Agricultural production site environment monitoring big data analysis method and system

Abstract

The present application relates to the technical field of agricultural origin analysis, and relates to an agricultural origin environment monitoring big data analysis method and system, which comprises the following steps: collecting multi-dimensional space-time environment data and crop growth state data of an agricultural origin, and preprocessing the collected data to form space-time feature sequence data containing a time stamp; constructing, training and optimizing a soil organic matter dynamic evolution prediction model, setting a dynamic control quantity of farming management measures, combining real-time input space-time feature sequence data, inputting into the soil organic matter dynamic evolution prediction model for simulation analysis, to predict the expected change trajectory and final state of the soil organic matter content of the agricultural origin in the entire management cycle after implementing the dynamic control quantity of the farming management measures, and generating an operation guidance scheme for the agricultural origin according to the simulation analysis result. The present application significantly improves the agricultural resource utilization efficiency and the sustainability of production management through big data analysis and deep learning intelligent prediction.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of agricultural production site analysis, in particular to an agricultural production site environment monitoring big data analysis method and system. BACKGROUND

[0002] Soil is the cornerstone of agricultural production, and soil organic matter content is a key core indicator for measuring soil fertility, health status and sustainable utilization capacity. It not only directly affects the growth and yield of crops, but also plays an irreplaceable role in carbon sequestration, soil and water conservation, and ecological environment protection. With the rapid development of modern agriculture towards precision and intelligence, real-time, dynamic and accurate monitoring and management of agricultural production site environment, especially effective prediction and regulation of such dynamic changes of key elements as soil organic matter, have become an urgent need to improve agricultural production efficiency, ensure the quality and safety of agricultural products, and promote sustainable development of agriculture. At present, in terms of soil organic matter management and agricultural decision support in agricultural production sites, the existing methods have some inherent limitations. On the one hand, soil organic matter monitoring often relies on periodic laboratory sampling analysis, which is not only costly, time-consuming and labor-intensive, but also difficult to capture the spatial heterogeneity and dynamic change characteristics of soil organic matter in time, resulting in insufficient representativeness and timeliness of the monitoring results. On the other hand, existing agricultural management recommendations mostly rely on experience or static and universal fertilization guidelines, lacking detailed consideration of dynamic environmental conditions, crop growth status and effects of different agricultural measures, and it is difficult for existing agricultural management recommendations to provide effective support for how to quantitatively evaluate the impact of different agricultural management measures on the future evolution trajectory of soil organic matter and make forward-looking optimization decisions, resulting in lagging adjustment of agricultural management measures and lack of sufficient scientific predictability and precision, which is difficult to meet the strict requirements of modern precision agriculture for resource efficient use and environmental protection. SUMMARY

[0003] The purpose of the present application is to provide an agricultural production site environment monitoring big data analysis method and system, which realizes dynamic and accurate prediction of soil organic matter in agricultural production sites and intelligent optimization decision of agricultural management through big data analysis and deep learning intelligent prediction, significantly improving the efficiency of agricultural resource utilization and the sustainability of production management.

[0004] The present application is realized by the following technical solutions:

[0005] An agricultural production site environment monitoring big data analysis method, the steps of the method comprising:

[0006] Collecting multi-dimensional spatio-temporal environmental data and crop growth state data covering agricultural production sites, and pre-processing the multi-dimensional spatio-temporal environmental data and crop growth state data to form time-stamped spatio-temporal feature sequence data;

[0007] constructing a soil organic matter dynamic evolution prediction model configured to receive the spatiotemporal feature sequence data and corresponding farming management measures as input and output a predicted value of soil organic matter content of the agricultural land;

[0008] driving the soil organic matter dynamic evolution prediction model to perform iterative optimization for a preset target change rate of soil organic matter content of the agricultural land, to determine a dynamic adjustment and control setting of the farming management measures that makes the predicted value of soil organic matter content reach the target change rate;

[0009] inputting the dynamic adjustment and control setting of the farming management measures, in combination with the real-time input spatiotemporal feature sequence data, into the soil organic matter dynamic evolution prediction model for simulation analysis, to predict an expected change trajectory and final achieved state of the soil organic matter content of the agricultural land in the entire management cycle after implementation of the dynamic adjustment and control setting of the farming management measures, and generating an operation guidance scheme for the agricultural land according to the simulation analysis result.

[0010] Optionally, the preprocessing of the multi-dimensional spatiotemporal environmental data and crop growth state data specifically includes:

[0011] performing spatial interpolation processing on the multi-dimensional spatiotemporal environmental data to generate a continuous data layer reflecting the spatial distribution characteristics of the agricultural land;

[0012] performing preprocessing and feature extraction on the crop growth state data to obtain time series features representing the crop growth process and its influence on soil organic matter;

[0013] unifying, gridding and fusing the continuous data layer and the time series features in the time and space dimensions, and performing missing value filling and feature transformation to form the spatiotemporal feature sequence data.

[0014] Optionally, the soil organic matter dynamic evolution prediction model is constructed specifically as follows:

[0015] constructing the soil organic matter dynamic evolution prediction model based on a deep spatiotemporal encoder-decoder neural network architecture;

[0016] The soil organic matter dynamic evolution prediction model includes an encoder and a decoder, wherein:

[0017] The encoder is configured to adopt a convolutional long short-term memory network unit to process the spatiotemporal feature sequence data and extract deep spatiotemporal dynamic features thereof;

[0018] The decoder is configured to generate the predicted value of the soil organic matter content based on the deep spatiotemporal dynamic features extracted by the encoder and the corresponding farming management measures in combination with an attention mechanism.

