Quinoa production increasing method and system based on artificial intelligence

Abstract

The present invention relates to the technical field of quinoa yield prediction, and in particular to a quinoa yield increasing method and system based on artificial intelligence, including: selecting quinoa planting area boundaries and establishing a spatiotemporal database; defining specific index data as characteristic variables based on the spatiotemporal database, defining quinoa variety unit area yield as target variable, and establishing quinoa variety adaptability rules; extracting matching data sets in the spatiotemporal database based on quinoa variety adaptability rules, and inputting the matching data sets into a preset quinoa yield hybrid prediction model to complete training, wherein the quinoa yield hybrid prediction model is specifically under the condition of optimal comprehensive adaptability of quinoa variety, with quinoa yield maximization as optimization target; based on the quinoa yield hybrid prediction model that has completed training, the real-time index data of quinoa planting areas are calculated to obtain the final yield increasing scheme of quinoa. The present invention can realize the precise management of quinoa planting, and can effectively guide farmers to increase production and increase efficiency.

Description

Technical Field

[0001] The present invention relates to the technical field of quinoa yield increasing scheme design based on artificial intelligence, and in particular, to a quinoa yield increasing method and system based on artificial intelligence. Background Art

[0002] As a food crop with high nutritional value, quinoa has the characteristics of good stress resistance and high nutritional value, and occupies an important position in global food security and sustainable agricultural development. At present, traditional quinoa planting decisions have many limitations. On the one hand, yield prediction models mostly rely on a single data modality, and fail to effectively integrate temporal meteorological changes and plot-specific characteristics, resulting in prediction results that deviate from actual production scenarios. On the other hand, variety selection is often based on empirical rules or simple statistical associations, lacking a systematic quantitative evaluation of the "soil-climate-variety" adaptation relationship, which can easily lead to variety recommendation deviations. In addition, existing optimization methods often isolate decision variables such as fertilization and irrigation, not only ignoring the nonlinear coupling effects between multiple factors, but also making it difficult to achieve cross-plot collaborative optimization under global resource constraints, resulting in insufficient feasibility of the solution. Summary of the invention

[0003] The purpose of the present invention is to provide a quinoa yield increasing method and system based on artificial intelligence, which improves the yield prediction accuracy through a hybrid prediction model, combines variety adaptability rules and resource constraint optimization decision-making plans, realizes precise management of quinoa planting, and can effectively guide farmers to increase production and efficiency.

[0004] The present invention is achieved through the following technical solutions:

[0005] A method for increasing quinoa production based on artificial intelligence, the method comprising the following steps:

[0006] Select the boundaries of quinoa planting areas, obtain real-time indicator data and historical monitoring data of quinoa planting areas, and combine and establish a spatiotemporal database;

[0007] Extract the historical yield data of quinoa varieties from the spatiotemporal database, and associate it with the specific indicator data of the corresponding plots. At the same time, define the specific indicator data as the characteristic variable, define the unit area yield of quinoa varieties as the target variable, and establish the adaptability rules of quinoa varieties through the CART regression tree;

[0008] Based on the adaptability rules of quinoa varieties, a matching data set in the spatiotemporal database is extracted, and the matching data set is divided into a training set and a test set. The training set is input into a preset quinoa yield hybrid prediction model for calculation, and the training results are verified by the test set to complete the training of the quinoa yield hybrid prediction model, wherein the quinoa yield hybrid prediction model specifically takes maximizing quinoa yield as the optimization goal under the condition of optimal comprehensive adaptability of quinoa varieties;

[0009] Based on the trained quinoa yield hybrid prediction model, the real-time indicator data of the quinoa planting area is calculated to obtain the final quinoa production increase plan.

[0010] Optionally, the spatiotemporal database is established by:

[0011] Select the boundaries of quinoa growing areas and deploy a variety of IoT detection devices in the quinoa growing areas to establish a quinoa growing area monitoring network;

[0012] Obtain real-time indicator data of quinoa planting areas based on the quinoa planting area monitoring network;

[0013] Collect historical monitoring data on quinoa growing areas;

[0014] Align the timestamps and geographic coordinates of real-time indicator data with historical monitoring data, and build a spatiotemporal database through PostgreSQL.

[0015] Optionally, before establishing the quinoa variety adaptability rules by using the CART regression tree, the method further includes:

[0016] Extract the historical yield data of quinoa varieties in the spatiotemporal database, associate it with the specific indicator data of the corresponding plots, calculate the yield variation coefficient of each quinoa variety under the same soil type, and screen out quinoa varieties whose yield variation coefficient is greater than the first threshold;

[0017] The yield variation coefficient is specifically:

[0018] Extract historical quinoa variety production data according to quinoa variety and specific indicator environment classification;

[0019] Calculate the mean yield and standard deviation of each quinoa variety;

[0020] The yield variation coefficient of quinoa varieties was solved based on the yield mean and standard deviation of each quinoa variety;

[0021] Based on the preset first threshold as the basis for judging the coefficient of variation of quinoa variety yield, quinoa varieties greater than or equal to the first threshold are screened out.

