阅读说明：本技术 一种基于qga-svr的冷害黄瓜psii潜在活性预测方法 (Cold damage cucumber PSII potential activity prediction method based on QGA-SVR ) 是由 胡瑾 卢苗 完香蓓 袁凯凯 高攀 李斌 于 2021-01-13 设计创作，主要内容包括：本发明通过分析低温对植物叶片生理状态变化的影响,以不同初始F-v/F-o值黄瓜幼苗为试验样本,测量在不同低温条件、持续时间下的F-v/F-o值变化数据,并构建建模样本集,采用量子遗传支持向量机算法建立低温环境下黄瓜叶片F-v/F-o值变化预测模型。模型训练集决定系数为0.9817,均方根误差为0.2141；测试集决定系数为0.9864,均方根误差为0.1741。结果表明,本发明方法可实现低温环境下的黄瓜叶片F-v/F-o的精准预测,为早期冷害胁迫和作物冷害无损诊断提供了新的研究方法。(The invention analyzes the influence of low temperature on the physiological state change of the plant leaf to obtain different initial F v /F o The cucumber seedlings are used as test samples, and F is measured under different low-temperature conditions and duration v /F o Changing data, constructing a modeling sample set, and establishing cucumber leaf F under a low-temperature environment by adopting a quantum genetic support vector machine algorithm v /F o A value change prediction model. The decision coefficient of the model training set is 0.9817, and the root mean square error is 0.2141; the test set decision factor is 0.9864 and the root mean square error is 0.1741. The results show that the invention is applied toThe method can realize the cucumber leaf F in the low-temperature environment v /F o The accurate prediction of the method provides a new research method for early cold damage stress and nondestructive diagnosis of crop cold damage.)

1. A cold damage cucumber PSII potential activity prediction method based on QGA-SVR is characterized in that different initial chlorophyll fluorescence parameters F are used_v/F_oCucumber seedlings of value are test specimens, and F is measured under different low temperature conditions and different durations_v/F_oChanging data, constructing a modeling sample set, and establishing cucumber leaf F under a low-temperature environment by adopting a quantum genetic support vector machine algorithm_v/F_oA value change prediction model according to which the pair F_v/F_oValues are predicted as F_v/F_oIndicating the potential activity of cold damage cucumber PSII, wherein F_vTo representVariable fluorescence, F_oIndicating minimal fluorescence.

2. The method for predicting PSII (cucumber plant growth regulator) potential activity of cucumber under cold damage based on QGA-SVR as claimed in claim 1, wherein the test sample is cucumber seedlings classified into 4 groups according to initial fluorescence parameters, and the test sample is tested in CO₂Carried out in a climatic chamber, CO₂The environmental parameters in the artificial climate box are set as follows: the photoperiod day/night is 14/10h, the temperature is set test temperature gradient and is respectively 8, 10, 12 and 14 ℃, the relative air humidity day/night is 60%/50%, and CO is₂The concentration is 400 mu mol & mol^-1Four phytotron cucumber seedling leaf fluorescence parameters are collected at 16:00 every day for 8 days continuously, wherein each cucumber plant takes two points on the same leaf position functional leaf to obtain 288 groups of cucumber seedling leaf fluorescence parameters.

3. The method for predicting PSII (latent Activity of cucumber) of cold injury based on QGA-SVR (QGA-SVR) as claimed in claim 2, wherein the fluorescence parameters are collected by a portable MINI-PAM-II type modulated chlorophyll fluorometer, the dark-adapted blade clamp is used to clamp the blade to be tested before the measurement, the dark-adapted blade clamp light barrier is opened after the dark adaptation is performed for 20 minutes, the fluorometer is started to obtain the minimum fluorescence F of the test sample_oMaximum fluorescence F_mAnd variable fluorescence F_vIn which F is_v＝F_m-F_o。

4. The method for predicting PSII (cucumber PSII) potential activity of cucumber with cold injury based on QGA-SVR as claimed in claim 1, wherein in the process of establishing the prediction model, the obtained original test data is first normalized, so that the fluctuation of the data is mapped to [0.2,0.8]]On the interval, dividing the data by 8:2 to generate a training data set and a prediction data set; secondly, selecting a radial basis kernel function by the SVR model, and optimizing model parameters c and g by using a quantum genetic algorithm and a prediction data set decision coefficient as a fitness evaluation index; finally, the environmental temperature, the low-temperature duration, the initial variable fluorescence and the initial minimum fluorescence are used as input, and the fluorescent ginseng is used as the inputNumber F_v/F_oAnd establishing a QGA-SVR cucumber fluorescence parameter prediction model for output, wherein the environmental temperature is the low temperature set by the experiment.

5. The method for predicting PSII (cucumber PSII) potential activity of cucumber with cold injury based on QGA-SVR as claimed in claim 4, wherein the normalization transformation function in the normalization process is

Wherein Y is the normalized data, X is the data to be normalized, X is_maxAnd X_minRespectively a maximum value and a minimum value in the same dimension data sequence, and the sample characteristics of the ith data areThe label is Y_iRepresents the current F of the ith sample_v/F_oThe value of the one or more of, the low temperature, low temperature duration, initial variable fluorescence and initial minimum fluorescence of the ith data sample, respectively.

