Rosa chinensis bud bearing prediction method based on nonlinear regression and multivariate probability distribution model

Abstract

The invention discloses a Chinese rose bud bearing prediction method based on nonlinear regression and a multivariate probability distribution model, and relates to the technical field of Chinese rose bud bearing prediction. According to the method, an application scene adopts a purple noxia variety of Chinese rose, an intelligent monitoring method oriented to greenhouse flower planting is constructed, the Chinese rose is used as a typical application scene, and a multi-dimensional prediction and probability model of environment and growth indexes is established based on multi-stage monitoring data. By identifying a nonlinear relationship among core indexes such as accumulated temperature, photosynthetic radiation and leaf area, a coupling mechanism is fitted by adopting a least square method, and prediction of key indexes is realized. And further constructing binary joint probability distribution based on different distribution characteristics of environment and plant indexes, and establishing a bracting probability calculation model in combination with a naive Bayesian algorithm. According to the method, missing data can be complemented based on real-time data or through a fitting relation, dynamic prediction of the bract bearing rate and the yield is achieved, decision support is provided for flowering phase regulation and control and production management, and greenhouse production intelligentization is promoted.

Description

Technical Field

[0001] This invention relates to the field of rose bud prediction technology, specifically to a rose bud prediction method based on nonlinear regression and multivariate probability distribution models. Background Technology

[0002] Rose bud formation prediction technology falls under the field of agricultural intelligent monitoring. Its core objective is to analyze environmental and plant growth data to determine the critical period for bud formation, thereby guiding production management. Currently, existing technologies in this field mainly focus on the post-bud formation stage. For example, some methods take a macro-sales perspective, using internet environmental data to predict the flowering period of different batches to coordinate supply and demand; others focus on hydroponics, constructing refined water and fertilizer management models by measuring indicators such as fresh weight and nitrogen content; still others use algorithms such as random forests and multinomial fitting after bud formation to predict and regulate the final quality and flowering period of the flowers.

[0003] However, these existing technologies have significant limitations. First, their predictive perspective is too macroscopic and lagging, failing to provide precise guidance for individual farmers during the critical growth stage before budding, resulting in a short window for production adjustments and difficulty in flexibly responding to market demands. Second, these methods typically rely heavily on complete, high-quality data after budding, making them poorly adaptable to ordinary greenhouse environments with incomplete monitoring equipment and missing indicators, and difficult to implement. Finally, existing models mostly perform deterministic analyses, failing to deeply explore the nonlinear coupling relationship and probability distribution characteristics between environmental and plant indicators, limiting the models' early warning capabilities and robustness under complex actual production conditions.

[0004] In summary, existing technologies have shortcomings in terms of timeliness of prediction, universality of application, and model refinement. Therefore, there is an urgent need for a method for predicting the probability of flower buds that can cover the early growth stages from pruning to bud formation and adapt to different monitoring conditions. This method is of vital importance and has significant value in improving the intelligence level of greenhouse production and the economic benefits for farmers. Summary of the Invention

[0005] The purpose of this invention is to provide a method for predicting rose bud formation based on nonlinear regression and multivariate probability distribution models, addressing the aforementioned problems.

[0006] The technical solution of the present invention is as follows: A method for predicting rose bud formation based on nonlinear regression and multivariate probability distribution models includes the following steps: Collect greenhouse environmental index data and plant growth index data from the pruning to bud formation stage; Correlation analysis was performed on environmental indicator data and plant growth indicator data to screen out core indicators that are strongly correlated with bud formation. Based on environmental index data and plant growth index data, a nonlinear regression model is constructed to predict at least one core index value for a specified future period. Based on the statistical distribution of core indicator values ​​at the time of bud formation in historical data, a multivariate probability distribution model is constructed. The core indicator values ​​predicted by the nonlinear regression model are input into the multivariate probability distribution model to calculate the probability of bud formation in a specified future period.

