Intelligent agricultural planting management method and system

Abstract

The invention discloses an intelligent agricultural planting management method and system, and belongs to the technical field of intelligent agriculture, and the method comprises the steps of data layer construction, water and fertilizer prediction layer construction, planting optimization layer construction and agricultural planting management. The space-time adaptive water and fertilizer prediction layer construction method combining the crop physiological load and the crop growth stage is adopted, the water demand and nitrogen, phosphorus and potassium nutrient demands of crops in the future stage are accurately predicted, the crop growth state and environment changes are dynamically responded, the water and fertilizer demand difference in different growth stages is considered, and the water and fertilizer prediction efficiency is improved. The accuracy and applicability of prediction are improved through time-space fusion, and accurate fertilization can be realized; a planting optimization layer construction method combining digital twinborn and multi-objective optimization is adopted, a water and fertilizer management strategy is subjected to refined and dynamic optimization under the condition of considering a crop growth stage, a soil nutrient condition and a future weather condition, and an optimal management strategy capable of improving the yield, saving water and fertilizer resources and reducing the nitrogen leaching risk is generated.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of intelligent agriculture, and particularly relates to an intelligent agricultural planting management method and system. BACKGROUND

[0002] The intelligent agricultural planting management utilizes multi-source agricultural data and intelligent decision-making technology to perform whole-process management on crops from monitoring and prediction to water and fertilizer regulation and control, realizes dynamic regulation and control of agricultural production, aims to improve crop yield and quality, improve water and fertilizer resource utilization efficiency, and reduce environmental risks, so as to promote the development of agricultural production in the direction of precision, efficiency and green.

[0003] However, in the existing intelligent agricultural planting management process, the traditional water and fertilizer management method cannot accurately reflect the actual physiological needs of crops, and the differences in water and nutrient requirements at different growth stages are ignored, resulting in uneven water and fertilizer application, resource waste, and unstable crop yield and quality; the traditional water and fertilizer management strategy cannot simultaneously consider crop yield, resource utilization efficiency and environmental protection, and cannot cope with future meteorological uncertainty and dynamic changes in crop growth state. SUMMARY

[0004] In view of the above problems, in order to overcome the defects of the prior art, the application provides an intelligent agricultural planting management method and system, which creatively adopts a water and fertilizer prediction layer construction method combining crop physiological load and crop growth stage, realizes accurate prediction of future water requirement and nitrogen, phosphorus and potassium nutrient requirements of crops, dynamically responds to crop growth state and environmental changes, takes into account the differences in water and fertilizer requirements at different growth stages, and improves the accuracy and applicability of prediction through spatio-temporal fusion, thereby helping to realize precise fertilization; and creatively adopts a planting optimization layer construction method combining digital twinning and multi-objective optimization, realizes fine and dynamic optimization of water and fertilizer management strategies considering crop growth stage, soil nutrient status and future weather conditions, thereby generating an optimal executable management strategy that can improve yield, save water and fertilizer resources, and reduce the risk of nitrogen leaching.

[0005] The technical scheme adopted by the application is as follows: the application provides an intelligent agricultural planting management method, which comprises the following steps:

[0006] Step S1: data layer construction;

[0007] Step S2: water and fertilizer prediction layer construction;

[0008] Step S3: planting optimization layer construction;

[0009] Step S4: agricultural planting management.

[0010] Further, in step S1, the data layer is constructed, specifically by collecting multi-channel remote sensing image data, field environment monitoring data, future weather forecast and crop growth information, and performing preprocessing, spatio-temporal alignment and standardization processing to obtain a spatio-temporal fusion data set.

[0011] Further, in step S2, the water and fertilizer prediction layer is constructed for predicting the water and nutrient requirements of crops, specifically based on the spatio-temporal fusion data set, a water and fertilizer prediction layer construction method combining crop physiological load and crop growth stage is adopted to obtain water and fertilizer requirement prediction information, including the following steps:

[0012] Step S21: Input sequence construction, specifically by calculating the crop physiological load factor and combining multi-channel remote sensing image data, environmental time series data and future weather forecast to construct spatial input sequence, time input sequence and weather forecast sequence;

[0013] The crop physiological load factor is used to represent the comprehensive degree of water stress, heat stress and evapotranspiration pressure of crops under current environmental conditions;

[0014] Step S22: Parallel feature extraction, for parallel extraction of spatial features and time series features, specifically by constructing a double-branch feature extraction network to parallelly extract spatial features and time series features, the double-branch feature extraction network includes a spatial branch and a time series branch; the spatial branch extracts convolutional features from the spatial input sequence through a lightweight convolutional neural network to obtain spatial features; the time series branch captures the trend of the time input sequence through a double-layer recurrent neural network to obtain a time series hidden state, and then splices the time series hidden state with the weather forecast sequence, and then maps through a fully connected layer to obtain a time series feature;

[0015] Step S23: Spatio-temporal consistency fusion, for generating globally spatio-temporally consistent fusion features, specifically by calculating the consistency of spatial features and time series features to construct a spatio-temporal adaptability coefficient, then concatenating the context vector composed of the one-hot encoding of the crop growth stage and the crop physiological load factor through a gating network to generate dynamic weights, and performing adaptive weight correction according to the spatio-temporal adaptability coefficient, finally weighting and fusing the spatial features and the time series features to obtain spatio-temporal fusion features;

[0016] Step S24: Multi-target prediction output, specifically by using a multi-task learning model to perform multi-output regression prediction on the spatio-temporal fusion features, and introducing agricultural proportion constraints and spatio-temporal consistency constraints to construct a comprehensive loss function to correct the prediction results to obtain water and fertilizer requirement prediction information;

[0017] The agricultural proportion constraint is used to ensure that the predicted values of nitrogen, phosphorus and potassium satisfy the agricultural empirical proportion relationship;

[0018] The spatio-temporal consistency constraint is used to maintain consistency between the spatial feature and the time sequence feature in the spatio-temporal fusion feature.