6. The method for predicting PSII (cucumber PSII) potential activity of cucumber with cold injury based on QGA-SVR as claimed in claim 4, wherein sample data x in SVR model is processed by nonlinear functionConversion to a high-dimensional feature space, denoted as

Wherein w is the coefficient of the self-varying function, w^TIs the transpose of w, and b is the offset.

7. The method for predicting PSII (cucumber PSII) potential activity of cucumber with cold injury based on QGA-SVR as claimed in claim 6, wherein w and b are obtained according to the principle of consistency of minimization of structural risk, and relaxation factor xi is introduced_iAndsupport vector regression optimization problem transformation

Wherein c is a penalty factor, | w |²For model complexity, m is the number of samples, ε is the insensitive loss parameter, and for the constraint, the dual-form lagrange polynomial is transformed into the optimization problem of the following formula:

in the formula a_i,a_j,Is Lagrange multiplier, K (x)_i,x_j) Is a kernel function.

8. The method for predicting PSII (cucumber PSII) potential activity of cucumber with cold injury based on QGA-SVR as claimed in claim 7, wherein the non-linear problem of input space is passed through kernel function K (x)_i,x_j) Mapping to a high-dimensional feature space, and constructing a linear discriminant function in the high-dimensional space so as to solve the nonlinear problem.

9. The method for predicting PSII potential activity of cucumber suffering from cold injury based on QGA-SVR as claimed in claim 7, wherein the penalty factor c and the kernel function parameter g are optimized by using quantum genetic algorithm.

Technical Field

The invention belongs to the technical field of intelligent agriculture, and particularly relates to a cold damage cucumber PSII (cucumber leaf photoreaction center II, PS, photosystem) potential activity prediction method based on QGA-SVR.

Background

Cucumbers are one of the important economic crops in the field of facility agriculture, are widely planted worldwide, and the off-season cultivation area is increased year by year. However, cucumbers like warm and bright, belong to typical cold sensitive crops, and meanwhile, in recent years, the frequency, degree and duration of extreme low temperature events are continuously enhanced, and the cucumbers are easy to suffer from low temperature stress in out-of-season cultivation to cause growth inhibition, so that the problems of influencing yield and quality are increasingly severe. The low temperature can cause phenotypical change, membrane damage, enzyme system disorder, osmotic adjustment substance imbalance, photosynthetic structure damage and the like of crops, so that the photosynthetic rate of the crops is reduced, the growth of the crops is inhibited, and even the crops die. Therefore, how to realize the real-time monitoring of the physiological state of the cucumbers under the cold damage condition and accurately evaluate the damage degree of the cucumbers becomes the key of the efficient planting of the out-of-season cucumber industry.

In order to solve the above problems, many scholars have studied the cold resistance of cucumber in terms of its growth and development state, physiological and biochemical resistance, and photosynthetic rate. Zhang hong Mei et al (2009) established a cold injury index with a low temperature resistance index as a standard, studied the cold tolerance differences of 3 different varieties of cucumbers, and proved that the identification of the cold tolerance of cucumber plants from the aspect of growth indexes alone is not accurate and needs to be comprehensively considered by other methods. Wan et al (2015) showed that the activity of superoxide dismutase (SOD) and Peroxidase (POD) in the active oxygen scavenging system in plant cells significantly correlated with the degree of cold injury; zhang et al (2018) showed that Malondialdehyde (MDA) as one of the end products of the oxidation reaction of unsaturated fatty acids in plant cell membranes has a significant negative correlation with the low temperature tolerance of plants. Yang et al (2019) showed that low temperature destroys the ultrastructure of plant photosynthetic organs, reducing photosynthetic efficiency. The research shows that the effective representation of various physiological parameter indexes of the cucumber on the cucumber state under the cold damage condition provides theoretical support for the parameter selection of a cucumber cold damage physiological state prediction model. However, most of the researches are destructive detection, the operation steps are complicated, and the portable nondestructive detection of the physiological state of the cucumber under the dynamic change of the environment cannot be realized.

It has been found that the fluorescence parameter F of chlorophyll_v/F_oThe PSII reaction center of the leaves of the crops has potential activity, is sensitive to the change of the growth state and the growth environment of the crops, can be used for representing the physiological state of the leaves under the cold damage condition, screening and identifying the cold resistance of the crops, effectively evaluating the damage degree of the crops under the cold damage condition and providing a theoretical basis for portable and accurate monitoring of the physiological state of the crops under the cold damage condition. However, since the above researches do not consider the change trend of the fluorescence parameters under the dynamic change of the environment and the crop self state, a dynamic change prediction model between the cold damage and the fluorescence parameters cannot be established in a fine-grained manner, and the fluorescence parameters cannot accurately describe the cold damage degree, so that the accurate prediction of the crop cold damage in the true sense is realized, and the establishment of the fluorescence parameter prediction model under the dynamic change of the environment is the key for realizing the early warning and nondestructive diagnosis of the crop cold damage.

