A Smart Agriculture Decision-Making Method and System Based on Multi-Source Data

Abstract

This invention relates to the field of data processing technology, and in particular to a smart agriculture decision-making method and system based on multi-source data. The method includes the following steps: a reference sample for the sample to be predicted corresponds to a set of soil nitrogen fertilizer application influencing factor data and corresponding nitrogen fertilizer application data; based on the variance difference of nitrogen fertilizer application data before and after removing the target reference sample from the reference sample set, and the difference between the target reference sample and the nitrogen fertilizer application data in the reference sample set, the extreme degree of nitrogen fertilizer application data is determined; based on the difference in soil nitrogen fertilizer application influencing factor data between the sample to be predicted and the target reference sample, the weight of the target reference sample is obtained; the initial predicted value of the sample to be predicted is determined; the decision tree in the GBDT algorithm is initialized to the initial predicted value and then iterated to obtain the predicted value of nitrogen fertilizer application data for the sample to be predicted, effectively improving the fertilization efficiency of smart agriculture.

Description

Technical Field

[0001] This invention relates to the field of data processing technology, and in particular to a smart agriculture decision-making method and system based on multi-source data. Background Technology

[0002] Nitrogen fertilizer is a core nutrient for crop growth. Insufficient application can lead to problems such as yellowing leaves, stunted growth, and low fruit set, resulting in reduced yield. Excessive application can cause excessive vegetative growth, lodging, and decreased fruit quality, while also increasing the risk of pests and diseases and even causing long-term environmental pollution. Accurately predicting nitrogen fertilizer application allows for control of nitrogen input at the source, ensuring that crops receive the appropriate amount of nitrogen at different growth stages. This guarantees yield while improving the quality of agricultural products, meeting market demand for high-quality produce.

[0003] Predicting crop nitrogen fertilizer application by processing multi-source data using intelligent algorithms can effectively improve fertilization efficiency. The relationship between nitrogen fertilizer application and its influencing factors is non-linear and interactive, involving the processing of multi-source data. The Gradient Boosting Decision Tree (GBDT) algorithm is an iterative decision tree algorithm that can adapt to complex multilinear agricultural data, making it suitable for predicting crop nitrogen fertilizer application.

[0004] However, in the initialization phase of predicting nitrogen fertilizer application for crops, the GBDT algorithm uses the average nitrogen fertilizer application of all reference samples as the initial prediction value for the weak learner. This initialization method leads to excessively large residuals in the initial prediction values, deviating too much from the true values. This necessitates more iterations to correct the residuals, thus reducing the prediction efficiency of nitrogen fertilizer application. Therefore, how to accurately set the initial prediction values ​​to improve prediction efficiency is a problem that needs to be solved. Summary of the Invention

[0005] To address the technical problem of how to accurately set initial forecast values ​​to improve forecast efficiency, this invention provides a smart agriculture decision-making method and system based on multi-source data.

[0006] In a first aspect, the present invention provides a smart agriculture decision-making method based on multi-source data, employing the following technical solution:

[0007] A smart agriculture decision-making method based on multi-source data includes the following steps:

[0008] Obtain a reference sample set for nitrogen fertilizer application of the sample to be predicted. Each reference sample in the reference sample set corresponds to a set of soil nitrogen fertilizer application influencing factor data and corresponding nitrogen fertilizer application data. Designate any reference sample as the target reference sample. Based on the variance difference of nitrogen fertilizer application data before and after removing the target reference sample from the reference sample set, and the difference between the maximum and minimum values ​​of nitrogen fertilizer application data between the target reference sample and the reference sample set, determine the extreme degree of nitrogen fertilizer application data of the target reference sample. Based on the difference in soil nitrogen fertilizer application influencing factor data between the sample to be predicted and the target reference sample, correct the extreme degree of nitrogen fertilizer application data of the target reference sample, obtaining the weight of the target reference sample in determining the nitrogen fertilizer application data of the sample to be predicted. The weight of the nitrogen fertilizer application data of the sample to be predicted is negatively correlated with the extreme degree of nitrogen fertilizer application data of the target reference sample. The initial predicted value of the sample to be predicted is obtained by multiplying the weight ratio of the nitrogen fertilizer application data of the sample to be predicted determined by each target reference sample with its nitrogen fertilizer application data. After initializing the decision tree in the GBDT algorithm to the initial predicted value, iterate to obtain the predicted value of nitrogen fertilizer application data of the sample to be predicted.

