A method and system for organic fertilizer application for dynamic management of nutrients in agricultural fields

Abstract

The application relates to the technical field of farmland nutrient management, and discloses an organic fertilizer application method and system for dynamic management of farmland nutrients, which comprises the following steps: obtaining original farmland environment data and standardizing the original farmland environment data, extracting features to generate a farmland feature vector, completing nutrient requirement identification and fertilizer type analysis, integrating data to generate an enhanced farmland matrix, performing multi-dimensional nutrient analysis, calculating application necessity and environmental impact value, and generating decision data after verification; constructing an enhanced fertilizer feature space based on an organic fertilizer library and the decision data, completing fertilizer matching, proportioning and application path optimization, and outputting pre-inspection report data; and collecting application data for abnormality detection and self-adaptive learning. The system comprises a processor and a memory, and the memory stores instructions which can realize the above method. Through multi-dimensional data integration and intelligent decision, the application improves the accuracy and scientificity of farmland nutrient management, and is suitable for the precise fertilization demand of modern agriculture.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of farmland nutrient management, in particular to an organic fertilizer application method and system for dynamic management of farmland nutrients. BACKGROUND

[0002] In agricultural production, farmland nutrient management is a key link to ensure the healthy growth of crops and improve yield and quality, especially for low-yield farmland nutrient monitoring, which is particularly important for improving farmland physical and chemical properties and increasing crop yield. Traditional organic fertilizer application methods are often based on experience or simple soil testing, lacking precise response to dynamic changes in farmland environment, resulting in prominent nutrient supply and demand imbalance. On the one hand, excessive fertilization may cause soil compaction, groundwater pollution and other environmental problems, increasing agricultural production costs; on the other hand, insufficient fertilization will limit crop growth and affect yield and quality. With the development of smart agriculture, how to use information technology to achieve dynamic and precise management of farmland nutrients has become a research hotspot in the current agricultural field.

[0003] In the prior art, most farmland fertilization management systems remain at the level of static data processing, and it is difficult to comprehensively consider multi-dimensional dynamic factors such as soil characteristics, crop growth stages, and climate conditions. For example, in the traditional method of extracting soil characteristics, only simple nutrient content detection can be performed, and a complete farmland feature vector cannot be constructed to fully reflect the farmland conditions; in the fertilization decision-making process, there is a lack of deep mining and multiple verification mechanisms for historical data, resulting in insufficient scientificity and reliability of the decision; in the aspect of fertilizer matching and application path optimization, it is difficult to realize the organic integration of dynamic characteristics and associated characteristics, and it is difficult to meet the individual needs of different farmland scenarios. In addition, the existing systems generally lack real-time monitoring and dynamic optimization capabilities for the fertilization process, making it difficult to respond to sudden changes in the field environment, resulting in poor adaptability of the fertilization scheme.

[0004] With the development of sensor technology, big data analysis and machine learning algorithms, the field of farmland nutrient management urgently needs an organic fertilizer application method and system that can integrate multi-source dynamic data and realize intelligent decision-making throughout the process. This method needs to solve the core problems of traditional technology such as extensive data processing, single decision-making model, and insufficient dynamic adaptability, by constructing a refined farmland feature analysis system, a multi-dimensional nutrient analysis model, and an intelligent fertilization decision-making process, to realize the precision, efficiency and sustainability of organic fertilizer application, to meet the multiple needs of modern agriculture for resource conservation, environmental friendliness and high yield and quality. SUMMARY

[0005] The present application aims to provide an organic fertilizer application method and system for dynamic management of farmland nutrients to solve the problems raised in the background.

[0006] To achieve the above object, the present application provides the following technical scheme: An organic fertilizer application method for farmland nutrient dynamic management, the method comprises:

[0007] S1, obtain farmland environment original data, and standardize the farmland environment original data to obtain standardized farmland data; based on the standardized farmland data, extract soil characteristics, calculate crop growth characteristics and climate characteristics, and generate a farmland feature vector; based on the farmland feature vector, perform nutrient demand identification and fertilizer type analysis to obtain preliminary analysis result data;

[0008] S2, integrate the preliminary analysis result data and the standardized farmland data to generate an enhanced farmland matrix; based on the enhanced farmland matrix, perform multi-dimensional nutrient analysis to obtain nutrient analysis result data; analyze the nutrient analysis result data, calculate the necessity score of organic fertilizer application and the environmental impact value, and form fertilizer decision data; multiple verification is performed on the fertilizer decision data to generate verified decision data;

[0009] S3, based on the pre-stored organic fertilizer library original information and the verified decision data, an enhanced fertilizer feature space is constructed; based on the enhanced fertilizer feature space, fertilizer matching and ratio optimization are performed to generate an optimized fertilizer selection scheme; the optimized fertilizer selection scheme is subjected to application path optimization to form an optimized application scheme; based on the optimized application scheme, multi-dimensional pre-inspection before application is performed, and finally pre-inspection report data is output.

[0010] Preferably, step S1 further comprises:

[0011] S11, obtain farmland environment original data containing soil samples, time stamps, plot identifiers and crop identifiers; convert the soil samples in the farmland environment original data into a unified coding format, remove abnormal values and redundant information, and obtain processed input data; based on the processed input data, perform data length standardization processing to generate standardized farmland data;

[0012] S12, based on the standardized farmland data, calculate soil parameter values, extract a key nutrient set, identify crop types, and generate basic feature data; based on the basic feature data, use a pre-configured growth analysis model to calculate the nutrient demand probability distribution and growth vector of the crops, and obtain growth characteristic data; obtain historical records of the farmland, and based on the historical records, extract plot state information; combine the plot state information and the growth characteristic data to form a farmland feature vector;

[0013] S13, identifying a nutrient demand type based on the farmland feature vector, calculating a demand priority score, and generating demand feature data according to a preset feature-demand mapping rule; analyzing the demand feature data, estimating a fertilizer resource demand value and a time window demand value, and obtaining resource demand data; identifying a dependency relationship between demands based on the demand feature data and the resource demand data, and constructing a dependency relationship graph; and generating preliminary analysis result data based on the nutrient demand type, the priority score, the resource demand data, and the dependency relationship graph.

[0014] Preferably, step S2 further comprises:

[0015] S21, converting the preliminary analysis result data into a feature matrix, converting the standardized farmland data into a vector representation, combining the feature matrix and the vector representation to form an initial farmland matrix, extracting relevant historical farmland records, calculating a historical information weight, and fusing the historical information weight with the initial farmland matrix to generate an enhanced farmland matrix;

[0016] S22, analyzing the enhanced farmland matrix, identifying a main nutrient target, decomposing the main nutrient target into a set of sub-targets, constructing a target dependency graph, and obtaining target structure data; calculating a resource demand vector and a target priority matrix for each sub-target, and generating target resource data; analyzing fertilizer feature demand based on the main nutrient target and historical farmland records, calculating a fertilizer importance weight, and integrating the target structure data and the target resource data into nutrient analysis result data based on the fertilizer importance weight;

[0017] S23, calculating an organic fertilizer application necessity score based on the nutrient analysis result data, evaluating an application risk value, and obtaining application evaluation data; determining an application opportunity based on the application evaluation data, and generating an application priority list; formulating an adjustment strategy based on the application priority list, and forming application strategy data; integrating the application evaluation data and the application strategy data, and generating an application path graph; calculating a confidence score based on the application path graph, and finally forming fertilization decision data;

[0018] S24, performing internal consistency verification on the fertilization decision data, and generating consistency verification data; verifying resource availability, checking technical constraints and seasonal restrictions based on the consistency verification data, and obtaining feasibility evaluation data; calculating a verification score based on the consistency verification data and the feasibility evaluation data, marking risk points, and generating optimization suggestions; and forming verified decision data based on the optimization suggestions.

[0019] Preferably, step S3 further comprises:

[0020] S31, based on the pre-stored organic fertilizer warehouse original information and the verified decision data, extract the function characteristic vector, attribute index vector and resource demand vector of each fertilizer, generate static characteristic data; calculate the historical effect matrix, average application time vector and resource consumption distribution of the fertilizer, form dynamic characteristic data; construct a fertilizer dependency graph, calculate the fertilizer compatibility matrix and fertilizer combination effect tensor, and obtain correlation characteristic data; integrate the static characteristic data, dynamic characteristic data and correlation characteristic data into an enhanced fertilizer characteristic space;

[0021] S32, based on the enhanced fertilizer characteristic space, calculate the function applicability; based on the function applicability, perform attribute constraint filtering to generate an initial candidate fertilizer set; obtain the context features of the current farmland, calculate the context correlation score; based on the context correlation score, adjust the candidate fertilizer weight and reorder the initial candidate fertilizer set to obtain an optimized candidate fertilizer set; construct a fertilizer feasible combination set, calculate the combination synergy score; based on the combination synergy score, select the optimal combination scheme from the optimized candidate fertilizer set to form the optimized fertilizer selection scheme;

[0022] S33, based on the optimized fertilizer selection scheme, construct an application dependency graph, calculate the critical path, and generate a parallel application scheme; based on the parallel application scheme, form the application sequence data; based on the application sequence data, construct a resource allocation matrix and optimize the application timing; based on the optimized application timing, construct a caching strategy to obtain resource optimization data; based on the resource optimization data, construct a failure handling strategy, an alternative scheme and a monitoring point set to generate fault tolerance mechanism data; integrate the application sequence data, resource optimization data and fault tolerance mechanism data into the optimized application scheme;

[0023] S34, perform online status check on the fertilizers in the optimized application scheme, verify resource sufficiency, test interface response, and generate availability verification data; based on the availability verification data, perform permission check, risk assessment and compliance verification to form security assessment data; based on the security assessment data, estimate response time, predict resource consumption, and calculate success probability to obtain performance prediction data; integrate the availability verification data, security assessment data and performance prediction data into pre-inspection report data.

