基于机器学习构建葡萄霜霉病预测模型及验证

doi:10.13304/j.nykjdb.2022.0866

摘要/Abstract

摘要：

葡萄霜霉病是葡萄上的主要病害之一，该病从葡萄苗期到果实成熟期都可发生，多雨年份常造成毁灭性损失。为了准确预测葡萄霜霉病的发生，最大限度地降低霜霉病对葡萄的危害，基于2020年葡萄生长期间气象数据和病害发生数据，结合4种机器学习算法（二项逻辑斯蒂、支持向量机、决策树、K最近邻）构建了葡萄霜霉病发生预测模型，并用2021年数据进行验证。结果表明，决策树模型在病害发生预测模型构建中的评价指标最优，其准确率达94%，预测发生的精准率、召回率、F₁分值分别为91%、90%、91%。经验证，决策树模型对葡萄霜霉病发生的预测精度及性能均优于其他3个模型。因此，可利用此模型进一步开发葡萄霜霉病预警系统，为生产上葡萄霜霉病的防治提供技术支持和决策指导。

关键词: 机器学习, 葡萄, 霜霉病, 预测模型

Abstract:

Grape downy mildew is one of the major diseases on grapes， which can occur from seedling to fruit maturity， often causing destructive losses in rainy years. To accurately predict the occurrence of grape downy mildew and minimize the hazards of grape downy mildew， based on the agrometeorological data and disease occurrence data during grape growth in 2020， a prediction model for the occurrence of grape downy mildew was developed by 4 machine learning algorithms （binary logistic regression， support vector machine， decision tree and K-nearest neighbors）， and the models were verified using the data in 2021. The results showed that the decision tree model had the best evaluation indicators in the disease prediction model， which the accuracy was 94%， and the precision， recall and F₁-score was 91%， 90% and 91%， respectively. The decision tree model’s performance was still better than the other models using validation data. Therefore， the decision tree model could be further used to develop early warning systems for grape downy mildew， which should provide technical support and decision-making guidance for controlling it in production.

Key words: machine learning, grape, downy mildew, prediction model

中图分类号:

S663.1

边凤霞, 刘凯歌, 容新民. 基于机器学习构建葡萄霜霉病预测模型及验证[J]. 中国农业科技导报, 2023, 25(8): 126-137.

Fengxia BIAN, Kaige LIU, Xinmin RONG. Development and Verification of Prediction Model for Grape Downy Mildew Based on Machine Learning[J]. Journal of Agricultural Science and Technology, 2023, 25(8): 126-137.

图/表 15

图1 SVM模型原理

Fig. 1 SVM model principle

图2 Decision Tree模型原理注：示意图仅表示模型处理问题时的决策过程，不代表以单一特征的决策结果作为预测结果。

Fig. 2 Decision tree model principleNote： The schematic diagram only shows the decision-making process of the model when dealing with the problem， and does not mean that the decision result of a single feature is used as the prediction result.

图3 KNN模型原理

Fig. 3 KNN model principle

表1 混淆矩阵列联表

Table 1 Confusion matrix contingency table

指标

Index

样本数Number of samples

预测不发病 Prediction label： 0

预测发病Prediction label： 1

实际不发病

True label： 0

真阴性 True negative： TN

假阳性 False positive： FP

实际发病

True label： 1

假阴性 False negative：FN

真阳性 True positive：TP

表2 k值一致性划分标准

Table 2 The k value consistency classification criteria

k值 k value	一致性结果 Consistent result
0≤k<0.2	极低Slight
0.2≤k<0.4	一般Fair
0.4≤k<0.6	中等Moderate
0.6≤k<0.8	高度Substantial
0.8≤k<1.0	几乎完全Almost perfect

图4 BLR模型分类预测指标A：ROC曲线；B：混淆矩阵

Fig. 4 BLR model classification prediction indicatorsA： ROC curve； B： Confusion matrix

图5 SVM模型分类预测指标A：ROC曲线；B：混淆矩阵

Fig. 5 SVM model classification prediction indicatorsA： ROC curve； B： Confusion matrix