[0007] Furthermore, the calculation of the probability of bud formation at a specified future period specifically includes: The Naive Bayes algorithm is used to calculate the posterior probability of bud formation by substituting the predicted values ​​of each core indicator into their respective probability density functions. The general formula for calculating the posterior probability of bud formation is as follows:

[0008] in Indicates bud formation / non-bud formation events. This indicates an event where accumulated temperature, accumulated light radiation, digital biomass, and leaf area reach a certain value.

[0009] Furthermore, the construction of the nonlinear regression model includes: By analyzing the trends in data visualization and combining the trend characteristics of various general functions, it was ultimately determined that the environmental indicators and plant indicators generally exhibit a non-linear correlation. And perform a nonlinear transformation. ; The least squares method was used to fit the nonlinear model and characterize the coupling mechanism among multiple indices. , in, This represents the actual index, i.e., solving for the sum of squared errors. The values ​​of each parameter when they reach their minimum; , , , These are the parameters obtained by fitting using the least squares method.

[0010] Furthermore, the multivariate probability distribution model is a binary log-normal distribution model, used to describe the joint probability distribution between the two indicators of leaf area and digital biomass; the construction steps include: remember The mean, For covariance (e.g.) This represents the covariance between the first and second variables. The number of indicators to be evaluated, The covariance matrix ( The inverse of the covariance matrix, The adjoint matrix of the covariance matrix, (where X is the determinant of the covariance matrix), and E(X) is the sample expectation. This represents the sample variance. based on The Gaussian distribution density function is: , when Then the covariance matrix Expectation and variance: , The inverse of a matrix is ​​equivalent to the product of the inverse of its determinant and its adjoint matrix: , The probability density function of the bivariate normal distribution is: .

[0011] Furthermore, the selection of the core indicators employs principal component analysis (PCA) for feature dimensionality reduction; specifically, this includes: Zero-mean data: , use The vector representation of the projection direction to be calculated. (represents the transpose of the projection direction vector) Sample Projected to direction Above, record Then all samples after projection are: The objective function is to maximize the variance of the projected samples. , , The accumulated temperature and light radiation, calculated using the Lagrange multiplier method, are the main factors affecting plant traits.

[0012] Furthermore, the selection of the core indicators also includes: Collect all environmental cumulative values ​​and plant trait data at the time of bud formation for all plants; perform frequency statistics and distribution analysis on each indicator, and calculate skewness, kurtosis, mean, and sample variance as measures; Analysis of the environmental control data of the bud formation revealed that the skewness was close to 0, indicating excellent symmetry; the kurtosis was significantly lower than 0; the data tails were thin; the distribution was more uniform; and the overall distribution exhibited a discrete normal distribution with the following probability density function: ; Leaf area, digital biomass, and other plant-related indicators were analyzed using the Shapiro-Wilk normality test and QQ plot visualization. The data exhibited a log-normal distribution, with the probability density function as follows: Its cumulative probability distribution can be expressed as: , .

[0013] in Indicates a sample, Represents the sample mean. Represents the sample variance, and is a calculation function. Regarding variables indefinite integrals, Introducing auxiliary variables .

[0014] Furthermore, when some core indicator data is missing, the existing nonlinear regression model is used to predict the missing indicator data based on the existing indicator data. Then, the directly measured and indirectly predicted indicator data are input into the multivariate probability distribution model to calculate the budding probability.

[0015] Furthermore, the core indicators include at least accumulated temperature, cumulative photosynthetically active radiation, leaf area, and digital biomass.

[0016] This application also includes an electronic device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement a method for predicting rose bud formation based on a nonlinear regression and multivariate probability distribution model.

[0017] This application also includes a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements a method for predicting rose bud formation based on a nonlinear regression and multivariate probability distribution model.