[0009] This invention uses the GBDT algorithm to predict the required nitrogen fertilizer application for crops, enabling precise fertilization based on the predicted values ​​and improving agricultural efficiency. During the prediction process, this invention addresses the potential for increased algorithm iteration overhead by directly using the average nitrogen fertilizer application data from all reference samples as the initial value. Therefore, this invention proposes a weighted method for the reference sample nitrogen fertilizer application data. By analyzing the differences between the reference sample nitrogen fertilizer application data and other reference samples, extreme reference samples can be accurately identified. Reducing the weight of such reference samples improves the algorithm's iteration efficiency. Furthermore, this invention analyzes the similarity of the environments in which the sample to be predicted and the reference sample data are located, adjusting the weights after correcting for the extreme nature of the reference samples. This effectively reduces prediction bias, making the initial predicted value of each sample more closely reflect its actual situation, resulting in faster convergence in subsequent iterations and thus significantly improving the efficiency of agricultural fertilization decisions.

[0010] According to the smart agriculture decision-making method based on multi-source data provided by the present invention, the step of obtaining a reference sample set for nitrogen fertilizer application of the sample to be predicted includes: collecting soil nitrogen content data, soil phosphorus content data, soil potassium content data, and precipitation data in the 30 days prior to collection in each region as data on soil nitrogen fertilizer application influencing factors; using the data on soil nitrogen fertilizer application influencing factors and the corresponding nitrogen fertilizer application data collected in different regions as samples to form a sample set; and randomly selecting a preset number of samples from the sample set as reference samples to obtain the reference sample set.

[0011] This invention uses soil data from different regions as samples when constructing the sample set, which allows the trained algorithm prediction model to better adapt to different regions and improve prediction accuracy.

[0012] According to the intelligent agricultural decision-making method based on multi-source data provided by the present invention, determining the extreme degree of nitrogen fertilizer application data of the target reference sample includes:

[0013] ;

[0014] The extreme level of nitrogen fertilizer application data for the j-th target reference sample. , These represent the variances of nitrogen fertilizer application data before and after removing the j-th target reference sample from the reference sample set. , Let be the absolute values ​​of the differences between the minimum and maximum values ​​of nitrogen fertilizer application data for the j-th target reference sample and the reference sample set, respectively. It is a linear normalized function. To find the minimum value function, This is a hyperparameter.

[0015] This invention takes into account that when the nitrogen fertilizer application data of the reference sample is too extreme, it is not advisable to set a large weight. Therefore, by analyzing the contribution of the reference sample data to the overall data variance and the degree to which it approaches the extreme values ​​of the reference sample set, the extreme prominence of the reference sample can be accurately assessed.

[0016] According to the smart agriculture decision-making method based on multi-source data provided by the present invention, the step of correcting the extreme degree of nitrogen fertilizer application data of the target reference sample based on the difference in soil nitrogen fertilizer application influencing factor data between the sample to be predicted and the target reference sample to obtain the weight of the target reference sample in determining the nitrogen fertilizer application data of the sample to be predicted includes: calculating the degree of dissimilarity between the sample to be predicted and the target reference sample based on the difference in soil nitrogen fertilizer application influencing factor data between the sample to be predicted and the target reference sample; correcting the extreme degree of nitrogen fertilizer application data of the target reference sample based on the degree of dissimilarity between the sample to be predicted and the target reference sample to obtain the weight of the target reference sample in determining the nitrogen fertilizer application data of the sample to be predicted.

[0017] According to the present invention, a smart agriculture decision-making method based on multi-source data is provided, wherein calculating the degree of dissimilarity between the sample to be predicted and the target reference sample includes:

[0018] ;

[0019] The degree of difference between the sample to be predicted and the j-th target reference sample. The number of dimensions in the data on factors affecting soil nitrogen fertilizer application. , The j-th target reference sample and the j-th target reference sample are respectively the sample to be predicted. Data on factors influencing soil nitrogen fertilizer application in multiple dimensions. It is a linear normalized function. It is the absolute value symbol.

[0020] This invention takes into account that reference soil data from similar regions can provide a better basis for the authenticity of the extreme degree of the reference soil data. Therefore, this invention corrects the extreme degree of the sample to be predicted by further analyzing the similarity between the sample to be predicted and the reference sample in the same region, so as to accurately obtain its weight.

[0021] According to the present invention, a smart agriculture decision-making method based on multi-source data is provided, wherein the method for obtaining the weight of the nitrogen fertilizer application data of the sample to be predicted by determining the target reference sample includes:

[0022] ;

[0023] For the j-th target reference sample, determine the weights of the nitrogen fertilizer application data of the sample to be predicted. The extreme level of nitrogen fertilizer application data for the j-th target reference sample. The degree of difference between the sample to be predicted and the j-th target reference sample. It is an exponential function with base e.