图6 DT模型分类预测指标A：ROC曲线；B：混淆矩阵

Fig. 6 DT model classification prediction indicatorsA： ROC curve； B： Confusion matrix

图7 KNN模型分类预测指标A：ROC曲线；B：混淆矩阵

Fig. 7 KNN model classification prediction indicatorsA： ROC curve； B： Confusion matrix

表3 不同预测模型的分类结果

Table 3 Classification results of different prediction models

模型 Model	类别标签 Classe label	分类结果 Classification result
模型 Model	类别标签 Classe label	准确率 Accuracy/%	精准率 Precision/%	召回率 Recall/%	F₁分值 F₁-score/%
BLR	0	86	87	93	90
BLR	1	86	83	70	76
SVM	0	86	86	94	90
SVM	1	86	85	69	76
DT	0	94	95	96	95
DT	1	94	91	90	91
KNN	0	93	96	94	95
KNN	1	93	88	91	89

图8 BLR 模型分类预测2021年验证指标A：ROC曲线；B：混淆矩阵

Fig. 8 BLR model classification prediction validation indicators in 2021A： ROC curve； B： Confusion matrix

图9 SVM模型分类预测2021年验证指标A：ROC曲线；B：混淆矩阵

Fig. 9 SVM model classification prediction validation indicators in 2021A： ROC curve； B： Confusion matrix

图10 DT模型分类预测2021年验证指标A：ROC曲线；B：混淆矩阵

Fig. 10 DT model classification prediction validation indicators in 2021A： ROC curve； B： Confusion matrix

图11 KNN模型分类预测2021年验证指标A：ROC曲线；B：混淆矩阵

Fig. 11 KNN model classification prediction validation indicators in 2021A： ROC curve； B： Confusion matrix

表4 不同预测模型分类验证结果

Table 4 Classification validation results of different prediction models

模型 Model	类别标签 Classe label	分类结果 Classification result
模型 Model	类别标签 Classe label	准确率 Accuracy/%	精准率 Precision/%	召回率 Recall/%	F₁分值 F₁-score/%
BLR	0	73	73	87	80
BLR	1	73	74	53	62
SVM	0	75	76	86	81
SVM	1	75	72	59	65
DT	0	90	92	91	92
DT	1	90	87	88	88
KNN	0	84	86	87	86
KNN	1	84	80	79	80