[0019] The soil organic matter dynamic evolution prediction model is configured to input the spatiotemporal feature sequence data into the encoder for processing, and input the farming management measures as conditional information into the decoder together with the deep spatiotemporal dynamic features output by the encoder for processing.

[0020] The decoder outputs the predicted value of the soil organic matter content of each predetermined spatial grid unit in the agricultural production site.

[0021] Optionally, the training process of the soil organic matter dynamic evolution prediction model comprises:

[0022] obtaining and dividing a training set and a test set containing historical soil organic matter content, corresponding spatiotemporal feature sequence data and farming management measures;

[0023] inputting the training set into the soil organic matter dynamic evolution prediction model in batches, obtaining the predicted value of the soil organic matter content of the data in this batch through forward propagation, and calculating the loss between the predicted value and the actual soil organic matter content corresponding to the training data in this batch according to a preset optimization objective function;

[0024] calculating the gradient of the soil organic matter dynamic evolution prediction model through back propagation, and iteratively updating the parameters of the soil organic matter dynamic evolution prediction model through an optimization algorithm;

[0025] until the iteration reaches a preset maximum number of times, evaluating the soil organic matter dynamic evolution prediction model through the test set, and completing the training of the soil organic matter dynamic evolution prediction model.

[0026] Optionally, the calculation of the loss between the predicted value and the actual soil organic matter content corresponding to the training data in this batch according to the preset optimization objective function specifically comprises:

[0027] calculating the prediction deviation of the predicted value of the soil organic matter content output by the soil organic matter dynamic evolution prediction model and the corresponding true soil organic matter content target value at each corresponding data point;

[0028] quantifying each prediction deviation based on a predefined mean square error loss function to obtain an individual error value of each data point;

[0029] aggregating the individual error values to jointly form an optimization objective function value for guiding the parameter update of the soil organic matter dynamic evolution prediction model.

[0030] Optionally, the driving soil organic matter dynamic evolution prediction model is iteratively optimized to determine the dynamic regulation and control setting quantity of the farming management measure that makes the predicted value of the soil organic matter content achieve the target change rate, and the specific process is as follows:

[0031] According to the allowed adjustment range and constraint conditions of the preset dynamic regulation and control setting quantity of the farming management measure, candidate dynamic regulation and control setting quantities of the farming management measure are generated;

[0032] For each group of candidate dynamic regulation and control setting quantities of the farming management measure processed in the current iteration period, the following is performed:

[0033] The candidate dynamic regulation and control setting quantity of the farming management measure is input into the soil organic matter dynamic evolution prediction model to obtain the expected change trajectory of the soil organic matter content of each predetermined spatial grid cell in the agricultural land under this group of candidate dynamic regulation and control setting quantities of the farming management measure;

[0034] According to the optimization objective function with the degree of achievement of the target change rate of the soil organic matter content in the agricultural land as the index, and in combination with the pre-defined evaluation function, the comprehensive performance score of this group of candidate dynamic regulation and control setting quantities of the farming management measure is calculated based on the expected change trajectory;

[0035] According to the comprehensive performance scores of each group of candidate dynamic regulation and control setting quantities of the farming management measure, new candidate dynamic regulation and control setting quantities of the farming management measure are generated under the allowed adjustment range and constraint conditions, and are returned to the soil organic matter dynamic evolution prediction model for iterative updating until the preset maximum number of iterations is reached;

[0036] The comprehensive performance score that reaches the optimum is selected as the dynamic regulation and control setting quantity of the farming management measure for output.

[0037] Optionally, the optimization objective function with the degree of achievement of the target change rate of the soil organic matter content in the agricultural land as the index has the following specific calculation formula:

[0038]

[0039] wherein, is the optimization objective function, is the total number of predetermined spatial grid cells in the agricultural land, is a preset normal adjustment factor, j is the index of the spatial grid cell, is the predicted soil organic matter content of the jth spatial grid cell, is the end time of the prediction period, is the initial soil organic matter content of the jth predetermined spatial grid cell, is the target change rate of the soil organic matter content of the jth predetermined spatial grid cell.

[0040] Optionally, the evaluation function has a calculation formula as follows:

[0041]

[0042] wherein, is an evaluation value, , are weight coefficients respectively, is an economic cost, is an environmental impact evaluation value, and m is a current candidate agricultural management measure dynamic control setting.

[0043] Optionally, the comprehensive performance score has a calculation formula as follows:

[0044]

[0045] wherein, is a comprehensive performance score of a candidate agricultural management measure dynamic control setting, , are weight coefficients respectively.

[0046] The agricultural production site environment monitoring big data analysis system comprises:

[0047] A data acquisition unit acquires multi-dimensional spatio-temporal environment data and crop growth state data covering an agricultural production site, and pre-processes the multi-dimensional spatio-temporal environment data and the crop growth state data to form spatio-temporal feature sequence data containing a time stamp;

[0048] A model construction unit constructs a soil organic matter dynamic evolution prediction model, which is configured to receive the spatio-temporal feature sequence data and corresponding agricultural management measures as input, and output a predicted value of soil organic matter content of the agricultural production site;

[0049] An optimization unit drives the soil organic matter dynamic evolution prediction model to perform iterative optimization for a preset target change rate of soil organic matter content of the agricultural production site, to determine an agricultural management measure dynamic control setting that makes the predicted value of soil organic matter content achieve the target change rate;

[0050] An analysis unit inputs the agricultural management measure dynamic control setting and real-time input spatio-temporal feature sequence data into the soil organic matter dynamic evolution prediction model for simulation analysis, to predict an expected change trajectory and a final achieved state of the soil organic matter content of the agricultural production site in a whole management cycle after implementation of the agricultural management measure dynamic control setting, and generates an operation guidance scheme for the agricultural production site according to a simulation analysis result.