[0019] Further, in step S3, the planting optimization layer is constructed to generate an optimal water and fertilizer management strategy considering yield, resource and environmental benefits, specifically, according to the water and fertilizer prediction information and the spatio-temporal fusion data set, an optimal water and fertilizer management strategy is obtained by using a planting optimization layer construction method combining digital twinning and multi-objective optimization, including the following steps:

[0020] Step S31: decision space initialization, used to construct an optimal decision space for controlling water and fertilizer in the presence of meteorological forecast errors, specifically, by constructing a water and fertilizer decision vector, setting a dynamic search interval, and introducing a weather scenario set to construct a multi-objective optimization function, setting multiple constraint conditions, and obtaining a multi-objective decision space model;

[0021] The water and fertilizer decision vector includes irrigation amount and nitrogen, phosphorus and potassium application amount;

[0022] The setting of the dynamic search interval constructs a feasible search range centered on the predicted value and adaptively adjusted with the growth stage;

[0023] The multi-objective optimization function is used to evaluate yield improvement, water resource utilization efficiency and nitrogen leaching risk simultaneously in the presence of meteorological forecast errors, including a minimum negative yield objective function, a minimum water consumption objective function and a minimum nitrogen leaching risk objective function;

[0024] The setting of multiple constraint conditions specifically constrains soil moisture content, nutrient allocation and equipment capacity;

[0025] Step S32: digital twinning deduction, used to deduce the crop growth process according to the candidate water and fertilizer management strategy, specifically, by constructing a digital twin including a crop growth mechanism model, a data agent model and a physical consistency correction module, receiving field environment monitoring data, future weather forecast and candidate water and fertilizer management strategy as input, performing time sequence deduction on yield and nutrient residual, and performing constraint correction through the physical consistency correction module, obtaining crop growth simulation results;

[0026] The crop growth mechanism model is used to simulate water transport, evapotranspiration process, soil moisture content change and potential growth state of crops;

[0027] The data agent model is a multi-output regression model based on machine learning, which performs preliminary prediction of yield and soil nutrient residual;

[0028] Step S33: multi-objective optimization solving, for obtaining a stable executable water and fertilizer management strategy in the presence of multiple constraint conditions and meteorological prediction errors, specifically, combining the decision space initialization and the digital twin deduction, on the basis of a multi-objective evolutionary algorithm, introducing a dynamic penalty mechanism, a robust fitness mechanism and a constraint-priority double-layer selection mechanism, adaptively optimizing the candidate water and fertilizer management strategy, and obtaining a candidate water and fertilizer management strategy set;

[0029] The dynamic penalty mechanism specifically adopts an exponential penalty factor that increases with the number of iterations to gradually suppress constraint violation behavior.

[0030] The robust fitness mechanism constructs a robust fitness by introducing a set of weather scenarios to evaluate the robustness of the candidate water and fertilizer management strategy under multiple weather scenarios.

[0031] The constraint-priority double-layer selection mechanism specifically selects water and fertilizer management strategies that fully satisfy multiple constraint conditions in the non-dominated sorting process, and then adaptively sorts and selects candidate solutions according to the comprehensive fitness, realizing the iterative evolution of the candidate water and fertilizer management strategy set.

[0032] Step S34: preference interactive decision making, for generating a final water and fertilizer management strategy from the strategy set according to user preferences, specifically, by obtaining user preference weights for yield, water saving and environmental risk, the candidate water and fertilizer management strategy set is weighted and reordered to select the optimal water and fertilizer management strategy that meets the preference constraints.

[0033] Further, in step S4, the agricultural planting management, specifically, by executing steps S1 to S3, integrates the data layer, the water and fertilizer prediction layer and the planting optimization layer to construct a water and fertilizer optimization model for agricultural planting management, performs water and fertilizer optimization, generates field-level control instructions according to the optimal water and fertilizer management strategy, issues them to the water and fertilizer integrated control system, and performs effect monitoring to generate a planting management feedback report.

[0034] The intelligent agricultural planting management system provided by the application comprises a data establishment module, a water and fertilizer prediction module, a planting optimization module and an agricultural planting management module.

[0035] The data establishment module is used for data layer construction, and through data layer construction, a spatio-temporal fusion data set is obtained and sent to the water and fertilizer prediction module and the planting optimization module.

[0036] The water and fertilizer prediction module is used for water and fertilizer prediction layer construction, and through water and fertilizer prediction layer construction, water and fertilizer demand prediction information is obtained and sent to the planting optimization module.

