Planting optimization method and system based on crop yield estimation in saline-alkali region

Abstract

The application discloses a kind of based on saline-alkali region crop yield estimation and planting optimization method and system, it is related to crop yield prediction and planting optimization field, this method includes: data acquisition fusion processing;Soil water salt migration dynamic prediction, realizes water salt space-time distribution prediction using the space-time cubic modeling adaptive model combined with physical correction;Crop salt tolerance modeling, based on physiological response fitting and threshold segmentation establishes multi-interval salt tolerance model and introduces covariant correction;Migration salt tolerance coupling modeling, introduce salt memory factor and stage sensitivity control factor, dynamically simulate the growth response of crop under salt stress;Crop yield estimation and planting optimization, based on coupling model output generation regional yield prediction and sowing configuration scheme, and form optimization strategy in combination with irrigation fertilization and the like measure.This application realizes the model-driven integrated optimization of crop in saline-alkali area from yield estimation to planting decision, with high precision and wide applicability.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of crop yield prediction and planting optimization, and particularly relates to a planting optimization method and system based on crop yield prediction in a saline-alkali region. BACKGROUND

[0002] The planting optimization method and system based on crop yield prediction in a saline-alkali region is an agricultural informatization technical scheme for saline-alkali farmland. By fusing remote sensing, monitoring and historical data, combining water-salt transport dynamic prediction and crop salt tolerance modeling, the fine prediction of the growth adaptability and yield level of crops in a saline-alkali environment is realized, and a scientific and reasonable planting optimization scheme is further generated. The role is to overcome the traditional extensive decision-making relying on experience and single indicators, realize the whole-process intelligent optimization from regional yield prediction to field management measures, and improve the scientific nature, adaptability and overall output benefit of agricultural production in the saline-alkali region.

[0003] However, in the existing planting optimization method based on crop yield prediction in a saline-alkali region, there are technical problems of disconnection between yield prediction and planting optimization, and over-experiential optimization strategy; in the existing soil water-salt transport dynamic prediction method, there are technical problems of lack of physical constraints in the model, and insufficient prediction accuracy and interpretability; in the existing crop salt tolerance modeling method, there are technical problems of rough threshold division, and difficulty in reflecting the nonlinear response of crops under different salt conditions; in the existing transport salt tolerance coupling modeling method, there are technical problems of only static coupling, and not considering the time sequence accumulation effect and stage sensitivity in the growth process of crops. SUMMARY

[0004] In view of the above, in order to overcome the defects of the prior art, the present application provides a planting optimization method and system based on crop yield estimation in saline and alkaline regions, which is aimed at the technical problems of disconnection between yield estimation and planting optimization and overly empirical optimization strategy in the existing planting optimization method based on crop yield estimation in saline and alkaline regions. The present scheme creatively adopts a comprehensive progressive modeling prediction method combining soil water and salt transport dynamic prediction, crop salt tolerance modeling and transport salt tolerance coupling modeling, and generates a comprehensive planting optimization strategy through yield estimation data, thereby realizing the transition from "passive empirical configuration" to "active optimization based on model deduction", effectively improving the adaptability and overall yield level of crop planting in saline and alkaline regions. In view of the technical problems of lack of physical constraints, insufficient prediction accuracy and interpretability in the existing soil water and salt transport dynamic prediction method, the present scheme creatively adopts a spatiotemporal cubic modeling self-adaptive prediction model combining physical correction, extracts water and salt spatiotemporal features through three-dimensional convolution and convolution LSTM combined with deconvolution decoding, and introduces a physical consistency loss function based on the water and salt diffusion equation for error constraint, realizing high-precision prediction considering data-driven and physical mechanism, and ensuring the physical rationality of the water and salt transport prediction results in long-term time series and spatial distribution. In view of the technical problems of rough threshold division and difficulty in reflecting the nonlinear response of crops under different salt conditions in the existing crop salt tolerance modeling method, the present scheme creatively adopts a salt tolerance response improvement method combining crop physiological response fitting and salt stress threshold segmentation modeling, automatically divides the salinity interval through variation point detection, and uses an exponential decay function and a covariant weighting correction factor within the interval to finely describe the crop yield response under different salinities, realizing the transition of salt tolerance modeling from single-threshold rough description to multi-interval and adjustable fine response modeling, and improving the adaptability of the model to different crops and different ecological environments. In view of the technical problems of only static coupling and not considering the time series accumulation effect and stage sensitivity in the growth process of crops in the existing transport salt tolerance coupling modeling method, the present scheme creatively adopts a time series coupling modeling method combining dynamic water and salt simulation and crop response regulation mechanism, introduces an exponential decay salt memory factor and a stage sensitivity regulation factor, so that the model can dynamically describe the accumulation and recovery process of salt stress on the time series scale, and dynamically call the crop salt tolerance response model, finally realizing dynamic response simulation of the whole growth process of crops in saline and alkaline regions, thereby significantly improving the yield estimation accuracy and reliability.

[0005] The technical scheme adopted by the present application is as follows: The present application provides a planting optimization method based on crop yield estimation in saline and alkaline regions, which comprises the following steps:

[0006] Step S1: data acquisition and fusion processing;

[0007] Step S2: soil water and salt transport dynamic prediction;

[0008] Step S3: crop salt tolerance modeling;

[0009] Step S4: coupling modeling of salt tolerance migration;

[0010] Step S5: crop yield estimation and planting optimization.

[0011] Further, in step S1, the data collection and fusion processing is used to collect multi-source saline-alkali soil and meteorological data and integrate the original data set, specifically, through multi-source data collection and standardization and processing flow, a standardized fusion data set is obtained.

[0012] The multi-source data collection includes remote sensing data collection, field monitoring data collection and historical statistical data retrieval.