[0051] The technical solution of the present application has at least the following advantages and beneficial effects:

[0052] This invention, on the one hand, collects and integrates multi-dimensional spatiotemporal environmental data and crop growth status data to construct a spatiotemporal feature sequence containing timestamps. This sequence can more comprehensively and dynamically reflect the actual conditions of agricultural production areas, laying a solid data foundation for subsequent accurate predictions. On the other hand, this invention constructs a dynamic evolution prediction model for soil organic matter based on a deep spatiotemporal encoder-decoder neural network architecture. This model can not only effectively process complex spatiotemporal feature sequence data and deeply understand its intrinsic correlations, but also combine agricultural management measures as conditional inputs. Furthermore, through advanced technologies such as attention mechanisms, it significantly improves the accuracy and spatial precision of soil organic matter content prediction. Moreover, by driving the prediction model to iteratively optimize, this invention can not only proactively explore and determine the optimal dynamic control settings of agricultural management measures to achieve the preset soil organic matter target change rate, but also combine real-time data for forward-looking simulation analysis to generate specific and operable operational guidance schemes. This represents a fundamental leap from traditional passive monitoring and experience-based decision-making to proactive prediction and intelligent optimization control. Attached Figure Description

[0053] Figure 1 A flowchart illustrating the big data analysis method for monitoring agricultural production environments provided by this invention;

[0054] Figure 2 A schematic diagram illustrating the principle of the agricultural production area environment monitoring big data analysis system provided by the present invention. Detailed Implementation

[0055] To make the objectives, technical solutions, and advantages of this invention clearer, the technical solutions of this invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, what is described is only a part of this invention, not all of it. The components of this invention, typically described and shown in the accompanying drawings, can be arranged and designed in various different configurations.

[0056] like Figure 1 As shown, Figure 1 This is a flowchart illustrating the big data analysis method for monitoring agricultural production environments provided by the present invention.

[0057] In one embodiment, the present invention provides a big data analysis method for monitoring agricultural production environment, the method comprising the following steps:

[0058] Collect multi-dimensional spatiotemporal environmental data and crop growth status data covering agricultural production areas, and preprocess the multi-dimensional spatiotemporal environmental data and crop growth status data to form spatiotemporal feature sequence data containing timestamps;

[0059] The soil organic matter dynamic evolution prediction model is configured to receive the spatio-temporal feature sequence data and corresponding farming management measures as input, and output a predicted value of soil organic matter content of the agricultural production site.

[0060] For a preset target change rate of soil organic matter content of the agricultural production site, the soil organic matter dynamic evolution prediction model is driven to perform iterative optimization to determine a dynamic adjustment and control setting of the farming management measures that makes the predicted value of soil organic matter content reach the target change rate.

[0061] The dynamic adjustment and control setting of the farming management measures is input into the soil organic matter dynamic evolution prediction model for simulation analysis in combination with real-time input spatio-temporal feature sequence data, to predict an expected change trajectory and final achieved state of the soil organic matter content of the agricultural production site in the entire management cycle after the dynamic adjustment and control setting of the farming management measures is implemented, and an operation guidance scheme for the agricultural production site is generated according to the simulation analysis result.

[0062] Specifically, in this embodiment, a soil sensor array deployed in the agricultural production site collects soil temperature, humidity and conductivity data in real time, a small automatic weather station collects air temperature, humidity, light and rainfall data, and a normalized vegetation index extracted from a satellite remote sensing image obtained periodically is used as crop growth state data, and the collected data is preprocessed to form spatio-temporal feature sequence data containing accurate time stamps. A soil organic matter dynamic evolution prediction model is constructed, which is based on a deep neural network architecture and receives the above spatio-temporal feature sequence data and historical farming management measures corresponding to the plot as input, and outputs a predicted value of soil organic matter content at a future specific time point for each grid in the plot. This embodiment mainly aims at a target change rate of 5% average increase in soil organic matter content in the next year set by the user, and uses a genetic algorithm to drive the constructed soil organic matter dynamic evolution prediction model to perform iterative optimization. The iterative process continuously adjusts the candidate setting of the farming management measures, and the goal is to find a set of farming management measures that can make the predicted value of soil organic matter content output by the prediction model closest to the target change rate. Finally, the dynamic adjustment and control setting of the farming management measures obtained by the genetic algorithm optimization is combined with the latest spatio-temporal feature sequence data and input into the soil organic matter dynamic evolution prediction model for simulation analysis for one year. This analysis predicts the monthly change trajectory and expected achieved state at the end of the year of the soil organic matter content under the farming management measures. According to the simulation analysis result, the system automatically generates an operation guidance scheme for the agricultural production site, which clearly lists the recommended fertilizer type, quantity and irrigation recommendation for each month.