参考文献 39

1	芦屹,努尔孜亚·亚力麦麦提,付文君,等.新疆伊犁河谷地区葡萄霜霉病流行与气候条件的关系[J].新疆农业科学,2020,57(10):1855-1862.
	LU Y, Yalimaimaiti Nuerziya, FU W J, et al.. Research the relationship between epidemic of grape downy mildew and climatic conditions in Ili Valley of Xinjiang [J]. Xinjiang Agric. Sci., 2020, 57(10): 1855-1862.
2	吉丽丽.石河子地区葡萄霜霉病菌孢子囊时空扩散动态及品种抗病性研究[D].石河子:石河子大学,2011.
	JI L L. The dynamics of sporangium of Plasmopara viticola and resistance of different grape varieties in Shihezi region [D]. Shihezi: Shihezi Univ., 2011.
3	张玮,燕继晔,刘梅,等.葡萄霜霉病流行与预测研究进展[J].中国果树,2020,44(3):11-15.
	ZHANG W, YAN J Y, LIU M, et al.. Research progress on prediction and epidemic of grapevine downy mildew [J]. China Fruits, 2020, 44(3):11-15.
4	GILLES T. Forecasting Downy Mildew Diseases in:Advances in Downy Mildew Research—Volume Ⅱ [M]. Dordrecht: Springer, 2004:35-67.
5	ORLANDINI S, CAPPUGI A. Calibration of agrometeorological model for the simulation of grapevine downy mildew in several viticultural areas of Europe [C]// Proceedins of Fifth European Conference on Application of Meteorology. Budapest Hungary, 2001:1-7.
6	张雪雪,王斌,田洋洋,等.作物病虫害预测机理与方法研究进展[J].中国农业科技导报,2019,21(5):110-120.
	ZHANG X X, WANG B, TIAN Y Y, et al.. Research progress on forecasting mechanism and methodology for crop disease and insect pest [J]. J. Agric. Sci. Technol., 2019, 21(5):110-120.
7	GESSLER C, PERTOT I, PERAZZOLLI M. Plasmopara viticola:a review of knowledge on downy mildew of grapevine and effective disease management [J]. Phytopathol. Mediterr., 2011, 50(1):3-44.
8	THIND T S, ARORA J K, MOHAN C, et al.. Epidemiology of Powdery Mildew, Downy Mildew and Anthracnose Diseases of Grapevine [M]. Dordrecht: Kluwer Academic Publishers, 2004:621-638.
9	FRANCESCA S, SIMONA G, FRANCESCO NICOLA T, et al.. Downy mildew (Plasmopara viticola) epidemics on grapevine under climate change [J]. Global Change Biol., 2006, 12(7):1299-1307.
10	ROSSI V, CAFFI T, GIOSUE S, et al.. A mechanistic model simulating primary infections of downy mildew in grapevine [J]. Ecol. Modell., 2008, 212(3-4):480-491.
11	白玉娇,马贵龙.葡萄霜霉病流行时间动态模型及防治指标[J].吉林农业大学学报,2020,42(3):286-292.
	BAI Y J, MA G L. Dynamic model of epidemic period and control index of grape downy mildew [J]. J. Jilin Agric. Univ., 2020, 42(3):286-292.
12	于舒怡,傅俊范,刘长远,等.沈阳地区葡萄霜霉病流行时间动态及其气象影响因子分析[J].植物病理学报,2016,46(4):529-535.
	YU S Y, FU J F, LIU C Y, et al.. Epidemic temporal dynamic of grape downy mildew and its meteorological influencing factors in Shenyang, Liaoning [J]. Acta Phytopathol. Sin., 2016, 46(4):529-535.
13	BLAISE P, GESSLER C. Development of a forecast model of grape downy mildew on a microcomputer [C]// Proceedings of International Symposium on Computer Modelling in Fruit Research and Orchard Management 276, 1989:63-70.
14	WILKS D S, SHEN K W. Threshold relative humidity duration forecasts for plant disease prediction [J]. J. Appl. Meteorol. Climatol., 1991, 30(4):463-477.
15	华来庆,熊林平,孟虹,等.AR-EGARCH模型在疾病指数时间序列建模中的应用研究[J].中国卫生统计,2006,23(6):482-485.
	HUA L Q, XIONG L P, MENG H, et al.. Application of AR-EGARCH model in establishing methods of disease index time series models [J]. Chin. J. Health Stat., 2006, 23(6):482-485.
16	POUZESHIMIYAB B, FANI S R. Epidemiology and aerobiology of Pseudoperonospora cubensis in northwest Iran [J]. Ital. J. Agrometeorol., 2020, 13(2):109-116.
17	NEUFELD K N, KEINATH A P, OJIAMBO P S. Evaluation of a model for predicting the infection risk of squash and cantaloupe by Pseudoperonospora cubensis [J]. Plant Dis., 2018, 102(5):855-862.
18	LIU Y, LI D, WAN S, et al.. A long short-term memory-based model for greenhouse climate prediction [J]. Int. J. Intell. Syst., 2022, 37(1):135-151.
19	张善文,王振,王祖良.结合知识图谱与双向长短时记忆网络的小麦条锈病预测[J].农业工程学报,2020,36(12):172-178.