[0024] Preferably, step S12 further comprises:

[0025] S121, read the soil segments in the standardized farmland data, calculate the parameter value, nutrient value and humidity value of each soil segment, and generate soil statistical data; based on the soil statistical data, use an analysis tool to segment the soil segments, obtain nutrient statistics, and generate nutrient characteristic data; combine the soil statistical data and nutrient characteristic data to construct complete basic feature data;

[0026] S122, obtain nutrient information in the basic feature data, calculate the environmental correlation strength of each nutrient based on the nutrient information, generate nutrient-level growth correlation data through the bidirectional correlation network of the model; based on the nutrient-level growth correlation data, construct a growth similarity matrix, calculate a key growth unit, and form growth unit data; map the growth unit data to a predefined demand space, calculate a demand probability distribution, and obtain growth feature data;

[0027] S123, obtain an operation sequence in the pre-stored farmland historical record, construct a time sequence feature vector, and generate historical farmland data; obtain and analyze the current plot state, including plot use time, operation round, and environmental continuity, to form plot state data; perform feature fusion on the growth feature data, the historical farmland data, and the plot state data, and output a final farmland feature vector.

[0028] Preferably, step S22 further comprises:

[0029] S221, read the nutrient description information in the enhanced farmland matrix, based on the nutrient description information, construct an environmental dependency tree using a model, extract core action nodes, and generate action sequence data; based on the action sequence data, identify key nutrient targets, calculate the logical relationship strength between the targets, and form target correlation data; combine the action sequence data and the target correlation data, and output nutrient target data;

[0030] S222, based on the nutrient target data, perform pattern matching using a predefined model target decomposition template library, identify decomposable sub-target units, and generate an initial sub-target set; analyze the execution conditions and completion standards of each sub-target in the initial sub-target set, construct a sub-target constraint relationship graph, and obtain target constraint data; based on the target constraint data, optimize and recombine the initial sub-target set, and output sub-target sequence data;

[0031] S223, based on the sub-target sequence data, extract the input-output dependency relationship of each sub-target, construct a data flow graph, and generate data dependency data; based on the data dependency data, analyze execution order constraints, identify parallel execution opportunities, construct a target execution network, and form execution dependency data; integrate the data dependency data and the execution dependency data, and construct a complete target dependency graph;

[0032] S224, obtain historical execution records, based on the sub-target sequence data and the historical execution records, calculate the processing complexity and resource consumption characteristics of each sub-target, generate resource feature data; based on the resource feature data, analyze the time sensitivity and priority factors of the sub-targets, construct a target scheduling weight matrix, and form scheduling feature data; combine the resource feature data, the scheduling feature data, and the target dependency graph, and output the final nutrient analysis result data.

[0033] Preferably, step S23 is further comprising:

[0034] S231, reading resource requirement information in the nutrient analysis result data, calculating resource utilization rate threshold based on the resource requirement information and pre-stored historical application records, generating resource evaluation data; based on the resource evaluation data, analyzing target completion time requirements, combining with the current system load state, calculating time pressure coefficient, forming time evaluation data; based on the resource evaluation data and the time evaluation data, calculating the organic fertilizer application necessity score matrix, outputting the application necessity data;

[0035] S232, based on the application necessity data, extracting the feature mode of the historical application failure case, constructing the risk feature vector, generating the risk mode data; based on the risk mode data, analyzing the similarity between the current target and the pre-stored historical high-risk scene, calculating the multi-dimensional risk coefficient, forming the risk evaluation data; based on the risk mode data and the risk evaluation data, constructing the risk-reward evaluation matrix, outputting the application risk data;

[0036] S233, based on the application necessity data and the application risk data, constructing the organic fertilizer application time sequence network, calculating the optimal application time window, generating the application time sequence data; based on the application time sequence data, analyzing the priority dependency relationship among fertilizers, establishing the application priority queue, forming the priority data; combining the application time sequence data and the priority data, constructing the application execution plan, outputting the application strategy data;

[0037] S234, based on the application strategy data and the pre-stored historical failure processing record, constructing the fault handling decision tree, generating the fault recovery data; based on the fault recovery data, constructing the multi-level adjustment scheme, including the alternative fertilizer chain and the simplified strategy, forming the adjustment strategy data; integrating the application strategy data, the fault recovery data and the adjustment strategy data, calculating the strategy reliability score, constructing the complete application path graph, finally outputting the fertilization decision data.

[0038] Preferably, step S32 is further comprising:

[0039] S321, based on the functional feature vector in the enhanced fertilizer feature space and the decision requirement, calculating the functional applicability score of each fertilizer through the model, generating the functional applicability data; based on the functional applicability data, using the attribute index vector to filter the constraint conditions, screening the fertilizer set meeting the attribute requirements, forming the attribute filtering data; combining the functional applicability data and the attribute filtering data, constructing the initial fertilizer list and its score matrix, outputting the initial candidate data;

[0040] S322, obtain the context features of the current farmland, including the seasonal window, the resource state and the target priority, generate context feature data; based on the context feature data, analyze the fertilizer use effect in the historical similar scene, construct a scene correlation matrix, form scene matching data; based on the context feature data and the scene matching data, calculate the context adjustment coefficient, and output the context score data;

[0041] S323, based on the initial candidate data and the context score data, adjust the fertilizer score by using a dynamic weight algorithm, generate adjustment weight data; obtain and update the fertilizer credibility score based on the historical success rate and stability index of the fertilizer, form credibility data; based on the adjustment weight data and the credibility data, reorder the preliminary fertilizer list, and output the optimized candidate data;

[0042] S324, based on the fertilizer combination feature tensor in the optimized candidate data, construct a set of feasible fertilizer combination schemes, generate combination scheme data; based on the combination scheme data, calculate the synergistic effect score of different combination schemes, including functional complementarity and attribute gain, form synergistic evaluation data; based on the synergistic evaluation data, analyze the complexity and risk factors of the combination scheme data, construct a comprehensive evaluation matrix, and obtain scheme evaluation data;

[0043] S325, based on the combination scheme data, the synergistic evaluation data and the scheme evaluation data, use a multi-objective optimization algorithm to calculate the comprehensive score of each combination scheme, generate optimization score data; based on the optimization score data, select the optimal combination scheme, construct a detailed fertilizer application sequence, form application sequence data; integrate the optimization score data and the application sequence data, and finally output the optimized fertilizer selection scheme.

[0044] Preferably, the method further comprises:

[0045] S4, collect the execution data stream and the historical monitoring data in the organic fertilizer application process, generate comprehensive monitoring data package; based on the comprehensive monitoring data package, perform abnormality detection and early warning analysis, output abnormality analysis result data; based on the abnormality analysis result data and the comprehensive monitoring data package, generate dynamic optimization strategy, form optimization strategy set data; based on the optimization strategy set data and the pre-stored historical optimization effect data, perform adaptive learning, and finally output optimization update data package.

[0046] Preferably, the application further comprises an organic fertilizer application system for dynamic management of farmland nutrients, the system comprising:

[0047] at least one processor;

[0048] and a memory in communication connection with the at least one processor;

[0049] The memory stores instructions executable by the processor, and the instructions are used to be executed by the processor to realize the organic fertilizer application method for dynamic management of farmland nutrients.

[0050] Compared with the prior art, the beneficial effects of the present application are:

[0051] At the data processing level, by acquiring original data of farmland environment and performing standardization processing, soil characteristics, crop growth characteristics and climate characteristics are extracted to generate comprehensive farmland feature vectors. This process not only realizes the unified coding and outlier removal of soil samples, but also calculates the nutrient demand probability distribution of crops through the growth analysis model, combines historical records to build a complete farmland feature vector, ensures the integrity and accuracy of the data, and provides a solid foundation for subsequent nutrient demand identification and fertilization type analysis.

[0052] At the decision analysis link, an enhanced farmland matrix is generated by integrating the preliminary analysis result data and standardized farmland data for multi-dimensional nutrient analysis. This process introduces historical information weight to realize data fusion, decomposes nutrient targets and builds target dependency graph, calculates resource demand vector and target priority matrix, and evaluates the necessity of organic fertilizer application and environmental impact value, ensuring the scientificity and reliability of the decision through multiple verification mechanisms. This multi-dimensional analysis framework can comprehensively consider the dynamic demand and potential risks of farmland, avoiding decision bias caused by single factor.