[0063] In a specific implementation, the multi-dimensional spatio-temporal environment data and the crop growth state data are preprocessed, which specifically comprises:

[0064] The multi-dimensional spatio-temporal environment data is subjected to spatial interpolation processing to generate a continuous data layer reflecting the spatial distribution characteristics of the agricultural production area;

[0065] The crop growth state data is preprocessed and feature extracted to obtain time series features representing the crop growth process and the influence of the soil organic matter;

[0066] The continuous data layer and the time series features are uniformly aligned, gridded and fused in the time and space dimensions, and are subjected to missing value filling and feature transformation to form the spatio-temporal feature sequence data.

[0067] In a specific implementation, the present embodiment is directed to multi-dimensional spatio-temporal environment data, including discrete point-shaped soil environment data obtained from a soil sensor array, which is subjected to spatial interpolation processing by ordinary Kriging interpolation method to be converted into a continuous soil property data layer covering the entire agricultural production area with a resolution of 10m x 10m. The present embodiment is directed to crop growth state data, including NDVI time series data obtained by satellite remote sensing, which is subjected to smoothing and denoising processing, and based on the smoothed NDVI time series, key phenological nodes such as the start date, peak date and end date of the growth season, as well as the cumulative NDVI value of the growth season, are extracted as time series features representing the crop growth process and the potential influence of the soil organic matter. The continuous soil data layer generated by the above processing and the extracted crop growth phenological time series features are uniformly aligned in the time and space dimensions. In space, all data are mapped to a unified 10m x 10m geographic grid cell, after alignment and fusion, if there are still missing values in the gridded data due to sensor failure or cloud cover, etc., a linear interpolation method based on adjacent time point data is used for filling, and all fused numerical features are subjected to minimum-maximum normalization processing to form the final spatio-temporal feature sequence data.

[0068] In a specific implementation of the present embodiment, the soil organic matter dynamic evolution prediction model is constructed, which specifically comprises:

[0069] The soil organic matter dynamic evolution prediction model is constructed based on a deep spatio-temporal encoder-decoder neural network architecture;

[0070] The soil organic matter dynamic evolution prediction model comprises an encoder and a decoder, wherein:

[0071] The encoder is configured to adopt a convolutional long short-term memory network unit to process the spatio-temporal feature sequence data and extract deep spatio-temporal dynamic features thereof;

[0072] The decoder is configured to generate the predicted value of the soil organic matter content based on the deep spatiotemporal dynamic features extracted by the encoder and the corresponding farming management measures in combination with an attention mechanism.

[0073] The soil organic matter dynamic evolution prediction model is constructed by inputting the spatiotemporal feature sequence data into the encoder for processing, and inputting the farming management measures as conditional information into the decoder together with the deep spatiotemporal dynamic features output by the encoder for processing.

[0074] The decoder outputs the predicted value of the soil organic matter content of each predetermined spatial grid unit in the agricultural production site.

[0075] In a specific implementation, the soil organic matter dynamic evolution prediction model of the present embodiment adopts a specific architecture based on a deep spatiotemporal encoder-decoder neural network, which specifically includes an encoder and a decoder. The encoder is stacked by three layers of ConvLSTM units. The spatiotemporal feature sequence data is received as input. Each layer of ConvLSTM units processes the spatial structure of the input feature map through its convolution kernel, and captures the dynamic changes of the time series through its LSTM gating mechanism, thereby extracting and encoding the deep spatiotemporal dynamic features in the data layer by layer, and finally outputting a fixed-length context vector as a compact representation of the historical spatiotemporal information. The decoder also adopts three layers of ConvLSTM units, and incorporates an attention mechanism at its input end. In the present embodiment, the Bahdanau attention mechanism is specifically adopted. The decoder receives the context vector output by the encoder, and at each decoding time step, selectively focuses on the output features of the encoder at different historical time states according to the weights calculated by the attention mechanism. Meanwhile, the decoder also receives the corresponding farming management measures as additional input. It can be understood that the corresponding farming management measures in the present embodiment are specifically a feature vector containing parameters such as fertilizer type, fertilizer amount, and irrigation amount. This vector is input as a conditional input at each prediction time step.

[0076] In a further implementation of the present embodiment, the spatiotemporal feature sequence data obtained by preprocessing is input into the encoder. After the encoder is processed, its final hidden state and output sequence at all time steps are passed. At each prediction time step, the decoder utilizes its current hidden state, the prediction output of the previous time step, the weighted encoder output calculated by the attention mechanism, and the current farming management measure conditional information, to generate the predicted value of the soil organic matter content of each 10m x 10m predetermined spatial grid unit in the agricultural production site at the next time point through its ConvLSTM unit and a subsequent convolution layer. The convolution layer is used to adjust the output channel and restore the spatial dimension.

[0077] In the specific implementation of the present embodiment, the training process of the soil organic matter dynamic evolution prediction model is as follows:

[0078] The training set and the test set are obtained and divided, which contain historical soil organic matter content, corresponding spatiotemporal feature sequence data and farming management measures;

[0079] The training set is input into the soil organic matter dynamic evolution prediction model in batches, the soil organic matter content prediction value of the batch data is obtained through forward propagation, and the loss between the prediction value and the actual soil organic matter content corresponding to the batch training data is calculated according to the preset optimization objective function;

[0080] The gradient of the soil organic matter dynamic evolution prediction model is calculated through back propagation, and the parameters of the soil organic matter dynamic evolution prediction model are iteratively updated through the optimization algorithm;

[0081] Until the preset maximum number of iterations is reached, the soil organic matter dynamic evolution prediction model is evaluated by the test set, and the training of the soil organic matter dynamic evolution prediction model is completed.