[0037] The planting optimization module is configured to construct a planting optimization layer, and obtain an optimal water and fertilizer management strategy through the construction of the planting optimization layer, and send the optimal water and fertilizer management strategy to the agricultural planting management module.

[0038] The agricultural planting management module is configured to perform agricultural planting management, and obtain a field-level control instruction and a planting management feedback report.

[0039] The above scheme has the following beneficial effects:

[0040] (1) In the existing intelligent agricultural planting management process, the traditional water and fertilizer management method cannot accurately reflect the actual physiological needs of crops, and ignores the differences in water and nutrient requirements at different growth stages, resulting in uneven water and fertilizer application, resource waste, and unstable crop yield and quality. The scheme creatively adopts a water and fertilizer prediction layer construction method combining crop physiological load and crop growth stage, realizes accurate prediction of future water requirement and nitrogen, phosphorus and potassium nutrient requirements of crops, dynamically responds to crop growth state and environmental changes, takes into account the differences in water and fertilizer requirements at different growth stages, and improves the accuracy and applicability of prediction through spatiotemporal fusion, thereby helping to realize precision fertilization.

[0041] (2) In the existing intelligent agricultural planting management process, the traditional water and fertilizer management strategy cannot simultaneously consider crop yield, resource utilization efficiency and environmental protection, and is difficult to cope with future meteorological uncertainty and dynamic changes in crop growth state. The scheme creatively adopts a planting optimization layer construction method combining digital twin and multi-objective optimization, realizes fine and dynamic optimization of water and fertilizer management strategy considering crop growth stage, soil nutrient status and future weather conditions, and generates an optimal executable management strategy that can improve yield, save water and fertilizer resources, and reduce the risk of nitrogen leaching. BRIEF DESCRIPTION OF DRAWINGS

[0042] Figure 1 A flowchart of an intelligent agricultural planting management method provided by the present application is shown.

[0043] Figure 2 A module diagram of an intelligent agricultural planting management system provided by the present application is shown.

[0044] Figure 3 A flowchart of step S2 is shown.

[0045] Figure 4 A flowchart of step S3 is shown.

[0046] The accompanying drawings are included to provide a further understanding of the present application, and constitute a part of the specification, which together with the embodiments of the present application, serve to explain the present application, and do not constitute a limitation of the present application. DETAILED DESCRIPTION

[0047] The technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all the embodiments of the present application. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative work fall within the scope of the present application.

[0048] In the description of the present application, it should be understood that the terms "upper", "lower", "front", "back", "left", "right", "top", "bottom", "inner", "outer" and the like indicate the orientation or positional relationship based on the orientation or positional relationship shown in the drawings, and are only for the purpose of facilitating the description of the present application and simplifying the description, and do not indicate or imply that the device or element referred to must have a particular orientation, be constructed and operated in a particular orientation, and therefore cannot be understood as a limitation on the present application.

[0049] Embodiment one, refer to Figure 1 The present application provides an intelligent agricultural planting management method, which comprises the following steps:

[0050] Step S1: data layer construction;

[0051] Step S2: water and fertilizer prediction layer construction;

[0052] Step S3: planting optimization layer construction;

[0053] Step S4: agricultural planting management.

[0054] Embodiment two, refer to Figure 1 The embodiment is based on the above-mentioned embodiment, in step S1, the data layer construction is used to provide unified and standardized perception data base for the water and fertilizer prediction layer and the planting optimization layer, specifically, by collecting multi-channel remote sensing image data, field environment monitoring data, future weather forecast and crop growth information, and performing preprocessing, space-time alignment and standardization processing, obtaining a space-time fusion data set;

[0055] The multi-channel remote sensing image data includes normalized vegetation index, red edge vegetation index and crop water stress index; the field environment monitoring data includes soil moisture content, soil conductivity, air temperature, air humidity and solar radiation intensity in the past 7 days; the future weather forecast includes rainfall prediction, reference evapotranspiration prediction and air temperature prediction in the next 24 hours, 48 hours and 72 hours; the crop growth information includes crop variety, crop growth stage and planting density;

[0056] The preprocessing includes data cleaning, abnormality elimination and missing value repair.

[0057] Embodiment three, refer to Figure 1 and Figure 3 , based on the above embodiment, in step S2, the water and fertilizer prediction layer is constructed for predicting the water and nutrient demand of crops, specifically, according to the spatio-temporal fusion data set, a water and fertilizer prediction layer construction method combining crop physiological load and crop growth stage is adopted to obtain water and fertilizer demand prediction information, including the following steps:

[0058] Step S21: input sequence construction, specifically, by calculating the crop physiological load factor, combining multi-channel remote sensing image data, environmental time series data and future weather forecast, constructing spatial input sequence, time input sequence and weather forecast sequence;

[0059] The crop physiological load factor is used to represent the comprehensive degree of water stress, heat stress and evapotranspiration pressure of crops under current environmental conditions, and the calculation formula is:

[0060] ;

[0061] In the formula, PSI is the crop physiological load factor, CWSI is the crop water stress index, which is used to represent the degree of crop water shortage, Tem is the heat stress weight, which is used to adjust the influence degree of crop temperature stress, Tem air is the air temperature, Tem opt is the optimal temperature of crops, VPD is the saturated water vapor pressure difference, which is used to represent the atmospheric dryness and crop evapotranspiration potential;

