High-stem cold-resistant plum blossom cultivation and grafting product consistency evaluation method

Abstract

The application discloses a high-pole cold-resistant plum blossom cultivation method, which comprises the following steps: (1) rootstock cultivation: selecting Mei Renmei which is consistent in growth, healthy and free from diseases and insect pests as the rootstock; (2) true plum selection and cultivation of scion: selecting a true plum variety which is free from diseases and insect pests, healthy in growth and pure in variety, selecting a branch which is developed and full from the periphery of the plum tree, and soaking the branch with auxin before grafting and keeping the humidity; and (3) multi-head grafting method and later management. The application further discloses a high-pole cold-resistant plum blossom grafting product consistency evaluation method, which takes growth rate and climate factors as input variables, combines a machine learning model GBDT, a SVM and a GBDT+SVM product evaluation model, and selects the GBDT+SVM model which can be used for the prediction and evaluation of the plum blossom product consistency. The seedling formed by the method has good cold resistance, consistent plant type and uniform specification, and greatly improves the product consistency.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of garden plants, in particular to a high-pole cold-resistant plum blossom cultivation and grafting product consistency evaluation method. BACKGROUND

[0002] Plum (Prunus mume) is an important ornamental plant in China, with important application value and cultural connotation. Plum is originally from the southwest of China and the Yangtze River Basin, and is a subtropical tree species. Low temperature is an important factor limiting the distribution and growth of plum. At present, there are few varieties that can be applied in northern China, and the cultivation of cold-resistant varieties has been an important direction of plum breeding.

[0003] It takes at least 8 years to cultivate a shaped plum with a breast diameter of about 10 cm, which greatly increases the time cost and economic cost for garden nursery practitioners. If fast-growing, consistent, and uniform product specifications are produced, it has always been one of the key technologies for garden practitioners. In addition, due to the cold in the north, plum application in the north is mainly apricot plum hybrid varieties, and there are very few true plum varieties. In addition, plum is not easy to survive in the north, and the plum that survives grows slowly, has a high mortality rate, and there are few cold-resistant plum varieties. Therefore, cultivating cold-resistant varieties is one of the important directions of plum breeding. SUMMARY

[0004] The technical problem to be solved by the present application is to provide a high-pole cold-resistant plum cultivation method which is simple to operate. In addition, machine learning is a cross-disciplinary subject involving multiple fields. It can automatically process the relationship between characteristic variables and target variables, mine rules from training data, and ultimately obtain learning experience from the structural description of these data. The present application proposes a plum product consistency evaluation method according to the growth rate of plum and climate factors.

[0005] A high-pole cold-resistant plum cultivation method, comprising the following steps:

[0006] (1) Rootstock cultivation and germination rate statistics: select healthy and disease-free Mei Mei as rootstock, and Mei Mei is obtained by seedling propagation;

[0007] The seed treatment method is as follows:

[0008] ① The undamaged and full seeds are disinfected with KMnO4 for 5 min, then put into 20*20 cm self-sealing bags, and treated in the dark and 0-4℃ environment for 40-60 days;

[0009] ② Soak the seeds in a mixture of 0.1 mg / L IAA and 2.0 mg / L GA for 2 days, then store in wet sand with humidity above 75% and temperature 0-8℃;

[0010] ③ 60 days later, the seeds are germinated and planted in a seedbed, and the humidity is maintained by spraying every day, so that the stock seedlings with consistent growth and robustness can be obtained.

[0011] 6 months later, the germination rate of the seeds is counted, and a traditional treatment method (only sand storage treatment) is used as a control, and the formula is as follows:

[0012] Germination rate = (number of germinated seeds / total number of treated seeds) * 100%.

[0013] (2) True plum scion selection and cultivation: selecting true plum varieties without diseases and pests, robust growth and pure varieties, selecting well-developed and full branches from the periphery of the plum tree, soaking with auxin before grafting and keeping humidity;

[0014] (3) Multi-head grafting method and post-management:

[0015] 1) Time selection: selecting warm weather and avoiding rainy days;

[0016] 2) Stock selection and adjustment: adjusting the main stem and skeleton, and removing other redundant branches;

[0017] 3) Grafting: grafting the scion treated with auxin by budding, coating the upper end of the scion with a tree wound healing paste (Happy Forest, Zhengzhou, Henan), and then tightly binding with a plastic bandage;

[0018] 4) Post-grafting management.