[0053] In terms of fertilizer matching and application optimization, based on the pre-stored original information of organic fertilizer library and the verified decision data, an enhanced fertilizer feature space containing static characteristics, dynamic characteristics and associated characteristics is constructed. Through function applicability calculation, attribute constraint filtering and context relevance analysis, precise matching and combination optimization of fertilizers are realized; at the same time, by constructing application dependency graph and fault tolerance mechanism, the application timing and resource allocation are optimized to ensure the efficient execution and risk resistance of the fertilization scheme. This intelligent matching and optimization process can significantly improve the fertilizer utilization rate, reduce resource waste and environmental risk.

[0054] In terms of system expandability, by collecting execution data flow and historical monitoring data during fertilization process, abnormality detection, dynamic optimization and self-adaptive learning are realized. This function enables the system to adjust the fertilization strategy in real time according to the changes of actual field environment, continuously optimize the decision model, and improve the adaptability and long-term performance of the system, providing a continuous improvement technical path for dynamic management of farmland nutrients.

[0055] The application realizes full-process automatic management from farmland feature analysis, nutrient demand identification to fertilizer matching and application optimization through a data-driven intelligent decision-making process, effectively solves the problems of extensive data processing, insufficient decision-making scientificity and poor dynamic adaptability in traditional methods, provides an efficient and reliable technical scheme for precision fertilization of modern agriculture, and has remarkable economic value and environmental benefits. BRIEF DESCRIPTION OF DRAWINGS

[0056] Fig. 1 A working principle diagram of the organic fertilizer application method for farmland nutrient dynamic management;

[0057] Fig. 2 A flowchart for farmland feature vector generation and preliminary analysis;

[0058] Fig. 3 A flowchart for enhanced farmland matrix analysis and fertilizer decision-making generation. DETAILED DESCRIPTION

[0059] The technical solutions in the embodiments of the application will be clearly and completely described below with reference to the drawings in the embodiments of the application. Obviously, the described embodiments are only part of the embodiments of the application, rather than all the embodiments of the application. Based on the embodiments in the application, all other embodiments obtained by those skilled in the art without creative labor fall within the protection scope of the application.

[0060] Please refer to Figs. 1-3 The organic fertilizer application method for farmland nutrient dynamic management according to the application specifically implements the following steps:

[0061] Step S1: First, obtain the farmland environment raw data containing soil samples, timestamps, plot identifiers, and crop identifiers, convert the soil samples to a unified encoding format, remove outliers and redundant information, obtain the processed input data, and then perform length standardization processing to generate standardized farmland data. Based on the standardized farmland data, calculate the soil parameter values, extract the key nutrient set, and identify the crop type to generate the basic feature data; through the pre-configured growth analysis model, calculate the nutrient demand probability distribution and growth vector of the crop to obtain the growth feature data; obtain the farmland historical records, extract the plot state information (such as plot use time, operation round, environmental continuity, etc.), and combine with the growth feature data to form the farmland feature vector. Further based on the farmland feature vector, identify the nutrient demand type using the pre-set feature-demand mapping rule, calculate the demand priority score, and generate the demand feature data; analyze the demand feature data, estimate the fertilizer resource demand value and time window demand value, and obtain the resource demand data; identify the dependency relationship between demands and construct a dependency graph, and finally combine the nutrient demand type, priority score, resource demand data, and dependency graph to generate preliminary analysis result data.

[0062] Step S2: Convert the preliminary analysis result data into a feature matrix, convert the standardized farmland data into a vector representation, and combine to form an initial farmland matrix; extract relevant historical farmland records and calculate the historical information weight, and fuse the initial farmland matrix to generate an enhanced farmland matrix. Analyze the enhanced farmland matrix, identify the main nutrient target, decompose it into a sub-target set, and construct a target dependency graph to obtain target structure data; calculate the resource demand vector and target priority matrix of each sub-target to generate target resource data; based on the main nutrient target and historical farmland records, analyze the fertilizer feature demand and calculate the fertilizer importance weight, integrate the target structure data and target resource data into nutrient analysis result data. Further based on the nutrient analysis result data, calculate the organic fertilizer application necessity score, evaluate the application risk value, determine the application timing, and generate the application priority list, and integrate the adjustment strategy into the fertilization decision data; perform internal consistency verification, resource availability verification, technical constraint and seasonal limit check on the fertilization decision data, calculate the verification score, mark the risk points, and generate the optimization suggestions, and finally form the verified decision data.

[0063] Step S3: Based on the pre-stored organic fertilizer library original information and the verified decision data, the functional characteristic vector, attribute index vector and resource demand vector of each fertilizer are extracted to generate static characteristic data; the historical effect matrix, average application time vector and resource consumption distribution of the fertilizer are calculated to form dynamic characteristic data; the fertilizer dependency graph is constructed to calculate the fertilizer compatibility matrix and combination effect tensor to obtain the correlation characteristic data, and the enhanced fertilizer characteristic space is integrated. Based on the enhanced fertilizer characteristic space, the functional applicability is calculated and the attribute constraint filtering is performed to generate an initial candidate fertilizer set; the context features (such as season window, resource state, target priority, etc.) of the current farmland are obtained, the context correlation score is calculated, the candidate fertilizer weight is adjusted and reordered to obtain an optimized candidate fertilizer set; the fertilizer feasible combination set is constructed, the combination synergy score is calculated, and the optimal combination scheme is selected to form the optimized fertilizer selection scheme. Further based on the optimized fertilizer selection scheme, the application dependency graph is constructed, the critical path is calculated and the parallel application scheme is generated to form the application sequence data; the resource allocation matrix is constructed to optimize the application timing, and the optimized application scheme is integrated by combining the cache strategy, failure handling strategy, alternative scheme and monitoring point set; the optimized application scheme is subjected to online state checking, resource sufficiency verification, interface response testing, permission checking, risk assessment, compliance verification, response time estimation, resource consumption prediction and success probability calculation to generate the pre-inspection report data.

[0064] The application will be further described below in conjunction with Examples 1 to 5: Example

[0065] This example further limits the specific implementation of step S1. When acquiring the original data of the farmland environment, multi-dimensional information including soil samples, time stamps, plot identifiers and crop identifiers needs to be collected completely. Among them, the soil samples cover soil samples at different depths (such as 0-20 cm plough layer and 20-40 cm subsoil layer), the time stamp is accurate to the minute level to record the sampling time, the plot identifier includes plot number and geographic location coordinates (longitude, latitude), and the crop identifier includes crop species (such as wheat and corn), variety name and growth cycle stage (such as seedling stage and tillering stage). The crop type identification when generating the basic characteristic data needs to be further refined to the variety and growth stage, for example, the image recognition technology is used to analyze the characteristics of the crop seedling stage leaf shape and color (input is the RGB image collected by the farmland camera, classified by the pre-trained ResNet model), verify the accuracy of the original crop identifier and supplement the variety information (such as “Ji Mai 22”). In addition, although the input data processing is mainly based on soil samples, the growth cycle stage information (such as “tillering stage”) in the crop identifier needs to be integrated when generating the standardized farmland data, and the time and space dimensional characteristic matrix is formed after being associated with the soil parameters.

[0066] When converting soil samples from raw farmland environmental data into a unified coding format, standardized coding is performed for physical and chemical property test parameters: Physical property parameters: soil bulk density (unit: g / cm³), porosity (%), and water content (%) are coded using floating-point number format; Chemical property parameters: pH value (dimensionless), organic matter content (g / kg), total nitrogen (g / kg), available phosphorus (mg / kg), and available potassium (mg / kg) are uniformly stored using the key-value pair format of "parameter name_value" (e.g., "pH_7.2" "organic matter_18.5").

[0067] When removing outliers using statistical methods, the criteria for outlier determination are the industry standard thresholds of the aforementioned detection parameters. For example, the reasonable range for organic matter content is set to ≤100g / kg (according to NY / T 1121.6-2006). Values ​​exceeding this range are considered outliers and are removed. Redundant information processing involves identifying repeated sampling records of the same plot at the same time point (using timestamps and plot identifiers as unique keys) and retaining the initial detection values ​​to obtain the processed input data.

[0068] When performing data length standardization based on the processed input data, a fixed-length sliding window technique is used to address the differences in crop types and sampling periods. For example, for annual crops, taking wheat with a growth cycle of 120 days as an example, the entire growth cycle is divided into a sampling window of 10 days. The soil parameters, crop growth indicators, and other data in each window are uniformly filled into a 12-dimensional vector, and any insufficient parts are completed using linear interpolation to generate standardized farmland data.

[0069] When calculating soil parameter values ​​based on standardized farmland data, physical and chemical property analyses are required for each soil sample. Specifically, soil parameters include eight indicators: physical properties (bulk density, porosity, water content) and chemical properties (pH, organic matter, nitrogen, phosphorus, and potassium content), which are filled into a vector using a fixed-length sliding window technique. For crop growth indicators (such as plant height and leaf area index), the data sources are crop identification information in the original data (such as manually entered growth monitoring data) or real-time data collected by IoT sensors. During standardization, these data are aligned with the soil parameter timestamps and merged to generate a multi-dimensional vector. For example, for wheat with a growth cycle of 120 days, the sampling window is divided into 12 sampling windows. The vector dimension of each window is "8 soil parameters + 3 growth indicators" (11 dimensions in total). The missing parts are supplemented by linear interpolation. The chemical property parameters include pH value (glass electrode method), organic matter content (potassium dichromate oxidation method), total nitrogen content (Kjeldahl nitrogen determination method), available phosphorus content (sodium bicarbonate extraction-molybdenum antimony colorimetric method), available potassium content (ammonium acetate extraction-flame photometry method), etc. The average value and standard deviation of each parameter are calculated as soil statistical characteristic values.