[0024] According to the intelligent agriculture decision-making method based on multi-source data provided by the present invention, the step of determining the weight ratio of the nitrogen fertilizer application data of the sample to be predicted and the product of the nitrogen fertilizer application data with the nitrogen fertilizer application data of each target reference sample to obtain the initial predicted value of the sample to be predicted includes:

[0025] ;

[0026] The initial predicted values ​​for the sample to be predicted. The number of reference samples in the reference sample set. For the j-th target reference sample, determine the weights of the nitrogen fertilizer application data of the sample to be predicted. The nitrogen fertilizer application data is for the j-th target reference sample.

[0027] According to the present invention, a smart agriculture decision-making method based on multi-source data includes the following steps: initializing the decision tree in the GBDT algorithm to the initial predicted value and then iterating to obtain the predicted value of nitrogen fertilizer application data for the sample to be predicted. The steps include: initializing the decision tree in the GBDT algorithm to the initial predicted value; obtaining the prediction error of each reference sample based on the difference between the initial predicted value and the nitrogen fertilizer application data of the reference samples; training a new decision tree to fit the prediction error based on the data of soil nitrogen fertilizer application influencing factors, and obtaining the predicted value of the new decision tree; continuing to obtain the prediction error; repeating this process until a trained prediction model is obtained; and inputting the sample to be predicted into the trained prediction model to obtain the predicted value of nitrogen fertilizer application data.

[0028] In initializing the initial prediction value, this invention quantifies the reference sample data to different degrees by weighting the reference sample nitrogen fertilizer application data, so that the obtained initial prediction value of nitrogen fertilizer application data is closer to the actual situation, reducing prediction deviation and improving the algorithm prediction iteration efficiency.

[0029] According to the present invention, a smart agriculture decision-making method based on multi-source data is provided, wherein inputting the sample to be predicted into the trained prediction model to obtain the predicted value of nitrogen fertilizer application data includes: adding the initial predicted value of the sample to be predicted with the prediction results of all decision trees to obtain the final predicted value of nitrogen fertilizer application.

[0030] Secondly, this invention provides a smart agricultural decision-making system based on multi-source data, employing the following technical solution:

[0031] A smart agriculture decision-making system based on multi-source data includes a processor and a memory, wherein the memory stores computer program instructions, and when the computer program instructions are executed by the processor, the aforementioned smart agriculture decision-making method based on multi-source data is implemented.

[0032] By adopting the above technical solution, a smart agriculture decision-making method based on multi-source data is generated into a computer program and stored in a memory for loading and execution by a processor. This allows for the creation of terminal devices based on the memory and processor, facilitating their use.

[0033] The present invention has the following technical effects:

[0034] Based on the above technical solutions, this invention provides a smart agriculture decision-making method and system based on multi-source data. It uses the GBDT algorithm to predict the required nitrogen fertilizer application for crops, enabling precise fertilization based on the predicted values ​​and improving agricultural efficiency. During the prediction process, this invention considers that directly using the average nitrogen fertilizer application data of all reference samples as the initial value may increase algorithm iteration losses. Therefore, this invention proposes a weighting method for the nitrogen fertilizer application data of the reference samples. By analyzing the differences between the nitrogen fertilizer application data of the reference samples and other reference samples, extreme reference samples can be accurately identified. By reducing the weight of such reference samples, the algorithm iteration efficiency is improved. Furthermore, this invention also analyzes the similarity of the environments in which the sample to be predicted and the reference sample data are located, and corrects the extreme nature of the reference samples to obtain weights. This effectively reduces prediction bias, making the initial predicted value of each sample more closely match its actual situation, resulting in faster convergence in subsequent iterations, thereby effectively improving the efficiency of agricultural fertilization decisions. Attached Figure Description

[0035] Figure 1 This is a flowchart illustrating a smart agriculture decision-making method based on multi-source data, provided in an embodiment of the present invention. Detailed Implementation

[0036] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are some embodiments of the present invention, but not all embodiments.

[0037] This invention discloses a smart agriculture decision-making method based on multi-source data. For details, please refer to [link / reference]. Figure 1 As shown, Figure 1 This is a flowchart illustrating a smart agriculture decision-making method based on multi-source data provided in an embodiment of the present invention. The method specifically includes the following steps:

[0038] S1: Obtain a reference sample set of nitrogen fertilizer application for the sample to be predicted. Each reference sample corresponds to a set of soil nitrogen fertilizer application influencing factors and corresponding nitrogen fertilizer application data.