[0062] The spatial input sequence is composed of multi-channel remote sensing image snapshots of normalized vegetation index, red edge vegetation index, crop water stress index and crop physiological load factor;

[0063] The time input sequence is composed of environmental change sequence of soil moisture, soil conductivity, air temperature, air humidity, solar radiation intensity and crop physiological load factor in the past 7 days;

[0064] The weather forecast sequence includes rainfall prediction, reference evapotranspiration prediction and air temperature prediction in the next 24 hours, 48 hours and 72 hours;

[0065] Step S22: parallel feature extraction, for extracting spatial features and time series features in parallel, specifically, constructing a double-branch feature extraction network to extract spatial features and time series features in parallel, the double-branch feature extraction network includes a spatial branch and a time series branch; the spatial branch extracts convolutional features of the spatial input sequence through a lightweight convolutional neural network to obtain spatial features; the time series branch captures the change trend of the time input sequence through a double-layer recurrent neural network to obtain time series hidden states, and splices the time series hidden states with the meteorological prediction sequence, and then maps through a fully connected layer to obtain time series features;

[0066] Preferably, the lightweight convolutional neural network adopts an EfficientNet-B0 architecture, including 5 convolutional layers and 3 fully connected layers, and the convolution kernel size is 3x3, and a ReLU activation function is used after each convolutional layer, and the output features are mapped through a fully connected layer;

[0067] Preferably, the double-layer recurrent neural network is a double-layer LSTM, each layer has 128 hidden units, and the random inactivation rate is 0.2;

[0068] The calculation formula of the time series branch is:

[0069] ;

[0070] ;

[0071] In the formula, h t is the time series hidden state of the last time step of the double-layer LSTM, T is the total length of the time step, LSTM(·) is a double-layer LSTM operation function, A temporal is the time input sequence, V temporal is the time series feature, w f is the mapping weight, [·||·] is the splicing operator, A futre is the meteorological prediction sequence, b f is the mapping bias term;

[0072] Step S23: spatio-temporal consistency fusion, for generating globally spatio-temporally consistent fusion features, specifically, calculating the consistency of the spatial features and the time series features, constructing a spatio-temporal adaptability coefficient, then splicing the context vector composed of the crop growth stage one-hot encoding and the crop physiological load factor through a gating network to generate a dynamic weight, and performing adaptive weight correction according to the spatio-temporal adaptability coefficient, finally weighting and fusing the spatial features and the time series features to obtain spatio-temporal fusion features;

[0073] The calculation formula for constructing the spatio-temporal adaptability coefficient is:

[0074] ;

[0075] wherein, is the spatio-temporal adaptability coefficient, exp(·) is the natural exponential function, ||·||2 is the L2 norm, is used to measure the difference between the spatial features and the temporal features, V spatial is the spatial feature;

[0076] The dynamic weight includes a spatial dynamic weight and a temporal dynamic weight, which are used to adjust the relative importance of the spatial features and the temporal features in the fusion process, and the calculation formula is:

[0077] ;

[0078] ;

[0079] wherein, C is the context vector, Stage OneHot is the one-hot encoding of the crop growth stage, is the spatial dynamic weight, Sig(·) is the Sigmoid activation function, is the first gating weight, is the first gating bias, is the second gating weight, is the temporal dynamic weight, is the second gating bias;

[0080] The calculation formula of the adaptive weight correction is:

[0081] ;

[0082] wherein, is the corrected spatial dynamic weight, is the corrected temporal dynamic weight;

[0083] The calculation formula of the weighted fusion of the spatial features and the temporal features is:

[0084] ;

[0085] wherein, V fused is the spatio-temporal fusion feature;

[0086] Step S24: multi-target prediction output, used to obtain the water and fertilizer demand prediction results in the next 24-72 hours, specifically, a multi-task learning model is used to perform multi-output regression prediction on the spatio-temporal fusion feature, an agricultural proportion constraint and a spatio-temporal consistency constraint are introduced to construct a comprehensive loss function, the prediction results are corrected to obtain water and fertilizer demand prediction information, and the water and fertilizer demand prediction information includes crop water demand and nitrogen, phosphorus and potassium nutrient demand in the next 24-72 hours;

[0087] The multi-task learning model adopts a structure form of a shared fully connected encoding layer and multiple prediction branches, first learns overall expression of the spatio-temporal fusion features through the shared fully connected encoding layer, extracts shared features, then inputs the shared features into independent water demand prediction branches, nitrogen prediction branches, phosphorus prediction branches and potassium prediction branches for multiple output regression prediction, and outputs non-negative constraint prediction results, each branch is composed of a fully connected layer with a ReLU activation function and a linear output layer;

[0088] The agricultural proportion constraint is used to ensure that the nitrogen, phosphorus and potassium prediction values meet the agricultural experience proportion relationship, prevent the prediction results from deviating from the actual agricultural management specification, and the calculation formula is:

[0089] ;

[0090] In the formula, L ratio is the agricultural proportion constraint, is the proportion constraint weight, is the predicted nitrogen nutrient requirement, is the predicted phosphorus nutrient requirement, is a small constant for avoiding the case that the denominator is zero, r NP is the nitrogen-phosphorus experience proportion, is the predicted potassium nutrient requirement, r NK is the nitrogen-potassium experience proportion;

[0091] The spatio-temporal consistency constraint is used to maintain the consistency of the spatial features and the time sequence features in the spatio-temporal fusion features, and reduce the influence of feature conflicts on the prediction results, and the calculation formula is:

[0092] ;

[0093] In the formula, is the consistency constraint weight;

[0094] The calculation formula of the comprehensive loss function is:

[0095] ;

[0096] In the formula, L final is the comprehensive loss function, L MSE is a multi-objective mean square error loss function.