Disclosure of Invention

In order to overcome the defects of the prior art, the invention aims to provide a cold damage cucumber PSII potential activity prediction method based on QGA-SVR, wherein seedling stage cucumber is adopted as a test material, initial variable fluorescence and initial minimum fluorescence of the cucumber, low temperature and low temperature duration are used as model input parameters, and the variation trend of the fluorescence parameters of cucumber leaves under the cold damage condition is predicted and researched based on a quantum genetic-regression support vector machine prediction model (QGA-SVR), so that a reference basis is provided for the analysis of physiological state in the actual production process of out-of-season cucumber.

In order to achieve the purpose, the invention adopts the technical scheme that:

method for predicting cold damage cucumber PSII potential activity based on QGA-SVR (quantitative genetic Algorithm-sequence regression) by using different initial chlorophyll fluorescenceOptical parameter F_v/F_oCucumber seedlings of value are test specimens, and F is measured under different low temperature conditions and different durations_v/F_oChanging data, constructing a modeling sample set, and establishing cucumber leaf F under a low-temperature environment by adopting a quantum genetic support vector machine algorithm_v/F_oA value change prediction model according to which the pair F_v/F_oValues are predicted as F_v/F_oIndicating the potential activity of cold damage cucumber PSII, wherein F_vDenotes variable fluorescence, F_oIndicating minimal fluorescence.

The test sample is cucumber seedlings divided into 4 groups according to the initial fluorescence parameter, and is tested in CO₂Carried out in a climatic chamber, CO₂The environmental parameters in the artificial climate box are set as follows: the photoperiod day/night is 14/10h, the temperature is set test temperature gradient and is respectively 8, 10, 12 and 14 ℃, the relative air humidity day/night is 60%/50%, and CO is₂The concentration is 400 mu mol & mol^-1Four phytotron cucumber seedling leaf fluorescence parameters are collected at 16:00 every day for 8 days continuously, wherein each cucumber plant takes two points on the same leaf position functional leaf to obtain 288 groups of cucumber seedling leaf fluorescence parameters.

The fluorescence parameters are collected by a portable MINI-PAM-II type modulation chlorophyll fluorescence instrument, a dark adaptation blade clamp is used for clamping a blade to be measured before measurement, after the dark adaptation is carried out for 20 minutes, the dark adaptation blade clamp light barrier is opened, the fluorescence instrument is started, and the minimum fluorescence F of a test sample is obtained_oMaximum fluorescence F_mAnd variable fluorescence F_vIn which F is_v＝F_m-F_o。

When a prediction model is established, firstly, normalization processing is carried out on the acquired original test data, so that the fluctuation of the data is mapped to [0.2,0.8]]On the interval, dividing the data by 8:2 to generate a training data set and a prediction data set; secondly, selecting a radial basis kernel function by the SVR model, and optimizing model parameters c and g by using a quantum genetic algorithm and a prediction data set decision coefficient as a fitness evaluation index; finally, the ambient temperature, the duration of the low temperature, the initial variable fluorescence and the initial minimum fluorescence are used as inputWith fluorescence parameter F_v/F_oAnd establishing a QGA-SVR cucumber fluorescence parameter prediction model for output, wherein the environmental temperature is the low temperature set by the experiment.

In the normalization process, the normalization transformation function is

Wherein Y is the normalized data, X is the data to be normalized, X is_maxAnd X_minRespectively a maximum value and a minimum value in the same dimension data sequence, and the sample characteristics of the ith data areThe label is Y_iRepresents the current F of the ith sample_v/F_oThe value of the one or more of, the low temperature, low temperature duration, initial variable fluorescence and initial minimum fluorescence of the ith data sample, respectively.

Sample data x in SVR model passes through nonlinear functionConversion to a high-dimensional feature space, denoted as

Wherein w is the coefficient of the self-varying function, w^TIs the transpose of w, and b is the offset.

Solving w and b according to the principle of minimizing consistency of structural risk, and introducing a relaxation factor xi_iAndsupport vector regression optimization problem transformation

Wherein c is a penalty factor, | w |²For model complexity, m is the number of samples, ε is the insensitive loss parameter, and for the constraint, the dual-form lagrange polynomial is transformed into the optimization problem of the following formula:

in the formula a_i,a_j,Is Lagrange multiplier, K (x)_i,x_j) Is a kernel function.

The invention passes the nonlinear problem of the input space through a kernel function K (x)_i,x_j) Mapping to a high-dimensional feature space, and constructing a linear discriminant function in the high-dimensional space so as to solve the nonlinear problem.

The invention optimizes the penalty factor c and the kernel function parameter g by using a quantum genetic algorithm.

Compared with the prior art, the invention has the beneficial effects that:

(1) by aligning different initial F_v/F_oThe cucumber seedlings with the fluorescence parameter F are subjected to low-temperature stress experiments under different environmental temperatures, and the fluorescence parameter F is found when the cucumber seedlings suffer cold damage_v/F_oThe change is rapid, the low temperature stress can be effectively monitored, and the capacity of the cucumber seedling to respond to the low temperature stress is influenced by the initial fluorescence parameter of the cucumber seedling.

