[1]卢宏亮,赵明松.基于神经网络模型的安徽省土壤pH预测[J].江苏农业学报,2019,(05):1119-1123.[doi:doi:10.3969/j.issn.1000-4440.2019.05.017]
　LU Hong-liang,ZHAO Ming-song.Prediction of soil pH in Anhui province based on RPROP and GRPROP algorithms[J].,2019,(05):1119-1123.[doi:doi:10.3969/j.issn.1000-4440.2019.05.017]
点击复制

基于神经网络模型的安徽省土壤pH预测()
分享到：

江苏农业学报[ISSN:1006-6977/CN:61-1281/TN]

卷:
期数:
2019年05期
页码:
1119-1123
栏目:
耕作栽培·资源环境
出版日期:
2019-10-31

文章信息/Info

Title:
Prediction of soil pH in Anhui province based on RPROP and GRPROP algorithms
作者:
卢宏亮1; 赵明松1; 2
(1.安徽理工大学测绘学院，安徽淮南232001; 2.中国科学院南京土壤研究所土壤与农业可持续发展国家重点实验室，江苏南京210008)
Author(s):
LU Hong-liang1; ZHAO Ming-song1; 2
(1.School of Geodesy and Geomatics, Anhui University of Science and Technology, Huainan 232001, China; 2.State Key Laboratory of Soil and Sustainable Agriculture, Institute of Soil Science, Chinese Academy of Sciences, Nanjing 210008, China)
关键词:
土壤pH; 空间预测; RPROP算法; GRPROP算法; 神经网络; 安徽省
Keywords:
soil pH; spatial prediction; RPROP algorithm; GRPROP algorithm; neural network; Anhui province
分类号:
S127
DOI:
doi:10.3969/j.issn.1000-4440.2019.05.017
文献标志码:
A
摘要:
以土壤pH为研究对象，利用一般反向传播(Back propagation，BP)神经网络模型、带回溯的弹性反向传播(Resilient back propagation with backtracking, RPROP-WB)神经网络模型、不带回溯的弹性反向传播(Resilient back propagation without backtracking, RPROP-OB)和最小绝对梯度反向传播(Smallest absolute gradient resilient back propagation, SAG-RPROP)神经网络模型进行安徽省土壤pH的预测及制图，选用均方根误差(RMSE)、绝对平均误差(MAE)及决定系数(R2)为评价标准，比较3种改进的神经网络模型与一般BP神经网络模型对于土壤pH的预测能力。结果表明：研究区域内，4种神经网络模型的拟合能力高低依次为：SAG-RPROP>RPROP-WB>RPROP-OB>BP。由建模集可以看出，RPROP-WB、RPROP-OB 2种模型与BP神经网络模型的预测精度一致，4种模型中预测精度最高的为SAG-RPROP，R2比其他3种模型提高0.07。对于验证集，预测能力高低依次为：SAG-RPROP>RPROP-WB>RPROP-OB>BP，预测精度和泛化能力最高的为SAG-RPROP模型，RMSE、MAE和R2分别为0.67、0.50及0.59。空间预测图结果显示，4种模型所得安徽省土壤pH空间分布基本类似，均呈“南酸北碱”趋势，一般BP神经网络对于土壤pH预测区分度较低，预测所得安徽省南部地区的土壤pH均集中在5.57至6.50之间， RPROP-WB、RPROP-OB及SAG- RPROP所得预测图则区分更为明显。综上所述，RPROP及其改进算法可以有效地进行土壤属性的预测，且精度均高于一般BP神经网络模型。
Abstract:
In this study, soil pH was taken as the research object, using general back propagation (BP) neural network model, resilient back propagation with backtracking (RPROP-WB) neural network model, resilient back propagation without backtracking (RPROP-OB) neural network model and smallest absolute gradient resilient back propagation (SAG-RPROP) neural network model for predicting and mapping soil pH in Anhui province, and the root mean square error (RMSE), mean absolute error (MAE) and coefficient of determination (R2) were used as evaluation criteria to compare the prediction ability of three improved neural network models and the general BP neural network model for soil pH. The results showed that: the fitting ability of the four neural network models in the study area from high to low was: SAG-RPROP>RPROP-WB>RPROP-OB>BP. It can be seen from the training set that the prediction accuracy of the RPROP-WB and RPROP-OB models was consistent with that of the BP neural network model. The highest prediction accuracy among the four models was SAG-RPROP, and the R2 was 0.07 higher than the other three models. For the verification set, the prediction ability was as follows: SAG-RPROP>RPROP-WB>RPROP-OB>BP. The SAG-RPROP model had the highest prediction accuracy and generalization ability, and RMSE, MAE and R2 were 0.67, 0.50 and 0.59. The results of spatial prediction showed that the spatial distribution of soil pH in Anhui province was similar in the four models, all of which showed the trend of “Southern acid North Base”. The general BP neural network had lower discrimination for soil pH prediction. The soil pH in the southern part of Anhui province was concentrated between 5.57 and 6.50. The prediction maps obtained by RPROP-WB, RPROP-OB and SAG-RPROP models were more obvious. In summary, RPROP and its improved algorithms can effectively predict soil properties, and the accuracy is higher than the general BP neural network model.