[0097] By performing the above operation, in view of the technical problems in the existing intelligent agricultural planting management process that the traditional water and fertilizer management method cannot accurately reflect the actual physiological needs of crops, ignores the differences in water and nutrient requirements at different growth stages, leading to uneven water and fertilizer application, resource waste and unstable crop yield and quality, the scheme creatively adopts the water and fertilizer prediction layer construction method combining crop physiological load and crop growth stage, realizes accurate prediction of the future water requirement and nitrogen, phosphorus and potassium nutrient requirements of crops, can dynamically respond to the growth state of crops and environmental changes, takes into account the differences in water and fertilizer requirements at different growth stages, and improves the accuracy and applicability of prediction through space-time fusion, thereby helping to realize precision fertilization.

[0098] In an embodiment, the planting optimization layer is constructed in step S3 to generate an optimal water and fertilizer management strategy considering yield, resource and environmental benefits, specifically, a planting optimization layer construction method combining digital twin and multi-objective optimization is used to obtain an optimal water and fertilizer management strategy based on water and fertilizer prediction information and a space-time fusion data set, including the following steps: Figure 1 Figure 4 In an embodiment, the planting optimization layer is constructed in step S3 to generate an optimal water and fertilizer management strategy considering yield, resource and environmental benefits, specifically, a planting optimization layer construction method combining digital twin and multi-objective optimization is used to obtain an optimal water and fertilizer management strategy based on water and fertilizer prediction information and a space-time fusion data set, including the following steps:

[0099] Step S31: decision space initialization, for constructing an optimal decision space for controlling water and fertilizer in the presence of meteorological forecast errors, specifically, a water and fertilizer decision vector is constructed, a dynamic search interval is set, a multi-objective optimization function is constructed by introducing a weather scenario set, multiple constraint conditions are set, and a multi-objective decision space model is obtained;

[0100] The water and fertilizer decision vector includes irrigation amount and nitrogen, phosphorus and potassium application amount, and the calculation formula is:

[0101] ;

[0102] In the formula, X i is the water and fertilizer decision vector of field i, indicating the water and fertilizer management strategy of field i, i is the field index, W i is the irrigation amount of field i, N i is the nitrogen application amount of field i, P i is the phosphorus application amount of field i, and K i is the potassium application amount of field i;

[0103] The dynamic search interval is set to construct a feasible search range centered on the prediction value and adaptively adjusted with the growth stage, and the calculation formula is:

[0104] ;

[0105] In the formula, X ij is the jth water and fertilizer decision component of field i, ​is a search width coefficient that varies dynamically with crop growth stages, and its value range is [0.1, 0.3], is the jth water-fertilizer prediction component of the ith field, which is specifically obtained from the water-fertilizer demand prediction information;

[0106] The multi-objective optimization function is used to simultaneously evaluate yield improvement, water resource utilization efficiency and nitrogen leaching risk in the presence of meteorological forecast errors, and includes a minimum negative yield objective function, a minimum water usage objective function and a minimum nitrogen leaching risk objective function.

[0107] The calculation formula of the negative yield objective function is:

[0108] ;

[0109] In the formula, f1(X) is the negative yield objective function, X is the water-fertilizer management strategy, the water-fertilizer management strategy is composed of water-fertilizer decision vectors of all fields, E[·] is a mathematical expectation operation, is a random weather scenario, which is subject to a probability distribution constructed based on future weather forecasts and their forecast errors, is the yield prediction value of the water-fertilizer management strategy under the random weather scenario ;

[0110] The calculation formula of the water usage objective function is:

[0111] ;

[0112] In the formula, f2(X) is the water usage objective function, and WaterUsage(X) is the water usage corresponding to the water-fertilizer management strategy;

[0113] The calculation formula of the nitrogen leaching risk objective function is:

[0114] ;

[0115] In the formula, f3(X) is the nitrogen leaching risk objective function, is the nitrogen leaching amount of the water-fertilizer management strategy under the random weather scenario , is a risk aversion coefficient, and Var[·] is a variance operation;

[0116] The calculation formula of the nitrogen leaching amount is:

[0117] ;

[0118] In the formula, is an empirical constant, is a nitrogen residual prediction value, ET is the evapotranspiration under the random weather scenario , and Rain is the rainfall under the random weather scenario the rainfall under the weather condition;

[0119] The multiple constraints are set, in particular, the soil water content, nutrient allocation and equipment capacity are constrained to ensure the agronomic feasibility of the strategy, and the calculation formula is:

[0120] ;

