A method, device and medium for regulating water, fertilizer and medicine in different zones for marigold

Abstract

The present invention belongs to the technical field of smart agriculture, and specifically discloses a method, device and medium for regulating the zoning of marigold water, fertilizer and medicine based on digital twins. The method disclosed in the present invention comprises: first, by collecting and clustering data, the marigold planting area is divided into a number of similar sub-areas; then, a time period is selected, and for each divided sub-area, the three feature evaluation methods of gradient boosting decision tree, Pearson coefficient and mutual information coefficient are respectively used, and the exponential decay law of time series data is considered to calculate the comprehensive time-varying characteristics within the time period; finally, the result prediction of the XGBoost model and the CNN‑LSTM model is integrated, and then all sub-areas are integrated to obtain the predicted optimal water, fertilizer and medicine dosage. The present invention can accurately predict the optimal dosage of marigold water, fertilizer and medicine based on time-varying characteristics, thereby improving the automation degree of marigold planting and improving the utilization rate of resources. The present invention can be widely used in the regulation of water, fertilizer and medicine for marigold planting.

Description

Technical Field

[0001] The present invention belongs to the technical field of smart agriculture, and specifically relates to a method, equipment and medium for zoning control of water, fertilizer and medicine for marigold. Background Art

[0002] In recent years, marigold has been planted on a large scale due to its fast growth, strong adaptability, high ornamental value and certain economic value. When marigold is planted on a large scale, factors such as regional and periodic differences in marigold growth, changes in weather and soil conditions will affect the growth of marigold. The traditional water, fertilizer and pesticide control method dominated by expert experience will lead to low resource utilization, destruction of soil nutrient balance and crop damage, which will in turn affect the economy and sustainability of marigold production.

[0003] At present, in addition to relying on expert decisions, statistics is the most widely used method in marigold cultivation. That is, using a small specific planting area and based on biological knowledge and statistical methods, observing the growth performance of crops over a period of time, so as to evaluate and analyze some influencing factors. Obviously, this method is based on a small number of samples and is extremely time-consuming. It is difficult to deeply explore the characteristics and laws, and cannot achieve automated regulation of large-scale marigold cultivation. Summary of the invention

[0004] The purpose of the present invention is to provide a method for regulating the use of water, fertilizer and pesticide in different regions of marigold, which can accurately predict the optimal use amount of water, fertilizer and pesticide for marigold based on time-varying characteristics, thereby improving the automation level of marigold planting and improving the utilization rate of resources;

[0005] The second object of the present invention is to provide a terminal device, which can implement a method for controlling water, fertilizer and pesticide in different areas of marigold when executing its own program;

[0006] The third object of the present invention is to provide a computer-readable storage medium for storing a corresponding computer program for a method for zoning control of water, fertilizer and pesticide for marigold.

[0007] In order to achieve the above object, the technical scheme adopted by the present invention is as follows:

[0008] A method for regulating water, fertilizer and pesticide by different zones for marigold comprises the following steps performed in sequence:

[0009] S1. By collecting and clustering marigold growth environment and growth characteristic data, the marigold planting area is divided into several similar sub-areas;

[0010] S2, select a time period, and for each divided sub-region, calculate the feature importance using three feature evaluation methods: gradient boosting decision tree, Pearson coefficient, and mutual information coefficient;

[0011] S3, based on the weighting coefficient, combined with the importance of the three features calculated by the three feature evaluation methods, calculate the comprehensive importance weight within the time period;

[0012] S4. Considering the exponential decay law of time series data, the comprehensive time-varying characteristics within the time period are calculated based on the comprehensive importance weight;

[0013] S5, input the calculated comprehensive time-varying features into the XGBoost model and the CNN-LSTM model respectively to obtain the prediction results of the two models;

[0014] S6. Weights are assigned to the prediction results of the two models to comprehensively obtain the optimal water, fertilizer and pesticide application amounts predicted for each sub-region;

[0015] S7. Integrate the optimal water, fertilizer and medicine usage predicted in all sub-regions to obtain the predicted water, fertilizer and medicine usage control data for the marigold planting area within the selected time period;

[0016] S8: Whether the data of the next time period is received;

[0017] If yes, the weight coefficient is updated by particle swarm algorithm and steps S2 to S7 are repeated;

[0018] If not, proceed to step S9;

[0019] S9. End.

