Crop per unit yield prediction method, system, equipment and medium

Abstract

The invention provides a crop per unit yield prediction method, system and device and a medium, and belongs to the field of agriculture and computer crossing, and the method comprises the following steps: constructing a least square structure equation PLS-SEM model; calculating an initial value of a factor weight w and a factor load i representing a relationship between an observation variable and a latent variable of the initial PLS-SEM model, and calculating a value of a goodness of fit index GOF of the initial PLS-SEM model; performing factor analysis on the initial PLS-SEM model according to the value of the GOF, and removing invalid observation variables, invalid latent variables and invalid paths among the latent variables of the initial PLS-SEM model to obtain a final PLS-SEM model; and inputting the scores of the plurality of latent variables of the final PLS-SEM model into a PSO-LSSVM to obtain a predicted value of the per unit yield of the crops. The method can improve the prediction precision of the per unit yield of crops.

Description

Technical Field

[0001] The present invention belongs to the cross field of agriculture and computers, and specifically relates to a method, system, equipment and medium for predicting crop yield. Background Art

[0002] Accurately predicting grain yields is crucial to regional food security, and how to improve prediction accuracy is a current technical challenge.

[0003] Existing research mainly uses statistical models or machine learning models to simulate or predict grain yields and analyze their main driving factors. Traditional statistical models, such as multivariate regression models, can use statistical data such as the sown area, fertilizer application, and climate factors of agricultural products to fit the yield of agricultural products. However, such models can only consider the linear relationship between statistical data and are powerless against nonlinear relationships, which limits the fitting accuracy of the model. Machine learning models solve the defects of traditional statistical models in nonlinear relationships, but the input data of machine learning models only considers the pairwise relationship between independent variables and dependent variables (direct relationship), and cannot consider how one independent variable affects the dependent variable through another independent variable (indirect relationship), resulting in low prediction accuracy of crop yields.

[0004] In summary, the existing crop yield prediction methods have the problem of low prediction accuracy. Summary of the invention

[0005] In order to overcome the problem of low accuracy in crop yield prediction, the present invention provides a method for crop yield prediction, comprising the following steps:

[0006] Determine the latent variables and observed variables that affect the crop yield in the study area; establish the directional relationship between the observed variables and the latent variables, and the directional relationship between the latent variables, and construct the initial PLS-SEM model based on the directional relationship; the observed variables are agricultural statistical data and remote sensing data; the latent variables are the quantities calculated from the observed variables;

[0007] Calculate the initial values ​​of factor weights w and factor loadings i that characterize the relationship between observed variables and latent variables of the initial PLS-SEM model, as well as the initial values ​​of the path coefficients r between the latent variables; obtain the value of the goodness-of-fit index GOF used to evaluate the overall fitting effect of the initial PLS-SEM model under the conditions of the initial values ​​of factor weights w, factor loadings i and path coefficients r; determine whether the value of the goodness-of-fit index GOF reaches the set threshold, if not, perform factor analysis on the initial PLS-SEM model, remove invalid observed variables, invalid latent variables and invalid paths between latent variables of the initial PLS-SEM model, and iteratively update the initial values ​​of factor weights w, factor loadings i and path coefficients r until the value of the goodness-of-fit index GOF corresponding to the updated values ​​reaches the set threshold, and obtain the final PLS-SEM model;

[0008] The scores of the corresponding latent variables are calculated by the weights of the observed variables in the final PLS-SEM model, and the scores of the latent variables are input into the least squares support vector machine model PSO-LSSVM to obtain the predicted value of crop yield.

[0009] Preferably, the step of establishing the directional relationship between the observed variables and the latent variables, and the directional relationship between the latent variables, to obtain the initial PLS-SEM model comprises the following steps:

[0010] The latent variable relationship matrix of the initial PLS-SEM model is defined as follows: the latent variables include crop yield, crop demand, climate conditions and human activities. According to the direct influence of crop demand, climate conditions and human activities on crop yield, the indirect influence of crop demand on crop yield by directly influencing human activities, and the indirect influence of climate conditions on crop yield by directly influencing human activities, the directional relationship between latent variables is established to obtain the latent variable relationship matrix;