[0019] The high-pole cold-resistant plum cultivation method, wherein in step (2), the selected branches are woody branches with a diameter of 0.6-1.0 cm, soaked with NAA (Aladdin, Shanghai) with a concentration of 0.1 mg / L for 10 min before grafting, bundled into 30-50 branches per bundle, with a length of 6-8 cm, and kept at 70% humidity. The true plum variety is 'Guli Hong', 'Beijing Yudie', 'Sanlun Lecal', 'Fenghou' or 'Fuxia'.

[0020] The high-pole cold-resistant plum cultivation method, wherein step (3) specifically comprises the following steps:

[0021] 1) Time selection: the grafting time is selected in May, the ground temperature is kept above 15℃, and the rainy weather is avoided;

[0022] 2) Stock selection and adjustment: selecting a stock with a height of 100-180 cm and a diameter of 6-8 cm; adjusting the main stem and skeleton of the selected stock to have a primary and secondary feeling, leaving 3-5 main branches with a length of 20-35 cm, and removing other redundant branches;

[0023] 3) Grafting: the scion treated with auxin is grafted by budding, 12-15 buds per plant, the upper end of the scion is smeared with a tree wound healing paste to prevent water and nutrient loss from the scion, and then a plastic bandage is tightly bound; each main branch is ensured to have a bud;

[0024] 4) Post-grafting management: after grafting, the scion is protected from wind and sun; at the same time, about 15 days later, the rootstock is touched to sprout, to prevent the rootstock from sprouting branches that compete for nutrients and water; in addition, the finished product is pinched to expand the crown, to promote branching, and to remove dead branches. On the basis of routine pruning and shaping, in order to eliminate differences between individual products, between April and June of the following year, the overgrown high-stem multi-headed product is sprayed with 800 mg / L of paclobutrazol once a week.

[0025] A plum grafting product consistency evaluation method, comprising the following steps:

[0026] The high-stem cold-resistant plum is cultivated by using the high-stem cold-resistant plum cultivation method in the application, and the consistency of the grafted product is evaluated in combination with meteorological data and growth rate; the machine learning models GBDT, SVM, and GBDT+SVM are used for data training and testing by taking the growth rate and the climate factor as input variables, to build a product evaluation model containing climate variables;

[0027] The specific evaluation method comprises the following steps:

[0028] (A) Meteorological data download: the planting area of the product is mainly in Xinyi, Jiangsu, and the meteorological observation data of the area is obtained from the China Meteorological Data Network (http: / / data.cma.cn / ), including rainfall, air temperature, and sunshine duration;

[0029] (B) Data normalization: the mapminmax normalization method is used;

[0030] (C) Machine learning model: the gradient boosting tree GBDT, the support vector machine SVM model, and the combination of the two are used;

[0031] (D) Model evaluation: 10-fold cross-validation is used to compare and optimize the model, that is, the data set is divided into 10 parts, 7 parts are used as training data, and the remaining 3 parts are used as validation data, which is repeated 10 times until all the data sets are involved in training and evaluation; the Pearson correlation coefficient statistical index is used to evaluate the simulation accuracy of the model.

[0032] The plum grafting product consistency evaluation method provided in the application, wherein the mapminmax normalization method is used in step (B), which is a row-by-row standardization of data, that is, all data values in each row are scaled to the range of [-1, 1];

[0033]

[0034] The formula is as above.