[0020] As a limitation, step S1 includes the following steps performed in sequence:

[0021] S11, collecting sensor data from each marigold planting area to form a data matrix;

[0022] S12. Standardize the data and give the same weight to different features;

[0023] S13. Use K-means clustering algorithm to divide the marigold planting area.

[0024] As a second limitation, the formula for calculating the comprehensive importance weight within the time period in step S3 is:

[0025]

[0026] In the formula, is the comprehensive importance weight of feature i in time period T, α, β and γ are weighting coefficients, is the feature importance of feature i in time period T obtained by gradient boosting decision tree, ρ X,YI(X;Y)(i,T) is the feature importance of feature i in time period T obtained by the Pearson coefficient, and I(X;Y)(i,T) is the feature importance of feature i in time period T obtained by the mutual information coefficient.

[0027] As a further limitation, the calculation formula of step S4 is:

[0028]

[0029] In the formula, I is the comprehensive time-varying feature, i k is the value of feature i at the kth time point, α k is the exponential decay time attenuation factor, and its calculation formula is:

[0030] α k =exp(-λ(tk))

[0031] Where λ is the decay rate and t is the current time.

[0032] As a third limitation, the calculation formula of step S6 is:

[0033]

[0034] In the formula, is the predicted optimal water, fertilizer and medicine dosage, F XGB 、F CNN-LSTM are the water, fertilizer and medicine usage predicted by the XGBoost model and the CNN-LSTM model, respectively. w1 and w2 are the weight coefficients of the XGBoost and CNN-LSTM models, respectively. The calculation formula is:

[0035]

[0036] In the formula, loss T,CNN-LSTM 、loss T,XGB They are the mean square prediction errors of the CNN-LSTM model and the XGBoost model in time period T.

[0037] As a further limitation, the calculation formula for updating the weight coefficient by the particle swarm algorithm in step S8 is:

[0038]

[0039] Where Y T , It represents the actual demand value and predicted value of water, fertilizer and medicine in time period T. Δα, Δβ and Δγ are the degrees of change of the three weighted coefficients after updating, and τ is the balance coefficient.

[0040] A terminal device includes a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, the above-mentioned method for controlling water, fertilizer and pesticide zoning of marigold is implemented.

[0041] A computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, it is used to implement the above-mentioned marigold water, fertilizer and pesticide zoning control method.

[0042] Due to the adoption of the above technical solution, the present invention has the following technical advances compared with the prior art:

[0043] (1) The present invention divides marigold planting areas into different zones, and integrates three feature evaluation methods and the exponential decay of time series data. The two prediction models complement each other to obtain the predicted optimal water, fertilizer and medicine dosages for different regions. The water, fertilizer and medicine dosages for marigold can be accurately regulated, thereby improving the automation level of marigold planting and the utilization rate of resources.

[0044] (2) The present invention uses the K-means clustering algorithm to divide the marigold planting areas according to similarity, so that the marigold data in each area has high similarity, while the marigold data between different areas are obviously different. This clear division is helpful for subsequent data analysis and processing;

[0045] (3) The present invention combines three feature evaluation methods, namely, gradient boosting decision tree, Pearson coefficient and mutual information coefficient, to calculate the comprehensive importance weight, thereby reducing the impact of the limitations of a single feature evaluation method;

[0046] (4) The present invention takes into account the exponential decay law of time series data during prediction when calculating the comprehensive time-varying features, thereby improving the accuracy of prediction;

[0047] (5) The present invention combines two different prediction models when predicting the optimal amount of water, fertilizer and medicine, and improves the accuracy of the prediction results by complementing the advantages of the two models;

[0048] (6) The present invention can adaptively change the weighting coefficients and the mean square prediction errors of the CNN-LSTM model and the XGBoost model when performing continuous prediction, thereby improving the accuracy of continuous prediction.

[0049] The present invention belongs to the technical field of smart agriculture, and can predict the optimal amount of water, fertilizer and medicine for marigold planting, accurately realize regulation, improve the degree of automation of marigold planting, and improve the utilization rate of resources. BRIEF DESCRIPTION OF THE DRAWINGS

[0050] The accompanying drawings are used to provide further understanding of the present invention and constitute a part of the specification. They are used to explain the present invention together with the embodiments of the present invention and do not constitute a limitation of the present invention.