(2) Cucumber F for constructing fusion environment factor and initial fluorescence parameter coupling_v/F_oValue prediction model, selecting different initial fluorescence parametersSelecting, with initial F_vValue, initial F_oThe value combination is used as an input to construct a model, and the performance is optimal.

(3) Establishing a cucumber leaf fluorescence parameter prediction model under a cold damage condition based on a quantum genetic support vector machine regression algorithm, wherein the model is R on a training set and a testing set²The fluorescent parameter of the cucumber leaf under the cold damage condition can be predicted with high precision by respectively 0.9817 and 0.9864, RMSE is 0.2141 and 0.1741, RMAE is 0.1490 and 0.1338. The QGA algorithm carries out automatic optimization selection aiming at SVR algorithm parameters, and the model prediction precision can be improved. Compared with intelligent algorithms SVR, RF and NLR, the QGA-SVR has higher generalization, stability and popularization in the aspect of cucumber fluorescence parameter prediction under the cold damage condition. Compared with the GA-SVR algorithm, the QGA-SVR algorithm has the advantages of less optimizing time and higher precision.

Drawings

FIG. 1 is a flow chart of fluorescence parameter prediction according to the present invention.

FIG. 2 is the results of the experiments in the examples of the present invention, in which (a) is the initial F_v/F_oTest results at less than 5, (b) is initial F_v/F_oTest results at > 5.

FIG. 3 is a graph showing the effect of different types of kernel functions on the model results, where (a) is the training set and (b) is the prediction set.

FIG. 4 is a graph of the effect of different SVR parameters on model test set RMSE, where (a) is parameter c and (b) is parameter g.

FIG. 5 is a quantum genetic algorithm optimization process.

FIG. 6 is a model prediction set mean square error surface under different c and g conditions.

FIG. 7 shows the model fitting results with different initial fluorescence parameters as inputs (without initial fluorescence as input), wherein (a) is the training set and (b) is the prediction set.

FIG. 8 shows the model fitting results with different initial fluorescence parameters as inputs (initial F)_v/F_oAs input), wherein (a) is a training set and (b) is a prediction set.

FIG. 9 shows the model fitting results with different initial fluorescence parameters as inputs (initial F)_v、F_oAs input), wherein (a) is a training set and (b) is a prediction set.

FIG. 10 shows F at different temperatures_v/F_oIdentical cucumber leaf F_v/F_oThe value changes tend to be as high as 8 ℃ in (a) and 14 ℃ in (b).

Detailed Description

The embodiments of the present invention will be described in detail below with reference to the drawings and examples.

The invention relates to a cold damage cucumber PSII potential activity prediction method based on QGA-SVR, which takes cucumber initial variable fluorescence and initial minimum fluorescence, low temperature and low temperature duration as model input parameters to realize cucumber F under the cold damage condition_v/F_oThe accurate prediction of the value provides a new research method for early cold damage stress and nondestructive diagnosis of crop cold damage.

Specifically, the invention firstly uses different initial chlorophyll fluorescence parameters F_v/F_oCucumber seedlings of value are test specimens, and F is measured under different low temperature conditions and different durations_v/F_oChanging data, constructing a modeling sample set, and establishing cucumber leaf F under a low-temperature environment by adopting a quantum genetic support vector machine algorithm_v/F_oA value change prediction model, and finally F_v/F_oValues are predicted as F_v/F_oIndicating the potential activity of cold damage cucumber PSII, wherein F_vDenotes variable fluorescence, F_oIndicating minimal fluorescence.

The following is a detailed description of the invention.

1 materials and methods

1.1 test materials and methods

1.1.1 test materials

The experiment is carried out in an agricultural Internet of things key laboratory (34 degrees 07 '39' N, 107 degrees 59 '50' E and an altitude of 648m) of agricultural rural area of northwest agriculture and forestry science and technology university, the experimental material is Bonai 14-3 cucumber seedlings, the seedlings are cultivated by adopting a Substrate (Pindstrup Substrate, Denmark), the seedlings are sowed in a seed tray in 11 months and 5 days in 2019, and when the cucumber seedlings grow to be two leaves and one heart, the seedlings are transplanted into a plastic flowerpot with the length of 10 multiplied by 9cm and then are planted in a CO flowerpot₂Artificial climate box (Daskatel, RGL-P500D-CO)₂) Culturing in the medium. CO 2₂The environmental parameters in the artificial climate box are set as follows: the photoperiod day/night is 14h/10h, and the illumination intensity is 140 mu mol.m^-2·s^-1Temperature day/night is 25 ℃/16 ℃, relative air humidity day/night is 60%/50%, CO₂The concentration is 400 mu mol & mol^-1. After culturing for 7 days, randomly selecting 16 cucumber plants, measuring the initial fluorescence parameters of the cucumber plants, averagely dividing the cucumber seedlings into 4 groups according to the initial fluorescence parameters, and culturing in CO₂The tests were carried out in a climatic chamber. CO 2₂The environmental parameters in the artificial climate box are set as follows: the photoperiod day/night is 14/10h, the temperature is set test temperature gradient and is respectively 8, 10, 12 and 14 ℃, the air relative humidity day/night is 60/50 percent, and CO is₂The concentration is 400 mu mol & mol^-1Four phytotron cucumber seedling leaf fluorescence parameters are collected at 16:00 every day for 8 days continuously, wherein each cucumber plant takes two points on the same leaf position functional leaf to obtain 288 groups of cucumber seedling leaf fluorescence parameters.