[0022] Optionally, the quinoa variety adaptability rules are established by using a CART regression tree, which are specifically:

[0023] Extract data from the spatiotemporal database to obtain soil and meteorological index data of quinoa planting areas, and associate historical yield data of various quinoa varieties;

[0024] Based on the extracted data in the spatiotemporal database, the soil index data and meteorological index data of the quinoa planting area are defined as feature variables, and the unit area yield of quinoa varieties is defined as the target variable, and the two are combined to form a training sample set;

[0025] Construct a CART regression tree model. In the training sample set, use the root node to represent all training samples, search for the best split point in each feature dimension, and use the best split point to divide the root node for the first time to obtain two child nodes.

[0026] For each child node, repeat the root node splitting process until the number of samples in the node is lower than the preset lower limit, and prune the CART regression tree model through the cost complexity pruning algorithm to complete the training of the CART regression tree model;

[0027] According to the trained CART regression tree model, the characteristic variable segmentation conditions and the corresponding predicted yields in each terminal node are extracted and combined to establish the adaptability rules of quinoa varieties.

[0028] Optionally, the process of determining the optimal segmentation point is:

[0029] In the training sample set, all training samples are represented by the root node, and the best segmentation point is searched in each feature dimension;

[0030] For each candidate feature and its candidate segmentation point, the mean square error of the two child nodes after segmentation is calculated and weighted by the proportion of the number of samples to obtain the weighted error;

[0031] Among all candidate segmentations, the segmentation point with the smallest weighted error is selected as the best segmentation point.

[0032] Optionally, the quinoa yield hybrid prediction model includes: a time series branch and a static branch; wherein the time series branch is specifically an LSTM neural network model, and the time series characteristics of the meteorological indicator data are learned through the LSTM neural network model;

[0033] The static branch is specifically a multi-layer perceptron model, and the soil index data is modeled by the multi-layer perceptron model;

[0034] The two branch outputs are passed through the feature fusion module to obtain the unit area yield prediction value of the quinoa planting area, the loss value between the unit area yield prediction value and the actual result is calculated, and the parameters of the quinoa yield hybrid prediction model are iteratively updated through the loss value to complete the training of the quinoa yield hybrid prediction model.

[0035] Optionally, the quinoa yield hybrid prediction model is specifically designed to maximize quinoa yield under the condition that the comprehensive adaptability of quinoa varieties is optimal, and the specific calculation process is as follows:

[0036] Set the number of plots K in the quinoa planting area and determine the decision combination for each plot;

[0037] By completing the training of the quinoa yield hybrid prediction model, based on the determined soil index data and quinoa varieties, the unit area yield prediction value and objective function value of each plot are calculated;

[0038] Under the set constraints, the decision combination of each plot is iteratively updated;

[0039] The predicted yield per unit area of ​​each plot is repeatedly calculated through the trained quinoa yield hybrid prediction model, and the objective function is updated until the maximum number of iterations is reached, and the decision combination result that maximizes the objective function value is output.

[0040] Optionally, the objective function of the quinoa yield hybrid prediction model is calculated as follows:

[0041]

[0042] Among them, F is the objective function, K is the total number of plots, k is the plot index, is the predicted yield per unit area, is the nitrogen application rate of the kth plot, is the irrigation water volume of the kth plot, is the quinoa variety of the kth plot, is the penalty coefficient, is the acceleration factor, For reference fertilizer amount.

[0043] Optionally, the set constraints include: an upper limit constraint on the amount of fertilizer applied to a single plot of land, an upper limit constraint on the amount of irrigation applied to a single plot of land, a constraint on the total amount of fertilizer resources, a constraint on the total amount of water resources, and a constraint on variety adaptability selection.

[0044] The artificial intelligence-based quinoa yield-increasing system includes:

[0045] A database establishment unit selects the boundaries of quinoa planting areas, obtains real-time indicator data and historical monitoring data of quinoa planting areas, and combines and establishes a spatiotemporal database;

[0046] The unit for establishing the adaptability rules of quinoa varieties extracts the historical yield data of quinoa varieties from the spatiotemporal database and associates the specific indicator data of the corresponding plots. At the same time, the specific indicator data is defined as the characteristic variable, the yield per unit area of ​​quinoa varieties is defined as the target variable, and the adaptability rules of quinoa varieties are established through the CART regression tree;

[0047] A model calculation unit extracts a matching data set in a spatiotemporal database based on the adaptability rules of quinoa varieties, divides the matching data set into a training set and a test set, inputs the training set into a preset quinoa yield hybrid prediction model for calculation, and verifies the training results through the test set to complete the training of the quinoa yield hybrid prediction model, wherein the quinoa yield hybrid prediction model specifically takes maximizing quinoa yield as an optimization goal under the condition that the comprehensive adaptability of quinoa varieties is optimal;

[0048] The solution output unit calculates the real-time indicator data of the quinoa planting area based on the trained quinoa yield hybrid prediction model to obtain the final quinoa production increase plan.