[0082] Specifically, the training process of the soil organic matter dynamic evolution prediction model is as follows: the historical soil organic matter content, corresponding spatiotemporal feature sequence data and corresponding farming management measure records of each plot in the past five years are obtained from the historical database. The data set is divided into a training set and a test set according to a proportion of 80%. When the training starts, the training set data is input into the initialized soil organic matter dynamic evolution prediction model in batches. For each batch of data, the soil organic matter dynamic evolution prediction model performs forward propagation to obtain the soil organic matter content prediction sequence corresponding to the batch data. The loss value between the prediction value output by the soil organic matter dynamic evolution prediction model and the actual soil organic matter content corresponding to the batch training data is calculated according to the preset optimization objective function. The gradient of the loss value with respect to the soil organic matter dynamic evolution prediction model is calculated through the back propagation algorithm. The model parameters are iteratively updated using the Adam optimization algorithm according to the calculated gradient and the preset learning rate of 0.001, until the preset maximum number of iterations is reached, the training is completed, and the generalization ability and prediction accuracy of the soil organic matter dynamic evolution prediction model are evaluated by the test set, and the training of the soil organic matter dynamic evolution prediction model is completed.

[0083] Further, the loss between the prediction value and the actual soil organic matter content corresponding to the batch training data according to the preset optimization objective function is specifically:

[0084] The prediction deviation of the soil organic matter content prediction value output by the soil organic matter dynamic evolution prediction model and the corresponding true soil organic matter content target value at each corresponding data point is calculated;

[0085] Each prediction deviation is quantified based on a predefined mean square error loss function to obtain an individual error value of each data point;

[0086] The individual error values are aggregated to collectively form an optimization objective function value for guiding the parameter updating of the soil organic matter dynamic evolution prediction model.

[0087] In implementation, after the soil organic matter dynamic evolution prediction model performs forward propagation on a batch of input spatiotemporal feature sequence data and farming management measures, it outputs the predicted value of soil organic matter content at each prediction time point and each spatial grid cell for each sample in the batch. For each predicted value output by the model and the actual soil organic matter content target value corresponding to it in the training data, the prediction deviation between the two is calculated, and each prediction deviation is quantified based on a predefined mean square error (MSE) loss function. Specifically, the square of the deviation is calculated to obtain the individual squared error value of the data point, and the individual squared error values of all data points in the batch are aggregated to obtain the average value, which is the mean square error loss value of the batch, representing the optimization objective function value for guiding the parameter updating of the soil organic matter dynamic evolution prediction model.

[0088] In the specific application of the present embodiment, the driving of the soil organic matter dynamic evolution prediction model for iterative optimization to determine the farming management measure dynamic adjustment setting amount that makes the predicted value of soil organic matter content achieve the target change rate is specifically:

[0089] According to the allowed adjustment range and constraint conditions of the preset farming management measure dynamic adjustment setting amount, a candidate farming management measure dynamic adjustment setting amount is generated;

[0090] For each group of candidate farming management measure dynamic adjustment setting amounts processed in the current iteration period, the following is performed:

[0091] The candidate farming management measure dynamic adjustment setting amount is input into the soil organic matter dynamic evolution prediction model to obtain the expected change trajectory of the soil organic matter content of each predetermined spatial grid cell in the agricultural production site under this group of candidate farming management measures;

[0092] According to the optimization objective function taking the achievement degree of the target change rate of the soil organic matter content in the agricultural production site as an index, and combining a predefined evaluation function, the comprehensive performance score of this group of candidate farming management measure dynamic adjustment setting amounts is calculated based on the expected change trajectory;

[0093] According to the dynamic adjustment of each group of candidate agricultural management measures, the comprehensive performance score of the set quantity is determined, and within the allowed adjustment range and constraint conditions, a new candidate agricultural management measure dynamic adjustment set quantity is generated, and returned to the input soil organic matter dynamic evolution prediction model to perform iterative updating until a preset maximum iteration number is reached.

[0094] The optimal comprehensive performance score is selected as the output of the dynamic adjustment set quantity of the agricultural management measures.

[0095] In specific implementation, the allowed adjustment range and constraint conditions of the preset dynamic adjustment set quantity of the agricultural management measures are defined according to agricultural expert knowledge. It can be understood that the allowed adjustment range of the embodiment includes a nitrogen fertilizer application amount range of 0-200 kg / ha, an organic fertilizer application amount of 0-3000 kg / ha, and an irrigation frequency of 0-5 times per growing season. The constraint condition is that the total nitrogen input does not exceed a specific environmental threshold. The embodiment uses a genetic algorithm for optimization. The GA population is initialized, where each individual represents a group of candidate dynamic adjustment set quantities of agricultural management measures. The initial population is randomly generated within the allowed range, and the population size is set to 50. Enter the iterative optimization loop, a total of 100 iterations. In each iteration cycle, for each group of candidate agricultural management measure set quantities in the population, the following operations are performed:

[0096] The group of candidate set quantities and the corresponding initial soil data and predicted period meteorological data are input into the trained soil organic matter dynamic evolution prediction model. The model outputs the expected value of the soil organic matter content in each predetermined spatial grid cell in the agricultural land at the end of the prediction period under the measure.