1.1.2 fluorescence parameter measurement method

The fluorescence parameter data of cucumber leaves in the experimental environment were collected using a portable MINI-PAM-II modulated chlorophyll fluorescence spectrometer (WALZ, Germany). Clamping the blade to be tested by using the dark adaptation blade clamp before the measurement, opening the light barrier of the dark adaptation blade clamp after the full dark adaptation is carried out for 20 minutes, starting the fluorometer, and obtaining the minimum fluorescence F of the test sample_oMaximum fluorescence F_mVariable fluorescence F_v(F_v＝F_m-F_o)_o。

1.2 model construction

A cucumber fluorescence parameter prediction model under a cold damage condition is constructed by using a support vector machine regression algorithm based on a test sample set, firstly, in order to avoid errors caused by different feature vectors, normalization processing is carried out on original test data, and fluctuation of the data is mapped to [0.2,0.8]]On the interval, dividing the data by 8:2 to generate a training data set and a prediction data set; secondly, the SVR model selects a radial basis kernel function, and a quantum genetic algorithm is utilized to take a model prediction set decision coefficient as adaptationEvaluating indexes to complete the optimization of model parameters c and g; finally, the initial variable fluorescence F is obtained according to the ambient temperature, the low temperature duration_vAnd initial minimum fluorescence F_oAs input, the fluorescence parameter F_v/F_oAnd establishing a QGA-SVR cucumber fluorescence parameter prediction model for output, wherein the environmental temperature is the low temperature set by the experiment. The specific algorithm flow is shown in fig. 1.

1.2.1 data preprocessing

Due to the different dimensions of the four-dimensional input data, there are significant differences in the numerical values. In order to eliminate the influence caused by different orders of magnitude among the characteristics and facilitate the establishment of a subsequent prediction model, the data set is normalized to a [0.2,0.8] interval, and the normalized transformation function is

Wherein Y is the normalized data, X is the data to be normalized, X is_maxAnd X_minRespectively a maximum value and a minimum value in the same dimension data sequence, and the sample characteristics of the ith data areThe label is Y_iRepresents the current F of the ith sample_v/F_oThe value of the one or more of, the low temperature, low temperature duration, initial variable fluorescence and initial minimum fluorescence of the ith data sample, respectively.

1.2.2 predictive model construction

The SVR constructs a kernel function based on the support vector, converts the problem of low-dimensional space nonlinear regression into the problem of high-dimensional space linear regression, realizes multi-factor regression by minimizing the expected error of a learning machine, and is suitable for the problem of nonlinear change of cucumber fluorescence parameters under the condition of cold damage. Sample data x in SVROver non-linear functionConversion to a high-dimensional feature space, which can be expressed as

Wherein w is the coefficient of the self-varying function, w^TIs the transpose of w, and b is the offset. Solving w and b according to the principle of minimizing consistency of structural risk, and introducing a relaxation factor xi_iAndthe support vector regression optimization problem is converted into:

wherein c is a penalty factor, | w |²For model complexity, m is the number of samples and ε is the insensitive loss parameter. For the constraint, the dual-form lagrange polynomial can be converted into an optimization problem of the following formula:

in the formula a_i,a_j,Is Lagrange multiplier, K (x)_i,x_j) Is a kernel function.

1.2.3 model Kernel function selection

In the cucumber fluorescence parameter prediction model, the SVR algorithm passes the nonlinear problem of the input space through a kernel function K (x)_i,x_j) Mapping into a high-dimensional feature space, in a high-dimensional spaceAnd (4) constructing a linear discriminant function so as to solve the nonlinear problem. Kernel function K (x)_i,x_j) The type of (2) has a large influence on the performance of the model, and the complexity of the model obtained by adopting different kernel functions is different. Linear (linear) kernel functions, polynomial (Poly) kernel functions, Sigmoid (Sigmoid) kernel functions, Radial Basis Function (RBF) kernel functions, etc. are commonly used, and the expressions are:

K(x_i,x_j)＝x_i·x_j

K(x_i,x_j)＝(γ×x_i·x_j+r)^d

K(x_i,x_j)＝tanh(γ×x_i·x_j+r)

K(x_i,x_j)＝exp(-g‖x_i-x_j‖²)

wherein gamma is the distribution of the original data in the high-dimensional feature space after being mapped to the high-dimensional data; r is a bias coefficient; d is the dimension of the mapping; g is a kernel function parameter.

1.2.4 model parameter optimization

In practical application, the performance of a prediction model established based on an SVR algorithm has a direct relation with a penalty factor c and a model kernel function parameter g value, wherein the penalty factor c directly influences the generalization and popularization capability of the SVR, the complexity and approximation error of the model are controlled, and the larger c is, the higher the data fitting degree is; the kernel function parameter g is related to the kernel function form and the number of the support vectors, the complexity of the final solution of the network is controlled, and the system generalization capability is deteriorated if g is too large or too small. Therefore, the selection of the kernel function parameter g and the penalty factor c is the key for modeling the SVR algorithm.