[0011] Set the observed variables corresponding to each latent variable in the latent variable relationship matrix, specifically: set the observed variable of crop yield to yield per unit area, set the observed variable of crop demand to crop consumption, GDP, population and crop export, set the observed variables of climate conditions to precipitation, evapotranspiration, maximum temperature and minimum temperature, and set the observed variables of human activities to N, P2O5, K2O and root water;

[0012] Point the arrow of the path between the latent variable and the observed variable to the observed variable to obtain the initial PLS-SEM model;

[0013] Preferably, before inputting the scores of the latent variables into the least squares support vector machine model PSO-LSSVM, the method further includes training the least squares support vector machine model PSO-LSSVM using a particle swarm optimization algorithm, including the following steps:

[0014] The scores of multiple latent variables of the final PLS-SEM model are used as the input of PSO-LSSVM, and the statistical values ​​of crop yield are used as the output of PSO-LSSVM;

[0015] Based on the Matlab platform, the data sets of input items and output items are imported into PSO-LSSVM, a pair of first-generation parameters c and θ are randomly given, and the first-generation parameters c and θ are substituted into the radial basis function of PSO-LSSVM;

[0016] The input and output data of PSO-LSSVM are used to train the generation parameters c and θ, and the trained PSO-LSSVM model is used to predict crop yields.

[0017] The first generation parameters c and θ are adjusted with the goal of reducing the error between the predicted value and the statistical value of crop yield, the first generation error is recorded, and the adjusted second generation parameters c and θ are used to replace the first generation parameters c and θ of PSO-LSSVM;

[0018] Repeat the crop yield prediction process for multiple iterations to select c and θ with the smallest error between the predicted value and the statistical value of crop yield, and obtain the optimized c and θ;

[0019] The PSO-LSSVM corresponding to the optimized c and θ is used as the trained PSO-LSSVM.

[0020] Preferably, the ranges of the parameters c and θ are set to [0.1, 500] and [0.1, 100] respectively, and the unit of parameter change in each iteration is 0.1.

[0021] Preferably, the calculation method of the score of each latent variable is specifically as follows:

[0022] Standardize the values ​​of multiple observed variables for each latent variable;

[0023] The values ​​of multiple observed variables of each latent variable after standardization are multiplied by the factor weight w of the observed variable and then added together; the scores of multiple latent variables input into the least squares support vector machine model PSO-LSSVM include crop demand scores, climate condition scores and human activity scores.

[0024] Preferably, the threshold value of the goodness of fit index GOF is greater than 0.4.

[0025] The present invention also provides a crop yield prediction system, comprising:

[0026] The initial model building module is used to determine the latent variables and observed variables that affect the crop yield in the study area; establish the directional relationship between the observed variables and the latent variables, and the directional relationship between the latent variables, and build the initial PLS-SEM model based on the directional relationship; the observed variables are agricultural statistical data and remote sensing data; the latent variables are the quantities calculated from the observed variables;

[0027] The model optimization module is used to calculate the initial values ​​of the factor weight w and factor loading i that characterize the relationship between the observed variables and latent variables of the initial PLS-SEM model, as well as the initial value of the path coefficient r between the latent variables; obtain the value of the goodness-of-fit index GOF used to evaluate the overall fitting effect of the initial PLS-SEM model under the conditions of the initial values ​​of the factor weight w, factor loading i and path coefficient r; determine whether the value of the goodness-of-fit index GOF reaches the set threshold, if not, perform factor analysis on the initial PLS-SEM model, remove invalid observed variables, invalid latent variables and invalid paths between latent variables of the initial PLS-SEM model, and iteratively update the initial values ​​of the factor weight w, factor loading i and path coefficient r until the value of the goodness-of-fit index GOF corresponding to the updated value reaches the set threshold, thereby obtaining the final PLS-SEM model;

[0028] The yield prediction module is used to calculate the scores of the corresponding latent variables through the weights of the observed variables of the final PLS-SEM model, and input the scores of the latent variables into the least squares support vector machine model PSO-LSSVM to obtain the predicted value of crop yield.

[0029] The present invention also provides a computer device, comprising a memory and a processor; the memory stores a computer program, and the processor is used to run the computer program in the memory to execute the crop yield prediction method.

[0030] The present invention also provides a computer-readable storage medium, wherein the computer-readable storage medium stores a computer program, and the computer program is suitable for being loaded by a processor to execute the crop yield prediction method.