[0035] In the plum grafting product consistency evaluation method, the gradient boosting decision tree (GBDT) model in step (C) is realized by using an additive model and a forward stepwise algorithm to learn and optimize, and is a model based on Booting method integration; in the GBDT regression algorithm, a regression tree is used as a weak learner, and when learning, the negative gradient of the loss function is used as the pseudo residual error of the current learner, and a regression tree is fitted according to the pseudo residual error, and the process is repeated until the requirement is met; the GBDT algorithm has fast convergence speed and is not prone to overfitting, and the principle expression is as follows:

[0036]

[0037] F(x) is a regression tree; x is an independent variable; M is the number of regression decision trees; J is the complexity of the regression tree, that is, the number of leaf nodes; C mj Loss function minimum value; I is a loss function; R mj is the leaf node area of the mth tree;

[0038] The GBDT algorithm package "gbm" in R language is used for calculation, and the required parameters are as follows: the distribution function is selected as "gaussian", the minimum number of tree terminal nodes n.minobsinnode is 10, the learning rate shrinkage is 0.01, the number of regression trees n.trees is 100, and the maximum depth of a single regression tree interaction.depth is 1 as a step in [0, 10].

[0039] In the plum grafting product consistency evaluation method, in step (D), the support vector machine (SVM) is a classic model for linear and nonlinear regression and classification, and in the feature space of sample distribution, an hyperplane is found to accurately separate two types of samples, and in the process of constructing the SVM, the penalty coefficient C and the sigma parameter are set; the greater the value of the parameter C, the farther the distance between the hyperplane and the data points on both sides, and the higher the accuracy obtained; the parameter sigma determines the distribution of the data after being mapped to a new feature space, and controls the influence of a single training sample, and the smaller the value, the greater the influence, thereby affecting the fitting degree and generalization ability of the model.

[0040] The main steps of the SVM regression algorithm are as follows:

[0041]

[0042] f(x) is a model target variable; ω is a weight vector; is a nonlinear function mapping the input spatial vector x to a high-dimensional feature space; b is a bias; the parameters ω and b are determined by the structural risk minimization theory;

[0043]

[0044] wherein, is a regularization term, and C is a penalty factor; L ε is an epsilon-insensitive loss function; f(x i ) is a predicted value; y i is a true value;

[0045] The model is constructed by using the R language "kernlab" package, the C value is set to 0.01, 0.1, 1 and 10, the sigma value is set to 0-1, and the interval is 0.1; the "caret" software package is used for 10-fold cross-validation training and evaluation of the model.

[0046] The plum grafting product consistency evaluation method, wherein, in step (D), to ensure the comparability of machine learning, the training set and the test set of each fold of all models should be kept consistent;

[0047]

[0048] y i and x i are the fitting value and the observed value of the relative meteorological product consistency model respectively; before model simulation, 70% of the observation data are randomly selected as the training data set, and 30% of the data are selected as the test data set; to reduce the influence of the maximum and minimum values of the model precision, the final precision of the model is the 50% quantile of the error of 100 times of operation.

[0049] The plum grafting product consistency evaluation method, wherein, the growth rate is calculated as follows: the consistency of the seedling mainly reflects on the growth rate of the branch, in order to promote the consistency of the high-pole multi-head plum product, the product with vigorous growth is sprayed with a concentration of 1000 mg / L of chlormequat, once a week, after one year of stable growth, more than 10 grafted products are randomly selected, the plum branch length is measured, the growth rate is calculated, and the calculation formula is as follows:

[0050] Growth rate = [(length of branch before treatment - average length of branch after treatment) / average length of branch after treatment] * 100%.

[0051] The high-pole cold-resistant plum cultivation and grafting product consistency evaluation method of the present application is different from the prior art in that:

[0052] The present application uses Prunus x blireiana with good resistance, vigorous growth and good plant type as a rootstock, and quickly forms a plum blossom through high grafting; the grafted seedling has good cold resistance, consistent plant type and uniform specification, and greatly improves the consistency of the product.

[0053] High-quality seedlings have strict requirements for flowering period and flower amount (the number of flowers on a tree), and product consistency is an important indicator of landscape seedlings. The present application mainly evaluates the consistency of grafted products in combination with meteorological data and growth rate, selects 'Gunihong' and Prunus x blireiana grafted products as input variables, and growth consistency rate as output variable, and constructs a product evaluation model containing climate variables.

[0054] The model constructed by meteorological and growth data can quickly predict and screen plum blossom products with consistent growth in the region, effectively avoiding the subjective differences of manual selection, and bringing certain convenience to the application of plants in the later period; at the same time, this selection model can also be extended to other woody landscape plants.