[0013] Further, in step S2, the soil water and salt migration dynamic prediction is used to build a spatio-temporal prediction model to simulate the dynamic distribution process of salt and water in the soil, specifically, according to the standardized fusion data set, a spatio-temporal cubic modeling adaptive prediction model combined with physical correction is used to predict the soil water and salt migration dynamics, to obtain water and salt migration dynamic data, including the following steps:

[0014] Step S21: data resampling, specifically, through uniform spatial resolution and time step, the original soil water and salt data is interpolated, missing value is repaired and rasterized, time and space standardization reconstruction is performed, and a standardized tensor sequence is obtained;

[0015] Step S22: spatio-temporal cubic structure modeling, specifically, by encapsulating the spatial grid tensor of continuous time steps into a spatio-temporal cubic input unit, a cubic data structure is constructed, and a spatio-temporal input tensor is obtained;

[0016] Step S23: spatio-temporal feature extraction and coding, specifically, the spatio-temporal cube is convolved by a three-dimensional convolution network to code the spatio-temporal features, and latent space state features are obtained;

[0017] Step S24: deconvolution long short-term memory network decoding, specifically, the latent space state features are input into a convolution long short-term memory network and decoded by combining a deconvolution layer, time series prediction recovery is performed, and future time step water and salt state prediction frames are obtained;

[0018] Step S25: constructing a physical constraint loss function, specifically, by introducing a water and salt diffusion equation and a conservation relationship, a regularization constraint based on mass conservation and diffusion gradient is constructed, prediction error correction is performed, and a physical consistency loss is obtained;

[0019] Step S26: soil water and salt transport dynamic prediction, specifically, network model training and reasoning are performed by jointly minimizing mean square error loss and physical consistency loss to obtain water and salt transport dynamic data.

[0020] Further, in step S3, the crop salt tolerance modeling is used to establish the growth adaptation model of different crops under different salt conditions according to the crop variety and growth data, specifically, the crop salt tolerance modeling is performed by using the salt tolerance response improvement method combining crop physiological response fitting and salt stress threshold segmentation modeling according to the standardized fusion data set, to obtain a crop salt tolerance response model, including the following steps:

[0021] Step S31: input factor extraction, specifically, the target crop yield value is extracted as a response variable, and the soil salt content, water content, air temperature, crop species code and sowing time are extracted to form an input matrix, the variables and factors are constructed to obtain the original input factor matrix and the observed response variable;

[0022] Step S32: salt stress threshold interval division, specifically, the soil salt level is divided into multiple threshold intervals by a variation point detection algorithm, the salt interval segmentation processing is performed, and a salt threshold set is obtained;

[0023] Step S33: physiological response factor function modeling, specifically, a nonlinear response function based on exponential decay and bottom correction is constructed in each salt interval, interval fitting calculation is performed, and a segmented salt response function is obtained;

[0024] Step S34: covariate coupling response correction, specifically, a weighted modulation factor containing water content, air temperature, sowing time and crop species is constructed to correct the salt response function, and a covariate coupling correction model is obtained;

[0025] Step S35: parameter optimization model training, specifically, the parameter optimization training is performed by minimizing the yield prediction error loss function and combining the regularization constraint to obtain the final crop salt tolerance response model.

[0026] Further, in step S4, the transport salt tolerance coupling modeling is used to fuse the soil water and salt changes with the crop salt tolerance model to simulate the response characteristics of crops in a dynamic salt environment, specifically, the coupling prediction model is obtained by using the time sequence coupling modeling method combining dynamic water and salt simulation and crop response regulation mechanism according to the water and salt transport dynamic data and the crop salt tolerance response model, including the following steps:

[0027] Step S41: dynamic water and salt state sequence construction, specifically, the salt and water data in the crop growth period are extracted to construct time step driven variables, and a water and salt driven matrix is obtained;

[0028] Step S42: physiological response memory factor design, specifically, by constructing a salt stress cumulative index and using an exponential decay weight function, modeling the memory effect of crop salt stress, obtaining a time step memory factor;

[0029] Step S43: stage sensitivity regulation factor design, specifically, by identifying different growth stages of crops and setting stage sensitivity coefficients, salt sensitivity correction is performed to obtain a stage regulation factor;

[0030] Step S44: dynamic coupling of crop response model, specifically, by calling the crop salt tolerance response model, combining the time step memory factor and the stage regulation factor, dynamic yield calculation is performed to obtain a dynamic response yield sequence;

[0031] Step S45: coupling prediction model construction, specifically, by minimizing the yield prediction error based on the principle of optimization, the dynamic response yield sequence is calibrated to obtain the final coupling prediction model.

[0032] Further, in step S5, the crop yield estimation and planting optimization is used to calculate the predicted yield of crops planted in a specific area based on the coupling model output and environmental parameters, specifically, using the coupling prediction model, crop yield estimation is performed to obtain regional yield estimation data, and by analyzing the spatial distribution of the regional yield estimation data, high-yield and low-yield difference areas are identified, and combined with the salt tolerance grade and yield threshold of different crop varieties, variety selection and matching are performed, and then according to the spatial distribution characteristics and variety matching results, a sowing area configuration scheme is generated, and irrigation, fertilization and salt drainage field management measures are optimized and scheduled, thereby comprehensive output regional planting optimization strategy is obtained, and planting optimization data is obtained.

[0033] The crop yield estimation and planting optimization system based on salt and alkali region provided by the application comprises a data acquisition and fusion processing module, a soil water and salt transport dynamic prediction module, a crop salt tolerance modeling module, a transport salt tolerance coupling modeling module and a planting optimization module.

[0034] The data acquisition and fusion processing module is used for data acquisition and fusion processing, and through data acquisition and fusion processing, a standardized fusion data set is obtained, and the standardized fusion data set is sent to the soil water and salt transport dynamic prediction module and the crop salt tolerance modeling module;

[0035] The soil water and salt transport dynamic prediction module is used for soil water and salt transport dynamic prediction, and through soil water and salt transport dynamic prediction, water and salt transport dynamic data is obtained, and the water and salt transport dynamic data is sent to the transport salt tolerance coupling modeling module;

[0036] The crop salt tolerance modeling module is configured to model crop salt tolerance, obtain a crop salt tolerance response model through the crop salt tolerance modeling, and send the crop salt tolerance response model to the transport salt tolerance coupling modeling module.