[0031] The crop yield prediction method provided by the present invention has the following beneficial effects:

[0032] The present invention can construct an initial PLS-SEM model by establishing a directional relationship between observed variables and latent variables, and a directional relationship between latent variables. The observed variables affecting crop yields can be linked to the latent variables through the initial PLS-SEM model. The initial values ​​of factor weights w and factor loads i that characterize the relationship between observed variables and latent variables of the initial PLS-SEM model are calculated, and the initial values ​​of path coefficients r between latent variables and the corresponding value of a goodness of fit index GOF for evaluating the overall fitting effect of the PLS-SEM model are calculated. The initial PLS-SEM model can be subjected to factor analysis and path analysis, and invalid observed variables, invalid latent variables and invalid paths between latent variables are removed, thereby solving the problem in the prior art that it is impossible to consider how an independent variable affects a dependent variable through another independent variable. The factor weights w, factor loads i and path coefficients r after removing invalid observed variables, invalid latent variables and invalid paths between latent variables are updated, and the scores of the latent variables calculated after the update are input into the PSO-LSSVM, so that the influence relationship between the variables can be taken into consideration, thereby making the prediction accuracy of the crop yield higher. BRIEF DESCRIPTION OF THE DRAWINGS

[0033] In order to more clearly illustrate the embodiment of the present invention and its design scheme, the following briefly introduces the drawings required for this embodiment. The drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0034] Figure 1 A schematic diagram of a process flow of a crop yield prediction method according to an embodiment of the present invention;

[0035] Figure 2 The theoretical framework for the initial PLS-SEM model;

[0036] Figure 3 are the parameter values ​​and fitting results of the adjusted PLS-SEM model;

[0037] Figure 4 is the fitting result of the final PLS-SEM model;

[0038] Figure 5 The results of predicting wheat yield using three different data input PSO-LSSVM models; Figure 5 (a) is the prediction result of taking all observed variables of the initial PLS-SEM model as the input of the PSO-LSSVM model. Figure 5 (b) is the prediction result of using the scores of the latent variables of the initial PLS-SEM model as the input of the PSO-LSSVM model; Figure 5(c) shows the prediction results of using the scores of the latent variables of the final PLS-SEM model as the input of the PSO-LSSVM model. DETAILED DESCRIPTION

[0039] In order to enable those skilled in the art to better understand the technical solution of the present invention and implement it, the present invention is described in detail below in conjunction with the accompanying drawings and specific embodiments. The following embodiments are only used to more clearly illustrate the technical solution of the present invention, and cannot be used to limit the scope of protection of the present invention.

[0040] In the description of the present invention, it is necessary to understand that the terms "PLS-SEM", "PSO-LSSVM", "Cronbach's alpha", "Dillon-Goldstein's rho", "GOF (Goodness-of-fit) goodness-of-fit index", "w (factor weight)", "i (factor loading)", "r (Pathway coefficient)", "R 2 The actual meanings of indications such as "(R-squared value)", "N (nitrogen fertilizer application)", "P2O5 (phosphate fertilizer application)", "K2O (potassium fertilizer application)", "GDP (gross domestic product)" are based on the contents shown in the accompanying drawings and are only for the convenience of describing the technical solution of the present invention and simplifying the description, rather than indicating or implying that the model name abbreviations, observed variables or model evaluation indicators referred to must have a specific representation method, and therefore cannot be understood as a limitation to the present invention.

[0041] In addition, the terms "first", "second", etc. are only used for descriptive purposes and cannot be understood as indicating or implying relative importance. In the description of the present invention, it should be noted that, unless otherwise clearly specified or limited, the term "adjustment model" should be understood in a broad sense, for example, it can be to adjust model parameters, or it can be to adjust model input and output items, or delete-add model paths. In addition, the observed variables of the latent variables "wheat demand", "climatic conditions", and "human activities" should be understood as indicators of human activities that affect changes in wheat demand, climate change, and crop yields, such as per capita income, dietary level, wind speed, light, and irrigation, etc., and should not be limited to the types mentioned in the present invention. For those of ordinary skill in the art, the specific meanings of the above terms in the present invention can be understood according to the specific circumstances. In the description of the present invention, unless otherwise specified, the meaning of "multiple" is two or more, which will not be described in detail here.