[0097] According to the optimization objective function with the agricultural land soil organic matter content target change rate as the index, the value of the individual is calculated.

[0098]

[0099] Wherein, is the optimization objective function, is the total number of predetermined spatial grid cells in the agricultural land, is a preset normal adjustment factor, j is the index of the spatial grid cell, is the predicted soil organic matter content of the jth spatial grid cell, is the end time of the prediction period, is the initial soil organic matter content of the jth predetermined spatial grid cell, is the soil organic matter content target change rate of the jth predetermined spatial grid cell. When calculating, for each grid cell j, first calculate the actual predicted relative change rate of the soil organic matter ​Then, the actual rate of change and the target rate of change are calculated. The absolute difference between them. This difference is multiplied by an adjustment factor. Add 1 to the end as the denominator. Take the reciprocal and apply it to all... The results of each grid cell are summed to obtain the final result. value. The larger the value, the more likely the candidate agricultural management measures are to bring the change in soil organic matter content in each grid closer to the target change rate.

[0100] Calculate this individual's score based on the evaluation function. value.

[0101]

[0102] in, As an evaluation value, , These are the weighting coefficients, For economic costs, is the environmental impact assessment value, and m is the dynamic adjustment setting of the current candidate agricultural management measures. The calculation steps are as follows: Based on the current candidate agricultural management measure m, identify all types of fertilizers to be used and their respective planned application quantities. Obtain the unit price of each planned fertilizer. For each fertilizer, multiply its planned application quantity by its unit price to obtain the cost of that fertilizer. Add up the costs of all types of fertilizers to obtain the total fertilizer cost. Based on the irrigation plan in measure m and the unit irrigation cost, estimate the total irrigation cost. Based on other agricultural operations that may be included in measure m, estimate the cost of the agricultural operations. Add up the total fertilizer cost, the total irrigation cost, and other related operation costs to obtain the total economic cost of the candidate agricultural management measure m. . The calculation steps are as follows: Based on the current candidate agricultural management measure m, calculate the total amount of nitrogen to be applied to the farmland through various fertilizers, i.e., the total nitrogen input. Based on the type of crop planted in the farmland and the expected yield level, estimate the amount of nitrogen that the crop is expected to absorb and utilize throughout its growth cycle, i.e., the estimated crop nitrogen absorption. Subtract the estimated crop nitrogen absorption from the total nitrogen input calculated in the first step to obtain the potential nitrogen surplus. Determine this nitrogen surplus: If the total nitrogen input is less than or equal to the estimated crop absorption, the nitrogen leaching risk under this measure is considered extremely low, and the environmental impact assessment value is [not specified]. Record it as zero or a very small preset positive value. If the total nitrogen input is greater than the estimated crop uptake, multiply this nitrogen surplus by an empirical leaching coefficient preset based on local soil type and rainfall conditions. The product is the environmental impact assessment value of the candidate agricultural management measure m. .

[0103] Calculate this individual's overall performance score based on the overall performance score formula.

[0104] The comprehensive performance score is calculated using the following formula:

[0105]

[0106] in, A comprehensive performance score is set for the dynamic control of candidate agricultural management measures. , These are the weighting coefficients. Calculated... and The ratio of the two values ​​is used to obtain the overall performance score. In the optimization process, the goal is to find the one that makes this... The highest score represents the dynamic adjustment setting of agricultural management measures. A high score indicates that the measure not only effectively achieves the expected target for soil organic matter, but also has relatively low economic and environmental costs.

[0107] After evaluating all individuals in the current generation, a new generation of population is generated based on the comprehensive performance scores of each group of individuals through operations such as crossover and mutation. This process then proceeds to the next iteration.

[0108] After reaching the preset maximum number of iterations, the agricultural management measure with the highest comprehensive effectiveness score is selected from all individuals and the set value is dynamically adjusted as the final optimization result.

[0109] like Figure 2 As shown, the present invention also provides another embodiment: an agricultural production area environment monitoring big data analysis system, comprising:

[0110] The data acquisition unit collects multi-dimensional spatiotemporal environmental data and crop growth status data covering agricultural production areas, and preprocesses the multi-dimensional spatiotemporal environmental data and crop growth status data to form spatiotemporal feature sequence data containing timestamps.

[0111] The model building unit constructs a soil organic matter dynamic evolution prediction model, which is configured to receive the spatiotemporal feature sequence data and corresponding agricultural management measures as input, and output the predicted value of soil organic matter content in agricultural production areas.

[0112] The optimization unit drives the soil organic matter dynamic evolution prediction model to iteratively optimize the target change rate of soil organic matter content in agricultural production areas in order to determine the dynamic adjustment setting of agricultural management measures that enable the predicted value of soil organic matter content to reach the target change rate.

[0113] The analysis unit inputs the dynamic regulation and setting quantity of the farming management measure, in combination with the real-time input of the time-space characteristic sequence data, into a soil organic matter dynamic evolution prediction model for simulation analysis, so as to predict the expected change trajectory and final state of the soil organic matter content of the agricultural land in the entire management cycle after the dynamic regulation and setting quantity of the farming management measure is implemented, and generate an operation guidance scheme for the agricultural land according to the simulation analysis result.

[0114] The above is only the preferred embodiment of the present application, and is not used to limit the present application. For those skilled in the art, the present application can have various modifications and changes. Any modification, equivalent replacement, improvement, etc. within the spirit and principle of the present application shall be included in the protection scope of the present application.