[0121] In the formula, is the soil water content under the water and fertilizer management strategy, is the wilting point water content, r min is the lower limit of nutrient application ratio, r max is the upper limit of nutrient application ratio, W max is the maximum water supply of the irrigation system;

[0122] Step S32: digital twin deduction, for deducing the crop growth process according to the candidate water and fertilizer management strategy, specifically, by constructing a digital twin including a crop growth mechanism model, a data agent model and a physical consistency correction module, receiving field environment monitoring data, future weather forecast and candidate water and fertilizer management strategy as input, time series deduction of yield and nutrient residual is carried out, and constraint correction is carried out through the physical consistency correction module to obtain the crop growth simulation result;

[0123] The crop growth mechanism model is used to simulate water transport, evaporation process, soil water content change and potential growth state of crops, specifically, a process simulation model based on crop, soil and atmosphere continuum can be used to realize the model, and models such as AquaCrop-OS, SWAP or DSSAT can be used to realize the model;

[0124] The data agent model is a multi-output regression model based on machine learning, which is used to preliminarily predict the yield and soil nutrient residual, and the calculation formula is:

[0125] ;

[0126] In the formula, is the yield prediction value, f ML is the data agent model, which can specifically use a multilayer perception model, A temporal is the time input sequence, A futre is the weather forecast sequence;

[0127] The physical consistency correction module is used to ensure that the deduced yield meets the crop physiological mechanism or water balance rule, and the calculation formula is:

[0128] ;

[0129] In the formula, Y is the corrected yield prediction value, is a soil water content correction weight, is a current soil water content derived from a crop growth mechanism model, is a soil water content reference value under ideal conditions for a crop growth mechanism model, is a crop growth correction weight, K c is a current crop growth coefficient derived from a crop growth mechanism model, is a crop growth coefficient reference value under ideal conditions for a crop growth mechanism model;

[0130] Step S33: Multi-objective optimization solution, for obtaining stable executable water and fertilizer management strategy in the presence of multiple constraint conditions and meteorological prediction error, specifically, combining the decision space initialization and the digital twin derivation, introducing dynamic penalty mechanism, robust fitness mechanism and constraint priority double selection mechanism on the basis of multi-objective evolutionary algorithm, self-adaptive optimization of candidate water and fertilizer management strategy, obtaining candidate water and fertilizer management strategy set;

[0131] The dynamic penalty mechanism, specifically, an exponential penalty factor is used, which increases with the number of iterations, to gradually suppress the constraint violation behavior, and the calculation formula is:

[0132] ;

[0133] In the formula, Penalty(t) is an exponential penalty factor, is a penalty coefficient, is a growth rate, t is the number of iterations, k is the constraint index, max(·) is the maximum value function, g k (X) is the violation amount of the kth constraint;

[0134] The robust fitness mechanism constructs the robust fitness by introducing a set of weather scenarios, and evaluates the robustness of the candidate water and fertilizer management strategy under multiple weather scenarios, and the calculation formula is:

[0135] ;

[0136] In the formula, F robust (X) is the robust fitness, m is the objective function index, M is the number of objective functions, w m is the weight of the mth objective function, f m (X) is the mth objective function, is an uncertainty sensitive parameter, is the variance of negative yield under different weather scenarios, used to measure the robustness;

[0137] The constraint-priority double-layer selection mechanism specifically selects a water and fertilizer management strategy that fully satisfies multiple constraint conditions in a non-dominated sorting process, and then performs adaptive sorting and selection on the candidate solutions according to the comprehensive fitness, so as to realize iterative evolution of the candidate water and fertilizer management strategy set.

[0138] The calculation formula of the comprehensive fitness is:

[0139] ;

[0140] In the formula, F total (X) is the comprehensive fitness;

[0141] Step S34: preference interaction decision, for generating a final water and fertilizer management strategy from the strategy set according to user preferences, specifically, by obtaining the user's preference weights for yield, water saving and environmental risk, the candidate water and fertilizer management strategy set is weighted and reordered, and the optimal water and fertilizer management strategy that satisfies the preference constraint is selected, and the calculation formula is:

[0142] ;

[0143] ;

[0144] In the formula, p is the preference weight, p1 is the first target function preference weight, specifically the negative yield target function preference weight, p2 is the second target function preference weight, specifically the water consumption target function, p3 is the third target function preference weight, specifically the nitrogen leaching risk target function, is the optimal water and fertilizer management strategy, is the water and fertilizer management strategy X selected from the candidate water and fertilizer management strategy set to minimize PaSet is the candidate water and fertilizer management strategy set, and p m is the mth target function preference weight, and f m (X) is the mth target function.

[0145] By performing the above operation, in view of the technical problems that in the existing intelligent agricultural planting management process, the traditional water and fertilizer management strategy is difficult to simultaneously consider crop yield, resource utilization efficiency and environmental protection, and is difficult to cope with future meteorological uncertainty and dynamic changes of crop growth state, the scheme creatively adopts a planting optimization layer construction method combining digital twin and multi-objective optimization, realizes fine and dynamic optimization of the water and fertilizer management strategy considering the crop growth stage, soil nutrient status and future weather conditions, and generates an optimal executable management strategy that can improve yield, save water and fertilizer resources and reduce nitrogen leaching risk.