[0049] The technical solution of the present invention has at least the following advantages and beneficial effects:

[0050] The present invention constructs a hybrid prediction model that integrates time-series deep learning and static feature modeling, which can simultaneously capture the impact of meteorological trends and soil management measures, and significantly improve the accuracy of yield prediction. A variety adaptability rule screening mechanism based on classification and regression trees is adopted to ensure that the optimization scheme is selected within a suitable variety range. An objective function that comprehensively considers yield benefits and resource constraints is designed, and the optimal decision combination is solved through an iterative optimization algorithm to ensure the yield increase effect and meet actual resource constraints. It supports dynamic adjustment of the original decision based on real-time monitoring data to improve the adaptability and reliability of the planting plan. The full process automation from data collection, model training to scheme optimization is realized, providing a precise management basis for quinoa planting. BRIEF DESCRIPTION OF THE DRAWINGS

[0051] Figure 1 A schematic diagram of the process of the method for increasing quinoa production based on artificial intelligence provided by the present invention;

[0052] Figure 2 A schematic diagram of the calculation steps of the objective function provided by the present invention;

[0053] Figure 3 This is a schematic diagram of the quinoa yield increasing system based on artificial intelligence provided by the present invention. DETAILED DESCRIPTION

[0054] In order to make the purpose, technical solutions and advantages of the present invention clearer, the technical solutions of the present invention will be clearly and completely described below in conjunction with the drawings in the present invention. Obviously, what is described is only a part of the present invention, not all of it. Generally, the components of the present invention described and shown in the drawings here can be arranged and designed in various different configurations.

[0055] Reference Figure 1 As shown, Figure 1 A schematic diagram of the process of the quinoa yield increasing method based on artificial intelligence provided by the present invention.

[0056] In one embodiment, the present invention provides a method for increasing quinoa production based on artificial intelligence, the method comprising the steps of:

[0057] Select the boundaries of quinoa planting areas, obtain real-time indicator data and historical monitoring data of quinoa planting areas, and combine and establish a spatiotemporal database;

[0058] Extract the historical yield data of quinoa varieties from the spatiotemporal database, and associate it with the specific indicator data of the corresponding plots. At the same time, define the specific indicator data as the characteristic variable, define the unit area yield of quinoa varieties as the target variable, and establish the adaptability rules of quinoa varieties through the CART regression tree;

[0059] Based on the adaptability rules of quinoa varieties, a matching data set in the spatiotemporal database is extracted, and the matching data set is divided into a training set and a test set. The training set is input into a preset quinoa yield hybrid prediction model for calculation, and the training results are verified by the test set to complete the training of the quinoa yield hybrid prediction model, wherein the quinoa yield hybrid prediction model specifically takes maximizing quinoa yield as the optimization goal under the condition of optimal comprehensive adaptability of quinoa varieties;

[0060] Based on the trained quinoa yield hybrid prediction model, the real-time indicator data of the quinoa planting area is calculated to obtain the final quinoa production increase plan.

[0061] In this embodiment, the present invention first selects the boundary of the quinoa planting area, and obtains real-time indicator data and historical monitoring data in the area. The real-time indicator data includes but is not limited to meteorological data (such as temperature, rainfall, etc.) and management data (such as irrigation amount, fertilization amount, etc.); the historical monitoring data mainly involves the physical and chemical properties of soil in different plots, the historical yield of quinoa varieties, and the records of management measures. The obtained data are sorted and integrated to establish a multi-source heterogeneous data spatiotemporal database covering spatiotemporal information, providing a data basis for subsequent data processing and model training. The historical yield data of quinoa varieties are extracted from the constructed spatiotemporal database, and the historical yield data are associated with the specific indicator data of the corresponding plot. Among them, the specific indicator data include soil nutrient content (such as nitrogen, phosphorus, potassium content), soil moisture, meteorological conditions (such as temperature, rainfall), management measures (such as fertilization and irrigation strategies), etc. These data are defined as characteristic variables, and the yield per unit area of ​​quinoa varieties is used as the target variable. Using these data, the quinoa variety adaptability rules are constructed by the CART regression tree method. The rules are fully supported by statistical data and objectively reflect the adaptability performance and yield stability of each quinoa variety under different environmental conditions. Based on the adaptability rules of quinoa varieties, a matching data set is extracted from the spatiotemporal database, and then the matching data set is divided into a training set and a test set. The training set data is input into a preset quinoa yield hybrid prediction model for calculation. The hybrid prediction model integrates the time series deep learning module and the static feature regression module, and can simultaneously model the dynamic changes of meteorology and soil management measures to improve the accuracy of yield prediction. The training results are verified by the test set to ensure that the model completes the training of the yield hybrid prediction model under the optimal conditions of the comprehensive adaptability of quinoa varieties, with the maximization of quinoa yield per unit area as the optimization goal. Using the trained quinoa yield hybrid prediction model, the real-time indicator data of the quinoa planting area is calculated to obtain the unit area yield prediction value of each plot under different management measures. By comprehensively evaluating the prediction results and checking the constraints (including the upper limit of fertilizer application for a single plot, the upper limit of irrigation for a single plot, global resource constraints, and variety adaptability selection, etc.), the optimal planting plan for increasing quinoa production is finally output. This plan not only provides an accurate decision-making basis for guiding actual planting, but also meets the requirements of rational resource utilization and environmentally friendly planting management.