[0037] The transport salt tolerance coupling modeling module is configured to model transport salt tolerance coupling, obtain a coupling prediction model through the transport salt tolerance coupling modeling, and send the coupling prediction model to the planting optimization module.

[0038] The planting optimization module is configured to estimate crop yield and optimize planting, obtain regional yield estimation data and planting optimization strategies through the estimation of crop yield and the optimization of planting.

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

[0040] (1) In the existing planting optimization method based on salt and alkali region crop yield estimation, there are technical problems such as disconnection between yield estimation and planting optimization, and over-experienced optimization strategies. For example, in traditional salt and alkali area improvement planting, usually only soil salt index or single crop test yield is used for calculation, lacking comprehensive consideration of salt and water dynamic changes and crop salt tolerance differences in different regions, resulting in that the regional scale sowing structure and field management scheme often lack scientificity and universality. The present scheme creatively adopts a comprehensive progressive modeling and prediction method combining soil water and salt transport dynamic prediction, crop salt tolerance modeling and transport salt tolerance coupling modeling, and generates a comprehensive planting optimization strategy through yield estimation data, thereby realizing the transition from "passive experience configuration" to "active optimization based on model deduction", and effectively improving the adaptability and overall yield level of crop planting in salt and alkali areas.

[0041] (2) In the existing soil water and salt transport dynamic prediction method, there are technical problems such as lack of physical constraints in the model, insufficient prediction accuracy and interpretability. For example, the prediction method based on conventional machine learning or simple regression model can only fit the static correlation between historical salt content and rainfall, groundwater depth, and cannot guarantee that the prediction result meets the salt conservation or water diffusion law, so the deviation will accumulate gradually in long-term prediction, and it is difficult to be used for agricultural decision-making. The present scheme creatively adopts a spatiotemporal cubic modeling self-adaptive prediction model combined with physical correction, extracts water and salt spatiotemporal features through three-dimensional convolution and convolution LSTM combined with deconvolution decoding, and introduces a physical consistency loss function based on water and salt diffusion equation for error constraint, realizing high-precision prediction considering data-driven and physical mechanism, and ensuring the physical rationality of water and salt transport prediction results in long-term time series and spatial distribution.

[0042] (3) In the existing crop salt tolerance modeling method, there is a technical problem that the threshold division is rough and it is difficult to reflect the nonlinear response of crops under different salt conditions. For example, traditional research usually divides whether crops are salt tolerant with a single "salinity critical value", ignoring the typical segmented characteristics that yield is almost not affected in the low salt interval, yield slowly decreases with salt in the medium salt interval, and yield sharply decreases in the high salt interval, resulting in that the model cannot accurately depict the yield-salt relationship. The scheme creatively adopts the salt tolerance response improvement method combining crop physiological response fitting and salt stress threshold segmented modeling, automatically divides the salinity interval through variation point detection, and uses exponential decay function and covariate weighted correction factor in the interval to finely depict the crop yield response under different salinity, realizes the salt tolerance modeling from single threshold rough description to multi-interval and adjustable fine response modeling, and improves the adaptability of the model to different crops and different ecological environments;

[0043] (4) In the existing migration salt tolerance coupling modeling method, there is a technical problem that only static coupling is made, and the time sequence cumulative effect and stage sensitivity in the growth process of crops are not considered. For example, part of the research only combines average salinity and yield regression, but fails to reflect the difference of crops in the stages of seedling stage, tillering stage and grain filling stage to salt stress, and also fails to introduce the historical memory effect of salt stress, so that the prediction result is often greatly different from the measured yield. The scheme creatively adopts the time sequence coupling modeling method combining dynamic water and salt simulation and crop response regulation mechanism, introduces the salt memory factor with exponential decay and the stage sensitivity regulation factor, so that the model can dynamically depict the accumulation and recovery process of salt stress in the time sequence scale, and dynamically call the crop salt tolerance response model, and finally realize the dynamic response simulation of the whole growth process of crops in saline-alkali area, thereby significantly improving the yield prediction accuracy and reliability. BRIEF DESCRIPTION OF DRAWINGS

[0044] Figure 1 The flowchart of the planting optimization method based on crop yield estimation in saline-alkali region provided by the present application is shown in the figure;

[0045] Figure 2 The schematic diagram of the planting optimization system based on crop yield estimation in saline-alkali region provided by the present application is shown in the figure;

[0046] Figure 3 The flowchart of the soil water and salt migration dynamic prediction of step S2 is shown in the figure;

[0047] Figure 4 The flowchart of the crop salt tolerance modeling of step S3 is shown in the figure;

[0048] Figure 5 The flowchart of the migration salt tolerance coupling modeling of step S4 is shown in the figure.

[0049] The accompanying drawings are included to provide a further understanding of the application, and are incorporated in and constitute a part of this specification, illustrate embodiments of the application, and together with the description serve to explain the application, and do not limit the application. DETAILED DESCRIPTION

[0050] 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 protection scope of the present application.

[0051] 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 used to facilitate the description of the present application and simplify 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 of the present application.

[0052] Embodiment one, refer to Figure 1 The present application provides a planting optimization method based on crop yield estimation in saline and alkaline regions, which comprises the following steps:

[0053] Step S1: data acquisition and fusion processing;

[0054] Step S2: dynamic prediction of soil water and salt transport;

[0055] Step S3: crop salt tolerance modeling;

[0056] Step S4: coupling modeling of transport and salt tolerance;

[0057] Step S5: crop yield estimation and planting optimization.

[0058] By performing the above operation, for the existing planting optimization method based on the salt and alkali area crop yield estimation, there are technical problems such as the disconnection between yield estimation and planting optimization, and the over-experiential optimization strategy. For example, in the traditional salt and alkali area improvement planting, usually only the soil salt content index or single crop test yield is relied on for calculation, and the comprehensive consideration of the dynamic changes of salt and water in different regions and the differences of crop salt tolerance is lacked, resulting in that the regional scale seeding structure and field management scheme often lack scientificity and universality. The present scheme creatively adopts a comprehensive progressive modeling and prediction method combining soil water and salt transport dynamic prediction, crop salt tolerance modeling and transport salt tolerance coupling modeling, and generates a comprehensive planting optimization strategy through yield estimation data, so as to realize the transformation from "passive experience configuration" to "active optimization based on model deduction", and effectively improve the adaptability and overall yield level of crop planting in salt and alkali areas.