[0146] Embodiment five, refer to Figure 1The embodiment is based on the above-mentioned embodiment, and in step S4, the agricultural planting management is specifically performed by integrating the data layer, the water and fertilizer prediction layer and the planting optimization layer to construct a water and fertilizer optimization model for agricultural planting management, performing water and fertilizer optimization, generating a field-level control instruction according to an optimal water and fertilizer management strategy, issuing the field-level control instruction to a water and fertilizer integrated control system, and performing effect monitoring to generate a planting management feedback report.

[0147] Embodiment six, refer to Figure 2 The embodiment is based on the above-mentioned embodiment, and the intelligent agricultural planting management system provided by the application includes a data establishment module, a water and fertilizer prediction module, a planting optimization module and an agricultural planting management module.

[0148] The data establishment module is used for data layer construction, and through the data layer construction, a spatio-temporal fusion data set is obtained and sent to the water and fertilizer prediction module and the planting optimization module.

[0149] The water and fertilizer prediction module is used for water and fertilizer prediction layer construction, and through the water and fertilizer prediction layer construction, water and fertilizer demand prediction information is obtained and sent to the planting optimization module.

[0150] The planting optimization module is used for planting optimization layer construction, and through the planting optimization layer construction, an optimal water and fertilizer management strategy is obtained and sent to the agricultural planting management module.

[0151] The agricultural planting management module is used for agricultural planting management, and a field-level control instruction and a planting management feedback report are obtained.

[0152] It should be noted that in this paper, relationship terms such as first and second are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between the entities or operations. Moreover, the term "includes", "contains" or any other variant thereof is intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements not only includes those elements, but also includes other elements not explicitly listed or inherent to such a process, method, article or device.

[0153] Although the embodiments of the application have been shown and described, it can be understood by those of ordinary skill in the art that various changes, modifications, replacements and variations can be made to the embodiments without departing from the principles and spirits of the application.

[0154] The above describes the present application and its embodiments, which are not limited, and the drawings only show one of the embodiments of the present application, and the actual structure is not limited thereto. In general, if a person skilled in the art is inspired thereby, without departing from the purpose of the present application, without creative design, similar structure and embodiments of the technical solution are not creative, and should belong to the protection scope of the present application.

Claims

1. An intelligent agricultural planting management method, characterized in that: The method comprises the following steps: Step S1: data layer construction; Step S2: water and fertilizer prediction layer construction, for predicting the water and nutrient requirements of crops, specifically, according to the spatio-temporal fusion data set, a water and fertilizer prediction layer construction method combining crop physiological load and crop growth stage is adopted to obtain water and fertilizer demand prediction information, comprising the following steps: step S21 input sequence construction, step S22 parallel feature extraction, step S23 spatio-temporal consistency fusion and step S24 multi-target prediction output; In step S24, the multi-target prediction output, specifically, a multi-output regression prediction is performed on the spatio-temporal fusion features by a multi-task learning model, and an agricultural proportion constraint and a spatio-temporal consistency constraint are introduced to construct a comprehensive loss function to correct the prediction results, and obtain the water and fertilizer demand prediction information; Step S3: construction of planting optimization layer, the construction of planting optimization layer is used to generate optimal water and fertilizer management strategies considering yield, resources and environmental benefits, specifically, according to the water and fertilizer prediction information and the spatio-temporal fusion data set, a planting optimization layer construction method combining digital twin and multi-objective optimization is adopted to obtain the optimal water and fertilizer management strategy, comprising the following steps: step S31 decision space initialization, step S32 digital twin deduction, step S33 multi-objective optimization solution and step S34 preference interactive decision; In step S33, the multi-objective optimization solution is used to obtain a stable and executable water and fertilizer management strategy in the presence of multi-constraint conditions and meteorological forecast errors; Step S4: agricultural planting management. 2.The intelligent agricultural planting management method according to claim 1, characterized in that: In step S23, the spatio-temporal consistency fusion is used to generate globally spatio-temporally consistent fusion features, specifically, by calculating the consistency of spatial features and time series features, constructing a spatio-temporal adaptability coefficient, then splicing the context vector composed of crop growth stage one-hot encoding and crop physiological load factor through a gating network to generate dynamic weights, and performing adaptive weight correction according to the spatio-temporal adaptability coefficient, finally weighting and fusing the spatial features and time series features to obtain the spatio-temporal fusion features; In step S24, the agricultural proportion constraint is used to ensure that the predicted values of nitrogen, phosphorus and potassium satisfy the agricultural empirical proportion relationship; and the spatio-temporal consistency constraint is used to maintain the consistency of spatial features and time series features in the spatio-temporal fusion features. 3.The intelligent agricultural planting management method according to claim 2, characterized in that: In step S21, the input sequence construction, specifically, by calculating the crop physiological load factor, combining multi-channel remote sensing image data, environmental time series data and future weather forecast, constructing spatial input sequence, time input sequence and weather forecast sequence; the crop physiological load factor is used to represent the comprehensive degree of water stress, heat stress and evapotranspiration pressure of crops under current environmental conditions; In step S22, the parallel feature extraction, specifically, by constructing a double-branch feature extraction network to extract spatial features and time series features in parallel, the double-branch feature extraction network comprises a spatial branch and a time series branch; the spatial branch extracts convolutional features from the spatial input sequence by a lightweight convolutional neural network to obtain spatial features; The time sequence branch captures the trend of the time input sequence through a double-layer recurrent neural network to obtain a time sequence hidden state, splices the time sequence hidden state with a meteorological prediction sequence, and maps through a fully connected layer to obtain a time sequence feature.