[0062] In a further implementation of this embodiment, the spatiotemporal database is established by:

[0063] Select the boundaries of quinoa growing areas and deploy a variety of IoT detection devices in the quinoa growing areas to establish a quinoa growing area monitoring network;

[0064] Obtain real-time indicator data of quinoa planting areas based on the quinoa planting area monitoring network;

[0065] Collect historical monitoring data on quinoa growing areas;

[0066] Align the timestamps and geographic coordinates of real-time indicator data with historical monitoring data, and build a spatiotemporal database through PostgreSQL.

[0067] During implementation, this embodiment first determines the boundary of the quinoa planting area, and selects the regional scope covering the main quinoa planting plots based on the established geographic information system data and regional planting plan. Subsequently, a variety of Internet of Things detection equipment (such as soil moisture sensors, temperature sensors, rainfall meters, light intensity meters, etc.) are deployed in the area to achieve full coverage monitoring of environmental indicators in the region, forming a quinoa planting area monitoring network. The monitoring network constructed in this embodiment can capture soil conditions and meteorological data in the quinoa planting area in real time. According to the quinoa planting area monitoring network, real-time indicator data can be collected in a timely manner. At the same time, combined with the long-term monitoring equipment in the region and the records of relevant departments, historical monitoring data of the quinoa planting area are collected. Using the real-time and historical data after alignment processing, the PostgreSQL database platform is used to build a spatiotemporal database, which not only supports the storage and query of basic data, but also can combine the spatiotemporal indexing function to efficiently manage multi-source heterogeneous data with spatiotemporal characteristics.

[0068] Specifically, before establishing the quinoa variety adaptability rules through the CART regression tree, the method further includes:

[0069] Extract the historical yield data of quinoa varieties in the spatiotemporal database, associate it with the specific indicator data of the corresponding plots, calculate the yield variation coefficient of each quinoa variety under the same soil type, and screen out quinoa varieties whose yield variation coefficient is greater than the first threshold;

[0070] The yield variation coefficient is specifically:

[0071] Extract historical quinoa variety production data according to quinoa variety and specific indicator environment classification;

[0072] Calculate the mean yield and standard deviation of each quinoa variety;

[0073] The yield variation coefficient of quinoa varieties was solved based on the yield mean and standard deviation of each quinoa variety;

[0074] Based on the preset first threshold as the basis for judging the coefficient of variation of quinoa variety yield, quinoa varieties greater than or equal to the first threshold are screened out.

[0075] The coefficient of variation of yield can measure the stability of the yield of a certain variety under the same or similar environment (such as the same soil type). A higher coefficient of variation indicates large fluctuations in yield, while a lower coefficient of variation indicates a more stable yield.

[0076] In the specific implementation of this embodiment, the quinoa variety adaptability rules are established by using the CART regression tree, which are specifically:

[0077] Extract data from the spatiotemporal database to obtain soil and meteorological index data of quinoa planting areas, and associate historical yield data of various quinoa varieties;

[0078] Based on the extracted data in the spatiotemporal database, the soil index data and meteorological index data of the quinoa planting area are defined as feature variables, and the unit area yield of quinoa varieties is defined as the target variable, and the two are combined to form a training sample set;

[0079] Construct a CART regression tree model. In the training sample set, use the root node to represent all training samples, search for the best split point in each feature dimension, and use the best split point to divide the root node for the first time to obtain two child nodes.

[0080] For each child node, repeat the root node splitting process until the number of samples in the node is lower than the preset lower limit, and prune the CART regression tree model through the cost complexity pruning algorithm to complete the training of the CART regression tree model;

[0081] According to the trained CART regression tree model, the characteristic variable segmentation conditions and the corresponding predicted yields in each terminal node are extracted and combined to establish the adaptability rules of quinoa varieties.

