| 引用本文: | 林 兴,王炳亮,陈淑娟,等.多种机器学习模型在超声波检测不同温度
条件下土壤含水率的应用研究[J].灌溉排水学报,2025,44(12):93-103. |
| LIN Xing,WANG Bingliang,CHEN Shujuan,et al.多种机器学习模型在超声波检测不同温度
条件下土壤含水率的应用研究[J].灌溉排水学报,2025,44(12):93-103. |
|
| 摘要: |
| 【目的】通过超声波检测技术结合机器学习模型和优化算法，精确模拟并预测不同温度条件下土壤含水率，为土壤物理特性预测及农业工程应用提供新的思路。【方法】通过室内试验，以宁夏石嘴山惠农区银河村土壤为样本，结合超声波检测技术，采用BP神经网络（BP）、长短期记忆网络（LSTM）、极限学习机（ELM）和随机森林（RF）等机器学习模型进行土壤超声波特性模拟与预测。为提高预测精度，引入蜣螂优化算法（DBO）、鹅优化算法（GOOSE）、粒子群优化算法（PSO）和麻雀优化算法（SSA）对ELM模型进行优化，同时对于超声波波速与土壤温度及含水率关系分析中的极端情况设置了2种方案，方案一为土壤冻结效果较差的情况，设置土壤温度为-5、-10、-15 ℃，含水率为10%、15%、20%，方案二为土壤冻结较为完全的情况，设定土壤温度为-10、-15 ℃，含水率为25%、30%。【结果】0 ℃以上时，随着土壤含水率的增加，超声波波速整体呈现略微递增的趋势，当含水率达到15%后有下降的趋势，而0 ℃以下，含水率为10%~20%时，波速与温度呈负相关，且出现骤减现象，反映出土壤结构较差。当含水率超过30%时，土壤趋近饱和，波速显著上升。在模拟过程中，发现ELM模型表现最佳，R2为0.96，MSE为0.01。引入不同优化算法后，麻雀优化算法（SSA）对ELM的优化效果最为突出，R2提升至0.98，MSE降至0.000 9，显示出较高的预测精度和稳定性。2种极端试验方案的模拟结果验证了模型在极端条件下的良好预测能力，R2分别为0.98和0.97，RMSE分别为0.02和0.06。【结论】本研究通过对比分析多种机器学习模型和优化算法在超声波检测不同温度条件下土壤含水率模拟中的表现，最终发现SSA-ELM的模拟效果最佳，且在土壤极端冻结条件下仍能满足模拟精度要求，具有重要的理论和实践意义。 |
| 关键词: 超声波波速；温度；含水率；土壤 |
| DOI：10.13522/j.cnki.ggps.2025015 |
| 分类号: |
| 基金项目: |
|
| Integration of multiple machine learning models and ultrasonic method to measure soil water content under freeing condition |
|
LIN Xing, WANG Bingliang, CHEN Shujuan, YANG Wu, TONG Yusen,
SHEN Fangsen, CHEN Bingtong, LI Wangcheng
|
|
1. School of Civil and Water Conservancy Engineering, Ningxia University, Yinchuan 750021, China; 2. Engineering Research Center for Efficient Utilization of Modern Agricultural Water Resources in arid Areas, Ministry of Education, Yinchuan 750021, China;
3. Key Laboratory of Digital Water Control of Yellow River Water Network, Ningxia, Yinchuan 750021, China;
4. Shizuishan Agricultural Technology Extension Service Center, Shizuishan 753000, China
|
| Abstract: |
| 【Objective】Methods for measuring soil moisture are many and vary, but most of them are either hindered by low accuracy, high cost, or sensitive to environment. Ultrasonic technology is a non-invasive method with high resolution; it has potential to overcome above limitations. Although ultrasound has been used to measure water content in unsaturated and frozen soils, there is a lack of models to accurately represent the influence of environmental factors on soil water measurement. This paper is to address this gap by integrating ultrasonic method and machine learning (ML) to optimize measurement of soil moisture content under the impact of different environmental conditions. 【Method】A laboratory experiment was conducted using soil samples taken from Yinhe Village in Huinong District of Shizuishan, Ningxia. Four ML models, BP neural networks (BP), long and short-term memory (LSTM), extreme learning machines (ELM), and random forests (RF), were combined with ultrasonic method to measured soil ultrasonic properties linked to moisture content. To improve modelling accuracy, the dung beetle optimization [DBO], goose optimization [GOOSE], particle swarm optimization [PSO] and sparrow search algorithm [SSA] were applied to optimize the ELM model. The relationships between ultrasonic wave velocity, temperature and moisture were analyzed in two extreme scenarios: poorly frozen soil with temperatures ranging from -5 to -15 ℃ and soil water content varying from 10% to 20%, heavily frozen soil with temperature ranging from -10 to -15 ℃ and soil water content varying from 25% to 30%.【Result】Above freezing point, the ultrasonic wave velocity slightly increased with soil water content but decreased once soil water content exceeds 15%. For temperature below freezing point and soil water content in the range of 10% to 20%, the ultrasonic velocity was negatively correlated with temperature and decreased sharply as soil water content increased. When soil water content exceeded 30%, the ultrasonic velocity increased significantly as soil water content increased. Among all ML models we compared, the ELM was most accurate prior to optimization, with R2=0.96 and MSE=0.01. After optimization, SSA-ELM was most accurate with R2=0.98 and MSE=0.000 9. For the two extreme scenarios, the SSA-ELM was most accurate; its statistical metric for Scenario 1 were R2=0.98 and RMSE=0.02, and R2=0.97 and RMSE=0.06 for Scenario 2. 【Conclusion】The SSA-optimized ELM model combined with ultrasonic technology is most accurate for estimating soil moisture content, even under extreme freezing condition. |
| Key words: ultrasonic wave velocity; temperature; water content; soil |