[0055] The high-pole cold-resistant plum blossom cultivation and grafted product consistency evaluation method of the present application will be further described below in combination with the drawings. DETAILED DESCRIPTION

[0056] Figure 1 Seed germination rate of the rootstock treated by the combined method of the present application; wherein, CT: represents the traditional sand storage method; ZH: represents the combined treatment method of the present application;

[0057] Figure 2 Effect of auxin treatment on plum graft survival rate in the present application; A: Gunihong / Prunus x blireiana grafting; B: Sanlunlvei / Prunus x blireiana grafting; C: Prunus x blireiana / Prunus x blireiana grafting; D: Beijingyudie / Prunus x blireiana grafting; E: Fenghou / Prunus x blireiana grafting; F: Fenxia / Prunus x blireiana grafting;

[0058] Figure 3 Product effect diagram after paclobutrazol treatment in the present application;

[0059] Figure 4 Growth rate diagram of high-pole multi-head plum blossom product after paclobutrazol treatment in the present application;

[0060] Figure 5 Rainfall (A), air temperature (B) and sunshine duration (C) change diagram in Xinyi, Jiangsu in the present application;

[0061] Figure 6 Technical route schematic diagram of model training in the present application;

[0062] Figure 7A schematic diagram of the prediction accuracy of the product consistency model based on meteorological and growth rate characteristic variables in the present application;

[0063] Figure 8 High-pole multi-head plum flower grafted by the method of the present application. DETAILED DESCRIPTION

[0064] I. Materials and Methods

[0065] (1) Rootstock cultivation

[0066] The Mei Mei is obtained by seedling breeding. In order to maintain the specifications and consistency of the product, Mei Mei with consistent growth, healthy and free of pests and diseases is selected as the rootstock.

[0067] (2) True plum scion selection and cultivation

[0068] Disease-free, healthy, and pure variety true plum varieties are selected. True plum varieties include Guli Hong, Beijing Yudie, Sanlun Lecal, Mei Mei, Fenghou, and Fenxia. Fully developed and full branches with a diameter of 0.6-1.0 cm are selected from the periphery of the plum tree. Before grafting, soak the branches in 0.1 mg / L NAA for 10 minutes, bundle them into 30-50 branches per bundle, with a length of 6-8 cm, and maintain a humidity of 70%.

[0069] (3) Multi-head grafting method and post-management

[0070] ① Time selection: The grafting time is selected in May, the ground temperature is maintained above 15℃, and the rainy weather is avoided. ② Rootstock selection and adjustment: Select rootstocks with a height of 100-180 cm and a diameter of 6-8 cm. Adjust the main trunk and skeleton of the selected rootstock to have a primary and secondary feeling, leaving 3-5 main trunk branches with a length of 20-35 cm, and removing other excess branches. ③ Grafting: The scion treated with auxin is grafted by budding, with 12-15 buds per plant. The upper end of the scion is smeared with a tree wound healing paste to prevent water and nutrient loss, and then tightly bound with plastic bandage; each main trunk branch is ensured to have buds. ④ Post-grafting management: After grafting, the scion is protected from wind and sun; at the same time, about 15 days after grafting, the rootstock is touched to prevent the rootstock from sprouting branches competing for nutrients and water. In addition, the finished product is pinched to expand the crown and promote branching; at the same time, dead branches are removed.

[0071] On the basis of daily pruning and shaping, in order to eliminate the differences between individual products, 800 mg / L of paclobutrazol is sprayed on the overgrown high-pole multi-head products between April and June of the following year, once a week.

[0072] (4) Grafting survival rate statistics

[0073] After grafting for one month, when the scion grew stably, the survival rate was counted:

[0074] Survival rate = (number of living scions / total number of grafts) x 100%.

[0075] (5) Growth rate determination

[0076] The uniformity of seedlings is mainly reflected in the growth rate of branches. In order to promote the uniformity of high-stem multi-head plum products, the vigorous products were sprayed with 1000 mg / L of chlormequat, once a week. After one year of stable growth, more than 10 grafted products were randomly selected, the plum branch length was measured, and the growth rate was calculated. The calculation formula is as follows:

[0077] Growth rate = [(branch length before treatment - average length of branches after treatment) / average length of branches after treatment] x 100%.