[0039] It should be noted that soil nitrogen, phosphorus, and potassium content data, as well as precipitation data within the 30 days prior to data collection, all affect nitrogen fertilizer application data. Lower soil nitrogen content indicates sufficient nitrogen content in the soil, thus requiring less nitrogen fertilizer. There is an indirect synergistic relationship between soil phosphorus and potassium content and nitrogen fertilizer application. When phosphorus and potassium are sufficient, crop roots are well-developed and metabolism is vigorous, resulting in higher nitrogen absorption and utilization efficiency, thus reducing nitrogen fertilizer waste and appropriately lowering the amount used. Excessive precipitation may cause the original nitrogen in the soil to be lost with water, requiring an appropriate increase in nitrogen fertilizer application; conversely, insufficient precipitation leads to soil drought, reduced crop absorption capacity, and excessive nitrogen fertilizer application can easily lead to accumulation, requiring a reduction in application or application in conjunction with irrigation.

[0040] Therefore, in this embodiment of the invention, when predicting the nitrogen fertilizer application data of a new area, a prediction model of the GBDT algorithm can be constructed by obtaining data on soil nitrogen fertilizer application influencing factors and corresponding nitrogen fertilizer application data, so as to realize the prediction of nitrogen fertilizer application data for the new area.

[0041] It should be further explained that there are common patterns in the relationship between soil properties and crop requirements across regions. When constructing a predictive model, analyzing soil data from different regions can reduce regional interference, extract the core relationship between soil physicochemical properties and crop nitrogen fertilizer application and nutrient requirements, avoid overfitting regional characteristics due to analysis in a single region, and effectively improve the universality of the model.

[0042] Based on this, in constructing the prediction model of the GBDT algorithm, the present invention can train the model by acquiring data on soil nitrogen fertilizer application influencing factors and corresponding nitrogen fertilizer application data in different regions. This allows the model to learn richer interaction patterns between soil and crops, ultimately enabling the model to predict reasonable nitrogen fertilizer application when facing unfamiliar regions.

[0043] For example, in an embodiment of the present invention, obtaining a reference sample set for nitrogen fertilizer application of the sample to be predicted includes: collecting soil nitrogen content data, soil phosphorus content data, soil potassium content data, and precipitation data within 30 days prior to collection in each region as data on factors affecting soil nitrogen fertilizer application; using each set of data on factors affecting soil nitrogen fertilizer application and the corresponding nitrogen fertilizer application data collected in different regions as samples to form a sample set; and randomly selecting a preset number of samples from the sample set as reference samples to obtain the reference sample set.

[0044] The number of reference samples can be set to 60, but the specific number can be adjusted according to actual needs.

[0045] It is understandable that the dimensions of the data on factors affecting soil nitrogen fertilizer application and the data on nitrogen fertilizer application may be different. Therefore, after collecting the corresponding data, preprocessing operations such as data cleaning, missing data interpolation, and dimension normalization can be performed on the data. The specific preprocessing method can be set according to actual needs, and this embodiment of the invention does not impose too many restrictions here.

[0046] The above steps can be used to obtain the soil nitrogen fertilizer application influencing factors and corresponding nitrogen fertilizer application data for each reference sample. However, the weights of the data for each reference sample to the model are different. If the data of the reference sample is too extreme, it means that its weight needs to be reduced during the initialization of the training model, so as to speed up the model training iteration process and improve the efficiency of soil fertilizer application decision-making. That is, the following steps should be performed.

[0047] S2: Designate any reference sample as the target reference sample. Based on the variance difference of nitrogen fertilizer application data before and after removing the target reference sample from the reference sample set, and the difference between the maximum and minimum values ​​of nitrogen fertilizer application data between the target reference sample and the reference sample set, determine the extreme degree of nitrogen fertilizer application data of the target reference sample.

[0048] It should be noted that if the nitrogen fertilizer application data of a reference sample is more unusual among all the nitrogen fertilizer application data of all reference samples, it indicates that its extreme degree is greater. In this case, the reference value of the nitrogen fertilizer application data of the reference sample is less meaningful when predicting the nitrogen fertilizer application data of the sample to be predicted.

[0049] Therefore, in the process of initializing the initial predicted value, the embodiments of the present invention can improve the efficiency of initialization by analyzing the extreme degree of each reference sample and weighting the nitrogen fertilizer application data of that reference sample.