[0051] In the attached picture:

[0052] Figure 1 This is a processing flow chart of Embodiment 1 of the present invention;

[0053] Figure 2 This is a sub-region distribution map of the marigold planting area in Example 1 of the present invention;

[0054] Figure 3 A comparison diagram of soil moisture ablation prediction between Example 1 of the present invention and three methods;

[0055] Figure 4 A comparison chart of the ablation prediction of the dosage of a certain brand of compound fertilizer in Example 1 of the present invention and three methods;

[0056] Figure 5 It is a comparison chart of Example 1 of the present invention and three methods in predicting the dosage ablation of a certain brand of medicine. DETAILED DESCRIPTION

[0057] The preferred embodiments of the present invention are described below in conjunction with the accompanying drawings. It should be understood that the preferred embodiments described herein are only used to illustrate and explain the present invention, and are not used to limit the present invention.

[0058] Example 1

[0059] This embodiment is a method for regulating water, fertilizer and medicine in different regions of marigold. This embodiment selects a data set of a marigold plant in Zhangjiakou area from January 1, 2017 to January 31, 2019. The plant covers an area of ​​100 mu, which is finely divided into 100 sub-regions per mu. Each sub-region shows significant differences due to its unique spatial distribution, climate environment, altitude and soil type. The key features related to the growth of marigold are selected: spatial location information of sub-regions (including central longitude, central latitude and altitude), meteorological conditions (date, wind speed, air humidity, wind direction, temperature and light intensity), soil element content (nitrogen, phosphorus, potassium, ammonia and moisture content), and soil moisture, a certain brand of compound fertilizer and a certain brand of medicine at the highest yield of marigold obtained by random sampling of each sub-region, totaling 17 dimensions of data features. The sampling frequency of the data is set to once every 6 hours.

[0060] In terms of data division, this embodiment selects the actual observation data from January 1, 2017 to December 31, 2018 as the training sample set for model training and optimization; and selects the data from January 1, 2019 to December 31, 2019 as the test set to evaluate the model's ability to predict soil moisture, a certain brand of compound fertilizer, and a certain brand of medicine demand in the next seven days. In order to quantify the prediction performance of the model, the root mean square error (RMSE) and the mean absolute percentage error (MAPE) are selected as prediction evaluation indicators.

[0061] like Figure 1 As shown, this embodiment includes the following steps performed in sequence:

[0062] S1. By collecting and clustering marigold growth environment and growth characteristic data, the marigold planting area is divided into several similar sub-areas.

[0063] Specifically, it is divided into the following steps which are performed in sequence:

[0064] S11. Collect sensor data from each marigold planting area to form a data matrix:

[0065]

[0066] Among them, x ij represents the jth feature of the ith area, n is the total number of planted areas, and d is the feature dimension.

[0067] At this point, it is known that one mu is a sub-region, that is, n is 100; the data features collected in the data set are the spatial location information, meteorological conditions, soil element content of the sub-region, and the soil moisture, dosage of a certain brand of compound fertilizer and dosage of a certain brand of medicine when the marigold yield is highest based on random sampling of each sub-region. The feature dimension d is 17.

[0068] S12. Standardize the data and give the same weight to different features:

[0069] The formula for standardization is:

[0070]

[0071] Among them, x' ij is the jth feature of the i-th region after normalization, μ j , σ j are the mean and standard deviation of the j-th feature, respectively.

[0072] At this time, the feature data in the training set are normalized one by one according to Formula 2, and the elements of each feature are mapped to [0,1].

[0073] S13. Use K-means clustering algorithm to divide the marigold planting area. The goal of K-means is to minimize the sum of the squares of the distances between the samples within each cluster and the cluster center. The clustering objective function is:

[0074]

[0075] Among them, K is the number of clusters, C k is the kth cluster, μ k is the centroid of the kth cluster, X i is the i-th data point belonging to the k-th cluster.

[0076] At this time, based on Formula 3, feature clustering is performed on the training set sub-region data to obtain three similar regions, which contain 32, 23, and 45 mu of marigolds respectively, and are not adjacent in spatial distribution, such as Figure 2 shown.

[0077] S2. Select a time period and calculate the feature importance for each sub-region after division using three feature evaluation methods: gradient boosting decision tree (GBDT), Pearson coefficient, and mutual information coefficient.

[0078] GBDT is a machine learning model that can discover nonlinear associations between features based on decision tree theory. The feature importance evaluation process based on the GBDT model is as follows:

[0079]

[0080] Among them, N trees is the number of trees in GBDT, ΔR node,f It is the reduction in model error brought by feature f on node splitting.