[0070] When extracting the key nutrient set, principal component analysis (PCA) is used to reduce the dimensionality of soil chemical property parameters, and the nutrient factors corresponding to the principal components with a cumulative contribution rate of more than 85% are selected as the key nutrients. For example, for a wheat field, after analysis, nitrogen, phosphorus, and potassium are determined as the key nutrient set, and the basic feature data containing the content and proportion of each nutrient is generated. Crop type recognition is achieved through image recognition technology, which uses a high-definition camera installed in the field to collect crop seedling leaf images, inputs a pre-trained convolutional neural network (CNN) model for classification, and outputs a crop type label (such as "winter wheat").

[0071] When calculating crop growth characteristics based on the basic feature data using a pre-configured growth analysis model, the growth characteristic data calculated by the pre-configured growth analysis model (such as the LSTM-GARCH hybrid model) includes nutrient demand probability distribution and growth vector:

[0072] Nutrient demand probability distribution: based on the crop growth period (such as the jointing stage), the demand probability of key nutrients such as nitrogen, phosphorus, and potassium (such as nitrogen fertilizer demand probability 75%) is calculated;

[0073] Growth vector: contains three indicators, including plant height growth rate (cm / day), leaf area expansion rate (cm² / day), and biomass accumulation rate (g / day), which are generated by the model through time series feature extraction of historical growth data.

[0074] When obtaining the historical records of the field, it is necessary to collect the fertilization records of the past 3-5 years (fertilizer type, application amount, application time), irrigation records (water quantity, time), crop yield data (unit area yield, quality indicators), and pest occurrence. When extracting the plot state information, the plot usage time is determined by calculating the number of consecutive planting years, the operation cycle is counted as the number of crops planted in a year (such as one crop per year, two crops per year), and the environmental continuity is analyzed by calculating the coefficient of variation of annual mean temperature and annual precipitation to evaluate the climate stability, generating plot state data containing time series features.

[0075] When combining plot state information and growth characteristic data to form a field feature vector, a multi-dimensional vector is constructed using feature splicing. For example, the vector dimensions include soil nutrient content (3 dimensions), growth indicators (2 dimensions), plot usage time (1 dimension), operation cycle (1 dimension), and environmental continuity indicators (2 dimensions), with a total dimension of 9. Each dimension of data is normalized (scaled to the [0, 1] interval) to eliminate the influence of dimensions.

[0076] When identifying nutrient demand based on farmland feature vectors, the pre-set feature-demand mapping rules are constructed by decision tree algorithm. The root node of the decision tree is soil nitrogen content, and the branch condition is whether it is lower than the critical value (such as 30 mg / kg). If it is lower, it is determined as nitrogen deficiency demand type, otherwise, it goes to the next layer node (soil phosphorus content) for judgment. After identifying the nutrient demand type (such as nitrogen and phosphorus deficiency type, potassium excess type) through this rule, the demand priority score is calculated by using the analytic hierarchy process (AHP), and the weight is assigned according to the urgency of nutrient demand in the growth stage of crops (such as the weight of nitrogen demand in the heading stage is 0.5), and the demand feature data is generated.

[0077] When analyzing demand feature data to estimate fertilizer resource demand value, based on crop nutrient absorption law (such as 3 kg of nitrogen, 1.2 kg of phosphorus and 2.5 kg of potassium are absorbed per 100 kg of wheat produced) and soil nutrient abundance and deficiency index, the fertilizer application amount is calculated in combination with the target yield. The determination of time window demand value is based on the crop growth calendar, for example, the nitrogen fertilizer best application time window is about 10 days at the jointing stage of wheat. When identifying the dependency relationship between demands, the nutrient application sequence is determined by causal relationship analysis (such as phosphorus fertilizer should be applied first to improve soil phosphorus basis, and then nitrogen fertilizer should be applied to promote growth), and the dependency relationship graph is constructed by using directed acyclic graph (DAG), the node is nutrient demand type, and the edge is dependency relationship (such as “phosphorus fertilizer application”→“nitrogen fertilizer application”).

[0078] When finally generating preliminary analysis result data, the nutrient demand type, priority score, resource demand value (such as urea application amount 15 kg / acre, superphosphate calcium application amount 20 kg / acre), time window (such as the 3rd-12th day of the jointing stage) and dependency relationship graph are integrated, stored in structured data format (such as JSON), including fields such as “demand type”, “priority score”, “fertilizer type”, “application amount”, “time window start date”, “dependency relationship” and the like, which provides a basis for subsequent data integration and fertilization decision-making.

[0079] In each link of data processing, data quality control mechanism is established. For example, three-point sampling method (block diagonal line top and middle point) is adopted during soil sample collection to ensure representativeness; abnormal value elimination log is recorded during data cleaning process for traceability; model prediction result is cross-validated by historical data to ensure the reliability of analysis result. The implementation process of the whole step S1 strictly follows the logical chain of data standardization, feature engineering and demand analysis, realizes the systematic processing from raw data to preliminary analysis result, and provides scientific basis for farmland nutrient dynamic management. EMBODIMENT

[0080] The embodiment further defines the specific implementation of step S2. In the process of integrating data and generating fertilization decisions, the preliminary analysis result data generated in step S1 needs to be structurally integrated with standardized farmland data. The preliminary analysis result data integrates demand characteristic data, resource demand data (such as fertilizer application amount), dependency relationship diagram and other information into a feature matrix, the rows of the matrix represent different farmland plots or sampling periods, and the time and space dimensions are standardized: if the plot is used as the row index, the column contains the characteristics of the plot at each sampling period (such as soil nitrogen content and demand priority at different time points), and the data is aligned by timestamp; if the sampling period is used as the row index, the column contains the characteristics of all plots at the period, and the missing values are interpolated by the adjacent plots. For example, for the demand priority data of 3 plots (A, B, C) in 4 sampling periods (T1-T4), a 3x4 matrix is constructed, the rows correspond to the plots, the columns correspond to the periods, and the values are the priority scores of each plot in the corresponding period, ensuring the structured mapping of data in different dimensions. The column dimension of the feature matrix includes “demand type”, “priority score”, “resource demand value” and the like. At the same time, the soil parameters, crop types and other information in the standardized farmland data are converted into vector representation, each vector corresponds to multi-dimensional data of a plot or time point, for example, a 10-dimensional vector containing soil pH value, organic matter content, crop growth stage and other indicators.

[0081] After combining the feature matrix and the vector representation to form the initial farmland matrix, historical farmland records need to be introduced to enhance the temporal correlation of the data. The historical information includes fertilization records (fertilizer type, application amount, time) in the past 3-5 years, soil nutrient monitoring data (nitrogen, phosphorus and potassium content detected every month), crop yield data (yield per unit area) and climate data (annual mean temperature, precipitation). The weight calculation of historical information uses the entropy weight method, and the specific steps are as follows:

[0082] Feature standardization: normalize historical soil nutrients, yield and other data to [0, 1];

[0083] Entropy calculation: calculate the information entropy of each feature dimension to reflect the degree of data dispersion;

[0084] Weight assignment: weight = 1 - entropy, the higher the weight, the greater the influence of the historical feature on the current decision. In matrix fusion, the historical information is reduced to the same dimension as the initial farmland matrix through principal component analysis (PCA), for example, the initial matrix is 10-dimensional, and the historical information is reduced to a 10-dimensional weight vector after dimension reduction, and the weighted fusion is realized through matrix point multiplication (such as enhanced farmland matrix = initial matrix x historical weight diagonal matrix), ensuring the dimension matching.

[0085] The historical information weight is calculated by the entropy weight method, and the weight value reflects the reference value of historical records to the current decision (for example, the weight of recent records is higher than that of long-term records). The historical information weight is weighted and fused with the initial farmland matrix, for example, the weight distribution is realized by matrix multiplication, to generate an enhanced farmland matrix containing historical trend characteristics, which can more comprehensively reflect the dynamic change law of farmland nutrient demand.

[0086] When performing multi-dimensional nutrient analysis on the enhanced farmland matrix, first, the main nutrient target is identified through clustering analysis or topic modeling. For example, for the enhanced farmland matrix of a certain corn planting area, the clustering result shows that 80% of the plots have nitrogen deficiency problem, and accordingly, “improving soil nitrogen content” is determined as the main nutrient target. Subsequently, the main nutrient target is decomposed into a set of sub-targets, such as “selecting appropriate nitrogen fertilizer type”, “determining the optimal application amount”, “planning the fertilization time window”, etc., and a target dependency graph is constructed through causal relationship analysis, in which the nodes are sub-targets and the edges are the logical relationships between sub-targets (such as “determining the application amount” needs to rely on the results of “soil nitrogen content detection”).