	ZHANG S W, WANG Z, WANG Z L. Prediction of wheat stripe rust disease by combining knowledge graph and bidirectional long short term memory network [J]. Trans. Chin. Soc. Agric. Eng., 2020, 36(12):172-178.
20	LIAKOS K G, BUSATO P, MOSHOU D, et al.. Machine learning in agriculture: a review [J]. Int. J. Sci. Technol. Res., 2018, 18(8):2674.
21	MEZEI I, LUKIC M, BERBAKOV L, et al.. Grapevine downy mildew warning system based on NB-IoT and energy harvesting technology [J]. Electronics, 2022, 11(3):356.
22	VOLPI I, GUIDOTTI D, MAMMINI M, et al.. Predicting symptoms of downy mildew, powdery mildew, and gray mold diseases of grapevine through machine learning [J]. Ital. J. Agrometeorol., 2021, 14(2):57-69.
23	CHEN M, BRUN F, RAYNAL M, et al.. Forecasting severe grape downy mildew attacks using machine learning [J/OL]. PLoS One, 2020, 15(3):e0230254 [2022-09-10]. .
24	吴宁.基于灰色关联分析和优化SVM的葡萄霜霉病短期预测[D].上海:上海海洋大学,2020.
	WU N. Short-term prediction of grape downy mildew based on grey correlation analysis and optimized SVM [J]. Shanghai: Shanghai Ocean University, 2020.
25	MENESATTI P, ANTONUCCI F, COSTA C, et al.. Multivariate forecasting model to optimize management of grape downy mildew control [J]. Vitis, 2013, 52(3):141-148.
26	FIRANJ SREMAC A, LALIC B, MARCIC M, et al.. Toward a weather-based forecasting system for fire blight and downy mildew [J/OL]. Atmosphere, 2018, 9(12):484 [2022-09-10]. .
27	边凤霞,王富霞,刘静,等.北疆葡萄设施延晚栽培病虫害防治技术[J].农业科技通讯,2019,49(9):344-345, 354.
28	王济川,郭志刚. Logistic回归模型——方法与应用[M].北京:高等教育出版社,2001:10-17.
29	JOSE C, GOYAL P, AGGRWAL P, et al.. Local deep kernel learning for efficient non-linear svm prediction [C]// Proceedings of International Conference on Machine Learning. PMLR, 2013:486-494.
30	SONG Y Y, YING L U. Decision tree methods:applications for classification and prediction [J]. Shanghai Arch. Psychiatry, 2015, 27(2):130.
31	GUO G, WANG H, BELL D, et al.. KNN model-based approach in classification [C]// Proceedings of OTM Confederated International Conferences on the move to Meaningful Internet Systems. Berlin, Heidelberg, Springer, 2003:986-996.
32	LIU K, ZHANG C, YANG X, et al.. Development of an occurrence prediction model for cucumber downy mildew in solar greenhouses based on Long Short-Term Memory neural network [J/OL]. Agronomy, 2022, 12(2):442 [2022-09-10]. .
33	宋旺.基于SVM的葡萄霜霉病病害发生预测研究[J].中小企业管理与科技,2018,11(5):104-105.
34	秦华.基于自适应神经网络的葡萄病害发生预测研究[D].保定:河北农业大学,2009.
	QIN H. Research on the prediction of grape disease based on adaptive neural networks [J]. Baoding: Hebei Agricultural University, 2009.
35	于舒怡,李柏宏,王辉,等.基于田间空气中病菌孢子囊浓度的葡萄霜霉病病情估计模型研究[J].果树学报,2021,38(10):1767-1777.
	YU S Y, LI B H, WANG H, et al.. Study on estimation model for grape downy mildew prediction based on airborne sporangium concentration of Plasmopara viticola in field [J]. J. Fruit Sci., 2021, 38(10):1767-1777.
36	RODRIGUEZ-RAJO F J, JATO V, FERNANDEZ-GONZALEZ M, et al.. The use of aerobiological methods for forecasting Botrytis spore concentrations in a vineyard [J]. Grana, 2010, 49(01):56-65.
37	CAFFI T, ROSSI V, LEGLER S E, et al.. A mechanistic model simulating ascosporic infections by Erysiphe necator, the powdery mildew fungus of grapevine [J]. Plant Pathol., 2011, 60(3):522-531.
38	GONZALEZ-DOMINGUEZ E, CAFFI T, CILIBERTI N, et al.. A mechanistic model of Botrytis cinerea on grapevines that includes weather, vine growth stage, and the main infection pathways [J/OL]. PloS ONE, 2015, 10(10):e0140444 [2022-09-10]. .
39	HILL G N, BERESFORD R M, EVANS K J. Automated analysis of aggregated datasets to identify climatic predictors of Botrytis bunch rot in wine grapes [J]. Phytopathology, 2019, 109(1):84-95.