[0050] For example, in an embodiment of the present invention, determining the extreme degree of nitrogen fertilizer application data of a target reference sample includes:

[0051] ;

[0052] The extreme level of nitrogen fertilizer application data for the j-th target reference sample. The variance of nitrogen fertilizer application data for the reference sample set, The variance of nitrogen fertilizer application data after removing the j-th target reference sample from the reference sample set. Let be the absolute value of the difference between the j-th target reference sample and the minimum value of nitrogen fertilizer application data in the reference sample set. Let be the absolute value of the difference between the j-th target reference sample and the maximum value of nitrogen fertilizer application data in the reference sample set. It is a linear normalized function. To find the minimum value function, This is a hyperparameter.

[0053] The hyperparameter can be set to 0.01 to avoid the case where the denominator is 0.

[0054] In this calculation method Indicates selection and The smaller the value, the closer the nitrogen fertilizer application data of the j-th target reference sample is to the minimum or maximum value of the nitrogen fertilizer application data of all reference samples in the reference sample set. The more special the performance of the nitrogen fertilizer application data of the j-th target reference sample, the greater its extreme degree. Therefore, the lower the reference significance of the nitrogen fertilizer application data of this reference sample when predicting the nitrogen fertilizer application data of the sample to be predicted.

[0055] This represents the variance difference in nitrogen fertilizer application data before and after removing the j-th target reference sample. The larger this value, the greater the contribution of the nitrogen fertilizer application data of the j-th target reference sample to the overall variance of the reference sample set. The more unique the value of the nitrogen fertilizer application data of the j-th target reference sample among all the nitrogen fertilizer application data of the reference samples, the greater the reliability. The greater the extreme degree of the data, the lower the reference value.

[0056] In summary, if the nitrogen fertilizer application data of the j-th target reference sample is closer to the minimum or maximum value of the nitrogen fertilizer application data of all reference samples in the reference sample set, and the variance difference of the nitrogen fertilizer application data before and after removing the j-th target reference sample is greater, it indicates that the extreme degree of the target reference sample is greater and the reference value is lower. The extreme degree of nitrogen fertilizer application data of each target reference sample can be obtained according to the above steps.

[0057] S3: Based on the differences in soil nitrogen fertilizer application influencing factors between the sample to be predicted and the target reference sample, the extreme degree of nitrogen fertilizer application data of the target reference sample is corrected to obtain the weight of nitrogen fertilizer application data of the sample to be predicted in the target reference sample. The weight is negatively correlated with the extreme degree of nitrogen fertilizer application data of the target reference sample.

[0058] It should be noted that by analyzing the differences between each target reference sample and other target reference samples according to the above steps, reference samples with large differences from other target reference samples can be obtained. However, it is impossible to determine the fit between the reference sample and the sample to be predicted. If the value of the soil nitrogen fertilizer application influencing factor data of the sample to be predicted is close to the value of the soil nitrogen fertilizer application influencing factor data of a reference sample, it means that the reference sample has greater reference significance for the sample to be predicted. Therefore, even if the value of the nitrogen fertilizer application data of this reference sample is relatively extreme, the nitrogen fertilizer application data of this reference sample needs to be given more weight when calculating the initial predicted value of the nitrogen fertilizer application data of the sample to be predicted.

[0059] Based on this, embodiments of the present invention can determine the degree of regional disparity between the sample to be predicted and the target reference sample by analyzing the numerical differences between the data on all soil nitrogen fertilizer application influencing factors of each reference sample and the data on all soil nitrogen fertilizer application influencing factors of the sample to be predicted. Combined with the extreme degree of nitrogen fertilizer application data of each reference sample, the weight of nitrogen fertilizer application data of each reference sample in the initial predicted value of nitrogen fertilizer application data of the sample to be predicted by the GBDT algorithm is calculated.

[0060] Understandably, the smaller the difference in values ​​between the data on all soil nitrogen fertilizer application influencing factors of a reference sample and the data on all soil nitrogen fertilizer application influencing factors of the sample to be predicted, the more similar the environment of the reference sample and the sample to be predicted is. The greater the reference significance of the nitrogen fertilizer application data of the reference sample when predicting the nitrogen fertilizer application data of the sample to be predicted, the greater the weight of the reference sample in the GBDT algorithm when predicting the initial predicted value of the nitrogen fertilizer application data of the sample to be predicted.