[0059] Embodiment two, refer to Figure 1 and Figure 2 This embodiment is based on the above embodiment, in step S1, the data acquisition and fusion processing is used to collect multi-source salt and alkali soil and meteorological data and integrate the original data set, specifically, through multi-source data acquisition and standardization and processing flow, a standardized fusion data set is obtained;

[0060] The multi-source data acquisition includes remote sensing data acquisition, field monitoring data acquisition and historical statistical data retrieval;

[0061] Preferably, the remote sensing data acquisition mainly includes surface reflectivity, vegetation index (NDVI), surface temperature and evapotranspiration; the field monitoring data acquisition mainly includes soil conductivity, pH value, salt content, water content, organic matter content, soil type, groundwater depth, soil physical and chemical characteristic parameters and real-time meteorological data, the real-time meteorological data includes rainfall, wind speed, evaporation and temperature and humidity;

[0062] The historical statistical data includes regional crop planting situation in recent years, annual meteorological change trend and historical yield data.

[0063] Embodiment three, refer to Figure 1 , Figure 2 and Figure 3 This embodiment is based on the above embodiment, in step S2, the soil water and salt transport dynamic prediction is used to build a spatio-temporal prediction model to simulate the dynamic distribution process of salt and water in soil, specifically, according to the standardized fusion data set, a spatio-temporal cubic modeling self-adaptive prediction model combining physical correction is adopted to perform soil water and salt transport dynamic prediction, to obtain water and salt transport dynamic data, including the following steps:

[0064] Step S21: data resampling, specifically, by unifying the spatial resolution and the time step, the original soil water and salt data is interpolated, missing value is repaired and rasterized, time and space standardization reconstruction is carried out, and a standardized tensor sequence is obtained;

[0065] Preferably, the unified spatial resolution, specifically using bilinear interpolation or Kriging interpolation method, unifies the soil salt and water observation values of different spatial resolutions to 30m*30m standard grid;

[0066] The time step is preferably 1 day or 1 week, and the missing value repair is repaired by linear interpolation and sliding window mean method;

[0067] The standardized tensor sequence specifically refers to the water and salt raster data after unifying the spatial grid;

[0068] Step S22: spatiotemporal cube structure modeling, specifically, by encapsulating the standardized tensor sequence of continuous time steps into a spatiotemporal cube input unit, a cube data structure is constructed, and a spatiotemporal input tensor is obtained;

[0069] Preferably, the spatiotemporal cube input unit is a data structure that unifies the time dimension, spatial dimension and feature dimension; the spatiotemporal input tensor encapsulates the soil salt distribution, water distribution and related meteorological and soil characteristic data of multiple continuous time points within a preset time window; in space, it is organized in the form of regular grid, each grid unit corresponds to a specific geographic location; in terms of features, each grid unit contains multiple observation indexes, including soil salt concentration, moisture content, conductivity, temperature and evapotranspiration; in terms of time, the spatiotemporal cube retains the dynamic changes of the above observation indexes at multiple continuous time points;

[0070] Through the above encapsulation, the spatiotemporal cube not only represents the spatial distribution state at a certain moment, but also reflects the evolution trend and dynamic change relationship between different time points, so as to realize the coupling modeling of time sequence information and spatial grid information, ensure that the subsequent three-dimensional convolution calculation can capture the time sequence dependent features and spatial correlation features, and provide complete spatiotemporal input information for the prediction of soil water and salt transport;

[0071] Step S23: spatiotemporal feature extraction and coding, specifically, the spatiotemporal cube is convolved by a three-dimensional convolution network, the spatiotemporal feature coding is carried out, and the latent space state feature is obtained;

[0072] Preferably, the structure parameter setting of the three-dimensional convolutional network is specifically that the convolution kernel size is 3x3x3, the convolution step is preferably 1, and the padding mode is Same padding to ensure that the spatial size of the output tensor after convolution is consistent with the input; the number of convolution layers is set to 3, and the number of output channels is 64, 128, and 128 in turn, and normalization processing and a nonlinear activation function can be selected between layers to enhance the stability and nonlinear expression ability of feature extraction;

[0073] Step S24: deconvolution long short-term memory network decoding, specifically, the latent space state feature is input into the convolution long short-term memory network, and decoding is performed by combining the deconvolution layer to recover the future time step water salt state prediction frame through time series prediction;

[0074] The calculation formula of the time series prediction recovery is:

[0075] ;

[0076] In the formula, is the predicted water salt state tensor sequence of the future h time steps, which is used to represent the future time step water salt state prediction frame, Deconv is a deconvolution layer operation, ConvLSTM is a convolution long short-term memory network operation, f 3D-CNN is a three-dimensional convolutional network operation, C t is a spatiotemporal input tensor, t is a time index, and h is a prediction time step index;

[0077] Preferably, the structure parameter setting of the convolution long short-term memory network is specifically that the number of network layers is set to 2, the number of hidden units of each layer is 128, the convolution kernel size is preferably 3x3, and the padding mode is Same padding to maintain the consistency of the time series features in the convolution operation; at the decoding end, 1 layer of deconvolution operation is adopted, the deconvolution kernel size is 3x3, and the step is 2, so as to restore the spatial resolution after downsampling to the original input size, thereby obtaining the water salt state prediction frame of the future time step;

[0078] Step S25: constructing a physical constraint loss function, specifically, by introducing a water salt diffusion equation and a conservation relationship, a regularization constraint based on mass conservation and diffusion gradient is constructed to correct the prediction error and obtain a physical consistency loss;

[0079] The calculation formula of the physical consistency loss is:

[0080] ;