[0082] The process of determining the optimal segmentation point is as follows:

[0083] In the training sample set, all training samples are represented by the root node, and the best segmentation point is searched in each feature dimension;

[0084] For each candidate feature and its candidate segmentation point, the mean square error of the two child nodes after segmentation is calculated and weighted by the proportion of the number of samples to obtain the weighted error;

[0085] Among all candidate segmentations, the segmentation point with the smallest weighted error is selected as the best segmentation point.

[0086] In the specific implementation, this embodiment selects the yield per unit area of ​​quinoa varieties as the target variable, selects soil index data and meteorological index data as feature variables, and constructs a data set to be modeled according to the corresponding relationship between the target variable and the feature variable; extracts soil and meteorological index data of the quinoa planting area from the spatiotemporal database, and associates it with the historical yield data of each variety; performs cleaning operations such as missing value processing, outlier detection, unit conversion and standardization on the extracted data to ensure data consistency in different plots and different growth periods; pre-processes the cleaned data, extracts or generates derived variables according to the characteristic requirements of soil and meteorological indicators, analyzes the distribution of the yield per unit area of ​​quinoa varieties and transforms them as needed, and divides the obtained data set into a training set and a validation set; in the training set, all training samples are represented by the root node, and the best segmentation point is searched in each feature dimension, and for each candidate feature and its segmentation point t, the mean square error of the child nodes S1 and S2 obtained after splitting is calculated, wherein, for the sample set S in the current node, its mean square error is: ;in, is the actual value of the sample, is the sample mean, and the weighted mean square error of the child nodes S1 and S2 after splitting is: ; Choose so that the error is reduced The largest split point is used for splitting; the root node is split at the selected optimal split point to obtain two child nodes; for each child node, the above process of searching for the optimal split point is repeated to achieve further binary division of the samples in the node until the number of samples in the node is lower than the preset lower limit, and the cost complexity pruning algorithm is used to prune the established regression tree, and the splitting conditions and corresponding predicted values ​​of each characteristic variable in each terminal node (i.e., the leaf node that is no longer split) are extracted according to the trained CART regression tree model, where each terminal node describes the statistical performance of the unit area yield of quinoa varieties within the threshold range of specific soil and meteorological indicators; the characteristic splitting conditions of these terminal nodes and their predicted yields are combined to form an adaptive rule, which describes the relationship between indicator data (such as soil pH, precipitation, sunshine hours, etc.) and the unit area yield of quinoa varieties, and then distinguishes the adaptability of quinoa varieties under high-yield or low-yield conditions.

[0087] In this embodiment, the quinoa yield hybrid prediction model includes: a time series branch and a static branch; wherein the time series branch is specifically an LSTM neural network model, and the time series characteristics of the meteorological index data are learned through the LSTM neural network model;

[0088] The static branch is specifically a multi-layer perceptron model, and the soil index data is modeled by the multi-layer perceptron model;

[0089] The two branch outputs are passed through the feature fusion module to obtain the unit area yield prediction value of the quinoa planting area, the loss value between the unit area yield prediction value and the actual result is calculated, and the parameters of the quinoa yield hybrid prediction model are iteratively updated through the loss value to complete the training of the quinoa yield hybrid prediction model.

[0090] In this embodiment, the extracted dynamic meteorological data is input into the time series branch to perform deep learning on the time series information. The time series branch is specifically a long short-term memory model, which is used to capture the trend and cycle of the weather series during the crop growth process; at the same time, the extracted static soil physical and chemical indicators and management measures such as varieties, fertilization, and irrigation are input into the static branch to perform regression modeling on the soil and management factors. The static branch is a multi-layer perceptron model, which is used to characterize the impact of different soils, fertilization and irrigation methods, and variety configurations on yield; the outputs of the two branches are feature fused to obtain a predicted value of the yield per unit area, the loss between the predicted result and the actual yield is calculated, and the training is completed by iteratively updating the model parameters to obtain a hybrid model that can predict the yield of different fertilizer amounts, irrigation amounts, and variety combinations.

[0091] Reference Figure 2 As shown, Figure 2 A schematic diagram of the calculation steps of the objective function provided by the present invention.

[0092] In a further implementation of this embodiment, the quinoa yield hybrid prediction model is specifically designed to maximize quinoa yield under the condition that the comprehensive adaptability of quinoa varieties is optimal, and the specific calculation process is as follows:

[0093] Set the number of plots K in the quinoa planting area and determine the decision combination for each plot;

[0094] By completing the training of the quinoa yield hybrid prediction model, based on the determined soil index data and quinoa varieties, the unit area yield prediction value and objective function value of each plot are calculated;

[0095] Under the set constraints, the decision combination of each plot is iteratively updated;

[0096] The predicted yield per unit area of ​​each plot is repeatedly calculated through the trained quinoa yield hybrid prediction model, and the objective function is updated until the maximum number of iterations is reached, and the decision combination result that maximizes the objective function value is output.