Claims

1. An agricultural origin environment monitoring big data analysis method, characterized in that, Steps of the method include: Collecting multi-dimensional spatio-temporal environment data and crop growth state data covering agricultural land, and preprocessing the multi-dimensional spatio-temporal environment data and crop growth state data to form spatio-temporal feature sequence data containing timestamps; A soil organic matter dynamic evolution prediction model is constructed, which is configured to receive the spatio-temporal feature sequence data and corresponding farming management measures as input, and output predicted values of soil organic matter content of the agricultural land; For a preset target change rate of soil organic matter content of the agricultural land, the soil organic matter dynamic evolution prediction model is driven to perform iterative optimization to determine a dynamic adjustment and control setting of the farming management measures that makes the predicted values of soil organic matter content reach the target change rate; The dynamic adjustment and control setting of the farming management measures is input into the soil organic matter dynamic evolution prediction model for simulation analysis in combination with real-time input spatio-temporal feature sequence data, to predict the expected change trajectory and final achieved state of the soil organic matter content of the agricultural land within the entire management period after implementing the dynamic adjustment and control setting of the farming management measures, and generate an operation guidance scheme for the agricultural land according to the simulation analysis result; The soil organic matter dynamic evolution prediction model is constructed as follows: The soil organic matter dynamic evolution prediction model is constructed based on a deep spatio-temporal encoder-decoder neural network architecture; The soil organic matter dynamic evolution prediction model includes an encoder and a decoder, wherein: The encoder is configured to use a convolutional long short-term memory network unit to process the spatio-temporal feature sequence data and extract deep spatio-temporal dynamic features thereof; The decoder is configured to generate the predicted values of soil organic matter content based on the deep spatio-temporal dynamic features extracted by the encoder and the corresponding farming management measures in combination with an attention mechanism; The soil organic matter dynamic evolution prediction model is constructed as follows: the spatio-temporal feature sequence data is input into the encoder for processing, and the farming management measures are input into the decoder as conditional information together with the deep spatio-temporal dynamic features output by the encoder for processing; The decoder outputs predicted values of soil organic matter content of each predetermined spatial grid unit in the agricultural land; The soil organic matter dynamic evolution prediction model is driven to perform iterative optimization to determine a dynamic adjustment and control setting of the farming management measures that makes the predicted values of soil organic matter content reach the target change rate, which is specifically as follows: According to a preset allowed adjustment range and constraint condition of the dynamic adjustment and control setting of the farming management measures, candidate dynamic adjustment and control settings of the farming management measures are generated; For each group of candidate dynamic adjustment and control settings of the farming management measures processed in the current iteration period, the following is performed: The candidate dynamic adjustment and control setting of the farming management measures is input into the soil organic matter dynamic evolution prediction model to obtain the expected change trajectory of soil organic matter content of each predetermined spatial grid unit in the agricultural land under this group of candidate dynamic adjustment and control settings of the farming management measures; According to the optimization objective function taking the degree of achievement of the target rate of change of the soil organic matter content in the agricultural production site as an index, and combining a predefined evaluation function, the comprehensive performance score of the dynamic regulation and control setting quantity of the candidate group of agricultural management measures is calculated based on the expected change trajectory; According to the comprehensive performance score of the dynamic regulation and control setting quantity of the candidate group of agricultural management measures, a new dynamic regulation and control setting quantity of the candidate group of agricultural management measures is generated under the allowed adjustment range and constraint conditions, and the input soil organic matter dynamic evolution prediction model is returned to perform iterative updating until a preset maximum number of iterations is reached; The dynamic regulation and control setting quantity of the agricultural management measures is output as the one with the optimal comprehensive performance score.

2. The agricultural geoenvironment monitoring big data analysis method according to claim 1, characterized in that, The multi-dimensional spatio-temporal environmental data and crop growth state data are preprocessed, specifically: The multi-dimensional spatio-temporal environmental data is subjected to spatial interpolation processing to generate a continuous data layer reflecting the spatial distribution characteristics of the agricultural production site; The crop growth state data is preprocessed and feature extracted to obtain time series features representing the crop growth process and its influence on soil organic matter; The continuous data layer and the time series features are uniformly aligned, gridded and fused in the time and space dimensions, and missing values are filled and features are transformed to form the spatio-temporal feature sequence data.

3. The agricultural geoenvironment monitoring big data analysis method according to claim 2, characterized in that, The training process of the soil organic matter dynamic evolution prediction model is as follows: A training set and a test set containing historical soil organic matter content, corresponding spatio-temporal feature sequence data and agricultural management measures are obtained and divided; The training set is input into the soil organic matter dynamic evolution prediction model in batches, the soil organic matter content prediction value of this batch of data is obtained through forward propagation, and the loss between the prediction value and the actual soil organic matter content corresponding to this batch of training data is calculated according to a preset optimization objective function; The gradient of the soil organic matter dynamic evolution prediction model is calculated through back propagation, and the parameters of the soil organic matter dynamic evolution prediction model are iteratively updated through an optimization algorithm; After a preset maximum number of iterations, the soil organic matter dynamic evolution prediction model is evaluated through the test set, and the training of the soil organic matter dynamic evolution prediction model is completed.