[0042] Example

[0043] The present invention provides a method for predicting crop yield per unit area, specifically taking wheat in Kazakhstan as the research object, such as Figure 1As shown, the following steps are included:

[0044] Step 1: Obtain statistical data and remote sensing data that affect crop yields in the study area, determine the latent variables and observed variables that affect crop yields based on the statistical data and remote sensing data; establish the directional relationship between observed variables and latent variables, and the directional relationship between latent variables, and construct the initial least squares structural equation PLS-SEM (Partial Least Squares Structural Equation Modeling) model based on the directional relationship.

[0045] The specific steps are as follows:

[0046] Statistical data include crop yield, crop area, crop consumption, crop export, GDP, population, fertilizer application (N, P2O5, K2O), etc. Remote sensing data include precipitation, evapotranspiration, temperature (maximum temperature, minimum temperature), root soil moisture content (root water), and other data. Quantities that can be directly observed are called observed variables; observed variables are calculated to quantify certain important concepts that cannot be directly observed, and these concepts are called latent variables.

[0047] The latent variable relationship matrix of the initial PLS-SEM model is defined as follows: the latent variables include crop yield, crop demand, climate conditions and human activities. According to the direct influence of crop demand, climate conditions and human activities on crop yield, the indirect influence of crop demand on crop yield by directly influencing human activities, and the indirect influence of climate conditions on crop yield by directly influencing human activities, the directional relationship between latent variables is established to obtain the latent variable relationship matrix.

[0048] Set the observed variables corresponding to each latent variable in the latent variable relationship matrix, specifically: set the observed variable of crop yield to yield per unit area, set the observed variable of crop demand to crop consumption, GDP, population, and crop export volume, set the observed variables of climate conditions to precipitation, evapotranspiration, maximum temperature, and minimum temperature, and set the observed variables of human activities to N, P2O5, K2O, and root water; the arrows of the paths between latent variables and observed variables point to the observed variables.

[0049] In this embodiment, in order to better understand the construction process of the initial PLS-SEM model, the present invention specifically selects wheat for illustration. Specifically, the latent variables include wheat yield, wheat demand, climate conditions and human activities; the latent variable "wheat yield" can be represented by the observed variable "yield per unit area", and its value is the ratio of total wheat yield to sown area; the change of the latent variable "wheat demand" can be reflected by the observed variables "wheat consumption", "GDP", "population", and "wheat export volume". The larger the value of the observed variable, the greater the demand for wheat; the latent variable "climate conditions" can be reflected by the observed variables "precipitation", "evapotranspiration", "maximum temperature", and "minimum temperature". The larger the value of the observed variable, the higher the climate condition score; the latent variable "human activities" can be reflected by the observed variables "N", "P2O5", "K2O" (reflecting fertilization) and "root water" (reflecting irrigation). The larger the value of the observed variable, the greater the impact of human activities on wheat yield. The initial PLS-SEM model of wheat finally obtained is as follows: Figure 2 shown.

[0050] Step 2: Calculate the initial values ​​of factor weights w and factor loadings i that characterize the relationship between observed variables and latent variables of the initial PLS-SEM model, as well as the initial values ​​of the path coefficients r between latent variables; obtain the value of the goodness-of-fit index GOF used to evaluate the overall fitting effect of the initial PLS-SEM model under the initial values ​​of factor weights w, factor loadings i, and path coefficients r; determine whether the value of the goodness-of-fit index GOF reaches the set threshold. If not, perform factor analysis on the initial PLS-SEM model to remove invalid observed variables, invalid latent variables, and invalid paths between latent variables of the initial PLS-SEM model, and iteratively update the initial values ​​of factor weights w, factor loadings i, and path coefficients r until the value of the goodness-of-fit index GOF corresponding to the updated value reaches the set threshold, and the final PLS-SEM model is obtained. The specific process is as follows:

[0051] Load the "plspm software package" of the RStudio platform, calculate the initial values ​​of the factor weights w and factor loadings i that characterize the relationship between the observed variables and latent variables of the initial PLS-SEM model; calculate the initial values ​​of the path coefficients r that characterize the relationship between the latent variables and latent variables of the initial PLS-SEM model, with negative values ​​indicating negative correlation and positive values ​​indicating positive correlation. The initial values ​​of the factor weights w and factor loadings i and the initial values ​​of the path coefficients r are as follows: Figure 2As shown in the figure; the goodness-of-fit index GoF (Goodness-of-fit) is an indicator to measure the overall fitting effect of the PLS-SEM model. The higher the value, the better the fitting effect (it is generally believed that a GOF value greater than 0.4 is acceptable); Cronbach's alpha is a statistical indicator to measure the internal consistency of the PLS-SEM model. The higher its value, the more consistent the observed variables are when measuring the same latent variable. Dillon-Goldstein's rho (DG.rho) and Cronbach's alpha (C.alpha) has similar functions (the range of C.alpha and DG.rho values ​​is between 0 and 1, and the higher the value, the more consistent the observed variables of the latent variable are in measuring the same latent variable concept); the significant P value is an indicator used in statistics to measure whether the observed results are statistically significant in hypothesis testing. Specifically, it indicates the probability of the observed results or more extreme results occurring under the premise that the null hypothesis is true. P values ​​less than 0.05, 0.01, and 0.001 indicate significant, relatively significant, and very significant, respectively. The stronger the significance, the less likely the correlation is determined by accidental factors. In the present invention, *** indicates that the P value is less than 0.001, ** indicates that the P value is less than 0.01, and * indicates that the P value is less than 0.05. R 2 The coefficient of determination indicates the degree to which the latent variable is explained by other latent variables. The goodness of fit index GOF calculated by the present invention evaluates the overall fitting effect of the initial PLS-SEM model, the P value of whether the path coefficient r is significant, and the exogenous latent variables (latent variables affected by other latent variables, such as Figure 2 R of "wheat yield", "human activities" 2 The value of Figure 2 As shown; according to the value of influencing factor load i, the value of factor weight w, and the value of path coefficient r, factor analysis is carried out to quantify the influence degree and action path of observed variables on wheat yield, remove invalid observed variables, invalid latent variables, and invalid paths between latent variables of the initial PLS-SEM model, retain and increase valid paths, and obtain a PLS-SEM model in which all evaluation indicators have improved after adjustment. In this embodiment, the designed paths all meet the requirements, so only the value of factor load i, the value of factor weight w, and the value of path coefficient r are adjusted.

[0052] Figure 3 are the parameter values ​​and fitting results of the adjusted PLS-SEM model. Negative values ​​of r, i, and w indicate negative correlation, and positive values ​​of r indicate positive correlation. The larger the absolute value, the stronger the correlation. Figure 3 As can be seen from the figure, the R 2=0.25, indicating that the information represented by the three latent variables of wheat demand, climate conditions, and human activities explains 25% of the variation in wheat yield. Similarly, the R 2 =0.35, indicating that the information represented by the two latent variables of wheat demand and climate conditions explains 35% of the change in wheat yield; wheat demand and climate conditions have positive (r=0.46) and negative (r=-0.42) effects on wheat yield respectively. As shown in Table 1, the values ​​of C.alpha and DG.rho are both greater than 0.6, indicating that the selected observed variables have good consistency in describing their latent variables.

[0053] Table 1 Values ​​of C.alpha and DG.rho of the adjusted PLS-SEM model

[0054] Evaluation indicators C.alpha DG.rho Wheat demand 0.62 0.80 Climate conditions 0.62 0.78 Human activities 0.65 0.79

[0055] The overall goodness of fit index GOF of the adjusted PLS-SEM model is 0.399, which is less than 0.4, indicating that the adjusted PLS-SEM model needs to be further adjusted. By observing the factor loading i and factor weight w, we can see that the influence of root water and wheat consumption on their latent variables is very weak, which may lead to poor fitting effect of the PLS-SEM model. Therefore, invalid observed variables are removed, and the factor loading i and factor weight w are updated to obtain a further adjusted PLS-SEM model. Figure 4 As shown, compared Figure 3 It can be seen that Figure 4 The GOF of the wheat yield is 0.402, and the overall fitting effect is improved. 2 =0.30, indicating that the explanation of the latent variable on wheat yield increased by 5%, and the model fitting accuracy was higher, but the explanation of wheat demand and climate conditions on wheat yield decreased. The values ​​of C.alpha and DG.rho were shown in Table 2, and they were greatly improved, indicating that after removing some observed variables with weak correlation, other observed variables were more consistent in describing their latent variables. The PLS-SEM model at this time was the final PLS-SEM model.