[0018] Compared with existing technologies, the advantages of this invention are: A smart monitoring method for greenhouse flower cultivation was developed, using roses as a typical application scenario. Based on multi-period monitoring data, a multi-dimensional prediction and probabilistic model of environmental and growth indicators was established. By identifying the nonlinear relationships among core indicators such as accumulated temperature, photosynthetic radiation, and leaf area, a least squares fitting coupling mechanism was employed to predict key indicators. Furthermore, based on the different distribution characteristics of environmental and plant indicators, a binary joint probability distribution was constructed, and a budding probability calculation model was established using a Naive Bayes algorithm. This method can dynamically predict budding rate and yield based on real-time data or by supplementing missing data through fitting relationships, providing decision support for flowering period regulation and production management, and promoting intelligent greenhouse production. This invention has currently been applied in a Floral Experiment and Modeling Analysis System. Attached Figure Description

[0019] Figure 1 This is a flowchart illustrating the selection of key indicators and nonlinear fitting predictions for the first stage of this application.

[0020] Figure 2 This is a flowchart illustrating the bract data distribution analysis and modeling for the second phase of this application.

[0021] Figure 3 The flowchart illustrates the practical application of the method proposed in this application, which integrates two models.

[0022] Figure 4 This is a partial result diagram of fitting environmental indicator data and plant indicator data.

[0023] Figure 5 This is a diagram illustrating the distribution of individual environmental indicators at the time of bud formation.

[0024] Figure 6 This is a log-Q plot selected from two indicators: leaf area and digital biomass.

[0025] Figure 7 This is a PDF image showing the visualization of individual plant data following a log-normal distribution.

[0026] Figure 8 A visualization analysis chart showing the correlation between the distributions of two plant indicators.

[0027] Figure 9 This is the interface of the analysis system currently used in this invention. Detailed Implementation

[0028] It should be noted that relational terms such as "first" and "second" are used merely to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0029] A smart monitoring system suitable for diverse greenhouse environments was developed to assist farmers in accurately monitoring the status of flower cultivation, especially enabling early assessment of growth trends and dynamic adjustment of agricultural strategies during the critical growth period before bud formation. Using greenhouse rose cultivation as a typical application scenario, the system is based on monitoring data from three stages—from pruning to bud formation—covering 59 environmental control indicators and 9 plant growth indicators. Through multi-source data fusion, probabilistic modeling, and predictive algorithms, it provides a reliable technical foundation for intelligent management of greenhouse flower production.

[0030] First, by combining agricultural expert experience with correlation analysis of indicators, significant positive correlations were identified among six core indicators: accumulated temperature, time, daily photosynthetically active radiation, cumulative radiation, leaf area, and digital biomass. Based on data trend analysis, a nonlinear programming model was used to model the relationship between environmental and plant indicators. The least squares method was used to fit the nonlinear model, characterizing the coupling mechanism among multiple indicators, and the accuracy of the model was verified by indicator prediction experiments over the next few days.

[0031] Secondly, statistical distribution analysis was conducted on the budding time and its corresponding environmental and plant indicators. Independence tests were performed on some indicators, and their correlations were examined using covariance matrix analysis and visualization techniques. A binary joint probability distribution model was then constructed to accurately describe the probabilistic characteristics of multiple indicators.

[0032] Finally, an index probability density function and a Naive Bayes algorithm are integrated to construct a general model for calculating the budding probability. This model can calculate the daily budding rate based on real-time greenhouse data, and its reliability is verified by comparing it with actual yield. For greenhouses where incomplete index monitoring exists, the model leverages the multi-index fitting relationship established in the first step to predict missing indicators based on existing data. By inputting both direct measurement and indirect prediction results into the budding probability model, daily yield can be predicted in advance, providing farmers with decision support for flowering period control and production management, ultimately achieving precise control of the flowering period.

[0033] The features and performance of the present invention will be further described in detail below with reference to embodiments.