[0087] When calculating the resource demand vector of each sub-target, it is necessary to clarify the type and quantity of resources required to achieve the sub-target. For example, the resource demand vector of “selecting appropriate nitrogen fertilizer type” includes a list of available fertilizer types, the supply cycle of each fertilizer, cost budget, etc.; the target priority matrix is sorted according to the influence of the sub-target on crop yield, and the priority coefficient is determined by expert scoring method or analytic hierarchy process (such as the priority coefficient of “fertilization time window planning” is 0.4, higher than that of “fertilizer cost control” which is 0.3). Based on the main nutrient target and historical farmland records, the fertilizer feature demand (such as slow-release property, adaptability to soil pH) is analyzed, and the importance weight of fertilizer is calculated by Delphi method, for example, the importance weight of composted organic fertilizer is 0.6 due to its long-term effect, and the importance weight of chemical fertilizer is 0.4. Finally, the target structure data, target resource data and fertilizer importance weight are integrated to form nutrient analysis result data containing nutrient target hierarchy, resource demand and fertilizer adaptability.

[0088] In the generation of fertilization decision data stage, the necessity score of organic fertilizer application is calculated based on the nutrient analysis result data. The specific method is as follows: combining resource demand information (such as the difference between current soil nitrogen content and target value) and historical application records, the resource utilization threshold is calculated by linear weighting model, for example, when the soil nitrogen gap exceeds 20 mg / kg, the utilization threshold triggers the necessity score calculation. At the same time, the time pressure coefficient is calculated by analyzing the target completion time requirement (such as fertilization must be completed before the crop jointing stage) and the current load state of the system (such as the available time length of fertilization equipment), and the time pressure coefficient calculation uses the formula:

[0089]

[0090] wherein "remaining time" is the number of days from current time to jointing stage, and "target completion time threshold" is the industry standard prescribed fertilization window (e.g. 7 days before jointing stage of wheat).

[0091] The resource utilization rate threshold is calculated by quantile of historical application records:

[0092]

[0093] The application necessity score matrix element is calculated as:

[0094]

[0095] wherein, , is the weight coefficient (wherein , ) is optimized by historical decision effect. The matrix row corresponds to different fertilizer types, and the column corresponds to the plot, and the numerical value reflects the necessity priority of applying different fertilizers in each plot.

[0096] The application risk value assessment is based on the feature mode of historical application failure cases, such as crop seedling burning events caused by improper fertilization time in the past, and the risk features such as soil humidity, air temperature, and fertilizer type are extracted to construct a risk feature vector. By analyzing the matching degree of the current target and the historical high-risk scene through the cosine similarity algorithm, the multi-dimensional risk coefficient (such as climate risk coefficient, operation risk coefficient) is calculated, a risk-reward assessment matrix is constructed, and the application risk data of each fertilizer scheme is output.

[0097] When determining the application time, an organic fertilizer application timing network is constructed based on the application necessity data and risk data, and the optimal time window is calculated by the critical path method. For example, the soil humidity of a certain plot needs to reach 60%-70% before fertilization, combined with weather forecast data, the best application period is determined to be 5-7 days in the future. When establishing the application priority queue, the priority dependency relationship between fertilizers (such as base fertilizer needs to be applied before sowing, and topdressing needs to be applied after seedling stage) is analyzed, and a priority scheduling algorithm is used to generate the queue, and the base fertilizer application priority in the queue is higher than that of topdressing. Combined with the timing data and priority queue, an application execution plan containing specific operation steps, responsible person and time node is constructed, for example, "the first day to purchase organic fertilizer, the third day to deep tillage, and the fifth day to spread fertilizer".

[0098] When building the failure handling mechanism, based on the application strategy data and historical failure records, a decision tree algorithm is used to generate the failure handling process. For example, if the fertilizer application equipment fails, the first level response is to enable the backup equipment, and the second level response is to adjust the fertilization method (such as manual application instead of mechanical application). Multi-level adjustment schemes include alternative fertilizer chains (such as switching to biogas residue fertilizer when the main selected organic fertilizer is out of stock) and simplified strategies (such as reducing the number of fertilization times but ensuring the total application amount remains unchanged). By integrating application strategy data, failure recovery data and adjustment strategy data, the reliability score of the strategy is calculated through reliability theory, for example, by considering factors such as the availability of backup equipment and the stability of alternative fertilizer supply, the score range is 0-100 points, and finally a complete application path diagram is constructed, including risk points, response measures and execution processes.

[0099] In the multi-verification link, first, the internal consistency of the fertilization decision data is verified, and the logical compatibility between sub-goals is checked. For example, "apply high phosphorus fertilizer" and "adjust soil pH to alkaline" are there any conflicts? The rule engine automatically checks and generates consistency verification data. Then, the resource availability is verified, including fertilizer inventory, equipment operating status, manpower allocation, etc., for example, confirming that the selected organic fertilizer is in sufficient stock and the fertilization machinery is in good condition, generating feasibility evaluation data. Based on the consistency and feasibility data, the verification score (such as through percentage scoring) is calculated, high-risk points (such as the backup fertilizer supplier is far away) are marked, optimization suggestions (such as signing an emergency supply agreement with the backup supplier in advance) are generated, and finally the verified decision data is formed to ensure the scientificity and executability of the fertilization scheme.

[0100] The implementation process of the whole step S2 runs through the logic of data integration, target decomposition, risk assessment and verification optimization. Through historical data-driven weight calculation, multi-dimensional target decomposition model and dynamically adjusted failure handling mechanism, the systematized derivation from data to decision is realized, providing reliable basis for the precision application of organic fertilizer in farmland.

[0101] Embodiment 3:

[0102] This embodiment further limits the specific implementation of step S3. When building the enhanced fertilizer feature space, the multi-dimensional features in the original information of the organic fertilizer warehouse and the verified decision data need to be integrated. Extract the static feature data of each fertilizer, including the functional feature vector (such as the ability to provide nitrogen, the effect of improving soil aggregate structure, etc.), the attribute index vector (pH value, water content, organic matter content, etc. physical and chemical indicators) and the resource demand vector (production cycle, transportation cost, storage condition, etc.). For example, the functional feature vector of matured sheep manure can be represented as [high organic matter, medium nitrogen content, acidic], and the attribute index vector includes pH value 5.5, water content 30%, etc.

[0103] The calculation of dynamic feature data is based on historical application records, including the historical effect matrix of fertilizers (such as crop yield increase rate and soil nutrient change value in different years and different plots), the average application time vector (the average number of days from procurement to completion of application), and the resource consumption distribution (the proportion of labor, machinery, and energy consumption). Taking commodity organic fertilizer as an example, its historical effect matrix may record that the average yield increase rate in wheat fields in the past three years is 8%-12%, the average application time vector is 5 days, and the proportion of mechanical operation in resource consumption is 60%.

[0104] The association feature data is realized by constructing a fertilizer dependency graph, in which the nodes are fertilizer types and the edges represent compatibility (such as sheep manure and superphosphate can be mixed, while ash and ammonium nitrate cannot be mixed). When calculating the fertilizer compatibility matrix, expert knowledge and historical mixed application effect data are combined, for example, the compatibility score of sheep manure and superphosphate is 90 (out of 100), and the compatibility score of ash and ammonium sulfate is 20. The fertilizer combination effect tensor quantifies the synergistic effect of combined application, such as the combination of sheep manure and cake fertilizer can increase the soil organic matter by 30%.

[0105] In the fertilizer matching and ratio optimization stage, the functional applicability is first calculated based on the enhanced fertilizer feature space. Through the cosine similarity algorithm, the specific steps are as follows:

[0106] Feature vectorization: Convert the context features of the current farmland (season window, resource status, target priority) into a multi-dimensional vector. For example:

[0107] Season window: Use one-hot encoding (spring = 100, summer = 010, autumn = 001, winter = 000);

[0108] Resource status: The sufficiency of organic fertilizer inventory is quantified as a value in the interval [0, 1] (such as sufficient = 1, moderate = 0.5, insufficient = 0);

[0109] Target priority: Emergency / High / Medium / Low are encoded as one-hot vectors [1, 0, 0, 0], [0, 1, 0, 0], etc.

[0110] After merging, an 8-dimensional context feature vector such as [1, 0, 0, 1, 0, 1, 0, 0] is formed . The nutrient demand feature vector of the current farmland (such as nitrogen deficiency and organic matter target improvement) is matched with the fertilizer function feature vector to generate a functional applicability score. For example, the demand feature of a certain plot is [nitrogen deficiency 20 mg / kg, organic matter target improvement 1%], the functional applicability score of decomposed sheep manure is 85, and the functional applicability score of chemical fertilizer urea is 70.