[0081] In the formula, is the physical consistency loss, t is a time index, is a predicted salinity field, used to represent a salt concentration prediction value, is the spatial gradient operator, D s is the salt equivalent diffusion coefficient, preferably in the range of 0.01 to 0.1 m² / day, obtained by laboratory non-fractional diffusion test, specifically, the diffusion profile of salt concentration in soil with time and space is measured under constant hydraulic conditions, and the effective diffusion coefficient is obtained by fitting the Fick diffusion equation, is the predicted water content field, D w is the hydraulic dispersion coefficient, preferably in the range of 0.1 to 1.0 m² / day, obtained by constant flow penetration test, specifically, inert tracer is injected under constant flow rate conditions, and the dispersion parameter is obtained by fitting the convection-dispersion equation to the collected outlet penetration curve, and then converted, is the L2 norm square operator;

[0082] Step S26: soil water and salt transport dynamic prediction, specifically, network model training and inference are performed by jointly minimizing the mean square error loss and the physical consistency loss, to obtain water and salt transport dynamic data;

[0083] Preferably, the calculation formula of the joint minimization of the mean square error loss and the physical consistency loss is:

[0084] ;

[0085] In the formula, L is the joint loss function, L MSE is the minimum mean square error loss, is the physical consistency loss weight, preferably in the range of 0.1 to 0.5;

[0086] Preferably, the table 1 is an example table of network model training parameters, as shown in the table, the spatial grid resolution of the uniform spatial resolution is 30m × 30m, to balance the availability of remote sensing data and the accuracy of field management; the time window length is set to 8 time steps, a cubic input unit is constructed by continuous stacking, which can cover the water and salt changes in a typical agricultural period; the prediction step is set to 14 days, which is used to connect the management period such as irrigation and salt drainage, to facilitate yield estimation and field scheduling;

[0087] In terms of network structure, 3 × 3 × 3 convolution kernel is used for three-dimensional convolution to capture local spatio-temporal features, and the number of convolution channels is 64, 128 and 128 in sequence, and the number of convolution long short-term memory network units is set to 128 to fully model the time series dependence relationship;

[0088] The training process adopts an Adam optimizer, the initial learning rate is set to 3e-4, and the batch size is set to 16 to balance the parameter update efficiency and memory consumption; at the same time, a physical consistency constraint is introduced into the loss function, the weight coefficient a is 0.3, which is used to ensure that the prediction result meets the water and salt conservation law; the upper limit of the training round is 60 rounds, and an early stopping mechanism is used, and the patience value is 8, when the validation set error does not decrease for a certain number of consecutive rounds, the training is terminated in advance, so as to ensure that the model converges stably while avoiding overfitting.

[0089] Table 1: Network model training parameter example table

[0090]

[0091] By performing the above operations, for the technical problems of lacking physical constraints, insufficient prediction accuracy and interpretability in existing soil water and salt transport dynamic prediction methods, for example, the prediction method based on conventional machine learning or simple regression model can only fit the static correlation between historical salt content and rainfall, groundwater depth, and cannot guarantee that the prediction result meets the salt conservation or water diffusion law, so the deviation will accumulate gradually in long-term prediction, and it is difficult to be used for agricultural decision-making; the scheme creatively adopts a spatio-temporal cubic modeling adaptive prediction model combined with physical correction, extracts water and salt spatio-temporal features through three-dimensional convolution and convolution LSTM combined with deconvolution decoding, and introduces a physical consistency loss function based on the water and salt diffusion equation to constrain the error, realizing high-precision prediction considering data-driven and physical mechanism, and ensuring the physical rationality of the water and salt transport prediction result in long-term time series and spatial distribution.

[0092] Embodiment four, refer to Figure 1 , Figure 2 and Figure 4 , this embodiment is based on the above embodiment, in step S3, the crop salt tolerance modeling is used to establish the growth adaptation model of different crops under different salt conditions according to the crop variety and growth data, specifically, the salt tolerance modeling is performed by using the salt stress threshold segmentation modeling method combined with crop physiological response fitting according to the standardized fusion data set, to obtain a crop salt tolerance response model, including the following steps:

[0093] Step S31: input factor extraction, specifically, the target crop yield value is extracted as a response variable, and soil salt content, water content, air temperature, crop species code and sowing time are extracted to form an input matrix, variable and factor construction is performed, and an original input factor matrix and an observed response variable are obtained;

[0094] Step S32: Salt stress threshold interval division, specifically, a multi-segment threshold division is performed on the soil salt level by a mutation point detection algorithm, a salt interval segmentation processing is performed, and a salt threshold set is obtained;

[0095] Preferably, the mutation point detection algorithm is implemented based on a method combining Bayesian information criterion and dynamic programming, is used for automatically identifying a significant turning point on a curve of crop yield change with salinity, and is used for determining typical threshold intervals such as low salinity, medium salinity and high salinity, preferably 3 to 5 segments, so as to ensure that the segmentation has biological significance and avoids overfitting;

[0096] Step S33: Physiological response factor function modeling, specifically, a nonlinear response function based on exponential decay and bottom line correction is constructed in each salinity interval, an interval fitting calculation is performed, and a segmented salinity response function is obtained;

[0097] The calculation formula of the segmented salinity response function is:

[0098] ;

[0099] In the formula, f(·) is the segmented salinity response function, S is a salinity index for representing salt concentration, k is the number of salinity segments, i is the salinity segment index, is an indicator function, is a salinity threshold set, wherein is the salinity threshold corresponding to the i th salinity segment, f i (S) is a physiological response function;

[0100] Preferably, the calculation formula of the physiological response function is:

[0101] ;

[0102] In the formula, f i (S) is a physiological response function, a i is an interval initial yield level, preferably an average observed yield at the starting point of the salinity interval, b i is a sensitivity coefficient of yield to salinity change, preferably in the range of 0.01 to 0.5, for adjusting the speed of exponential decline, S is a salinity index for representing salt concentration, and c i is a bottom line correction term, preferably greater than zero, to ensure that when the salinity continues to rise, the predicted yield will not tend to negative, but will be stable at a certain minimum physiological yield level;

[0103] Step S34: Covariate coupling response correction, specifically, a weighted modulation factor containing moisture content, air temperature, sowing time and crop species is constructed, the salinity response function is corrected, and a covariate coupling correction model is obtained;

[0104] The calculation formula of the weighted modulation factor is:

[0105] ;

[0106] In the formula, is a weighted modulation factor, w1 is a water content weight, the default value is 0.3, W is a water content factor, w2 is a temperature weight, the default value is 0.3, T is a temperature factor, w3 is a sowing time weight, the default value is 0.2, Ts is a sowing time factor, w4 is a crop type weight, the default value is 0.2, and C is a crop type factor;

[0107] The calculation formula of the covariate coupling correction model is:

[0108] ;

[0109] In the formula, Y pred is the predicted output of the covariate coupling correction model, f(·) is a piecewise salinity response function, is a coupling strength coefficient, preferably the value range is 0.1 to 0.5, is a weighted modulation factor;

[0110] Step S35: parameter optimization model training, specifically, by minimizing the yield prediction error loss function and combining regularization constraints for parameter optimization training, the final crop salt tolerance response model is obtained;

[0111] Preferably, the minimum yield prediction error loss function adopts mean square error (MSE), the regularization constraint adopts L2 regularization to prevent overfitting, the optimizer adopts Adam optimizer, the learning rate is preferably 1e-3, the training batch size is preferably 16 to 32, the training round is preferably 50 to 100, and the early stop strategy is introduced, when the validation set error does not decrease in continuous rounds, the training is terminated in advance, thereby improving the generalization performance of the model.

[0112] By performing the above operation, for the technical problem that in the existing crop salt tolerance modeling method, there is a threshold division, which is rough and difficult to reflect the nonlinear response of crops under different salt conditions. For example, traditional research usually divides whether crops are salt tolerant with a single "salinity critical value", ignoring the typical segmented characteristics that yield is almost unaffected in the low salt interval, yield slowly decreases with salt in the medium salt interval, and yield sharply decreases in the high salt interval, resulting in that the model cannot accurately depict the yield-salt relationship. The present scheme creatively adopts a salt tolerance response improvement method combining crop physiological response fitting and salt stress threshold segmented modeling, automatically divides the salinity interval through variation point detection, and uses an exponential decay function and a covariate weighted correction factor within the interval to finely depict crop yield response under different salinity, realizing the transition of salt tolerance modeling from single threshold rough description to multi-interval and adjustable fine response modeling, and improving the adaptability of the model to different crops and different ecological environments.

[0113] Embodiment five, refer to Figure 1 、 Figure 2 and Figure 5 , this embodiment is based on the above embodiment, in step S4, the migration salt tolerance coupling modeling is used to fuse the soil water and salt change and the crop salt tolerance model to simulate the response characteristics of crops in a dynamic salt environment, specifically, a time sequence coupling modeling method combining dynamic water and salt simulation and crop response regulation mechanism is used to perform migration salt tolerance coupling modeling according to the dynamic water and salt simulation data and the crop salt tolerance response model, to obtain a coupling prediction model, including the following steps:

[0114] Step S41: dynamic water and salt state sequence construction, specifically, by extracting salt data and water data in the crop growth period, time step driving variable construction is performed to obtain a water and salt driving matrix;

[0115] Preferably, the water and salt driving matrix uses the same 30 m x 30 m grid as step S21 in space, uses a time window consistent with the crop growth period in time, and uses a daily or weekly scale as a time step; wherein the driving variables of each time step include root zone soil salt concentration (represented by electrical conductivity EC), water content and rainfall evaporation factor;

[0116] Step S42: physiological response memory factor design, specifically, by constructing a salt stress cumulative index and using an exponential decay weight function, a crop salt stress memory effect modeling is performed to obtain a time step memory factor;

[0117] The calculation formula of the time step memory factor is:

[0118] ;

[0119] In the formula, R tis a time step memory factor, used to represent the delayed effect of past high-salinity periods on current growth, t is a time index, is a time step memory index, is a time step of the time step weight, is a time step of the root zone salinity, specifically represented by electrical conductivity EC, f(·) is a piecewise salinity response function;

[0120] wherein the calculation formula of the time step weight is:

[0121]

[0122] in the formula, is a decay coefficient, preferably ranging from 0.01 to 0.1, used to control the rate of attenuation of historical salinity stress over time;

[0123] Step S43: stage sensitivity regulation factor design, specifically, by identifying different growth stages of crops and setting stage sensitivity coefficients, the salinity sensitivity is corrected to obtain a stage regulation factor;

[0124] Preferably, the growth stages are segmented according to agronomic segmentation methods, specifically into seedling stage, heading stage and mature stage; the sensitivity coefficients of each stage are preferably in the range of 0.6 to 0.8 for the seedling stage, 0.4 to 0.6 for the heading stage, and 0.2 to 0.3 for the mature stage, to reflect the differential sensitivity of different growth stages to salinity stress;

[0125] Step S44: dynamic coupling of crop response model, specifically, by calling the crop salt-tolerant response model, combining the time step memory factor and the stage regulation factor, dynamic yield calculation is performed to obtain a dynamic response yield sequence;

[0126] Preferably, the calculation formula of the dynamic yield calculation is:

[0127]

[0128] in the formula, Y t is a predicted yield contribution value, is the output of the crop salt-tolerant response model under the current salinity condition, S t is the salinity index at time t, used to represent the current salinity condition, is a memory effect weight coefficient, preferably ranging from 0.1 to 0.5, R t is a time step memory factor, is a stage sensitivity regulation factor

[0129] ​​Step S45: coupling prediction model construction, specifically, calibrating and optimizing the dynamic response yield sequence based on the yield prediction error minimization principle to obtain the final coupling prediction model;

[0130] Preferably, the yield prediction error is measured by root mean square error (RMSE) or mean absolute percentage error (MAPE), and the parameters are iteratively updated by Adam optimizer, and the learning rate is preferably 1e-3; At the same time, a regularization term is introduced, preferably L2 regularization, and the regularization coefficient ranges from 1e-4 to 1e-2, to prevent model overfitting.