[0056] Table 2 Values ​​of C.alpha and DG.rho of the final PLS-SEM model

[0057] Evaluation indicators C.alpha DG.rho Wheat demand 0.88 0.93 Climate conditions 0.62 0.78 Human activities 0.81 0.89

[0058] Step 3: Calculate the scores of the corresponding latent variables through the weights of the observed variables of the final PLS-SEM model, input the scores of the latent variables into the least squares support vector machine model PSO-LSSVM, and obtain the predicted value of crop yield.

[0059] (1) The least squares support vector machine model PSO-LSSVM (Particle Swarm Optimization-Least Squares Support Vector Machine) is trained using the particle swarm optimization algorithm, including the following steps:

[0060] The scores of multiple latent variables of the final PLS-SEM model are used as the input of PSO-LSSVM, and the statistical values ​​of crop yield are used as the output of PSO-LSSVM; based on the Matlab platform, the data sets of input and output items are imported into PSO-LSSVM, and a pair of first-generation parameters c and θ are randomly given and substituted into the radial basis function of PSO-LSSVM.

[0061] The input and output data of the PSO-LSSVM model are used to train c and θ; and the trained PSO-LSSVM model is used to predict crop yields.

[0062] The first-generation parameters c and θ are adjusted with the goal of reducing the error between the predicted value and the statistical value of crop yield, the first-generation error is recorded, and the adjusted second-generation parameters are used to replace the first-generation parameters c and θ of PSO-LSSVM.

[0063] Repeat the crop yield prediction process for multiple iterations to select c and θ with the smallest error between the predicted value and the statistical value of the crop yield, that is, to obtain the optimized c and θ.

[0064] The PSO-LSSVM corresponding to the optimized c and θ is used as the trained PSO-LSSVM.

[0065] (2) The values ​​of the observed variables corresponding to wheat demand, climate conditions, and human activities in the final PLS-SEM model were standardized to eliminate the dimensionality effect. The values ​​of the observed variables corresponding to each standardized latent variable were multiplied by the factor weight of the observed variable and then added up to obtain the scores of wheat demand, climate conditions, and human activities. Finally, the scores of the latent variables wheat demand, climate conditions, and human activities were input into the least squares support vector machine model PSO-LSSVM to predict wheat yield.

[0066] Finally, in order to verify the prediction effect of the method of the present invention, the present invention uses the traditional method to predict the wheat yield per unit area, and the specific process is: based on the MATLAB platform, the PSO-LSSVM model is established and run, the data from 1992 to 2011 are used as training samples, the data from 2012 to 2019 are used as prediction data, all the observed variables of the initial PLS-SEM model or the scores of the latent variables of the initial PLS-SEM model are used as the input items of the PSO-LSSVM model (i.e., the input items of the traditional method), and the wheat yield per unit area is used as the output item to obtain the prediction results. The prediction results include the optimal PSO-LSSVM parameters (c and θ), the predicted values ​​of wheat yield per unit area from 2012 to 2019, and the error between the predicted value of wheat yield and the actual value. Figure 5 The results of predicting wheat yield using three different data input PSO-LSSVM models; Figure 5 (a) is the prediction result of taking all observed variables of the initial PLS-SEM model as the input of the PSO-LSSVM model. Figure 5 (b) is the prediction result of using the scores of the latent variables of the initial PLS-SEM model as the input of the PSO-LSSVM model; Figure 5 (c) is the prediction result of using the final PLS-SE. The scores of the latent variables of the M model as the input items of the PSO-LSSVM model are given by Figure 5 From (a), we can see that when only all the observed variables of the initial PLS-SEM model are used as the input items of the PSO-LSSVM model, the total error in 8 years is 0.93 tons / hectare, and the annual fitting error is 0.12 tons / hectare (traditional prediction result); Figure 5 (b) shows that when the scores of the latent variables of the initial PLS-SEM model are used as the input items of the PSO-LSSVM model, the total error in 8 years is 0.86 tons / hectare, and the annual fitting error is 0.11 tons / hectare (traditional prediction result); Figure 5 (c) shows that when the scores of the latent variables of the final PLS-SEM model are used as the input items of the PSO-LSSVM model, the total error in 8 years is 0.77 tons / hectare, and the fitting error each year is 0.10 tons / hectare. It can be seen that the present invention uses the scores of the latent variables of the final PLS-SEM model as the input items of the PSO-LSSVM model, and the obtained wheat yield prediction accuracy is higher.