参考文献/References:

[1]吕贻忠,李保国，李跃进，等.土壤学[M].北京:中国农业出版社,2005:115.
[2]LAUBER C L, HAMADY M, KNIGHT R, et al. Pyrosequencing-based assessment of soil pH as a predictor of soil bacterial community structure at the continental scale[J].Applied and Environmental Microbiology,2009,75(15):5111-5120.
[3]蒋欢，彭玉梅，董代文，等. 土壤酸碱度与榨菜根肿病的关系分析[J]．南方农业学报，2017，48(9)：1617-1623.
[4]HOSSNER L R. Field pH[M].Netherlands:Springer,2008:271-272.
[5]KIM M, GILLEY J E. Artificialneural network estimation of soil erosion and nutrient concentrations in runoff from land application areas[J].Computers & Electronics in Agriculture,2008,64(2):268-275.
[6]WERE K, BUI D T, OYSTEIN B D , et al. A comparative assessment of support vector regression,artificial neural networks,and random forests for predicting and mapping soil organic carbon stocks across an Afromontane landscape[J].Ecological Indicators,2015,52:394-403.
[7]董敏,王昌全,李冰,等.基于GARBF神经网络的土壤有效锌空间插值方法研究[J].土壤学报,2010,47(1):42-50.
[8]郑立华,李民赞,潘娈,等.基于近红外光谱技术的土壤参数BP神经网络预测[J].光谱学与光谱分析,2008,28(5):1160-1164.
[9]江叶枫,郭熙,叶英聪,等. 应用集成 BP 神经网络模型预测土壤有机质空间分布[J].江苏农业学报,2017,33(5):1044-1050.
[10]RIEDMILLER M, BRAUN H . A direct adaptive method for faster backpropagation learning:the RPROP algorithm[J].Chemical Product & Process Modeling,1993,6(A8):586-591.
[11]ANASTASIADIS A D, MAGOULAS G D, VRAHATIS M N , et al. New globally convergent training scheme based on the resilient propagation algorithm[J].Neurocomputing,2005,64:253-270.
[12]李德成，张甘霖，王华等.中国土系志·安徽卷[M].北京：科学出版社，2017:3-24.
[13]RUMELHART D E, MCCLELLAND J L . Parallel distributed processing[M].Cambridge,Massachusetts:The MIT Press,1986:45-76.
[14]MEULEAU N, DORIGO M . Ant colony optimization and stochastic gradient descent[J].Artificial Life,2002,8(2):103.
[15]NESTEROV Y . Gradient methods for minimizing composite functions[J].Mathematical Programming,2013,140(1):125-161.
[16]WEI C H, LI Y, HUANG P J , et al. Estimation of breast density:an adaptive moment preserving method for segmentation offibroglandular tissue in breast magnetic resonance images.[J].European Journal of Radiology,2012,81(4):618-624.
[17]杨克俭,梁佳成,王艳秋.基于RPROP神经网络算法的旋转机械故障诊断模型[J].硅谷,2008(23):116-116.

备注/Memo

备注/Memo:
收稿日期：2018-11-13 基金项目：国家自然科学基金项目(41501226)；安徽省高校自然科学研究项目(KJ2015A034)；土壤与农业可持续发展国家重点实验室开发基金项目(Y412201431) 作者简介：卢宏亮(1993-)，男，安徽铜陵人，硕士研究生，从事数字土壤制图研究。(E-mail) 17355481665@163.com 通讯作者：赵明松，(E-mail) zhaomingsonggis@163.com
更新日期/Last Update: 2019-11-11