[0034] Please see Figure 1-8 This application presents a method for predicting rose bud formation based on nonlinear regression and multivariate probability distribution models. The aim is to use the growth status of the plant at different growth stages as a basis for regulating plant growth stages, thereby controlling the timing of flower bud formation and ensuring timely and accurate fulfillment of local needs, maximizing the interests of individual farmers. The application scenario uses the Purple Fairy rose variety. The specific steps include: The environmental data collected by the greenhouse environmental control sensors is subdivided and processed into environmental index data, resulting in measured greenhouse temperature (meas grh temp), measured humidity (meas HD), measured relative humidity (meas RH), cumulative photosynthetically active radiation (PAR sum), ventilation temperature (vent temp), measured fertilizer and irrigation temperature (meas F&P temp), actual fertilizer and irrigation temperature (temp F&P act), measured heating temperature (meas heat t), measured shading (meas lee), measured ventilation temperature (meas vent t), measured wind speed (meas wind), and actual The data includes 59 environmental control indicators, such as total application amount (act dos total), daily cumulative photosynthetically active radiation (meas PAR sum td), cumulative radiation (radiation sum), actual photosynthetically active radiation (used PAR), used radiation (used radiation), calculated cooling temperature (calc cool temp), calculated humidity (calc HD), calculated relative humidity (calc RH), used humidity measurement (used HD meas), used relative humidity measurement (used RH meas), and used temperature measurement (used temp meas). The plant state parameters collected by the three-dimensional laser sensor are integrated to obtain nine plant trait indicators, including plant height, maximum plant height, leaf area, projected area, circumscribed rectangle area, digital biomass, light penetration depth, leaf uprightness, and 3D leaf area index. Using environmental control indicators as input and plant performance traits as output, correlation analysis was performed on each, and a correlation feature matrix was constructed based on the indicators with strong correlations. , in Indicates the first The correlation vector between each environmental control index and each plant trait index; Due to the large number of relevant indicators, principal component analysis (PCA) and feature reduction were performed. The specific steps included: Zero-mean data: , Sample Projected to direction Above, record Then all samples after projection are: The objective function is to maximize the variance of the projected samples. , , Calculations using the Lagrange multiplier method show that accumulated temperature and light radiation are indeed the main factors affecting plant traits. By analyzing the trends in data visualization and combining the trend characteristics of various general functions, it was ultimately determined that the environmental indicators and plant indicators generally exhibit a non-linear correlation. Therefore, a nonlinear transformation is performed. .

[0035] The parameters are solved using the least squares method: , in This represents the actual index, i.e., solving for the sum of squared errors. The values ​​of each parameter when they reach their minimum. See the partial fitting result diagram for reference (...). Figure 4 ); Alternatively, the combined effects of multiple environmental control indicators on a specific plant indicator can be analyzed to establish a multiple regression model for data prediction; or an autoregressive approach can be used to analyze only the historical variation patterns of plant indicators and further describe and predict their near-term trends. Data on cumulative environmental values ​​and plant traits at the time of bud formation were collected for all plants. Frequency statistics and distribution analysis were performed on each indicator, and skewness, kurtosis, mean, sample variance, and other metrics were calculated. Analysis of the environmental control data of the bud formation revealed that the skewness was close to 0, indicating excellent symmetry; the kurtosis was significantly lower than 0; the data tails were thin; the distribution was more uniform; and the overall distribution exhibited a discrete normal distribution. The probability density function (PDF) was as follows: ( Figure 5 ); Shapiro-Wilk normality test and QQ plot were performed on plant-related indicators such as leaf area and digital biomass. Figure 6 Visual analysis shows the data follows a log-normal distribution with a probability density function (PDF). ( Figure 7 Its cumulative probability distribution (CDF) can be expressed as: , , Because the plant index data are relatively continuous and their distributions have high similarity, further analysis of the correlation between plant indices is conducted, attempting to establish a multivariate distribution model. By calculating the covariance matrix, a large covariance is observed, which is then combined with a log-space scatter plot (…). Figure 8 The distribution of leaf area and digital biomass confirmed a strong positive correlation, thus establishing a binary log-normal distribution model: based on The Gaussian distribution density function of the meta-Gaussian distribution: , when , denote the covariance matrix Expectation and variance: , The inverse of a matrix is ​​equivalent to the product of the inverse of its determinant and its adjoint matrix. , The probability density function of the bivariate normal distribution is: , By combining the probability distributions of multiple indicators using the Naive Bayes algorithm, a general formula for calculating the posterior probability of flower bud formation is obtained: , in Indicates bud formation / non-bud formation events. This indicates an event where various data such as accumulated temperature, accumulated light radiation, digital biomass, and leaf area reach a certain value.