[0078] (6) Determination of cold tolerance related indicators

[0079] ① Conductivity determination

[0080] Cut the branches into small pieces of 1-2 mm thick, mix each group, weigh 1.5 g into a calibrated test tube, add 15 ml of deionized water, stand for 12 h, and measure the conductivity Cl with a conductivity meter (Lei magnet, DDS-11A). Boil in water bath for 30 min, cool to room temperature, measure the conductivity C2, and measure the conductivity of distilled water Co. Each grafting group is set with 3 replicates.

[0081]

[0082] In the formula, L is the relative conductivity, C1 is the initial conductivity after low temperature treatment, C2 is the final conductivity after boiling, and Co is the conductivity of distilled water.

[0083] ② Determination of semilethal temperature

[0084] The treatment temperature and relative conductivity are fitted by using the Logistic equation:

[0085]

[0086] y represents the cell damage rate, K is the saturation capacity of cell damage rate, x is the treatment temperature, a and b are equation parameters. The values of a and b are obtained from the equation ln[(K-y) / y]=lna-bx. The semilethal temperature is:

[0087]

[0088] (7) Product uniformity evaluation:

[0089] High-quality nursery stock, especially for the requirements of flowering period, flower quantity (referring to the number of flowers on a tree), product consistency is an important indicator of landscape nursery. In this study, the consistency of grafted products was evaluated by combining meteorological data and growth rate. The growth rate and climate factors were selected as input variables, and the growth consistency rate was selected as output variable. The grafted products of Gushuli Hong and Meiren Mei were selected to construct the product evaluation model containing climate variables.

[0090] ① Downloading meteorological data

[0091] The planting area of this product is mainly in Xinyi, Jiangsu. The meteorological observation data of this area comes from China Meteorological Data Network (http: / / data.cma.cn / ), including rainfall, air temperature, sunshine hours, etc.

[0092] ② Data normalization

[0093] Data normalization is a common preprocessing step that scales input data to a specific range, thereby improving the stability and convergence speed of the model. The mapminmax normalization method is used in this invention. This method standardizes data row by row, which scales all data values in each row to the range of [-1, 1]. The formula is as follows:

[0094]

[0095] ③ Machine learning model

[0096] Gradient boosting decision tree (GBDT): This model uses additive model and forward step algorithm to realize learning optimization, which is an integrated model based on Booting method. In GBDT regression algorithm, regression tree is used as weak learner. During learning task, the negative gradient of loss function is used as pseudo residual error of current learner. Regression tree is fitted according to pseudo residual error, and the process is repeated until the requirement is met. GBDT algorithm has fast convergence speed and is not prone to overfitting. Its principle expression is as follows:

[0097]

[0098] F(x) is the regression tree; x is the independent variable; M is the number of regression decision trees; J is the complexity of regression tree, i.e. the number of leaf nodes; C mj Loss function minimum value; I is the loss function; R mj is the leaf node area of the mth tree.

[0099] The parameters required for calculation by using the GBDT algorithm package "gbm" in R language are as follows: the distribution function is selected as "gaussian", the minimum number of tree end nodes n.minobsinnode is 10, the learning rate shrinkage is 0.01, the number of regression trees n.trees is 100, and the maximum depth of a single regression tree interaction.depth is 1 as a step in [0, 10].

[0100] Support Vector Machines (SVM): a classic model for linear and nonlinear regression and classification, mainly finding a hyperplane to accurately separate two classes of samples in the feature space of sample distribution. In the process of constructing SVM, the penalty coefficient C and the sigma parameter are set. The greater the value of parameter C, the farther the distance between the hyperplane and the data points on both sides, and the higher the accuracy obtained; parameter sigma determines the distribution of data mapping to the new feature space, controls the influence of a single training sample, and the smaller the value, the greater the influence, thereby affecting the fitting degree and generalization ability of the model.

[0101] The main steps of the SVM regression algorithm are as follows:

[0102]

[0103] f(x) is the model target variable; ω is the weight vector; is a kernel function, which is a nonlinear function that maps the input space vector x to a high-dimensional feature space; b is a bias. Parameters ω and b are determined by structural risk minimization theory.