[0111] The attribute constraint filtering link filters fertilizers according to conditions such as soil pH value and crop type. For example, when the soil pH value is 8.0 (alkaline), exclude acidic fertilizers with a pH value lower than 5.0 in the attribute index, and retain neutral or weakly alkaline fertilizers (such as grass ash with a pH value of 10.0 and compost with a pH value of 7.5), to generate an initial candidate fertilizer set. Obtain the context features of the current farmland (such as the seasonal window being the spring sowing period, the resource state being sufficient organic fertilizer inventory, and the target priority being urgent), match the current farmland with similar historical scenarios (such as fertilizer use records of similar plots in the same period in the past three years), calculate the context correlation score using a weighted cosine similarity algorithm: first, convert the seasonal window, resource state, and other features of the current farmland into a multi-dimensional vector, calculate the similarity with the historical scenario vector, and introduce a time decay factor and a historical effect weight, finally obtain the correlation score to adjust the candidate fertilizer weight. For example, the demand for base fertilizer is high during the spring sowing period, and the historical use rate of mature sheep manure as base fertilizer is 70%, so the context correlation score can be increased by 15%.

[0112] When constructing the feasible combination set of fertilizers, exclude combinations with compatibility scores below a threshold value (such as below 60 points), and generate a list of schemes that can be mixed and applied. When calculating the combination synergy score, integrate functional complementarity (such as the combination of nitrogen fertilizer and organic fertilizer to meet the short-term and long-term nutrient needs of crops) and attribute gain (such as the adjustment of soil pH value through mixed application), and use the analytic hierarchy process and fuzzy comprehensive evaluation model: functional complementarity accounts for 60%, calculate the scores of indicators such as nutrient release period matching degree through judgment matrix; attribute gain accounts for 40%, quantify the improvement degree of mixed application on soil pH, organic matter, etc., and finally integrate the weights to obtain the synergy score. For example, the combination synergy score of sheep manure and urea is 88 points, which is significantly higher than the single application score.

[0113] In the application path optimization phase, based on the optimized fertilizer selection scheme, construct the application dependency graph to determine the fertilization sequence and parallel opportunities. For example, base fertilizer application needs to be completed before sowing, and topdressing can be done after seedling stage and synchronized with irrigation, and the shortest application period is calculated by the critical path method. When constructing the resource allocation matrix, allocate manpower and mechanical resources according to time nodes, such as arranging mechanical transportation of fertilizers on the first day, manual spreading of base fertilizer on the second day, and sowing on the third day, to maximize resource utilization.

[0114] The development of the caching strategy is based on the fertilizer supply cycle and the application timing, and the long-purchase-cycle fertilizers (such as customized organic fertilizers) are reserved in advance, and the safety stock quantity is set. The fault tolerance mechanism includes failure handling strategies (such as enabling backup devices in case of mechanical failure), alternative schemes (such as switching to equivalent fertilizers when the main selected fertilizer is out of stock), and a set of monitoring points (such as real-time monitoring sensors for fertilizer quantity and soil moisture detectors), to ensure the stability of the application process.

[0115] The pre-verification before application includes online status check (verifying the inventory quantity of fertilizers, the running status of equipment), resource sufficiency verification (calculating whether the total application amount exceeds the inventory), interface response test (instruction receiving delay of the fertilizer application equipment control system); the safety assessment covers permission check (qualification audit of the operator), risk assessment (impact of weather changes on fertilizer application effect), compliance verification (whether it meets environmental protection standards); performance prediction includes response time estimation, resource consumption prediction (such as fuel consumption, man-hours), success probability calculation (based on the historical execution success rate of similar schemes). For example, through pre-verification, it is found that the inventory of a certain fertilizer is insufficient, triggering the switching of an alternative scheme, and regenerating the pre-verification report data.

[0116] The whole step S3 realizes accurate mapping from the fertilizer feature space to the specific application scheme through multi-dimensional feature modeling, dynamic weight adjustment and path optimization, and ensures the feasibility and reliability of the scheme in combination with the pre-verification mechanism, thereby providing technical support for the scientific application of farmland organic fertilizer.

[0117] Embodiment 4:

[0118] This embodiment further limits the specific implementation mode of step S12. In the process of generating the farmland feature vector, the standardized farmland data needs to be analyzed and fused in multiple dimensions. The soil segments in the standardized farmland data are read, and the physical parameter values (such as bulk density, porosity), chemical nutrient values (such as nitrogen, phosphorus, and potassium content) and humidity values of each segment are calculated, and soil statistical data is generated through descriptive statistics, including the mean, standard deviation and distribution range of each parameter. For example, the soil statistical data of a certain 0-20 cm soil layer shows that the average organic matter content is 15 g / kg, the standard deviation is 2.3 g / kg, and it presents a normal distribution characteristic.

[0119] Based on the soil statistical data, a sliding window algorithm or a density clustering algorithm is used to segment the soil segments, and the soil is divided into high-nutrient zones, medium-nutrient zones and low-nutrient zones. For example, through K-means clustering, the plot is divided into 3 nutrient regions, and the soil nutrient values in each region differ by less than 10%, the nutrient statistics of each region (such as the average nitrogen content of the high-nutrient zone is 120 mg / kg, the medium-nutrient zone is 80 mg / kg, and the low-nutrient zone is 50 mg / kg) are obtained, and the nutrient feature data is generated. Combined with the soil statistical data and the nutrient feature data, a complete basic feature data including the soil physical structure, chemical composition and spatial distribution is constructed, for example, stored in a three-dimensional matrix form, the rows represent the sampling points, the columns represent the parameter types, and the third dimension represents different soil layer depths.

[0120] After obtaining the nutrient information in the basic feature data, the environmental correlation strength of each nutrient is calculated through a bidirectional correlation network model. The model input includes soil nutrient values, temperature and humidity, light duration, and other environmental factors, and the output is a correlation coefficient matrix of nutrients and environmental factors. For example, the correlation coefficient between nitrogen fertilizer demand and temperature is 0.6, and the correlation coefficient between nitrogen fertilizer demand and precipitation is -0.3, indicating that an increase in temperature may increase nitrogen fertilizer demand, while an increase in precipitation may reduce nitrogen fertilizer effectiveness. Based on the nutrient level growth correlation data, a growth similarity matrix is constructed, and key growth units are extracted through matrix decomposition algorithms such as singular value decomposition, for example, identifying that a combination of nitrogen, phosphorus, and potassium in a ratio of 2:1:1 is a key unit affecting crop seedling growth, forming growth unit data. Map the growth unit data to the predefined demand space (such as nutrient demand intervals according to crop growth stages), calculate the demand probability distribution, for example, during the jointing stage, the demand probability of the key unit of nitrogen, phosphorus, and potassium is 75%, and obtain growth feature data.

[0121] Obtain the operation sequence (such as the time points and types of fertilization, irrigation, and tillage) in the pre-stored historical records of farmland, and construct a time series feature vector through time series analysis. For example, the operation sequence in the past three years shows that base fertilizer is applied 7 days before spring planting each year, and topdressing is carried out during the seedling stage in summer, forming a periodic time series feature. The generated historical farmland data contains operation type, time interval, cumulative number of times, and other information, for example, organic fertilizer has been applied 4 times in the past three years in a certain plot, with an average interval of 120 days.

[0122] When analyzing the current plot state, the plot use time is determined by the number of consecutive planting years, such as 5 years of continuous wheat planting; the operation round is counted as the number of planting crops within the year, such as 1 round under the one-crop-a-year system; and the environmental continuity is evaluated by calculating the coefficient of variation of climate data (annual mean temperature, precipitation) in the past three years, with a smaller coefficient of variation indicating a more stable environment. For example, the coefficient of variation of annual mean temperature is 5%, and the coefficient of variation of precipitation is 8%, indicating that the environmental continuity is high, forming plot state data.

[0123] When feature fusion is performed on growth feature data, historical farmland data, and plot state data, a multi-layer perceptron (MLP) model is used for non-linear transformation. The input layer includes growth features (10 dimensions), historical time series features (5 dimensions), and plot state features (3 dimensions), totaling 18 dimensions of data; the hidden layer extracts high-order features through a ReLU activation function, and the output layer generates a 12-dimensional farmland feature vector. For example, in the fused vector, the feature weight related to fertilization frequency is 0.3, and the feature weight related to soil nutrient spatial distribution is 0.4, highlighting key influencing factors.

[0124] In the feature fusion process, a data quality control mechanism is established. The chemical analysis of soil samples uses standard laboratory methods to ensure data accuracy; historical operation records are stored through blockchain technology to prevent data tampering; model parameters are optimized through cross-validation to ensure the representativeness of the feature vector. The entire step S12 forms a feature vector that can fully reflect the current state and historical trends of farmland through soil feature extraction, growth feature calculation, historical data integration, and multi-source feature fusion, providing key evidence for subsequent nutrient demand identification and fertilization decision-making.

[0125] Embodiment 5:

[0126] This embodiment further defines the specific implementation of step S22. When performing multi-dimensional nutrient analysis on the enhanced farmland matrix, first read the nutrient description information in the matrix, such as soil nutrient content target value, nutrient demand description of crop growth stage, etc. Based on this information, use natural language processing model to construct environment dependency tree, the root node of the tree is the core nutrient target (such as "improve soil nitrogen content"), the child nodes are the environmental factors that affect the target (such as temperature, precipitation, soil pH value) and operation nodes (such as "apply nitrogen fertilizer" "adjust soil humidity"). By recursively traversing the tree structure, extract the core action nodes (such as "select nitrogen fertilizer type" "determine fertilization amount"), generate action sequence data, which reflects the key step order to achieve the nutrient target.