[0036] Based on the established predictive model, various indicators can be predicted for the next few days. Finally, the predicted values ​​of each indicator are applied to the density distribution function of each indicator and the general formula for calculating the posterior probability of flower bud formation, thereby obtaining the bud formation rate for the next few days, thus assisting farmers in controlling the flowering period. Furthermore, this model has strong versatility; even when data collection for some indicators is limited in greenhouses, the model can still indirectly measure some important indicators by utilizing the relationships between indicators, greatly reducing the difficulty of implementing the model in various greenhouses.

[0037] like Figure 9 The diagram shows an analysis system in which the method of this application is applied.

[0038] By combining historical data indicators to predict various data indicators for the next few days, we can monitor plant growth and guide agricultural operations. By using predictive indicator data to forecast the probability of bud formation, farmers can be guided to adjust their environmental control strategies in advance and regulate the bud formation date.

[0039] When the collection of some indicators in greenhouses is limited, indirect measurement of indicators that cannot be directly measured can be achieved through the correlation of indicators, which is highly adaptable to the complex conditions of greenhouses.

[0040] Although this application is based on research conducted under greenhouse conditions, the same method can be used for modeling in open-air environments. The difference is that greenhouse environments are human-controlled; in open-air environments, it is possible to integrate a weather forecasting system to predict future weather changes and combine this with sensor measurement data for modeling. In the modeling stage of indicator correlation analysis, the combined effect of multiple environmental control indicators on a certain indicator of the plant can be analyzed to establish a multiple regression model for data prediction; a deep neural network can even be constructed to use vectors of all relevant indicators as input for training and fitting; or an autoregressive approach can be used to analyze only the historical change patterns of plant indicators to further describe and predict the recent trends. In the stage of selecting the main influencing indicators, the PCA algorithm can also be replaced by other algorithms such as random forest to analyze the main environmental control indicators affecting plant growth.

[0041] The embodiments described above merely illustrate specific implementation methods of this application, and while the descriptions are detailed and specific, they should not be construed as limiting the scope of protection of this application. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of the technical solution of this application, and these modifications and improvements all fall within the scope of protection of this application.

Claims

1. A method for predicting rose bud formation based on nonlinear regression and multivariate probability distribution models, characterized in that, Includes the following steps: Collect greenhouse environmental index data and plant growth index data from the pruning to bud formation stage; Correlation analysis was performed on environmental indicator data and plant growth indicator data to screen out core indicators that are strongly correlated with bud formation. Based on environmental index data and plant growth index data, a nonlinear regression model is constructed to predict at least one core index value for a specified future period. Based on the statistical distribution of core indicator values ​​at the time of bud formation in historical data, a multivariate probability distribution model is constructed. The core indicator values ​​predicted by the nonlinear regression model are input into the multivariate probability distribution model to calculate the probability of bud formation in a specified future period.

2. The method for predicting rose bud formation based on nonlinear regression and multivariate probability distribution models according to claim 1, characterized in that, The calculation of the probability of bud formation at a specified future period specifically includes: The Naive Bayes algorithm is used to calculate the posterior probability of bud formation by substituting the predicted values ​​of each core indicator into their respective probability density functions. The general formula for calculating the posterior probability of bud formation is as follows: in Indicates bud formation / non-bud formation events. This indicates an event where accumulated temperature, accumulated light radiation, digital biomass, and leaf area reach a certain value.