4. The agricultural geoenvironment monitoring big data analysis method according to claim 3, characterized in that, The loss between the prediction value and the actual soil organic matter content corresponding to this batch of training data is calculated according to the preset optimization objective function, specifically: The prediction deviation of the soil organic matter content prediction value output by the soil organic matter dynamic evolution prediction model and the corresponding true soil organic matter content target value at each corresponding data point is calculated; Based on a predefined mean square error loss function, each prediction deviation is quantified to obtain individual error values at each data point; The individual error values are aggregated to form an optimization objective function value for guiding the parameter update of the soil organic matter dynamic evolution prediction model.

5. The agricultural geoenvironment monitoring big data analysis method according to claim 4, characterized in that, The specific calculation formula of the optimization objective function taking the degree of achievement of the target rate of change of the soil organic matter content in the agricultural production site as an index is as follows: wherein, is an optimization objective function, is a total number of predetermined spatial grid cells within an agricultural production site, is a predetermined normal number adjustment factor, j is an index of a spatial grid cell, is a predicted soil organic matter content of the jth spatial grid cell, is an ending time of a prediction period, is an initial soil organic matter content of the jth predetermined spatial grid cell, is a target change rate of soil organic matter content of the jth predetermined spatial grid cell.

6. The agricultural geoenvironment monitoring big data analysis method according to claim 5, characterized in that, The calculation formula of the evaluation function is as follows: wherein, is the evaluation value, , are weight coefficients, respectively, is the economic cost, is the environmental impact evaluation value, and m is the current candidate farming management measure dynamic control setting quantity.

7. The agricultural geoenvironment monitoring big data analysis method according to claim 6, characterized in that, The calculation formula of the comprehensive performance score is as follows: wherein, setting a quantitative comprehensive performance score for dynamic regulation and control of the candidate farming management measures, , are weight coefficients, respectively.

8. An agricultural origin environment monitoring big data analysis system, characterized in that, It comprises: The data acquisition unit collects multi-dimensional spatio-temporal environment data and crop growth state data covering agricultural lands, and pre-processes the multi-dimensional spatio-temporal environment data and the crop growth state data to form spatio-temporal feature sequence data containing time stamps; The model construction unit constructs a soil organic matter dynamic evolution prediction model, which is configured to receive the spatio-temporal feature sequence data and corresponding farming management measures as input, and output soil organic matter content prediction values of the agricultural lands; The optimization unit drives the soil organic matter dynamic evolution prediction model to perform iterative optimization for a preset target change rate of soil organic matter content of the agricultural lands, to determine a dynamic adjustment and control setting of the farming management measures that makes the soil organic matter content prediction values achieve the target change rate; The analysis unit inputs the dynamic adjustment and control setting of the farming management measures and real-time input spatio-temporal feature sequence data into the soil organic matter dynamic evolution prediction model for simulation analysis, to predict the expected change trajectory and final achieved state of the soil organic matter content of the agricultural lands in the entire management period after implementing the dynamic adjustment and control setting of the farming management measures, and generates an operation guidance scheme for the agricultural lands according to the simulation analysis result; The soil organic matter dynamic evolution prediction model is constructed based on a deep spatio-temporal encoder-decoder neural network architecture; The soil organic matter dynamic evolution prediction model includes an encoder and a decoder, wherein: The encoder is configured to adopt a convolutional long short-term memory network unit to process the spatio-temporal feature sequence data and extract deep spatio-temporal dynamic features thereof; The decoder is configured to generate the soil organic matter content prediction values based on the deep spatio-temporal dynamic features extracted by the encoder and the corresponding farming management measures in combination with an attention mechanism; The soil organic matter dynamic evolution prediction model is constructed by inputting the spatio-temporal feature sequence data into the encoder for processing, and inputting the farming management measures as conditional information into the decoder together with the deep spatio-temporal dynamic features output by the encoder for processing; The decoder outputs soil organic matter content prediction values of each predetermined spatial grid unit in the agricultural lands; The driving of the soil organic matter dynamic evolution prediction model to perform iterative optimization to determine the dynamic adjustment and control setting of the farming management measures that makes the soil organic matter content prediction values achieve the target change rate is specifically: According to a preset allowed adjustment range and constraint condition of the dynamic adjustment and control setting of the farming management measures, candidate dynamic adjustment and control settings of the farming management measures are generated; For each group of candidate dynamic adjustment and control settings of the farming management measures processed in the current iteration period, the following operations are respectively performed: The candidate dynamic adjustment and control setting of the farming management measures is input into the soil organic matter dynamic evolution prediction model to obtain the expected change trajectory of the soil organic matter content of each predetermined spatial grid unit in the agricultural lands under this group of candidate dynamic adjustment and control settings of the farming management measures; ​ According to an optimization objective function taking the degree of achievement of the target change rate of the soil organic matter content of the agricultural production site as an index, and combining a predefined evaluation function, the comprehensive performance score of the dynamic regulation and control setting quantity of the candidate group of agricultural management measures is calculated based on the expected change trajectory; According to the comprehensive performance score of the dynamic regulation and control setting quantity of the candidate group of agricultural management measures, and within the allowed adjustment range and constraint conditions, a new dynamic regulation and control setting quantity of the candidate group of agricultural management measures is generated, and the soil organic matter dynamic evolution prediction model is returned to for iterative updating until a preset maximum number of iterations is reached; The comprehensive performance score reaching the optimum is selected as the dynamic regulation and control setting quantity of the agricultural management measures for output.

Images