[0131] By performing the above operation, for the existing migration salt tolerance coupling modeling method, there is a technical problem that only static coupling is done, and the time sequence cumulative effect and stage sensitivity in the growth process of crops are not considered, for example, some studies only combine average salinity with yield regression, but fail to reflect the difference of crops in the stages of seedling stage, tillering stage, and grain filling stage under salt stress, and also fail to introduce the historical memory effect of salt stress, so the prediction result often has a large gap with the measured yield; The scheme creatively adopts a time sequence coupling modeling method combining dynamic water and salt simulation and crop response regulation mechanism, introduces an exponentially decaying salt memory factor and a stage sensitivity regulation factor, so that the model can dynamically describe the accumulation and recovery process of salt stress on the time scale, and combine the crop salt tolerance response model for dynamic calling, and finally realize the dynamic response simulation of the whole growth process of crops in saline-alkali area, thereby significantly improving the yield estimation accuracy and reliability.

[0132] Embodiment six, refer to Figure 1 and Figure 2 This embodiment is based on the above-mentioned embodiments, in step S5, the crop yield estimation and planting optimization is used to calculate the predicted yield of crops planted in a specific area based on the coupling model output and environmental parameters, specifically, using the coupling prediction model to estimate the yield of crops, obtaining regional yield estimation data, and identifying high-yield and low-yield difference areas by spatial distribution analysis of the regional yield estimation data, and combining the salt tolerance grade and yield threshold of different crop varieties to select and match the varieties, and then generating a sowing area configuration scheme according to the spatial distribution characteristics and the variety matching result, and scheduling and optimizing the irrigation, fertilization and salt drainage field management measures based on the dynamic data of water and salt migration, thereby comprehensively outputting the regional planting optimization strategy to obtain planting optimization data.

[0133] Preferably, the field management scheduling optimization is quantitatively modeled by constructing a multi-objective optimization function, so as to improve the total yield of the region, reduce the resource investment cost and maintain the water-salt balance as the main optimization objectives, and the optimization function is constructed; the optimization function is iterated under the constraint conditions, and the constraint conditions specifically include that the total cultivated land area of the region needs to be controlled within the planning range, the sowing proportion of various crops should meet the agricultural structure requirements, the total amount of irrigation and fertilization should not exceed the available resource threshold, and the salinity level needs to be lower than the upper limit of the crop tolerance;

[0134] Under the premise of meeting the above constraint conditions, the generated sowing area configuration scheme is preferably iteratively solved by using an improved multi-objective evolutionary algorithm, so as to obtain a field management scheme that takes into account yield, cost and risk.

[0135] Embodiment seven, refer to Figure 1 and Figure 2 Based on the above-mentioned embodiments, the present application provides a planting optimization system based on crop yield estimation in saline-alkali regions, which comprises a data acquisition and fusion processing module, a soil water and salt transport dynamic prediction module, a crop salt tolerance modeling module, a transport salt tolerance coupling modeling module and a planting optimization module.

[0136] The data acquisition and fusion processing module is used for data acquisition and fusion processing, and through data acquisition and fusion processing, a standardized fusion data set is obtained, and the standardized fusion data set is sent to the soil water and salt transport dynamic prediction module and the crop salt tolerance modeling module.

[0137] The soil water and salt transport dynamic prediction module is used for soil water and salt transport dynamic prediction, and through soil water and salt transport dynamic prediction, water and salt transport dynamic data are obtained, and the water and salt transport dynamic data are sent to the transport salt tolerance coupling modeling module.

[0138] The crop salt tolerance modeling module is used for crop salt tolerance modeling, and through crop salt tolerance modeling, a crop salt tolerance response model is obtained, and the crop salt tolerance response model is sent to the transport salt tolerance coupling modeling module.

[0139] The transport salt tolerance coupling modeling module is used for transport salt tolerance coupling modeling, and through transport salt tolerance coupling modeling, a coupling prediction model is obtained, and the coupling prediction model is sent to the planting optimization module.

[0140] The planting optimization module is used for crop yield estimation and planting optimization, and through crop yield estimation and planting optimization, regional yield estimation data and planting optimization strategies are obtained.

[0141] It is to be understood that the terminology used herein is for the purpose of describing particular embodiments only and is not intended to be limiting; it is not intended to exclude myriad other embodiments of the present application that other presenters can develop. It is also possible, however, that only a single element can be present. Terms such as "a", "an", and "the" are not intended to refer to only a singular entity but include the general class of which a single element is only one species, unless otherwise indicated. Furthermore, the use of the terms "primary" and "secondary", "first" and "second", etc., designate different Stages in the process, and are not intended to otherwise limit the number of stages which can be employed. The terminology includes the words specifically noted above, derivatives thereof, and words of similar import. The designation of a component as "optional" indicates that the component is "optional" and can or can not be present or used in the practice of the application, but that when it is present or used, it can be used in varying embodiments of the present application.

[0142] While the embodiments of the application have been shown and described, it is to be understood that various further modifications and changes can be made thereto without departing from the spirit and scope of the present application.

[0143] The above description of the application and its embodiments is not intended to limit the application, as described by the appended claims, to the one embodiment shown and described above. Rather, the scope of the present application is intended to cover all embodiments falling within the scope of the appended claims, and the scope of the claims themselves, including any claims which can issue from this application, as equivalents to the recited embodiments.