[0081] Mutual information measures the nonlinear relationship between two variables from a probability statistics perspective and is also applicable to feature evaluation. The feature importance calculation formula of mutual information I(X;Y) is:

[0082]

[0083] Where p(x,y) is the joint probability distribution of X and Y. p(x) and p(y) are the marginal probability distributions of X and Y respectively.

[0084] Mutual information reflects the dependency between two variables by calculating the amount of information shared between them. The larger the value, the stronger the dependency.

[0085] The Pearson coefficient is used to measure the linear correlation between two feature variables. The calculation formula for the Pearson correlation coefficient between feature X and target variable Y is:

[0086]

[0087] where cov(X,Y) is the covariance between X and Y, σ X and σ Y are the standard deviations of X and Y respectively.

[0088] Based on Formula 4, Formula 5 and Formula 6, each feature of the training set data is evaluated to obtain the corresponding evaluation results.

[0089] S3. Based on the weighting coefficient, the importance of the three features calculated by the three feature evaluation methods is combined to calculate the comprehensive importance weight within the time period.

[0090] Combining the above three evaluation indicators, we can get the comprehensive importance weight within time period T:

[0091]

[0092] In the formula, is the comprehensive importance weight of feature i in time period T, α, β and γ are weighting coefficients, is the feature importance of feature i in time period T obtained by gradient boosting decision tree, ρ X,Y I(X;Y)(i,T) is the feature importance of feature i in time period T obtained by the Pearson coefficient, and I(X;Y)(i,T) is the feature importance of feature i in time period T obtained by the mutual information coefficient.

[0093] Initialize the α, β, and γ parameters to 0.33 respectively, and obtain the comprehensive importance weights of 14 features including the spatial location information of the sub-region (including central longitude, central latitude and altitude), meteorological conditions (date, wind speed, air humidity, wind direction, temperature and light intensity), and soil element content (nitrogen, phosphorus, potassium, ammonia and water content) based on Formula 7.

[0094] S4. Considering the exponential decay law of time series data, the comprehensive time-varying characteristics within the time period are calculated based on the comprehensive importance weight.

[0095] The three feature evaluation indicators evaluate the importance of each feature in the time period T from multiple dimensions, but time series data has the time series characteristics of "small near and large far", that is, the newer the feature data point, the greater the impact on the use effect of the current model, and the degree of influence changes exponentially. Based on the comprehensive importance weights in the time period T obtained by the three indicators, based on the exponential decay law of time series data, the features of each time point are weighted by an exponential function, so that the older the feature points, the lower the importance, and vice versa.

[0096] The calculation formula for the comprehensive time-varying characteristics taking into account the time series is:

[0097]

[0098] In the formula, I is the comprehensive time-varying feature, i k is the value of feature i at the kth time point, α k is the exponential decay time attenuation factor, and its calculation formula is:

[0099] α k =exp(-λ(tk)) (9)

[0100] Where λ is the decay rate and t is the current time.

[0101] Based on Formula 8 and Formula 9, exponential decay is performed according to the distance of the time point, and T is set to a fixed 52, and finally a comprehensive time-varying feature that takes into account the time series is obtained.

[0102] S5. Input the calculated comprehensive time-varying features into the XGBoost model and the CNN-LSTM model respectively to obtain the prediction results of the two models.

[0103] XGBoost is a traditional machine learning model, and CNN-LSTM is a deep neural network model that can discover the temporal correlation of features. Both will predict the amount of water, fertilizer and medicine from different mechanisms.

[0104] S6. Weights are assigned to the prediction results of the two models to comprehensively obtain the optimal water, fertilizer and pesticide application amounts predicted for each sub-region.

[0105] After the XGBoost model and CNN-LSTM model complete the prediction, they are assigned weights respectively to obtain the optimal water, fertilizer and medicine dosage for each sub-region. The calculation formula is:

[0106]

[0107] In the formula, is the predicted optimal water, fertilizer and medicine dosage, F XGB 、F CNN-LSTM are the water, fertilizer and medicine usage predicted by the XGBoost model and the CNN-LSTM model, respectively. w1 and w2 are the weight coefficients of the XGBoost and CNN-LSTM models, respectively. The calculation formula is:

[0108]

[0109] In the formula, loss T,CNN-LSTM 、loss T,XGB They are the mean square prediction errors of the CNN-LSTM model and the XGBoost model in time period T.