[0097] In the specific implementation, this embodiment formulates an initial solution, assigns an initial decision combination of "nitrogen application amount", "irrigation amount" and "quinoa variety" to each plot, and uses a hybrid prediction model to calculate the yield prediction value of each plot and the overall objective function value. Subsequently, the decision variables are iteratively optimized using an evolutionary algorithm, and the values ​​of each decision variable are continuously adjusted while keeping the various resources and adaptability constraints unchanged. The hybrid prediction model is used to re-evaluate the objective function value in each round of iteration. When the objective function reaches convergence or the number of iterations reaches a preset upper limit, the output makes the objective function F F The largest optimal combination.

[0098] Specifically, the objective function of the quinoa yield hybrid prediction model is calculated as follows:

[0099]

[0100] Among them, F is the objective function, K is the total number of plots, k is the plot index, is the predicted yield per unit area, is the nitrogen application rate of the kth plot, is the irrigation water volume of the kth plot, is the quinoa variety of the kth plot, is the penalty coefficient, is the acceleration factor, For reference fertilizer amount.

[0101] The set constraints of this embodiment include: an upper limit constraint on the amount of fertilizer applied to a single plot of land, an upper limit constraint on the amount of irrigation applied to a single plot of land, a constraint on the total amount of fertilizer resources, a constraint on the total amount of water resources, and a constraint on the adaptability of varieties.

[0102] The upper limit constraint of the fertilizer application amount for a single plot of land is calculated as follows:

[0103]

[0104] The upper limit constraint of the single-plot irrigation volume is calculated as follows:

[0105]

[0106] The calculation formula of the total amount of fertilizer resources constraint is:

[0107]

[0108] The calculation formula of the total water resources constraint is:

[0109]

[0110] The variety adaptability selection constraint is calculated as follows:

[0111]

[0112] in, , They are the upper limits of fertilizer and irrigation for a single field, is the area of ​​the kth plot, , are the global total upper limits of available fertilizer and water resources, is a feasible set.

[0113] Reference Figure 3 As shown, Figure 3 This is a schematic diagram of the quinoa yield increasing system based on artificial intelligence provided by the present invention.

[0114] In another embodiment, a quinoa yield increasing system based on artificial intelligence comprises:

[0115] A database establishment unit selects the boundaries of quinoa planting areas, obtains real-time indicator data and historical monitoring data of quinoa planting areas, and combines and establishes a spatiotemporal database;

[0116] The unit for establishing the adaptability rules of quinoa varieties extracts the historical yield data of quinoa varieties from the spatiotemporal database and associates the specific indicator data of the corresponding plots. At the same time, the specific indicator data is defined as the characteristic variable, the yield per unit area of ​​quinoa varieties is defined as the target variable, and the adaptability rules of quinoa varieties are established through the CART regression tree;

[0117] A model calculation unit extracts a matching data set in a spatiotemporal database based on the adaptability rules of quinoa varieties, divides the matching data set into a training set and a test set, inputs the training set into a preset quinoa yield hybrid prediction model for calculation, and verifies the training results through the test set to complete the training of the quinoa yield hybrid prediction model, wherein the quinoa yield hybrid prediction model specifically takes maximizing quinoa yield as an optimization goal under the condition that the comprehensive adaptability of quinoa varieties is optimal;

[0118] The solution output unit calculates the real-time indicator data of the quinoa planting area based on the trained quinoa yield hybrid prediction model to obtain the final quinoa production increase plan.