4. The intelligent agricultural planting management method according to claim 3, characterized in that: In step S31, the decision space is initialized, specifically by constructing a water and fertilizer decision vector, setting a dynamic search interval, introducing a weather scenario set to construct a multi-objective optimization function, setting multiple constraint conditions, and obtaining a multi-objective decision space model. The water and fertilizer decision vector includes irrigation amount and nitrogen, phosphorus and potassium application amount; the dynamic search interval is set to construct a feasible search range centered on the predicted value and adaptively adjusted with the growth stage; The multi-objective optimization function is used to evaluate yield improvement, water resource utilization efficiency and nitrogen leaching risk simultaneously in the presence of meteorological prediction errors, including a minimum negative yield objective function, a minimum water consumption objective function and a minimum nitrogen leaching risk objective function; The multiple constraint conditions are specifically constrained by soil water content, nutrient allocation and equipment capacity.

5. The intelligent agricultural planting management method according to claim 4, characterized in that: In step S33, the multi-objective optimization is solved, specifically by combining the decision space initialization and the digital twin deduction, introducing a dynamic penalty mechanism, a robust fitness mechanism and a constraint-priority double-layer selection mechanism based on a multi-objective evolutionary algorithm, adaptively optimizing the candidate water and fertilizer management strategies, and obtaining a candidate water and fertilizer management strategy set; The dynamic penalty mechanism specifically uses an exponential penalty factor that increases with the number of iterations to gradually suppress constraint violation behavior; the robust fitness mechanism constructs a robust fitness by introducing a weather scenario set to evaluate the robustness of the candidate water and fertilizer management strategies under multiple weather scenarios; the constraint-priority double-layer selection mechanism specifically selects water and fertilizer management strategies that fully meet the multiple constraint conditions in the non-dominated sorting process, and then adaptively sorts and selects the candidate solutions according to the comprehensive fitness, realizing the iterative evolution of the candidate water and fertilizer management strategy set.

6. The intelligent agricultural planting management method according to claim 5, characterized in that: In step S32, the digital twin deduction is used to deduce the crop growth process according to the candidate water and fertilizer management strategy, specifically by constructing a digital twin including a crop growth mechanism model, a data proxy model and a physical consistency correction module, receiving field environment monitoring data, future weather forecasts and candidate water and fertilizer management strategies as inputs, performing time sequence deduction on yield and nutrient residue, and performing constraint correction through the physical consistency correction module to obtain crop growth simulation results; The crop growth mechanism model is used to simulate water transport, evapotranspiration process, soil water content change and potential growth state of crops; The data proxy model is a multi-output regression model based on machine learning for preliminary prediction of yield and soil nutrient residue; In step S34, the preference interactive decision is used to generate a final water and fertilizer management strategy from the strategy set according to user preferences, specifically by obtaining user preference weights for yield, water saving and environmental risk, weighting and reordering the candidate water and fertilizer management strategy set, and selecting the optimal water and fertilizer management strategy that meets the preference constraints.

7. The intelligent agricultural planting management method according to claim 6, characterized in that: In step S4, the agricultural planting management, specifically, by executing steps S1 to S3, the data layer, the water and fertilizer prediction layer and the planting optimization layer are integrated to construct a water and fertilizer optimization model for agricultural planting management, water and fertilizer optimization is performed, optimal water and fertilizer management strategies are generated, field-level control instructions are issued to the water and fertilizer integrated control system, and execution effect monitoring is performed to generate planting management feedback reports. In step S1, the data layer is constructed, specifically, by collecting multi-channel remote sensing image data, field environment monitoring data, future weather forecasts and crop growth information, and performing preprocessing, spatio-temporal alignment and standardization processing to obtain a spatio-temporal fusion data set.

8. An intelligent agricultural planting management system for implementing the intelligent agricultural planting management method according to any one of claims 1-7, characterized in that: It comprises a data establishment module, a water and fertilizer prediction module, a planting optimization module and an agricultural planting management module. 9.The intelligent agricultural planting management system according to claim 8, characterized in that: The data establishment module is used for data layer construction, and through data layer construction, a spatio-temporal fusion data set is obtained, and the spatio-temporal fusion data set is sent to the water and fertilizer prediction module and the planting optimization module; The water and fertilizer prediction module is used for water and fertilizer prediction layer construction, and through water and fertilizer prediction layer construction, water and fertilizer demand prediction information is obtained, and the water and fertilizer demand prediction information is sent to the planting optimization module; The planting optimization module is used for planting optimization layer construction, and through planting optimization layer construction, optimal water and fertilizer management strategies are obtained, and the optimal water and fertilizer management strategies are sent to the agricultural planting management module; The agricultural planting management module is used for agricultural planting management to obtain field-level control instructions and planting management feedback reports.

Images