[0110] Initialize the weights w1 and w2 to 0.5 respectively, and use Formula 10 and Formula 11 to get the predicted optimal water, fertilizer and pesticide dosage for each sub-region.

[0111] S7. By integrating the optimal water, fertilizer and medicine dosages predicted in all sub-regions, the predicted water, fertilizer and medicine dosage control data for the marigold planting area within the selected time period can be obtained.

[0112] S8: Whether the data of the next time period is received;

[0113] If yes, the weight coefficient is updated by particle swarm algorithm and steps S2 to S7 are repeated;

[0114] If not, proceed to step S9.

[0115] After completing the prediction of time period T, this embodiment can continue to receive data for the next time period and make predictions for the next time period. Since the importance of different features in different time periods will change over time, it is necessary to update the weighting coefficients through the particle swarm algorithm. The calculation formula is:

[0116]

[0117] Where Y T , It represents the actual demand value and predicted value of water, fertilizer and medicine in time period T. Δα, Δβ and Δγ are the degrees of change of the three weighted coefficients after updating, and τ is the balance coefficient.

[0118] When the weight coefficient is adaptively changed, τ is set to 0.02. At the same time, the weights of w1 and w2 will also change with the mean square prediction error of the CNN-LSTM model and the XGBoost model in the next time period.

[0119] After the weighting coefficient and the weights w1 and w2 are changed, the prediction for the next time period can be performed. If no prediction is performed, step S9 can be executed.

[0120] S9. End.

[0121] In order to verify the prediction effect of this embodiment, an ablation test was conducted using the third year data of a marigold plantation to predict the soil moisture, the amount of a certain brand of compound fertilizer, and the amount of a certain brand of medicine when the marigold yield is the highest. This comparison compares the method proposed in this embodiment with the following three methods: a single-index feature method without clustering and without considering time series (Model_three), a single-index feature method with clustering but without considering time series (Model_one), and a single-index feature method with clustering and considering time series (Model_two). Figure 3 , Figure 4 and Figure 5The prediction visualization results show that the method in this embodiment shows higher accuracy in predicting soil moisture demand, the dosage of a certain brand of compound fertilizer, and the dosage of a certain brand of medicine. Using the root mean square error (RMSE) and the mean absolute percentage error (MAPE) as prediction evaluation indicators, Table 1, Table 2, and Table 3 list the prediction evaluation index results of soil moisture demand, the dosage of a certain brand of compound fertilizer, and the dosage of a certain brand of medicine, respectively. The results show that the method proposed in this embodiment shows higher prediction accuracy.

[0122] The root mean square error (RMSE) and mean absolute percentage error (MAPE) are calculated as follows:

[0123]

[0124] Where m is the number of load sample points; is the predicted value; i is the actual value.

[0125] Table 1 Results of soil moisture demand prediction evaluation indicators

[0126]

[0127] Table 2 Results of the prediction and evaluation indexes of a brand of compound fertilizer dosage

[0128]

[0129]

[0130] Table 3 Results of the evaluation index for the dosage prediction of a certain brand of drug

[0131]

[0132] In summary, the method used in this embodiment can accurately regulate the amount of water, fertilizer and medicine used for marigold, improve the degree of automation in marigold planting, and improve the utilization rate of resources.

[0133] Example 2

[0134] This embodiment is a terminal device, which includes a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it is used to implement a marigold water, fertilizer, and pesticide zoning control method according to Embodiment 1.

[0135] Example 3

[0136] The present embodiment is a computer-readable storage medium. The computer-readable storage medium in the present embodiment stores a computer program. When the computer program is executed by a processor, it is used to implement a marigold water, fertilizer and pesticide zoning control method according to embodiment 1.

[0137] Among them, the computer-readable storage medium can be a computer storage medium or a communication medium. The communication medium includes any medium that facilitates the transmission of a computer program from one place to another. The computer storage medium can be any available medium that can be accessed by a general-purpose or special-purpose computer. For example, a computer-readable storage medium is coupled to a processor so that the processor can read information from the computer-readable storage medium and write information to the computer-readable storage medium. Of course, the computer-readable storage medium can also be a component of the processor. The processor and the computer-readable storage medium can be located in an application-specific integrated circuit (ASIC). In addition, the ASIC can be located in a user device. Of course, the processor and the computer-readable storage medium can also exist in a communication device as discrete components. Specifically, the computer-readable storage medium can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as a static random access memory (SRAM), an electrically erasable programmable read-only memory (EEPROM), an erasable programmable read-only memory (EPROM), a programmable read-only memory (PROM), a read-only memory (ROM), a magnetic memory, a flash memory, a magnetic disk or an optical disk, etc. The storage medium can be any available medium that can be accessed by a general-purpose or special-purpose computer.