[0127] Based on the action sequence data, identify the key nutrient target through semantic similarity calculation. For example, match "improve nitrogen content" in the action sequence with "wheat jointing stage nitrogen supplement" in the historical target library to determine the current main target. When calculating the logical relationship strength between targets, use association rule algorithm, for example, the support of "improve nitrogen content" and "promote tillering" is 80%, the confidence is 90%, forming target association data. Combine action sequence data and target association data to output nutrient target data containing target hierarchy and logical relationship, for example, store target name, sub-target list and association strength matrix in JSON format.

[0128] Based on the nutrient target data, a predefined target decomposition template library is called for pattern matching. The template library contains target decomposition patterns for different crop types and growth stages, such as the "balanced nutrient supply" template for corn seedling stage can be decomposed into "base fertilizer application", "seedling stage topdressing", "micronutrient supplementation" and other sub-target units. Through template matching, decomposable sub-targets are identified, and an initial sub-target set is generated, such as "improve soil nitrogen content" is decomposed into "soil nitrogen content detection", "nitrogen fertilizer type selection", "fertilization time determination" and other sub-targets. The execution conditions (such as the detection result needs to be completed within 24 hours before fertilization) and completion standards (such as the nitrogen content reaches 120 mg / kg) of each sub-target are analyzed, and a sub-target constraint relationship graph is constructed, for example, "nitrogen fertilizer type selection" needs to be executed after "soil nitrogen content detection" is completed and the results meet the standards, and target constraint data is obtained. Based on the constraint data, the initial sub-target set is optimized and reorganized, for example, the execution order of the sub-targets is adjusted to "detection -> selection -> time determination", and logical and coherent sub-target sequence data is output.

[0129] Based on the sub-target sequence data, the input-output dependency relationship of each sub-target is extracted. For example, the output of "soil nitrogen content detection" is the detection report, which is used as the input of "nitrogen fertilizer type selection", a data flow direction graph is constructed, and data dependency data is generated. Analyzing the execution order constraints, identifying sub-targets that can be executed in parallel, such as "fertilization equipment debugging" and "soil detection" have no dependency relationship and can be performed simultaneously, constructing a target execution network, and forming execution dependency data. Integrating data dependency data and execution dependency data, a complete target dependency graph is constructed, with nodes representing sub-targets and directed edges representing data flow or execution order relationships.

[0130] Historical execution records are obtained, including actual processing time, resource consumption (such as man-hours, fertilizer cost) and other data of each sub-target. Based on the sub-target sequence data and historical records, the processing complexity of each sub-target is calculated, for example, "nitrogen fertilizer type selection" needs to compare multiple supplier data, with a complexity score of 7 points (out of 10 points); resource consumption characteristics include unit cost (such as 2000 yuan per ton of organic fertilizer) and time consumption (such as 48 hours for detection), and resource characteristic data is generated. Analyzing the time sensitivity (such as "fertilization time determination" needs to be completed before the rainy season, with high sensitivity) and priority factors (such as the weight of the impact on yield is 0.6) of the sub-targets, a target scheduling weight matrix is constructed, with the priority coefficients of the sub-targets (such as 0.4, 0.5, 0.6) as the matrix elements, and scheduling characteristic data is formed. Combining resource characteristic data, scheduling characteristic data and target dependency graph, the final nutrient analysis result data is output, such as a multi-dimensional data set containing sub-target priority, resource demand, and execution network.

[0131] In the entire process, only the construction of the target scheduling weight matrix is involved, and its mathematical expression is:

[0132]

[0133] wherein, denotes the target scheduling weight matrix, is the number of sub-targets, is the priority influencing factor dimension (such as time sensitivity, resource consumption, yield impact); denotes the weight coefficient of the th sub-target under the th influencing factor, the value range is , which is determined by the analytic hierarchy process or expert scoring. The matrix is used to quantify the priority of the sub-targets, guide the resource allocation and timing arrangement of the subsequent fertilization decision, and ensure the scientificity and efficiency of the nutrient management scheme.

[0134] It should be noted that, in this article, relational terms such as first and second are used merely to distinguish one entity or action from another, without necessarily requiring or implying any such actual relationship or order between or among the entities or actions. Also, the terms "comprises", "comprising", or any other variations thereof, are intended to cover a non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements does not include only those elements but can include other elements not expressly listed or inherent to such process, method, article, or apparatus.

[0135] Although embodiments of the present application have been shown and described, it is to be understood that various modifications, substitutions, replacements and variations can be made to these embodiments without departing from the principles and spirit of the present application, and the scope of the present application is defined by the appended claims and their equivalents.

Claims

1. A method for applying organic fertilizer for dynamic management of farmland nutrients, characterized in that, Includes the following steps: S1. Obtain raw farmland environmental data and standardize it to obtain standardized farmland data; based on the standardized farmland data, extract soil characteristics, calculate crop growth characteristics and climate characteristics to generate farmland feature vectors; based on the farmland feature vectors, identify nutrient requirements and analyze fertilization types to obtain preliminary analysis results data. S2. Integrate the preliminary analysis results data and standardized farmland data to generate an enhanced farmland matrix; Based on the enhanced farmland matrix, multi-dimensional nutrient analysis is performed to obtain nutrient analysis results data; the nutrient analysis results data are analyzed to calculate the necessity score of organic fertilizer application and the environmental impact value, forming fertilization decision data; the fertilization decision data is then subjected to multiple verifications to generate verified decision data; S3. Based on the pre-stored original information of the organic fertilizer library and the decision data after verification, construct an enhanced fertilizer feature space; Based on the enhanced fertilizer feature space, fertilizer matching and ratio optimization are performed to generate an optimized fertilizer selection scheme; the application path of the optimized fertilizer selection scheme is optimized to form an optimized application scheme; based on the optimized application scheme, a multi-dimensional pre-test is performed before application, and finally the pre-test report data is output. Step S2 further includes: S21. Convert the preliminary analysis results data into a feature matrix, convert the standardized farmland data into a vector representation, and combine the feature matrix and vector representation to form an initial farmland matrix; extract relevant historical farmland records and calculate historical information weights; merge the historical information weights with the initial farmland matrix to generate an enhanced farmland matrix; S22. Analyze the enhanced farmland matrix to identify the main nutrient targets; decompose the main nutrient targets into a set of sub-targets, construct a target dependency graph, and obtain target structure data; Calculate the resource requirement vector and target priority matrix for each sub-target to generate target resource data; Based on key nutrient targets and historical farmland records, we analyze fertilizer characteristic requirements and calculate fertilizer importance weights. Based on the importance weight of fertilizers, target structure data and target resource data are integrated into nutrient analysis results data; S23. Based on the nutrient analysis results, calculate the necessity score for organic fertilizer application, assess the application risk value, and obtain application assessment data; based on the application assessment data, determine the application timing and generate an application priority list; based on the application priority list, formulate adjustment strategies and form application strategy data; integrate the application assessment data and application strategy data to generate an application path diagram; based on the application path diagram, calculate the confidence score and finally form fertilization decision data. S24. Perform internal consistency verification on the fertilization decision data and generate consistency verification data; Based on the consistency verification data, resource availability is verified, technical constraints and seasonal limitations are checked, and feasibility assessment data is obtained. Based on the consistency verification data and feasibility assessment data, the verification score is calculated, risk points are marked, and optimization suggestions are generated. Based on the optimization suggestions, validated decision data is generated. Step S3 further includes: S31. Based on the pre-stored original information of the organic fertilizer database and the post-verification decision data, extract the functional feature vector, attribute index vector, and resource demand vector of each fertilizer to generate static feature data; calculate the historical effect matrix, average application time vector, and resource consumption distribution of the fertilizer to form dynamic feature data; construct a fertilizer dependency graph, calculate the fertilizer compatibility matrix and fertilizer combination effect tensor to obtain associated feature data; integrate the static feature data, dynamic feature data, and associated feature data into an enhanced fertilizer feature space; S32. Calculate the functional applicability based on the enhanced fertilizer feature space; Based on functional suitability, attribute constraint filtering is performed to generate an initial candidate fertilizer set; the context features of the current farmland are obtained, and the context relevance score is calculated. Based on the context relevance score, the weights of candidate fertilizers are adjusted, and the initial candidate fertilizer set is reordered to obtain an optimized candidate fertilizer set; a feasible fertilizer combination set is constructed, and the combination synergy score is calculated. Based on the combined synergistic score, the optimal combination scheme is selected from the optimized candidate fertilizer set to form an optimized fertilizer selection scheme; S33. Based on the optimized fertilizer selection scheme, construct an application dependency graph, calculate the critical path, and generate a parallel application scheme. Based on the parallel application scheme, application sequence data is generated; based on the application sequence data, a resource allocation matrix is ​​constructed to optimize the application timing; based on the optimized application timing, a caching strategy is constructed to obtain resource optimization data. Based on resource optimization data, failure handling strategies, alternative solutions, and a set of monitoring points are constructed to generate fault tolerance mechanism data; application sequence data, resource optimization data, and fault tolerance mechanism data are integrated into an optimized application plan. S34. Perform online status checks on the fertilizer in the optimized application plan to verify resource sufficiency, test interface response, and generate availability verification data. Based on availability verification data, permission checks, risk assessments, and compliance verifications are performed to generate security assessment data. Based on security assessment data, response time is estimated, resource consumption is predicted, and success probability is calculated to obtain performance prediction data. The availability verification data, security assessment data, and performance prediction data are integrated into the pre-test report data.