[1]	卢倩倩, 阿布都外力·阿不力米提, 侯毅兴, 李志慧, 王爽, 周龙. 复合盐碱胁迫下7个鲜食葡萄品种光合特性研究[J]. 中国农业科技导报, 2023, 25(7): 63-76.
[2]	王爽, 侯毅兴, 冯琳骄, 卢倩倩, 周龙. 干旱胁迫对鲜食葡萄叶片解剖结构的影响[J]. 中国农业科技导报, 2023, 25(6): 40-49.
[3]	周宇, 李佳玉, 王乐, 贾晓爽, 高思琦, 王潇, 焦健. 金黄色葡萄球菌小菌落突变体诱导筛选及特性研究[J]. 中国农业科技导报, 2023, 25(5): 147-157.
[4]	贾毅男, 张光弟, 张浩宇, 许昌, 张昆明, 王江龙, 侯晓健. 根区施用硅肥对‘玫瑰香’葡萄果实品质的影响[J]. 中国农业科技导报, 2023, 25(5): 215-223.
[5]	张娜, 闫亚茹, 武运, 张宇宏, 张伟. 信号肽优化提高葡萄糖氧化酶在毕赤酵母中的表达量[J]. 中国农业科技导报, 2023, 25(2): 211-219.
[6]	吴楠, 杨君, 张艳, 孙正文, 张冬梅, 李丽花, 吴金华, 马峙英, 王省芬. 过表达棉花葡萄糖醛酸激酶基因GbGlcAK促进拟南芥细胞伸长[J]. 中国农业科技导报, 2022, 24(6): 36-46.
[7]	刘阳阳, 潘越, 王世伟, 虎海防. 不同山葡萄品种光响应模型拟合及综合评价[J]. 中国农业科技导报, 2022, 24(2): 104-114.
[8]	朱燕芳, 白耀栋, 王元元, 李玉斌, 马麒龙, 郝燕, 赵明新. 遮阴处理对酿酒葡萄‘西拉’果实皱缩的影响[J]. 中国农业科技导报, 2022, 24(1): 54-62.
[9]	吴婕, 宫江宁. 大孔树脂纯化甜茶多酚及其对 α-葡萄糖苷酶抑制活性和DPPH·抗氧化性的研究[J]. 中国农业科技导报, 2021, 23(6): 113-119.
[10]	阿依图拉·拜各吐木尔, 李洁, 陈存坤, 刘慧, 李相阳, 林少华. 葡萄柚籽提取物和聚乙烯包装对香椿贮藏品质的影响[J]. 中国农业科技导报, 2021, 23(5): 116-123.
[11]	孙艳,韩斌,袁军伟,刘长江,李敏敏,尹勇刚,贾楠,郭紫娟,赵胜建*. 不同抗寒性葡萄品种根系形态结构及生理性状差异分析[J]. 中国农业科技导报, 2021, 23(4): 47-57.
[12]	孔德谦1,何振嘉2*,刘全祖3,吴永杰1. 施肥对贺兰山东麓滴灌条件下‘赤霞珠’葡萄产量和品质的影响[J]. 中国农业科技导报, 2021, 23(3): 148-155.
[13]	王东1,曹源倍2,吉遥芳3,傅渝亮3*. 不同滴灌量对红寺堡区酿酒葡萄生长和品质的影响[J]. 中国农业科技导报, 2021, 23(1): 154-161.
[14]	尹勇刚,袁军伟,刘长江,韩斌,李敏敏,孙艳,贾楠,郭紫娟,赵胜建*. NaCl胁迫对葡萄砧木光合特性与叶绿素荧光参数的影响[J]. 中国农业科技导报, 2020, 22(8): 49-55.
[15]	张艳丽1,2,任柳3,张宇宏1*,张伟1. 棉铃虫葡萄糖氧化酶HaGOX在毕赤酵母中的重组表达和性质研究[J]. 中国农业科技导报, 2020, 22(8): 56-63.