[0061] For example, in an embodiment of the present invention, based on the differences in soil nitrogen fertilizer application influencing factor data between the sample to be predicted and the target reference sample, the extreme degree of nitrogen fertilizer application data of the target reference sample is corrected to obtain the weight of the target reference sample in determining the nitrogen fertilizer application data of the sample to be predicted. This includes: calculating the degree of regional disparity between the sample to be predicted and the target reference sample based on the differences in soil nitrogen fertilizer application influencing factor data between the sample to be predicted and the target reference sample; correcting the extreme degree of nitrogen fertilizer application data of the target reference sample based on the degree of regional disparity between the sample to be predicted and the target reference sample to obtain the weight of the target reference sample in determining the nitrogen fertilizer application data of the sample to be predicted.

[0062] For example, in an embodiment of the present invention, calculating the degree of difference between the sample to be predicted and the target reference sample includes:

[0063] ;

[0064] The degree of difference between the sample to be predicted and the j-th target reference sample. The number of dimensions in the data on factors affecting soil nitrogen fertilizer application. For the sample to be predicted Data on factors influencing soil nitrogen fertilizer application in multiple dimensions. For the j-th target reference sample Data on factors influencing soil nitrogen fertilizer application in multiple dimensions. It is a linear normalized function. It is the absolute value symbol.

[0065] In this calculation method The larger the value, the less similar the collection environment is between the j-th target reference sample and the sample to be predicted, and the less significant the nitrogen fertilizer application data of the j-th reference sample is in predicting the nitrogen fertilizer application data of the sample to be predicted.

[0066] After obtaining the degree of disparity between the sample to be predicted and the j-th target reference sample according to the above steps, the extreme degree of nitrogen fertilizer application data of the target reference sample can be corrected according to the degree of disparity, and the weight of the target reference sample can be accurately obtained.

[0067] For example, in an embodiment of the present invention, determining the weights of the nitrogen fertilizer application data of the sample to be predicted by determining the target reference sample includes:

[0068] ;

[0069] For the j-th target reference sample, determine the weights of the nitrogen fertilizer application data of the sample to be predicted. The extreme level of nitrogen fertilizer application data for the j-th target reference sample. The degree of difference between the sample to be predicted and the j-th target reference sample. It is an exponential function with base e.

[0070] In this calculation method The larger the value, the more unique the nitrogen fertilizer application data of the j-th target reference sample is among all the nitrogen fertilizer application data of all target reference samples. In this case, if the collection environment of the sample to be predicted is less similar to that of the j-th target reference sample, the weight of the nitrogen fertilizer application data of the j-th target reference sample in predicting the nitrogen fertilizer application data of the sample to be predicted needs to be smaller.

[0071] Based on the above steps, the weight of the nitrogen fertilizer application data of the sample to be predicted can be determined for each target reference sample. Using the weight of each target reference sample in the process of initializing the decision tree can effectively improve the iteration efficiency, thereby improving the soil fertilization efficiency.

[0072] S4: Determine the weight ratio of the nitrogen fertilizer application data of the sample to be predicted based on each target reference sample and multiply it with the nitrogen fertilizer application data to obtain the initial predicted value of the sample to be predicted; initialize the decision tree in the GBDT algorithm to the initial predicted value and then iterate to obtain the predicted value of the nitrogen fertilizer application data of the sample to be predicted.

[0073] Understandably, the smaller the weight of a reference sample's nitrogen fertilizer application data in the GBDT algorithm's prediction of the nitrogen fertilizer application data of the sample to be predicted, the more extreme the nitrogen fertilizer application data of this reference sample is, the greater the difference between its soil nitrogen fertilizer application influencing factor data and the soil nitrogen fertilizer application influencing factor data of the sample to be predicted, and the lower its reference significance.

[0074] For example, in an embodiment of the present invention, the initial predicted value of the sample to be predicted is obtained by multiplying the weight ratio of the nitrogen fertilizer application data of the sample to be predicted with the nitrogen fertilizer application data of each target reference sample, including:

[0075] ;

[0076] The initial predicted values ​​for the sample to be predicted. The number of reference samples in the reference sample set. For the j-th target reference sample, determine the weights of the nitrogen fertilizer application data of the sample to be predicted. The nitrogen fertilizer application data is for the j-th target reference sample.