[0104]

[0105] In the formula, is a regularization term, C is a penalty factor; L ε is an epsilon-insensitive loss function; f(x i ) is a predicted value; y i is a true value.

[0106] The model is constructed by using the "kernlab" package of R language, the value of C is set to: 0.01, 0.1, 1, 10, the value of sigma is set to 0-1, and the interval is 0.1; the "caret" software package is used for 10-fold cross-validation training and evaluation of the model.

[0107] (4) Model evaluation

[0108] The present application adopts 10-fold cross-validation to compare and evaluate the model, that is, the data set is divided into 10 parts, 7 parts are used as training data, and the remaining 3 parts are used as validation data, repeated 10 times, until all the data sets are involved in training and evaluation. The Pearson correlation coefficient is used as a statistical index to evaluate the simulation accuracy of the model. To ensure the comparability of machine learning, the training set and test set of all models should be consistent.

[0109]

[0110] y i and x i are the relative meteorological product consistency model fitting value and observed value respectively; n is the sample number. Before model simulation, 70% of the observation data is randomly selected as the training data set, and 30% of the data is selected as the test data set. In order to reduce the influence of the maximum and minimum values of the model precision, the final precision of the model is the 50% quantile of the error of 100 runs.

[0111] II. Results and analysis:

[0112] (1) Statistics of the germination rate of the stock

[0113] As shown in Figure 1 , the traditional seed treatment method of plum blossom needs to go through the sand storage stage, and can germinate in May of the next year, with a very low germination rate of only 31.34%. Through the combined method of the present application, the germination rate of the Mei Ren Mei seed can be significantly improved to 76.29%, and the germination time is advanced by 2 months.

[0114] (2) Statistics of the survival rate of grafting

[0115] After one month of grafting, when the scion grows stably, the survival rate of 5 Mei Ren Mei grafting combinations is counted Figure 2 . In the Guni Hong / Mei Ren Mei grafting combination, the survival rate of the untreated is 80.6%, and the survival rate after treatment is 91.3% Figure 2 A); in the Sanlun Lè / Mei Ren Mei combination, the survival rate of the untreated is 83.1%, and the survival rate after treatment is 88.0% Figure 2 B); in the Beijing Yudie / Mei Ren Mei combination, the survival rate of the untreated is 86.2%, and the survival rate after treatment is 92.2% Figure 2 C); in the Fenghou / Mei Ren Mei combination, the survival rate of the untreated is 79.0%, and the survival rate after treatment is 85.8% Figure 2 D); in the Fenxia / Mei Ren Mei combination, the survival rate of the untreated is 85.5, and the survival rate after treatment is 91.7% Figure 2E). In summary, compared with the untreated group, the auxin treatment significantly improved the grafting survival rate.

[0116] (3) Effect of paclobutrazol on product growth rate

[0117] In addition to routine pruning and shaping, to eliminate individual differences among the plants, from April to June, excessively vigorous, tall, multi-headed plants should be sprayed with 1000 mg / L paclobutrazol once a week. Figure 3 and Figure 4 As shown, paclobutrazol not only inhibited the growth rate of grafted seedling branches, but also significantly improved the uniformity of product growth. The growth rate among the treated products was less than 10%. At the same time, during the flowering period, the flowering difference among the treated grafted products was less than 2%, which significantly improved the flowering uniformity of tall, multi-headed products.

[0118] (3) Comparison of cold resistance of grafted combinations

[0119] Low-temperature stress primarily affects cell membranes, leading to damage to the protoplasmic structure, increased cell membrane permeability, and significant leakage of intracellular substances. The amount of electrolyte leakage can be used as an indicator of a plant's cold resistance. The greater the electrolyte leakage, the greater the damage to the cell membrane and the weaker the cold resistance. The amount of electrolyte leakage can be measured using relative conductivity. Relative conductivity is negatively correlated with a plant's cold resistance.