[0119] The above is only the preferred embodiment of the present invention and is not intended to limit the present invention. For those skilled in the art, the present invention may have various modifications and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. A method for increasing quinoa production based on artificial intelligence, characterized in that: The steps of the method include: Select the boundaries of quinoa planting areas, obtain real-time indicator data and historical monitoring data of quinoa planting areas, and combine and establish a spatiotemporal database; Extract the historical yield data of quinoa varieties from the spatiotemporal database, and associate it with the specific indicator data of the corresponding plots. Define the specific indicator data as the characteristic variable, define the yield per unit area of ​​quinoa varieties as the target variable, and establish the adaptability rules of quinoa varieties through the CART regression tree. The specific indicator data include soil nutrient content, soil moisture, meteorological conditions, and management measures. Based on the adaptability rules of quinoa varieties, a matching data set in the spatiotemporal database is extracted, and the matching data set is divided into a training set and a test set. The training set is input into a preset quinoa yield hybrid prediction model for calculation, and the training results are verified by the test set to complete the training of the quinoa yield hybrid prediction model, wherein the quinoa yield hybrid prediction model specifically takes maximizing quinoa yield as the optimization goal under the condition of optimal comprehensive adaptability of quinoa varieties; Based on the trained quinoa yield hybrid prediction model, the real-time indicator data of the quinoa planting area is calculated to obtain the final quinoa yield increase plan; The quinoa yield hybrid prediction model includes: a time series branch and a static branch; wherein the time series branch is specifically an LSTM neural network model, and the time series characteristics of the meteorological indicator data are learned through the LSTM neural network model; The static branch is specifically a multi-layer perceptron model, and the soil index data is modeled by the multi-layer perceptron model; The outputs of the two branches are passed through the feature fusion module to obtain the unit area yield prediction value of the quinoa planting area, the loss value between the unit area yield prediction value and the actual result is calculated, and the parameters of the quinoa yield hybrid prediction model are iteratively updated through the loss value to complete the training of the quinoa yield hybrid prediction model; The extracted dynamic meteorological data is input into the time series branch to perform deep learning on the time series information. The time series branch is specifically a long short-term memory model, which is used to capture the trend and cycle of the weather series during the crop growth process; at the same time, the extracted static soil physical and chemical indicators and varieties, fertilization, and irrigation management measures are input into the static branch to perform regression modeling on the soil and management elements. The static branch is a multi-layer perceptron model, which is used to characterize the impact of different soils, fertilization and irrigation methods, and variety configurations on yield; the outputs of the two branches are feature fused to obtain the predicted value of the yield per unit area, the loss between the predicted result and the actual yield is calculated, and the training is completed by iteratively updating the model parameters to obtain a hybrid model for yield prediction for different fertilization amounts, irrigation amounts, and variety combinations; The quinoa yield hybrid prediction model is specifically designed to maximize quinoa yield under the condition that the comprehensive adaptability of quinoa varieties is optimal, and the specific calculation process is as follows: Set the number of plots K in the quinoa planting area and determine the decision combination for each plot; By completing the training of the quinoa yield hybrid prediction model, based on the determined soil index data and quinoa varieties, the unit area yield prediction value and objective function value of each plot are calculated; Under the set constraints, the decision combination of each plot is iteratively updated; The predicted yield per unit area of ​​each plot is repeatedly calculated through the trained quinoa yield hybrid prediction model, and the objective function is updated until the maximum number of iterations is reached, and the decision combination result that maximizes the objective function value is output.

2. The method for increasing quinoa production based on artificial intelligence according to claim 1, characterized in that: The establishment process of the spatiotemporal database is as follows: Select the boundaries of quinoa growing areas and deploy a variety of IoT detection devices in the quinoa growing areas to establish a quinoa growing area monitoring network; Obtain real-time indicator data of quinoa planting areas based on the quinoa planting area monitoring network; Collect historical monitoring data on quinoa growing areas; The real-time indicator data and historical monitoring data are aligned with timestamps and geographic coordinates, and a spatiotemporal database is built using PostgreSQL.

3. The method for increasing quinoa production based on artificial intelligence according to claim 2, characterized in that: Before establishing the quinoa variety adaptability rules by using the CART regression tree, the method further includes: Extract the historical yield data of quinoa varieties in the spatiotemporal database, associate it with the specific indicator data of the corresponding plots, calculate the yield variation coefficient of each quinoa variety under the same soil type, and screen out quinoa varieties whose yield variation coefficient is greater than the first threshold; The yield variation coefficient is specifically: Extract historical quinoa variety production data according to quinoa variety and specific indicator environment classification; Calculate the mean yield and standard deviation of each quinoa variety; The yield variation coefficient of quinoa varieties was solved based on the yield mean and standard deviation of each quinoa variety; Based on the preset first threshold as the basis for judging the coefficient of variation of quinoa variety yield, quinoa varieties greater than or equal to the first threshold are screened out.

4. The method for increasing quinoa production based on artificial intelligence according to claim 3, characterized in that: The quinoa variety adaptability rules are established by using the CART regression tree, which are specifically: Extract data from the spatiotemporal database to obtain soil and meteorological index data of quinoa planting areas, and associate historical yield data of various quinoa varieties; Based on the extracted data in the spatiotemporal database, the soil index data and meteorological index data of the quinoa planting area are defined as feature variables, and the unit area yield of quinoa varieties is defined as the target variable, and the two are combined to form a training sample set; Construct a CART regression tree model. In the training sample set, use the root node to represent all training samples, search for the best split point in each feature dimension, and use the best split point to divide the root node for the first time to obtain two child nodes. For each child node, repeat the root node splitting process until the number of samples in the node is lower than the preset lower limit, and prune the CART regression tree model through the cost complexity pruning algorithm to complete the training of the CART regression tree model; According to the trained CART regression tree model, the characteristic variable segmentation conditions and the corresponding predicted yields in each terminal node are extracted and combined to establish the adaptability rules of quinoa varieties.

