| BP神经网络和SVM模型对施加生物炭土壤水分预测的适用性 |
| |
| 引用本文： | 王彤彤,翟军海,何欢,郑纪勇,涂川.BP神经网络和SVM模型对施加生物炭土壤水分预测的适用性[J].水土保持研究,2017(3):86-91. |
| |
| 作者姓名： | 王彤彤 翟军海 何欢 郑纪勇 涂川 |
| |
| 作者单位： | 1. 西北农林科技大学资源环境学院,陕西杨凌,712100;2. 陕西省农业厅,西安,710003;3. 西北农林科技大学理学院,陕西杨凌,712100;4. 西北农林科技大学资源环境学院,陕西杨凌712100;中国科学院水利部水土保持研究所黄土高原土壤侵蚀与旱地农业国家重点实验室,陕西杨凌712100;5. 重庆邮电大学计算机科学与技术学院,重庆,400065 |
| |
| 基金项目： | 国家自然科学基金“生物炭对黄土高原不同质地土壤水文过程影响及机理的定位研究”(41571225) |
| |
| 摘 要： | 生物炭作为土壤改良剂对半干旱区土壤水分有良好的吸持作用,为确定施加生物炭对土壤水分预测模型适用性的影响,依托黄土高原半干旱区固原生态站开展了小区定位试验。向土壤中施加不同种类及比例的生物炭,定期监测土壤水分含量;考虑土壤含水量的非线性特征以及生物炭对土壤水分的影响,选取BP神经网络和SVM支持向量机两种模型,建立施加生物炭土壤水分预测模型。计算预测值,并与实测值对比,分析相对误差;利用RMSE、MRE、MAE和R2评估BP神经网络和SVM模型的精度。结果表明;BP神经网络预测值的平均相对误差为3.78%,最大误差为13.14%;SVM模型的平均相对误差为0.56%,最大误差为2.42%。SVM模型的RMSE、MRE、MAE值(分别为0.34~0.17,0.07,0.56~1.27)均小于BP神经网络的(分别为1.04~1.16,0.47~0.68,3.78~4.57),且决定系数R2值SVM模型(0.96~0.99)大于BP神经网络(0.56~0.64)。BP神经网络和SVM模型均能很好地预测施加生物炭的土壤水分,但SVM模型预测结果更加稳定,精度较高,更适于施加生物炭土壤水分的预测。该研究可为半干旱地区生物炭还田土壤水分的预测及管理提供理论依据。
|
| 关 键 词： | 土壤水分 生物炭 模型预测 SVM模型 BP神经网络 |
Applicability of BP Neural Network Model and SVM Model to Predicting Soil Moisture Under incorporation of Biochar into Soils |
| |
| Abstract: | As a soil amendment,biochar has a good effect on the soil moisture in the semi-arid area.In order to know the effect of adding biochar on soil water content prediction model,a district positioning experiment was carried out in semi-arid Guyuan ecology research station on the Loess Plateau.In the experiment,different kinds and amounts of biochar were added to soil and the soil water contents were monitored regularly.In consideration of soil water nonlinear characteristic and random effect of adding biochar,BP Neural Network model and SVM (Support Vector Machine) model were selected to build water content prediction model for biochar-added soil and the applicability of the two models were finally evaluated according to the measured data and predicted data by using RMSE,MRE,MAE and R2 to assess the precision.The results showed that the average relative error value of BP Neural Network model was 3.78％ and the max relative error value was 13.14％,while the average relative error value of SVM model was 0.56％ and the max relative error value was 2.42％,respectively.The RMSE,MRE,MAE value of SVM model(0.34～0.17,0.07 and 0.56～ 1.27,respectively) were less than BP Neural Network model(1.04～1.16,0.47～0.68 and 3.78～4.57 respectively),and the R2 value of SVM model (0.96～0.99) were greater than BP Neural Network model (0.56～0.64),respectively.BP Neural Network model and SVM model both performed well in predicting soil water content and the prediction results of SVM model were more steady and precise.So the SVM model is the appropriate model to predict water content in biochar-added soil.The reuslt can provide theoretical evidence for prediction and management of moisture in the biochar-added soil in the semi-arid area. |
| |
| Keywords: | soil moisture biochar prediction model SVM model BP neural network model |
| 本文献已被 CNKI 万方数据 等数据库收录！ |
|
