| 引用本文: | 唐茂淞,张楠,李国辉,等.基于机器学习算法的棉田土壤钾、钠离子量预测[J].灌溉排水学报,0,():-. |
| Tang Maosong,Zhang Nan,Li Guohui,et al.基于机器学习算法的棉田土壤钾、钠离子量预测[J].灌溉排水学报,0,():-. |
|
| |
|
|
| 本文已被：浏览 728次 下载 0次 |
|
|
|
| 基于机器学习算法的棉田土壤钾、钠离子量预测 |
|
唐茂淞1, 张楠1, 李国辉1, 赵泽艺1, 李明发2, 王兴鹏1
|
|
1.塔里木大学水利与建筑工程学院;2.新疆生产建设兵团第一师水文水资源管理中心
|
|
| 摘要: |
| 【目的】比较4种机器学习方法对南疆棉田土壤K+、Na+量的预测结果，确定一种预测准确度较高的机器学习模型作为可供参考的方法。【方法】采用支持向量回归（SVR）、随机森林回归（RFR）、K-最近邻回归（KNNR）和梯度提升回归树（GBRT）4种机器学习算法，2020年棉田土壤K+、Na+量数据训练模型，2021年实测数据进行测试验证。使用平均绝对误差（MAE）、均方根误差（RMSE）和决定系数（R2）对模型预测结果进行评估。【结果】4种模型(SVR、RFR、KNNR和GBRT)对测试样本K+量预测的MAE分别为0.100、0.169、0.169 g/kg和0.167 g/kg；RMSE分别为0.119、0.218、0.218 g/kg和0.223 g/kg；R2分别为0.687、0.437、0.430和0.395。对测试样本Na+量预测的MAE分别为0.841、2.841、2.826 g/kg和2.856 g/kg；RMSE分别为1.154、3.658、3.630 g/kg和3.650 g/kg；R2分别为0.838、0.299、0.219和0.200。将测试样本K+、Na+量分别按4个土层深度（0~10、10~20、20~30和30~40 cm）进行预测时，SVR模型的误差值最小，其对K+量按照4个深度预测的MAE分别为0.122、0.114、0.056 g/kg和0.106 g/kg，RMSE分别为0.135、0.135、0.069 g/kg和0.126 g/kg；对Na+量预测的MAE分别为0.540、0.619、0.835 g/kg和1.371 g/kg，RMSE分别为0.636、0.748、1.198 g/kg和1.710 g/kg。【结论】SVR预测K+、Na+量的精度最高，可推荐作为南疆棉田土壤钾、钠离子量预测的一种方法。 |
| 关键词: 南疆棉田；土壤盐分离子；机器学习；回归预测模型 |
| DOI： |
| 分类号:TP181 |
| 基金项目:十四五国家重点研发计划（2022YFD1900505）；兵团重大科技项目(2021AA003)；塔里木大学研究生科研创新项目(TDGRI202143) |
|
| Prediction of Soil K+ and Na+ Contents in Cotton Field Based on Machine Learning Algorithms |
|
Tang Maosong1, Zhang Nan1, Li Guohui1, Zhao Zeyi1, Li Mingfa2, Wang Xingpeng1
|
|
1.College of Water Resource and Architecture Engineering,Tarim University;2.Hydrology and Water ResourcesManagement Center of the First Division of Xinjiang Production and Construction Corps
|
| Abstract: |
| 【Objective】To compare the prediction results of four machine learning methods for K+ and Na+ content in cotton field soil in southern Xinjiang, and determine a machine learning model with high prediction accuracy as a reference method. 【Methods】 Four machine learning algorithms, namely support vector regression (SVR), random forest regression (RFR), K-nearest neighbor regression (KNNR), and gradient lifting regression tree (GBRT), were used to train the model of soil K+ and Na+ in cotton fields in 2020 and test and verify the measured data in 2021. The model prediction results were evaluated using mean absolute error (MAE), root mean square error (RMSE), and determination coefficient (R2). 【Results】The MAEs predicted by the four models (SVR, RFR, KNNR, and GBRT) for the K+ content of the test sample were 0.100, 0.169, 0.169, and 0.167 g/kg, respectively; The RMSE was 0.119, 0.218, 0.218 g/kg, and 0.223 g/kg, respectively; R2 was 0.687, 0.437, 0.430, and 0.395, respectively. The MAEs predicted for the Na+ content of the test sample were 0.841, 2.841, 2.826 g/kg, and 2.856 g/kg, respectively; The RMSE was 1.154, 3.658, 3.630 g/kg, and 3.650 g/kg, respectively; R2 was 0.838, 0.299, 0.219, and 0.200, respectively. When predicting the amount of K+ and Na+ in the test sample at four soil depths (0~10, 10~20, 20~30, and 30~40 cm), the error value of the SVR model is the smallest, and its MAEs for predicting the amount of K+ at four depths are 0.122, 0.114, 0.056 g/kg, and 0.106 g/kg, respectively, while the RMSE is 0.135, 0.135, 0.069 g/kg, and 0.126 g/kg, respectively; The MAE predicted for the amount of Na+ was 0.540, 0.619, 0.835 g/kg, and 1.371 g/kg, respectively, while the RMSE was 0.636, 0.748, 1.198 g/kg, and 1.710 g/kg, respectively. 【Conclusion】SVR has the highest accuracy in predicting K+ and Na+ amounts, and can be recommended as a method for predicting soil potassium and sodium ions in cotton fields in southern Xinjiang. |
| Key words: South Xinjiang cotton field; soil salt ions; machine learning; regression prediction model |
|
|
|
|