2. The method for applying organic fertilizer for dynamic management of farmland nutrients according to claim 1, characterized in that, Step S1 further includes: S11. Obtain raw farmland environmental data containing soil samples, timestamps, plot identifiers, and crop identifiers; convert the soil samples in the raw farmland environmental data into a unified encoding format, remove outliers and redundant information, and obtain processed input data; based on the processed input data, perform data length standardization processing to generate standardized farmland data. S12. Based on standardized farmland data, calculate soil parameter values, extract key nutrient sets, identify crop types, and generate basic feature data; based on the basic feature data, use a pre-configured growth analysis model to calculate the nutrient requirement probability distribution and growth vector of crops to obtain growth feature data; obtain historical records of farmland, and extract plot status information based on historical records; combine plot status information with growth feature data to form farmland feature vectors. S13. Based on farmland feature vectors, identify nutrient demand types using preset feature-demand mapping rules, calculate demand priority scores, and generate demand feature data; analyze the demand feature data, estimate fertilizer resource demand values ​​and time window demand values, and obtain resource demand data; based on the demand feature data and resource demand data, identify the dependencies between demands and construct a dependency graph; based on nutrient demand types, priority scores, resource demand data, and dependency graph, generate preliminary analysis results data.

3. The method for applying organic fertilizer for dynamic management of farmland nutrients according to claim 1, characterized in that, Step S12 further includes: S121. Read soil segments from standardized farmland data, calculate parameter values, nutrient values, and moisture values ​​for each soil segment, and generate soil statistical data; based on the soil statistical data, use analysis tools to segment the soil segments, obtain nutrient statistics, and generate nutrient characteristic data; combine the soil statistical data and nutrient characteristic data to construct complete basic characteristic data; S122. Obtain nutrient information from the basic feature data. Based on the nutrient information, calculate the environmental correlation strength of each nutrient through the bidirectional correlation network of the model to generate nutrient-level growth correlation data. Based on the nutrient-level growth correlation data, construct a growth similarity matrix, calculate key growth units, and form growth unit data. Map the growth unit data to a predefined demand space, calculate the demand probability distribution, and obtain growth feature data. S123. Obtain the operation sequence from the pre-stored farmland history records, construct the time-series feature vector, and generate historical farmland data; obtain and analyze the current plot status, including plot usage time, operation rounds, and environmental coherence, to form plot status data; fuse the growth feature data, historical farmland data, and plot status data to output the final farmland feature vector.

4. The method for applying organic fertilizer for dynamic management of farmland nutrients according to claim 1, characterized in that, Step S22 further includes: S221. Read the nutrient description information in the enhanced farmland matrix. Based on the nutrient description information, construct an environment dependency tree using a model, extract core action nodes, and generate action sequence data. Based on action sequence data, key nutrient targets are identified, the strength of logical relationships between targets is calculated, and target association data is formed; the action sequence data and target association data are combined to output nutrient target data. S222. Based on nutrient target data, pattern matching is performed using a predefined model target decomposition template library to identify decomposable sub-target units and generate an initial sub-target set; the execution conditions and completion criteria of each sub-target in the initial sub-target set are analyzed to construct a sub-target constraint relationship graph and obtain target constraint data. Based on the target constraint data, the initial sub-target set is optimized and reorganized to output the sub-target sequence data; S223. Based on the sub-target sequence data, extract the input-output dependency relationship of each sub-target, construct a data flow graph, and generate data dependency data; Based on data dependency data, we analyze execution order constraints, identify parallel execution opportunities, construct a target execution network, and form execution dependency data; we then integrate the data dependency data and execution dependency data to construct a complete target dependency graph. S224. Obtain historical execution records. Based on the sub-target sequence data and historical execution records, calculate the processing complexity and resource consumption characteristics of each sub-target and generate resource characteristic data. Based on resource characteristic data, the time sensitivity and priority factors of sub-objectives are analyzed, and a target scheduling weight matrix is ​​constructed to form scheduling characteristic data. The resource characteristic data, scheduling characteristic data and target dependency graph are combined to output the final nutrient analysis results data.

5. The method for applying organic fertilizer for dynamic management of farmland nutrients according to claim 1, characterized in that, Step S23 further includes: S231. Read the resource demand information from the nutrient analysis results data, and calculate the resource utilization rate threshold based on the resource demand information and the pre-stored historical application records to generate resource assessment data. Based on resource assessment data, the target completion time requirement is analyzed, and combined with the current system load status, the time pressure coefficient is calculated to form time assessment data; Based on resource assessment data and time assessment data, calculate the organic fertilizer application necessity score matrix and output application necessity data; S232. Based on application necessity data, extract feature patterns from historical application failure cases, construct risk feature vectors, and generate risk pattern data. Based on risk pattern data, the similarity between the current target and pre-stored historical high-risk scenarios is analyzed, multi-dimensional risk coefficients are calculated, and risk assessment data is formed. Based on risk pattern data and risk assessment data, a risk-return assessment matrix is ​​constructed, and application risk data is output. S233. Based on application necessity data and application risk data, construct an organic fertilizer application time series network, calculate the optimal application time window, and generate application time series data; based on the application time series data, analyze the priority dependence relationship between fertilizers, establish an application priority queue, and form priority data; combine the application time series data and priority data to construct an application execution plan and output application strategy data. S234. Based on the application strategy data and pre-stored historical failure handling records, construct a fault handling decision tree and generate fault recovery data; Based on fault recovery data, a multi-level adjustment scheme is constructed, including alternative fertilizer chains and simplified strategies, forming adjustment strategy data; By integrating application strategy data, fault recovery data, and adjustment strategy data, a strategy reliability score is calculated, a complete application path map is constructed, and fertilization decision data is finally output.

6. The method for applying organic fertilizer for dynamic management of farmland nutrients according to claim 1, characterized in that, Step S32 further includes: S321. Based on the functional feature vector and decision requirements in the enhanced fertilizer feature space, calculate the functional applicability score of each fertilizer through the model to generate functional applicability data; based on the functional applicability data, use attribute index vectors to filter the constraint conditions, select the set of fertilizers that meet the attribute requirements, and form attribute filtering data. By combining functional applicability data and attribute filtering data, a preliminary fertilizer list and its scoring matrix are constructed, and the initial candidate data is output. S322. Obtain the contextual features of the current farmland, including seasonal window, resource status and target priority, and generate contextual feature data; Based on contextual feature data, analyze the fertilizer application effects under similar historical scenarios, construct a scenario correlation matrix, and form scenario matching data; Based on context feature data and scene matching data, calculate the context adjustment coefficient and output context score data; S323. Based on the initial candidate data and contextual scoring data, the fertilizer score is adjusted using a dynamic weighting algorithm to generate adjusted weighting data; the fertilizer credibility score is updated based on the fertilizer's historical success rate and stability indicators to form credibility data. Based on the adjusted weight data and confidence data, the initial fertilizer list is reordered, and optimized candidate data is output. S324. Based on the fertilizer combination feature tensor in the optimized candidate data, construct a set of feasible fertilizer combination schemes and generate combination scheme data; Based on the combined scheme data, the synergy score of different combined schemes is calculated, including functional complementarity and attribute gain, to form synergy evaluation data; Based on collaborative evaluation data, the complexity and risk factors of combined scheme data are analyzed, a comprehensive evaluation matrix is ​​constructed, and scheme evaluation data is obtained. S325. Based on the combined scheme data, collaborative evaluation data, and scheme evaluation data, a multi-objective optimization algorithm is used to calculate the comprehensive score of each combined scheme and generate optimization score data. Based on the optimized scoring data, the optimal combination scheme is selected, a detailed fertilizer application sequence is constructed, and application sequence data is formed. The optimized scoring data and application sequence data are integrated to finally output the optimized fertilizer selection scheme.

7. The method for applying organic fertilizer for dynamic management of farmland nutrients according to claim 6, characterized in that, Also includes: S4. Collect execution data streams and historical monitoring data during the organic fertilizer application process to generate a comprehensive monitoring data package; Based on comprehensive monitoring data packets, anomaly detection and early warning analysis are performed, and anomaly analysis results are output. Based on anomaly analysis results and comprehensive monitoring data, dynamic optimization strategies are generated, forming an optimization strategy set data. Based on the optimized strategy set data and pre-stored historical optimization effect data, adaptive learning is performed, and finally an optimized update data package is output.

8. An organic fertilizer application system for dynamic management of farmland nutrients, characterized in that, include: At least one processor; And, a memory communicatively connected to at least one of the processors; The memory stores instructions that can be executed by the processor to implement the organic fertilizer application method for dynamic management of farmland nutrients as described in any one of claims 1-7.

Images