[0067] The present invention also provides a crop yield prediction system, including an initial model construction module, a model optimization module and a yield prediction module. The initial model construction module is used to determine the latent variables and observed variables that affect the crop yield in the study area; establish a directional relationship between observed variables and latent variables, and a directional relationship between latent variables, and construct an initial PLS-SEM model based on the directional relationship; the observed variables are agricultural statistical data and remote sensing data; the latent variables are quantities calculated from the observed variables; the model optimization module is used to calculate the initial values ​​of the factor weights w and factor loadings i that characterize the relationship between the observed variables and the latent variables of the initial PLS-SEM model, as well as the initial values ​​of the path coefficients r between the latent variables; obtain the goodness of fit index GO for evaluating the overall fitting effect of the initial PLS-SEM model under the conditions of the initial values ​​of the factor weights w, factor loadings i and path coefficients r. The value of F is obtained; whether the value of the goodness of fit index GOF reaches the set threshold is determined; if not, the initial PLS-SEM model is subjected to factor analysis to remove invalid observed variables, invalid latent variables and invalid paths between latent variables of the initial PLS-SEM model, and the initial values ​​of factor weight w, factor loading i and path coefficient r are iteratively updated until the value of the goodness of fit index GOF corresponding to the updated value reaches the set threshold, and the final PLS-SEM model is obtained; the yield prediction module is used to calculate the score of the corresponding latent variable through the weight of the observed variable of the final PLS-SEM model, and input the score of the latent variable into the least squares support vector machine model PSO-LSSVM to obtain the predicted value of crop yield.

[0068] The present invention also provides a computer device, comprising a memory and a processor; the memory stores a computer program, and the processor is used to run the computer program in the memory to execute the crop yield prediction method.

[0069] The present invention also provides a computer-readable storage medium, which stores a computer program. The computer program is suitable for being loaded by a processor to execute the crop yield prediction method.

[0070] The above embodiments are only preferred specific implementation methods of the present invention, and the protection scope of the present invention is not limited thereto. Any simple changes or equivalent replacements of the technical solutions that can be obviously obtained by any technician familiar with the field within the technical scope disclosed in the present invention belong to the protection scope of the present invention.

Claims

1. A method for predicting crop yield, characterized in that: The steps include: Determine the latent variables and observed variables that affect the crop yield in the study area; establish the directional relationship between the observed variables and the latent variables, and the directional relationship between the latent variables, and construct the initial PLS-SEM model based on the directional relationship; the observed variables are agricultural statistical data and remote sensing data; the latent variables are the quantities calculated from the observed variables; Calculate the initial values ​​of factor weights w and factor loadings i that characterize the relationship between observed variables and latent variables of the initial PLS-SEM model, as well as the initial values ​​of the path coefficients r between the latent variables; obtain the value of the goodness-of-fit index GOF used to evaluate the overall fitting effect of the initial PLS-SEM model under the conditions of the initial values ​​of factor weights w, factor loadings i and path coefficients r; determine whether the value of the goodness-of-fit index GOF reaches the set threshold, if not, perform factor analysis on the initial PLS-SEM model, remove invalid observed variables, invalid latent variables and invalid paths between latent variables of the initial PLS-SEM model, and iteratively update the initial values ​​of factor weights w, factor loadings i and path coefficients r until the value of the goodness-of-fit index GOF corresponding to the updated values ​​reaches the set threshold, and obtain the final PLS-SEM model; The scores of the corresponding latent variables are calculated by the weights of the observed variables in the final PLS-SEM model, and the scores of the latent variables are input into the least squares support vector machine model PSO-LSSVM to obtain the predicted value of crop yield.

2. The method for predicting crop yield per unit area according to claim 1, characterized in that: The step of establishing the directional relationship between the observed variables and the latent variables, and the directional relationship between the latent variables, and obtaining the initial PLS-SEM model, includes the following steps: The latent variable relationship matrix of the initial PLS-SEM model is defined as follows: the latent variables include crop yield, crop demand, climate conditions and human activities. According to the direct influence of crop demand, climate conditions and human activities on crop yield, the indirect influence of crop demand on crop yield by directly influencing human activities, and the indirect influence of climate conditions on crop yield by directly influencing human activities, the directional relationship between latent variables is established to obtain the latent variable relationship matrix; Set the observed variables corresponding to each latent variable in the latent variable relationship matrix, specifically: set the observed variable of crop yield to yield per unit area, set the observed variable of crop demand to crop consumption, GDP, population and crop export, set the observed variables of climate conditions to precipitation, evapotranspiration, maximum temperature and minimum temperature, and set the observed variables of human activities to N, P2O5, K2O and root water; The arrow of the path between the latent variable and the observed variable is pointed to the observed variable to obtain the initial PLS-SEM model.