5. The method for increasing quinoa production based on artificial intelligence according to claim 4, characterized in that: The process of determining the optimal segmentation point is as follows: In the training sample set, all training samples are represented by the root node, and the best segmentation point is searched in each feature dimension; For each candidate feature and its candidate segmentation point, the mean square error of the two child nodes after segmentation is calculated and weighted by the proportion of the number of samples to obtain the weighted error; Among all candidate segmentations, the segmentation point with the smallest weighted error is selected as the best segmentation point.

6. The method for increasing quinoa production based on artificial intelligence according to claim 5, characterized in that: The objective function of the quinoa yield hybrid prediction model is calculated as follows: Among them, F is the objective function, K is the total number of plots, k is the plot index, is the predicted yield per unit area, is the nitrogen application rate of the kth plot, is the irrigation water volume of the kth plot, is the quinoa variety of the kth plot, is the penalty coefficient, is the acceleration factor, For reference fertilizer amount.

7. The method for increasing quinoa production based on artificial intelligence according to claim 6, characterized in that: The set constraints include: upper limit constraint on the amount of fertilizer applied to a single plot of land, upper limit constraint on the amount of irrigation applied to a single plot of land, total amount constraint on fertilizer resources, total amount constraint on water resources and constraint on variety adaptability selection.

8. The quinoa yield increasing system based on artificial intelligence is characterized by: include: A database establishment unit selects the boundaries of quinoa planting areas, obtains real-time indicator data and historical monitoring data of quinoa planting areas, and combines and establishes a spatiotemporal database; The unit for establishing the adaptability rules of quinoa varieties extracts the historical yield data of quinoa varieties from the spatiotemporal database and associates the specific indicator data of the corresponding plots. At the same time, the specific indicator data is defined as the characteristic variable, and the yield per unit area of ​​the quinoa variety is defined as the target variable. The adaptability rules of quinoa varieties are established through the CART regression tree. The specific indicator data include soil nutrient content, soil moisture, meteorological conditions, and management measures. A model calculation unit extracts a matching data set in a spatiotemporal database based on the adaptability rules of quinoa varieties, divides the matching data set into a training set and a test set, inputs the training set into a preset quinoa yield hybrid prediction model for calculation, and verifies the training results through the test set to complete the training of the quinoa yield hybrid prediction model, wherein the quinoa yield hybrid prediction model specifically takes maximizing quinoa yield as an optimization goal under the condition that the comprehensive adaptability of quinoa varieties is optimal; The solution output unit calculates the real-time indicator data of the quinoa planting area based on the trained quinoa yield hybrid prediction model to obtain the final quinoa yield increase plan; The quinoa yield hybrid prediction model includes: a time series branch and a static branch; wherein the time series branch is specifically an LSTM neural network model, and the time series characteristics of the meteorological indicator data are learned through the LSTM neural network model; The static branch is specifically a multi-layer perceptron model, and the soil index data is modeled by the multi-layer perceptron model; The outputs of the two branches are passed through the feature fusion module to obtain the unit area yield prediction value of the quinoa planting area, the loss value between the unit area yield prediction value and the actual result is calculated, and the parameters of the quinoa yield hybrid prediction model are iteratively updated through the loss value to complete the training of the quinoa yield hybrid prediction model; The extracted dynamic meteorological data is input into the time series branch to perform deep learning on the time series information. The time series branch is specifically a long short-term memory model, which is used to capture the trend and cycle of the weather series during the crop growth process; at the same time, the extracted static soil physical and chemical indicators and varieties, fertilization, and irrigation management measures are input into the static branch to perform regression modeling on the soil and management elements. The static branch is a multi-layer perceptron model, which is used to characterize the impact of different soils, fertilization and irrigation methods, and variety configurations on yield; the outputs of the two branches are feature fused to obtain the predicted value of the yield per unit area, the loss between the predicted result and the actual yield is calculated, and the training is completed by iteratively updating the model parameters to obtain a hybrid model for yield prediction for different fertilization amounts, irrigation amounts, and variety combinations; The quinoa yield hybrid prediction model is specifically designed to maximize quinoa yield under the condition that the comprehensive adaptability of quinoa varieties is optimal, and the specific calculation process is as follows: Set the number of plots K in the quinoa planting area and determine the decision combination for each plot; By completing the training of the quinoa yield hybrid prediction model, based on the determined soil index data and quinoa varieties, the unit area yield prediction value and objective function value of each plot are calculated; Under the set constraints, the decision combination of each plot is iteratively updated; The predicted yield per unit area of ​​each plot is repeatedly calculated through the trained quinoa yield hybrid prediction model, and the objective function is updated until the maximum number of iterations is reached, and the decision combination result that maximizes the objective function value is output.

Images