The Quantum Genetic Algorithm (QGA) is an emerging Algorithm for performing biomimetic evolution simulation by using the idea of Quantum computing, inherits the advantages of the classical Genetic Algorithm and can effectively fill in the defects of part of the classical Genetic Algorithm. The core of the algorithm is quantum bit coding and quantum gate updating, the chromosome in the genetic algorithm is represented by proper quantum state, the updating evolution operation is completed by the quantum gate rotation, so that the quantum genetic algorithm can fully exert the quantum computing characteristic, in the iteration process, the superposition state of each quantum bit can be collapsed to a determined state, thereby tending to be stable, achieving convergence, realizing optimization, and finally selecting the individual with the highest fitness value, namely the optimal solution of the problem. Because the quantum genetic algorithm adopts a unique coding mode and an updating mode, the quantum genetic algorithm has richer population diversity, faster convergence speed and higher convergence precision compared with the traditional genetic algorithm. Therefore, the present invention optimizes the penalty factor c and the kernel function parameter g using QGA.

1.2.5 model Performance evaluation index

Three error evaluation indicators are used to measure the performance of the prediction model involved, including coefficient of determination (R)²) Root Mean Square Error (RMSE) and Mean Absolute Error (MAE), R²And (3) representing the degree of closeness of the relationship between the actual value and the predicted value of the model, wherein the closer the decision coefficient is to 1, the higher the model interpretation degree is, and the better the performance is. RMSE is the arithmetic square root of the square sum of the difference between the predicted value and the actual value of the model and the expected value, MAE is the average value of the absolute errors of the model, the actual situation of the error of the predicted value is reflected, and the smaller the value is, the higher the performance of the model is.

2 results and analysis

2.1 fluorescence parameter test results

In order to analyze the change of the cucumber fluorescence parameters under the coupling change of the low-temperature, the low-temperature duration and the initial fluorescence parameters, the invention compares that the low-temperature environment temperature is 8-14 ℃ (the step length is 2) and the initial fluorescence parameters are different (F is used)_v/F_oBorderline 5), the fluorescence parameter of cucumber leaves has a trend of increasing with the duration of low temperature, and the results of part of the test are shown in fig. 2.

As can be seen from FIG. 2, in the low temperature growth environment, the change of the fluorescence parameter of cucumber leaves is not only related to the low temperature and the duration, but also related to the initial F of cucumber before being cold-damaged_v/F_oThe sizes are closely related. Cucumber seedling F in low temperature growth environment_v/F_oThe values are all fast within 0-2 daysThe decrease is slowly decreased along with the increase of the duration, the lower the environmental temperature is, the faster the decrease speed is, the early stage of low-temperature stress is shown, the stress temperature and the fluorescence parameter F_v/F_oThe degree of attenuation has a significant correlation. Treating the cucumber seedlings at low temperature for 2-4 days, wherein the cucumber seedlings F_v/F_oInflection point appears in response curve, different initial F_v/F_oThe position coordinates of the inflection point are different under the values and the low-temperature treatment level. As the ambient temperature decreases, the low temperature duration corresponding to the location of the inflection point in fig. 2 continues to shift to the right, F_v/F_oThe value is continuously shifted downwards, which shows that when the cucumber seedlings are subjected to low-temperature stress, the larger the difference between the stress temperature and the normal growth temperature is, the longer the adaptation time of the leaves to the low-temperature stress is, and the more serious the damage is.

Further comparing the effect of different initial fluorescence on the decay process of (a) and (b) in FIG. 2 at the same ambient temperature, it can be found that the initial F of cucumber seedlings_v/F_oValue and cucumber seedling F under low temperature treatment_v/F_oThere is a correlation between the values. It can be seen from the figure that the initial fluorescence F is obtained at low temperatures of 8 ℃ and 10 DEG C_v/F_oSize pair F_v/F_oThe inflection point of the response curve has little influence, the abscissa of the inflection point is around 3d, and F is treated at 8 DEG C_v/F_oThe value drops to around 0 and no change follows; at 10 ℃ treatment, F_v/F_oThe value decreased to about 1 and then slowly decreased to about 0.7 at 8 d. And at low temperatures of 12 ℃ and 14 ℃, the initial F_v/F_oSize pair influences corresponding F at inflection point_v/F_oThe values had a significant effect. At 12 ℃ treatment, F_v/F_oThe response curve has an inflection point in 2-3 d and is reduced to the minimum at 8d, wherein the initial F_v/F_oHigh cucumber seedling Current F_v/F_oValue 1.47, and initial F_v/F_oLow cucumber seedling Current F_v/F_oThe value is only 1.01; at 14 ℃ treatment, F_v/F_oThe response curve has an inflection point at about 2d, and is reduced to the minimum after 8d treatment, namely initial F_v/F_oHigh and low cucumber seedlings F_v/F_oThe values are 2.21 and 1.38, respectively.