[0077] For example, in this embodiment of the invention, after initializing the decision tree in the GBDT algorithm to the initial predicted value, iteratively obtaining the predicted value of nitrogen fertilizer application data for the sample to be predicted includes: initializing the decision tree in the GBDT algorithm to the initial predicted value; obtaining the prediction error of each reference sample based on the difference between the initial predicted value and the nitrogen fertilizer application data of the reference sample; training a new decision tree to fit the prediction error based on the data of soil nitrogen fertilizer application influencing factors, and obtaining the prediction value of the new decision tree; continuing to obtain the prediction error, and repeating this process until a trained prediction model is obtained; and inputting the sample to be predicted into the trained prediction model to obtain the predicted value of nitrogen fertilizer application data.

[0078] For example, in an embodiment of the present invention, inputting the sample to be predicted into the trained prediction model to obtain the predicted value of nitrogen fertilizer application data includes: adding the initial predicted value of the sample to be predicted to the prediction results of all decision trees to obtain the final predicted value of nitrogen fertilizer application.

[0079] The above steps can be used to obtain the predicted nitrogen fertilizer application value of the sample to be predicted. Fertilization can be carried out in the collection area of ​​the sample to be predicted based on the predicted nitrogen fertilizer application value, thereby realizing precision agriculture and improving the efficiency of agricultural operations.

[0080] When fertilizing a region based on the predicted nitrogen fertilizer application value of the sample to be predicted, fertilizer can be applied to the soil using drones, fertilizer applicators, etc. The specific method can be set according to actual needs.

[0081] For example, after fertilization, the effect of fertilization can be verified by soil nitrogen testing and crop growth observation. If the nitrogen content deviates significantly from the prediction, the prediction value can be updated by combining new sample data, and the subsequent fertilization amount can be dynamically corrected. The specific settings can be made according to actual needs.

[0082] As can be seen, in the embodiments of the present invention, when realizing intelligent agricultural fertilization decisions based on multi-source data, it is possible to...

[0083] A reference sample set for nitrogen fertilizer application of the sample to be predicted is obtained. Each reference sample in the reference sample set corresponds to a set of soil nitrogen fertilizer application influencing factor data and corresponding nitrogen fertilizer application data. Any reference sample is designated as the target reference sample. Based on the variance difference of nitrogen fertilizer application data before and after removing the target reference sample from the reference sample set, and the difference between the maximum and minimum values ​​of nitrogen fertilizer application data between the target reference sample and the reference sample set, the extreme degree of nitrogen fertilizer application data of the target reference sample is determined. Based on the difference in soil nitrogen fertilizer application influencing factor data between the sample to be predicted and the target reference sample, the extreme degree of nitrogen fertilizer application data of the target reference sample is corrected, and the weight of the nitrogen fertilizer application data of the sample to be predicted is obtained from the target reference sample. The weight of the nitrogen fertilizer application data of the sample to be predicted is negatively correlated with the extreme degree of nitrogen fertilizer application data of the target reference sample. The initial predicted value of the sample to be predicted is obtained by multiplying the weight ratio of the nitrogen fertilizer application data of the sample to be predicted determined by each target reference sample with its nitrogen fertilizer application data. The decision tree in the GBDT algorithm is initialized with the initial predicted value and then iterated to obtain the predicted value of nitrogen fertilizer application data of the sample to be predicted, which effectively improves the fertilization efficiency of smart agriculture.

[0084] This invention also discloses a smart agriculture decision-making system based on multi-source data, including a processor and a memory. The memory stores computer program instructions, which, when executed by the processor, implement a smart agriculture decision-making method based on multi-source data provided by this invention.

[0085] The system also includes other components well known to those skilled in the art, such as communication buses and communication interfaces, the settings and functions of which are known in the art and will not be described in detail here.

[0086] In this invention, the aforementioned memory can be any tangible medium that contains or stores a program that can be used by or in conjunction with an instruction execution system, apparatus, or device.

[0087] The above are all preferred embodiments of the present invention and are not intended to limit the scope of protection of the present invention. Therefore, all equivalent changes made in accordance with the structure, shape and principle of the present invention should be covered within the scope of protection of the present invention.

