyperparameter tuning.【Conclusion】The SCSSA-CNN- BiLSTM model is accurate for predicting soil moisture in the 0-20 cm root zone of winter wheat. It can be used for real-time irrigation management. |
| Key words: neural network; sparrow optimization algorithm; soil moisture prediction; winter wheat |