Claims

1. A method for regulating water, fertilizer and pesticide by different zones for marigold, characterized in that: The process includes the following steps: S1. By collecting and clustering marigold growth environment and growth characteristic data, the marigold planting area is divided into several similar sub-areas; S2, select a time period, and for each divided sub-region, calculate the feature importance using three feature evaluation methods: gradient boosting decision tree, Pearson coefficient, and mutual information coefficient; S3, based on the weighting coefficient, combined with the importance of the three features calculated by the three feature evaluation methods, calculate the comprehensive importance weight within the time period; S4. Considering the exponential decay law of time series data, the comprehensive time-varying characteristics within the time period are calculated based on the comprehensive importance weight; S5, input the calculated comprehensive time-varying features into the XGBoost model and the CNN-LSTM model respectively to obtain the prediction results of the two models; S6. Weights are assigned to the prediction results of the two models to comprehensively obtain the optimal water, fertilizer and pesticide application amounts predicted for each sub-region; S7. Integrate the optimal water, fertilizer and medicine usage predicted in all sub-regions to obtain the predicted water, fertilizer and medicine usage control data for the marigold planting area within the selected time period; S8: Whether the data of the next time period is received; If yes, the weight coefficient is updated by particle swarm algorithm and steps S2 to S7 are repeated; If not, proceed to step S9; S9. End.

2. A method for controlling water, fertilizer and pesticide in different regions of marigold according to claim 1, characterized in that: Step S1 includes the following steps performed in sequence: S11, collecting sensor data from each marigold planting area to form a data matrix; S12. Standardize the data and give the same weight to different features; S13. Use K-means clustering algorithm to divide the marigold planting area.

3. A method for controlling water, fertilizer and pesticide in different regions of marigold according to claim 1, characterized in that: The formula for calculating the comprehensive importance weight within the time period in step S3 is: In the formula, is the comprehensive importance weight of feature i in time period T, α, β and γ are weighting coefficients, is the feature importance of feature i in time period T obtained by gradient boosting decision tree, ρ X,Y I(X;Y)(i,T) is the feature importance of feature i in time period T obtained by the Pearson coefficient, and I(X;Y)(i,T) is the feature importance of feature i in time period T obtained by the mutual information coefficient.

4. A method for controlling water, fertilizer and pesticide in different regions of marigold according to claim 3, characterized in that: The calculation formula of step S4 is: In the formula, I is the comprehensive time-varying feature, i k is the value of feature i at the kth time point, α k is the exponential decay time attenuation factor, and its calculation formula is: a k =exp(-λ(tk)) Where λ is the decay rate and t is the current time.

5. The method for controlling water, fertilizer and pesticide in different regions of marigold according to claim 1, characterized in that: The calculation formula of step S6 is: In the formula, is the predicted optimal water, fertilizer and medicine dosage, F XGB 、F CNN-LSTM are the water, fertilizer and medicine usage predicted by the XGBoost model and the CNN-LSTM model, respectively. w1 and w2 are the weight coefficients of the XGBoost and CNN-LSTM models, respectively. The calculation formula is: In the formula, loss T,CNN-LSTM 、loss T,XGB They are the mean square prediction errors of the CNN-LSTM model and the XGBoost model in time period T.

6. A method for controlling water, fertilizer and pesticide in different regions of marigold according to claim 5, characterized in that: The calculation formula for updating the weighted coefficient by the particle swarm algorithm in step S8 is: Where Y T , It represents the actual demand value and predicted value of water, fertilizer and medicine in time period T. Δα, Δβ and Δγ are the degrees of change of the three weighted coefficients after updating, and τ is the balance coefficient.

7. A terminal 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, the method for zoning control of water, fertilizer and pesticide for marigold is implemented as described in any one of claims 1 to 6.

8. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, it is used to implement the method for zoning control of water, fertilizer and pesticide for marigold as described in any one of claims 1 to 6.

Images