The above phenomena indicate that under low temperature treatment of 10 ℃ and below, the photochemical activity of the PSII reaction center of the cucumber leaf is obviously reduced, irreversible damage is likely to occur, and the reaction is compared with that of the original F_v/F_oThe size and temperature of the cucumber seedling F_v/F_oKey element of the response curve, the lower the temperature, F_v/F_oThe faster the decay. When the environmental temperature is higher than 10 ℃, the cucumber seedling F is not damaged by cold_v/F_oInfluencing F together with temperature and duration_v/F_oResponse curve, initial F_v/F_oThe larger the value, the less the reaction center of the cucumber leaf PSII is damaged when the cucumber leaf is subjected to low-temperature stress, and the stronger the capacity of the cucumber leaf to respond to low-temperature environment. In summary, when cucumber seedlings suffer cold damage, the fluorescence parameter F_v/F_oThe rapid change can effectively monitor the low temperature stress, which is similar to the literature^[23]The obtained results are consistent, so that the cucumber fluorescence parameter F under the cold damage condition is constructed_v/F_oThe prediction model is the key for establishing the effective cold damage early warning of the cucumbers.

2.2 construction of QGA-SVR model

2.2.1 QGA-SVR model Kernel function selection

In order to compare the influence of different kernel functions on the accuracy of the prediction model, the kernel functions of Linear, Poly, Sigmoid and RBF are respectively adopted, and the model is constructed by the same method. Comparing R of measured values and predicted values of different model test samples²RMSE and MAE, the results are shown in fig. 3.

Compared with the limitation that the linear kernel function is only suitable for processing the linear problem, the RBF kernel function can process the nonlinear problem between the independent variable and the dependent variable of the model, and in addition, the model fitting effect of applying the RBF kernel function is superior to that of the polynomial kernel function and the Sigmoid kernel function, and the number of parameters required to be set is small. As shown in FIG. 3, the training set and the test set R are selected when the kernel function is RBF²The model is more than 98 percent and is far larger than other three groups of kernel functions, the RMSE of the training set and the prediction set are 0.2141 and 0.1741 respectively, the MAE of the training set and the prediction set are 0.1490 and 0.1338 respectively, and the RMSE of the training set and the prediction set are far smaller than other three groups of kernel functions, which shows that the RBF kernel function has highest modeling precision and minimum error, so the RBF is selected as the cucumber fluorescence parameter prediction modelA kernel function of type.

2.2.2 QGA-SVR model core parameter tuning

The parameters needing to be optimized by the QGA-SVR prediction model are a penalty factor c and a kernel function parameter g. In order to improve the QGA algorithm optimizing precision and speed, an optimizing interval of the QGA algorithm needs to be determined firstly. Setting the initial optimizing interval of the parameter c as [1,100] and the step length as 0.1; the parameter g is the initial optimization interval [0.05,10], and the step length is 0.05. And (3) obtaining the RMSE change condition of the SVR model prediction set under different c and g conditions by adopting a traversal method, as shown in FIG. 4.

The influence of the changes of the parameters c and g on the model precision can be observed, and when the value of the parameter c is within 10-50 and the value of the parameter g is within 2-7, the model prediction set RMSE can obtain the minimum value. Therefore, setting a parameter optimization interval: c belongs to [10,50] and g belongs to [2,7], and the number of the c chromosome gene and the g chromosome gene is respectively set to be 12 and 7, the population number is 20, the maximum iteration step number is 100, the cross probability is 0.8, and the variation probability is 0.1. When the iteration optimization is terminated, the optimal penalty factor c and the kernel function parameter g of the SVR model are obtained, and the quantum genetic algorithm optimization process is shown in FIG. 5.

As shown in fig. 5, the convergence rate is high in the process of the QGA algorithm training, which is optimal in about 60 generations, which shows that the QGA has a significant optimization effect on the parameters c and g, and the optimal c value of the obtained SVR prediction model is 42.93, and the optimal g value is 4.24. In order to verify the optimization effect of the QGA algorithm on the parameters c and g, the grid traversal and the GA algorithm are adopted to carry out optimization in the same method, the optimization intervals of the c and the g are consistent with the QGA algorithm, the optimization step length of the grid traversal method parameter c and the optimization step length of the g are respectively set to be 0.5 and 0.05, the GA algorithm parameters are consistent with the QGA algorithm, the population size is 20, the maximum iteration step number is 100, the cross probability is 0.8, and the variation probability is 0.2. And constructing a model prediction set RMSE value change curved surface obtained by a grid traversal method, and simultaneously marking the final result coordinate values obtained by the three methods on the curved surface, as shown in FIG. 6. It can be found that the RMSE minimum points of the model prediction sets obtained by the three methods are not very different, namely, QGA (c-42.93, g-4.24, RMSE-0.1744), GA (c-44.69, g-4.19, RMSE-0.1752), grid traversal (c-48.5, g-4.10, RMSE-0.1746), where the RMSE obtained by the QGA algorithm is minimum, and the time required for the grid traversal and the grid traversal is 65.04s and 2071s, respectively, and the time required for the QGA to find the optimum is 50.81s, which is smaller than the former two.