3. The method for predicting crop yield per unit area according to claim 1, characterized in that: Before inputting the scores of the latent variables into the least squares support vector machine model PSO-LSSVM, the least squares support vector machine model PSO-LSSVM is trained by using a particle swarm optimization algorithm, including the following steps: The scores of multiple latent variables of the final PLS-SEM model are used as the input of PSO-LSSVM, and the statistical values ​​of crop yield are used as the output of PSO-LSSVM; Based on the Matlab platform, the data sets of input items and output items are imported into PSO-LSSVM, a pair of first-generation parameters c and θ are randomly given, and the first-generation parameters c and θ are substituted into the radial basis function of PSO-LSSVM; The input and output data of PSO-LSSVM are used to train the generation parameters c and θ, and the trained PSO-LSSVM model is used to predict crop yields. The first-generation parameters c and θ are adjusted with the goal of reducing the error between the predicted value and the statistical value of crop yield, the first-generation error is recorded, and the adjusted second-generation parameters c and θ are used to replace the first-generation parameters c and θ of PSO-LSSVM; Repeat the crop yield prediction process for multiple iterations to select c and θ with the smallest error between the predicted value and the statistical value of crop yield, and obtain the optimized c and θ; The PSO-LSSVM corresponding to the optimized c and θ is used as the trained PSO-LSSVM.

4. The method for predicting crop yield per unit area according to claim 3, characterized in that: The ranges of the parameters c and θ are set to [0.1, 500] and [0.1, 100] respectively, and the unit of parameter change in each iteration is 0.

1.

5. The method for predicting crop yield per unit area according to claim 2, characterized in that: The calculation method of the score of the latent variable is specifically as follows: Standardize the values ​​of multiple observed variables for each latent variable; The values ​​of multiple observed variables of each latent variable after standardization are multiplied by the factor weight w of the observed variable and then added up; the scores of the latent variables input into the least squares support vector machine model PSO-LSSVM include the crop demand score, climate condition score and human activity score.

6. The method for predicting crop yield per unit area according to claim 1, characterized in that: The threshold value of the goodness of fit index GOF is greater than 0.

4.

7. A crop yield prediction system, characterized in that: include: The initial model building module was used to determine the latent and observed variables that affect crop yields in the study area; Establishing the directional relationship between observed variables and latent variables, and the directional relationship between latent variables, and constructing the initial PLS-SEM model according to the directional relationship; the observed variables are agricultural statistical data and remote sensing data; the latent variables are quantities calculated from the observed variables; The model optimization module is used to calculate the initial values ​​of the factor weight w and factor loading i that characterize the relationship between the observed variables and latent variables of the initial PLS-SEM model, as well as the initial value of the path coefficient r between the latent variables; obtain the value of the goodness-of-fit index GOF used to evaluate the overall fitting effect of the initial PLS-SEM model under the conditions of the initial values ​​of the factor weight w, factor loading i and path coefficient r; determine whether the value of the goodness-of-fit index GOF reaches the set threshold, if not, perform factor analysis on the initial PLS-SEM model, remove invalid observed variables, invalid latent variables and invalid paths between latent variables of the initial PLS-SEM model, and iteratively update the initial values ​​of the factor weight w, factor loading i and path coefficient r until the value of the goodness-of-fit index GOF corresponding to the updated value reaches the set threshold, thereby obtaining the final PLS-SEM model; The yield prediction module is used to calculate the scores of the corresponding latent variables through the weights of the observed variables of the final PLS-SEM model, and input the scores of the latent variables into the least squares support vector machine model PSO-LSSVM to obtain the predicted value of crop yield.

8. A computer device, characterized in that: It comprises a memory and a processor; the memory stores a computer program, and the processor is used to run the computer program in the memory to execute the crop yield prediction method according to any one of claims 1 to 6.

9. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a computer program, and the computer program is suitable for being loaded by a processor to execute the crop yield prediction method according to any one of claims 1-6.

Images