Claims

1. A method for intelligent agricultural decision-making based on multi-source data, characterized in that, include: Obtain a reference sample set for nitrogen fertilizer application of the sample to be predicted. Each reference sample in the reference sample set corresponds to a set of soil nitrogen fertilizer application influencing factors and corresponding nitrogen fertilizer application data. Designate any reference sample as the target reference sample. Based on the variance difference of nitrogen fertilizer application data before and after removing the target reference sample from the reference sample set, and the difference between the maximum and minimum values ​​of nitrogen fertilizer application data between the target reference sample and the reference sample set, determine the extreme degree of nitrogen fertilizer application data of the target reference sample, including: ； an extreme degree of nitrogen fertilizer application data of the jth target reference sample, 、 respectively, the variance of the nitrogen fertilizer application data before and after the jth target reference sample is removed from the reference sample set, 、 respectively, the absolute value of the difference between the jth target reference sample and the minimum value and the maximum value of the nitrogen fertilizer application data of the reference sample set, is a linear normalization function, is a minimum value taking function, is a hyperparameter; Based on the differences in soil nitrogen fertilizer application influencing factors between the sample to be predicted and the target reference sample, the extreme degree of nitrogen fertilizer application data in the target reference sample is corrected to obtain the weights of the target reference sample in determining the nitrogen fertilizer application data of the sample to be predicted, including: Based on the differences in soil nitrogen fertilizer application influencing factors between the sample to be predicted and the target reference sample, the degree of regional disparity between the sample to be predicted and the target reference sample is calculated, including: ； is the degree of different region of the to-be-predicted sample and the jth target reference sample, is the number of dimensions of the soil nitrogen fertilizer use amount influencing factor data, , is the soil nitrogen fertilizer use amount influencing factor data of the jth dimension of the to-be-predicted sample and the jth target reference sample, is the soil nitrogen fertilizer use amount influencing factor data of the jth dimension of the to-be-predicted sample and the jth target reference sample, is a linear normalization function, is an absolute value symbol; The extreme degree of nitrogen fertilizer application data in the target reference sample is adjusted based on the degree of dissimilarity between the sample to be predicted and the target reference sample, resulting in the weights of the target reference sample in determining the nitrogen fertilizer application data of the sample to be predicted, including: ； determining a weight of the to-be-predicted sample nitrogen application rate data for the jth target reference sample, an extreme degree of the nitrogen application rate data for the jth target reference sample, is an exponential function with base e; The weight ratio of nitrogen fertilizer application data of the sample to be predicted is determined by the product of the nitrogen fertilizer application data of the sample to be predicted and the target reference sample. The initial predicted value of the sample to be predicted is obtained by initializing the decision tree in the GBDT algorithm to the initial predicted value and then iterating to obtain the predicted value of nitrogen fertilizer application data of the sample to be predicted. 2.The method of claim 1, wherein, The process of obtaining a reference sample set for nitrogen fertilizer application of the sample to be predicted includes: Soil nitrogen content, soil phosphorus content, soil potassium content, and precipitation data within the 30 days prior to collection were collected in each region to serve as data on factors influencing soil nitrogen fertilizer application. Each set of data on factors influencing soil nitrogen fertilizer application and the corresponding nitrogen fertilizer application data collected from different regions were used as samples to form a sample set. A predetermined number of samples were randomly selected from the sample set as reference samples to obtain a reference sample set. 3.The wisdom agriculture decision-making method based on multi-source data according to claim 1, characterized in that, The step of determining the weighting percentage of the nitrogen fertilizer application data of the sample to be predicted based on each target reference sample and multiplying it with the nitrogen fertilizer application data to obtain the initial predicted value of the sample to be predicted includes: ； an initial prediction value of the to-be-predicted sample, a number of reference samples in the reference sample set, determining a weight of the nitrogen fertilizer use data of the to-be-predicted sample for the jth target reference sample, the nitrogen fertilizer use data corresponding to the jth target reference sample. 4.The method of claim 1, wherein, The process of initializing the decision tree in the GBDT algorithm with initial predicted values ​​and then iterating to obtain predicted values ​​for nitrogen fertilizer application data of the sample to be predicted includes: The decision tree in the GBDT algorithm is initialized with the initial prediction value. The prediction error of each reference sample is obtained based on the difference between the initial prediction value and the nitrogen fertilizer application data of the reference sample. A new decision tree is trained to fit the prediction error based on the data of soil nitrogen fertilizer application influencing factors, and the prediction value of the new decision tree is obtained. The prediction error is obtained again and this process is repeated until the training prediction model is obtained. The sample to be predicted is input into the training prediction model to obtain the predicted value of nitrogen fertilizer application data. 5.The method of claim 4, wherein, The step of inputting the sample to be predicted into the trained prediction model to obtain the predicted value of nitrogen fertilizer application data includes: The initial predicted value of the sample to be predicted is added to the prediction results of all decision trees to obtain the final predicted value of nitrogen fertilizer application.

6. A smart agriculture decision system based on multi-source data, characterized in that, include: A processor and a memory, the memory storing computer program instructions which, when executed by the processor, implement a method for intelligent agriculture decision making based on multi-source data according to any one of claims 1-5.

Images