[0120] After grafting, this invention compared and analyzed the relative electrical conductivity and proline content of grafted combinations of *Prunus mume* and different cultivated plum varieties. Table 1 shows that, compared with *Prunus mume* varieties, the relative electrical conductivity of the five grafted combinations of *Prunus mume* decreased after treatment at -10℃, 20℃, and 30℃, indicating that the cold resistance was improved after grafting. The half-lethal temperature is the critical temperature for cold resistance, which is the temperature at which more than 50% of the branches freeze to death. The electrical conductivity method combined with the Logistic equation is a commonly used method for calculating the half-lethal temperature. In this experiment, the half-lethal temperatures of the five grafted combinations were calculated using the Logistic regression equation, as shown in Table 2. *Gulihong* / *Prunus mume* -31.80℃, *Beijing Yudie* / *Prunus mume* -32.79℃, *Sanlun Lü'e* / *Prunus mume* -29.64℃, *Fenghou* / *Prunus mume* -29.63℃, *Fenxia* / *Prunus mume* -30.55℃. The semi-lethal temperatures of the five scion groups were: *Gulihong* -28.51℃, *Beijing Yudie* -25.66℃, *Sanlun Lv'e* -23.38℃, *Fenghou* -24.54℃, and *Fenxia* -27.76℃. This indicates that grafting with *Prunus mume* can improve the cold resistance of grafted seedlings.

[0121] Table 1. Relative conductivity of Prunus mume and different grafting combinations

[0122]

[0123] Table 2. Semi-lethal temperatures of Prunus mume and different grafting combinations

[0124]

[0125] (4) Consistency evaluation of high-pole multi-head plum blossom products

[0126] ① Changes in meteorological variables

[0127] This study analyzes meteorological data from 2021 to 2023 in Xinyi, Jiangsu Province. Rainfall in the region is concentrated in July, while maximum temperatures are concentrated in June, July, and August. Overall, summer has longer sunshine hours. Figure 5 ).

[0128] ② Model training parameter settings and model building

[0129] like Figure 6 As shown, to construct an evaluation and prediction model with small errors and high accuracy, this invention selects the most important meteorological characteristic variables—rainfall, temperature, sunshine duration, and growth rate—as the input feature variable combination for machine learning models (GBDT, SVM, GBDT+SVM) to construct a product consistency prediction model. The construction process mainly includes data normalization, partitioning the training and test sets, selecting kernel functions and parameters, and model training to ultimately achieve accurate predictions. The dataset is randomly partitioned into training and test sets in a 7:3 ratio using the cvpartition function. After partitioning the dataset, data normalization is performed. Next, to select a suitable kernel function, the prediction performance of each model is compared.

[0130] ③ Model Evaluation

[0131] like Figure 7 As shown, the model accuracy was evaluated using the correlation coefficient (r). A higher r value indicates a more stable model. The data points were evenly distributed on both sides of the 1:1 line: r = 0.63 for the GBDT model, r = 0.58 for the SVM model, and r = 0.81 for the GBDT+SVM model. The simulation results show that the simulation accuracy exhibits a changing trend. Overall, the product consistency prediction model based on the GBDT+SVM machine learning algorithm can be used for product consistency evaluation. Furthermore, based on climatic factors and growth rate, a paclobutrazol concentration of 800 mg / L indicates good consistency in grafted products of *Prunus armeniaca* and *Prunus mume*. This model can also be used to predict and evaluate the consistency of other plum blossom products.

[0132] The above-described embodiments are merely intended to describe the preferred embodiments of the present application, and are not intended to limit the scope of the present application. Various modifications and improvements to the present application made by those skilled in the art are intended to fall within the scope of the present application defined in the claims.