3. The method for predicting rose bud formation based on nonlinear regression and multivariate probability distribution models according to claim 1, characterized in that, The construction of the nonlinear regression model includes: By analyzing the trends in data visualization and combining the trend characteristics of various general functions, it was ultimately determined that the environmental indicators and plant indicators generally exhibit a non-linear correlation. And perform a nonlinear transformation. ; The least squares method was used to fit the nonlinear model to characterize the coupling mechanism among multiple indices: ， in, This represents the actual index, i.e., solving for the sum of squared errors. The values ​​of each parameter when they reach their minimum; , , , These are the parameters obtained by fitting using the least squares method.

4. The method for predicting rose bud formation based on nonlinear regression and multivariate probability distribution models according to claim 1, characterized in that, The multivariate probability distribution model is a binary log-normal distribution model, used to describe the joint probability distribution between leaf area and digital biomass; the construction steps include: remember The mean, For covariance, such as This represents the covariance between the first and second variables. The number of indicators to be evaluated, Let covariance matrix be the variance matrix. The inverse of the covariance matrix, The adjoint matrix of the covariance matrix, Let E(X) be the determinant of the covariance matrix, and E(X) be the sample expectation. This represents the sample variance. based on The Gaussian distribution density function of the meta-Gaussian distribution: ， when , denote the covariance matrix Expectation and variance: , The inverse of a matrix is ​​equivalent to the product of the inverse of its determinant and its adjoint matrix: ， The probability density function of the bivariate normal distribution is: 。 5. The method for predicting rose bud formation based on nonlinear regression and multivariate probability distribution models according to claim 1, characterized in that, The selection of the core indicators employs principal component analysis (PCA) for feature dimensionality reduction; specifically, it includes: Zero-mean data: ， use The vector representation of the projection direction to be calculated. This represents the transpose of the projection direction vector; Sample Projected to direction Above, record Then all samples after projection are: The objective function is to maximize the variance of the projected samples. ， ， The accumulated temperature and light radiation, calculated using the Lagrange multiplier method, are the main factors affecting plant traits.

6. The method for predicting rose bud formation based on nonlinear regression and multivariate probability distribution models according to claim 5, characterized in that, The selection of the core indicators also includes: Collect all environmental cumulative values ​​and plant trait data at the time of bud formation for all plants; perform frequency statistics and distribution analysis on each indicator, and calculate skewness, kurtosis, mean, and sample variance as measures; Analysis of the environmental control data of the bud formation revealed that the skewness was close to 0, indicating excellent symmetry; the kurtosis was significantly lower than 0, indicating fewer data tails and a more uniform distribution. Overall, the data exhibited a discrete normal distribution with the following probability density function: ; Leaf area, digital biomass, and other plant-related indicators were analyzed using the Shapiro-Wilk normality test and QQ plot visualization. The data exhibited a log-normal distribution, with the following probability density function: Its cumulative probability distribution can be expressed as: ， ， in Indicates a sample, Represents the sample mean. Represents the sample variance, and is the calculation function. Regarding variables indefinite integrals, Introducing auxiliary variables .

7. The method for predicting rose bud formation based on nonlinear regression and multivariate probability distribution models according to claim 1, characterized in that, When some core indicator data is missing, the existing nonlinear regression model is used to predict the missing indicator data based on the existing indicator data. Then, the directly measured and indirectly predicted indicator data are input into the multivariate probability distribution model to calculate the budding probability.

8. The method for predicting rose bud formation based on nonlinear regression and multivariate probability distribution models according to claim 1, characterized in that, The core indicators include at least accumulated temperature, cumulative photosynthetically active radiation, leaf area, and digital biomass.

9. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the computer program, it implements a method for predicting rose bud formation based on nonlinear regression and multivariate probability distribution models as described in any one of claims 1 to 8.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements a method for predicting rose bud formation based on nonlinear regression and multivariate probability distribution models as described in any one of claims 1 to 8.

Images