2.3 QGA-SVR model validation and analysis

2.3.1 fitting Effect and analysis of the model

Proposed for validating the invention with an initial F_vAnd F_oThe prediction performance of the fluorescence parameters of the cucumber leaves under the cold damage condition of a QGA-SVR algorithm model with the input of low temperature and duration time, and the initial fluorescence parameters which are not added and the initial F are respectively established_v/F_oA reference comparison was made for the QGA-SVR algorithm model with single fluorescence parameters as input and low temperature and duration as inputs. R for training set and prediction set of each model²Fig. 7, 8, and 9 show RMSE and MAE.

As can be seen from a comparison of FIGS. 7, 8 and 9, the fluorescence parameter F was added to the model with only the low temperature and duration as inputs_v/F_oAnd simultaneously adding F_vAnd F_oLater model training set and prediction set R²Increasing in order, decreasing in order RMSE, MAE. Wherein, F_v/F_oIs a reflection of the potential activity of the PSII reaction center, as can also be seen in FIG. 2, in the case of mild cold injury, the initial F_v/F_oCucumber seedlings with different values, as the duration of the low temperature increases, F_v/F_oThe trend of the value changes is greatly different, so that the model accuracy is improved when the trend of the value changes is introduced into a prediction model compared with a model only considering temperature and time.

And minimal fluorescence of cucumber leaves F_oThe fluorescence yield when the PSII reaction center of the cucumber leaf is completely opened is reflected, and the value of the fluorescence yield is related to the chlorophyll concentration of the leaf; cucumber leaf variable fluorescence F_vMaximum fluorescence F_mWith minimal fluorescence F_oThe difference reflects the reduction of the primary electron acceptor QA in the PSII reaction center, and the value is related to the relative activity of the leaf PSII reaction center. Due to the change of the two under the condition of low temperatureThe rates of change are also different, resulting in cucumber seedlings at the initial F_v/F_oThe values are the same and F_o、F_vUnder different conditions, when it is subjected to low temperature stress, F_v/F_oThe tendency of the change in value was different, and the result was consistent with that shown in fig. 2. The above phenomenon, namely, the initial F_oAnd F_vCombination compared to F_v/F_oThe physiological state of the cucumber leaf can be reflected better, which shows that the physiological state of the cucumber leaf before suffering cold injury is represented more accurately by the combined fluorescence parameter information, and is consistent with the verification result shown in figure 10, so the initial F is selected by the invention_oInitial F_vAnd the low temperature and the duration are used as the input of a cucumber fluorescence parameter prediction model.

2.3.2 Performance evaluation of the model

In order to verify the prediction performance of the QGA-SVR model provided by the invention on cucumber fluorescence parameters under cold damage conditions, Non-linear regression (NLR) and Random Forest (RF) models are selected to train and predict test data, and the performance of the algorithm is shown in Table 1.

TABLE 1 comparison of Performance of different algorithms

As can be seen from Table 1, the accuracies of the nonlinear regression model training set and the prediction set are both less than those of the other three machine learning algorithms because the fitting ability of machine learning to nonlinear samples is strong, the accuracy of the RF model training set is higher than that of the prediction set, and R is²The SVR model is smaller than the SVR model, which shows that the SVR model has greater advantages in small sample training. Cucumber leaf F established by QGA-SVR model in all models_v/F_oThe prediction model works best, its training set R²0.9817, RMSE 0.2141, MAE 0.1490; test set R²0.9864, RMSE 0.1741, MAE 0.1338 and short training time, so the cucumber fluorescent parameter prediction model is selected as a cucumber fluorescent parameter prediction model under the condition of cold damage, the precision is improved, the tedious manual parameter adjustment is avoided, and the parameter optimization is betterEfficiency and predictive performance.

3 conclusion

(1) By aligning different initial F_v/F_oThe cucumber seedlings with the fluorescence parameter F are subjected to low-temperature stress experiments under different environmental temperatures, and the fluorescence parameter F is found when the cucumber seedlings suffer cold damage_v/F_oThe change is rapid, the low temperature stress can be effectively monitored, and the capacity of the cucumber seedling to respond to the low temperature stress is influenced by the initial fluorescence parameter of the cucumber seedling.

(2) Cucumber F for constructing fusion environment factor and initial fluorescence parameter coupling_v/F_oValue prediction model, in different initial fluorescence parameter selections, with initial F_vValue, initial F_oThe value combination is used as an input to construct a model, and the performance is optimal.

(3) Establishing a cucumber leaf fluorescence parameter prediction model under a cold damage condition based on a quantum genetic support vector machine regression algorithm, wherein the model is R on a training set and a testing set²The fluorescent parameter of the cucumber leaf under the cold damage condition can be predicted with high precision by respectively 0.9817 and 0.9864, RMSE is 0.2141 and 0.1741, RMAE is 0.1490 and 0.1338. The QGA algorithm carries out automatic optimization selection aiming at SVR algorithm parameters, and the model prediction precision can be improved. Compared with intelligent algorithms SVR, RF and NLR, the QGA-SVR has higher generalization, stability and popularization in the aspect of cucumber fluorescence parameter prediction under the cold damage condition. Compared with the GA-SVR algorithm, the QGA-SVR algorithm has the advantages of less optimizing time and higher precision.