Claims

1. A method for evaluating the consistency of a product of a pinching graft, characterized by: Comprising the following steps: The cultivation of high-pole cold-resistant plum blossoms is carried out, meteorological data and growth rate data are combined, machine learning models GBDT, SVM and GBDT+SVM are used for data training and testing, and a plum grafting product evaluation model containing climate variables is constructed; The specific evaluation method comprises the following steps: (A) Meteorological data download: the planting area of the product is in Xinyi, Jiangsu, and the meteorological observation data of the area is from the China Meteorological Data Network, including rainfall, air temperature and sunshine duration; (B) Data normalization: the mapminmax normalization method is adopted; (C) Machine learning model: gradient boosting tree GBDT, support vector machine SVM model and their combination are adopted; (D) Model evaluation: 10-fold cross-validation is adopted to compare and optimize the model, that is, the data set is divided into 10 parts, 7 parts are used as training data, and the remaining 3 parts are used as validation data, which is repeated 10 times until all the data sets are involved in training and evaluation; the Pearson correlation coefficient statistical index is used to evaluate the simulation accuracy of the model; The calculation method of the growth rate is as follows: the consistency of the seedling is reflected in the growth rate of the branches, the products with vigorous growth are sprayed with 1000 mg / L of chlormequat, once a week, and after one year of stable growth, more than 10 grafting products are randomly selected, the plum branch length is measured, the growth rate is calculated, and the calculation formula is as follows: Growth rate = 【(treatment before branch growth length-treatment after branch average length) / treatment after branch average length】x 100%.

2. The evaluation method of the consistency of the pin grafting products according to claim 1, characterized in that: In step (B), the mapminmax normalization method is adopted, which standardizes the data row by row, that is, all data values in each row are scaled to [-1, 1]; ， The formula is as follows.

3. The method for evaluating the consistency of the pin grafting product according to claim 2, characterized in that: In step (C), the gradient boosting tree GBDT model adopts additive model and forward step algorithm to realize learning optimization, which is a model based on the Booting method; in the GBDT regression algorithm, the regression tree is used as a weak learner, the negative gradient of the loss function is used as the pseudo residual error of the current learner during the learning task, and the regression tree is fitted according to the pseudo residual error until the requirement is met to stop repeating the process; GBDT algorithm has fast convergence speed and is not prone to overfitting, and its principle expression is as follows: ， F(x) is a regression tree; x is an independent variable; M is the number of regression decision trees; J is the complexity of the regression tree, i.e. the number of leaf nodes; C mj Loss function minimum value; I is a loss function; R mj is the leaf node area of the mth tree The GBDT algorithm package "gbm" in R language is used for calculation, and the required parameters are as follows: the distribution function is selected as "gaussian", the minimum number of tree terminal nodes n.minobsinnode = 10, the learning rate shrinkage = 0.01, the number of regression trees n.trees = 100, and the maximum depth of single regression tree interaction.depth is taken as 1 step in [0, 10].

4. The plum grafting product consistency evaluation method according to claim 3, wherein: In step (D), the support vector machine (SVM) is a classic model for linear and nonlinear regression and classification, which finds a hyperplane to accurately separate two classes of samples in the feature space of sample distribution. In the process of constructing the SVM, the penalty coefficient C and the sigma parameter are set. The greater the value of the parameter C, the farther the distance between the hyperplane and the data points on both sides, and the higher the accuracy obtained. The parameter sigma determines the distribution of the data after being mapped to the new feature space and controls the influence of a single training sample. The smaller the value, the greater the influence, thereby affecting the fitting degree and generalization ability of the model. The steps of the SVM regression algorithm are as follows: ， f(x) is the target variable of the model; ω is the weight vector; φ(x) is the kernel function, which is a nonlinear function that maps the input space vector x to a high-dimensional feature space; b is the bias; the parameters ω and b are determined by the structural risk minimization theory; ， ， In the formula, is a regularization term, C is a penalty factor; L ε is an ε-insensitive loss function; f(x i ) is a predicted value; y i is a true value; The R language "kernlab" package is used to construct the model, the C value is set to 0.01, 0.1, 1, and 10, and the sigma value is set to 0-1 with an interval of 0.1; the "caret" software package is used for 10-fold cross-validation training and evaluation of the model.

5. The evaluation method of the plum grafting product consistency according to claim 4, characterized in that: In step (D), to ensure the comparability of machine learning, the training set and test set of each fold of all models should be kept consistent. ， y i and x i are the relative meteorological product consistency model fitting value and observation value, respectively; n is the sample number; before model simulation, 70% of the observation data is randomly selected as the training data set, and 30% of the data is selected as the test data set; in order to reduce the influence of the maximum and minimum values of the model precision, the final precision of the model is the 50% quantile of the error of 100 times of operation.

Images