Claims

1. A planting optimization method based on crop yield prediction in saline-alkali areas, characterized by: The method comprises the following steps: Step S1: data acquisition and fusion processing, obtaining a standardized fusion data set through multi-source data acquisition and standardization and processing flow; Step S2: soil water and salt transport dynamic prediction, based on the standardized fusion data set, using a spatiotemporal cubic modeling adaptive prediction model combined with physical correction to predict the soil water and salt transport dynamics, and obtaining water and salt transport dynamic data; Step S3: crop salt tolerance modeling, based on the standardized fusion data set, using an improved salt tolerance response method combining crop physiological response fitting and salt stress threshold segmentation modeling to model crop salt tolerance, and obtaining a crop salt response model, comprising the following steps: Step S31: input factor extraction; Step S32: salt stress threshold interval division; Step S33: physiological response factor function modeling; Step S34: covariate coupled response correction; Step S35: parameter optimization model training; Step S4: transport salt tolerance coupled modeling, based on the water and salt transport dynamic data and the crop salt response model, using a time sequence coupled modeling method combining dynamic water and salt simulation and crop response regulation mechanism to model the transport salt tolerance coupling, and obtaining a coupled prediction model; Step S5: crop yield prediction and planting optimization, using the coupled prediction model to predict crop yield, obtaining regional yield prediction data, and comprehensively outputting regional planting optimization strategies to obtain planting optimization data. 2.The method of claim 1, wherein the method further comprises: determining a crop yield prediction based on the soil salinity data and the crop yield data; and determining a planting optimization based on the crop yield prediction. In step S1, the multi-source data acquisition includes remote sensing data acquisition, field monitoring data acquisition and historical statistical data retrieval. 3.The method of claim 2, wherein the method further comprises: determining a crop yield prediction based on the soil salinity and the soil water content. In step S2, the soil water and salt transport dynamic prediction is used to build a spatiotemporal prediction model to simulate the dynamic distribution process of salt and water in the soil, specifically, based on the standardized fusion data set, using a spatiotemporal cubic modeling adaptive prediction model combined with physical correction to predict the soil water and salt transport dynamics, and obtaining water and salt transport dynamic data, comprising the following steps: Step S21: data resampling, specifically, through uniform spatial resolution and time step, interpolating, filling missing values and rasterizing the original soil water and salt data, performing spatiotemporal standardization reconstruction, and obtaining a standardized tensor sequence; Step S22: spatiotemporal cubic structure modeling, specifically, by encapsulating the spatial grid tensor of continuous time steps into a spatiotemporal cubic input unit, constructing a cubic data structure, and obtaining a spatiotemporal input tensor; Step S23: spatiotemporal feature extraction and coding, specifically, by performing convolution calculation on the spatiotemporal cubic through a three-dimensional convolution network, performing spatiotemporal feature coding, and obtaining latent space state features; Step S24: deconvolution long short-term memory network decoding, specifically, by inputting the latent space state features into a convolution long short-term memory network and combining a deconvolution layer for decoding, performing time series prediction recovery, and obtaining a future time step water and salt state prediction frame; Step S25: constructing a physical constraint loss function, specifically, by introducing a water and salt diffusion equation and a conservation relationship, constructing a regularization constraint based on mass conservation and diffusion gradient, and performing prediction error correction to obtain a physical consistency loss; Step S26: soil water and salt transport dynamic prediction, specifically, network model training and reasoning are performed by jointly minimizing mean square error loss and physical consistency loss, to obtain water and salt transport dynamic data. 4.The method of claim 3, wherein the method further comprises: determining a crop yield prediction based on the soil salinity and the soil water content. In step S31, the input factor is extracted, specifically, the target crop yield value is extracted as the response variable, and the soil salt content, water content, air temperature, crop type code and sowing time are extracted to form an input matrix, the variables and factors are constructed, and the original input factor matrix and observed response variable are obtained; In step S32, the salt stress threshold interval is divided, specifically, the soil salt level is divided into multiple threshold intervals by a variation point detection algorithm, the salt interval is segmented, and a salt threshold set is obtained. In step S33, the physiological response factor function modeling is performed, specifically, a nonlinear response function based on exponential decay and bottom correction is constructed in each salt interval, interval fitting calculation is performed, and a segmented salt response function is obtained. 5.The method of claim 4, wherein the method further comprises: determining a crop yield prediction based on the soil salinity and the soil water content. In step S34, the covariate coupled response correction is performed, specifically, the water content, air temperature, sowing time and crop type are used to construct a weighted modulation factor, the salt response function is corrected, and a covariate coupled correction model is obtained. In step S35, the parameter optimization model training is performed, specifically, the yield prediction error loss function is minimized, and parameter optimization training is performed combined with regularization constraint, and a final crop salt tolerance response model is obtained. 6.The method of claim 5, wherein the method further comprises: determining a crop yield prediction based on the soil salinity and the soil water content. In step S4, the transport salt tolerance coupling modeling is used to fuse soil water and salt changes with crop salt tolerance model to simulate the response characteristics of crops in a dynamic salt environment, specifically, according to the water and salt transport dynamic data and the crop salt tolerance response model, a time sequence coupling modeling method combining dynamic water and salt simulation and crop response regulation mechanism is used for transport salt tolerance coupling modeling, a coupled prediction model is obtained, including the following steps: Step S41: dynamic water and salt state sequence construction, specifically, salt and water data in the crop growth period are extracted, time step driven variable construction is performed, and a water and salt driven matrix is obtained; Step S42: physiological response memory factor design, specifically, a salt stress cumulative index is constructed and an exponential decay weight function is used to model the crop salt stress memory effect, and a time step memory factor is obtained; Step S43: stage sensitivity regulation factor design, specifically, different growth stages of crops are identified and stage sensitivity coefficients are set, salt sensitivity correction is performed, and a stage regulation factor is obtained; Step S44: dynamic coupling of crop response model, specifically, the crop salt tolerance response model is called, combined with the time step memory factor and the stage regulation factor, dynamic yield calculation is performed, and a dynamic response yield sequence is obtained; Step S45: coupled prediction model construction, specifically, the dynamic response yield sequence is calibrated and optimized based on the principle of minimizing yield prediction error, and a final coupled prediction model is obtained. 7.The method of claim 6, wherein the method further comprises: determining a crop yield prediction based on the soil salinity and the soil water content. In step S5, the crop yield estimation and planting optimization are used to calculate the predicted yield of crops planted in a specific area based on the coupling model output and environmental parameters. Specifically, the crop yield estimation is performed using the coupling prediction model to obtain regional yield estimation data. By analyzing the spatial distribution of the regional yield estimation data, high-yield and low-yield difference areas are identified. Combined with the salt tolerance levels and yield thresholds of different crop varieties, variety selection and matching are performed. Then, based on the spatial distribution characteristics and variety matching results, a sowing area configuration scheme is generated. Combined with the dynamic water and salt transport data, irrigation, fertilization, and salt drainage field management measures are optimized, thereby comprehensively outputting the regional planting optimization strategy to obtain planting optimization data.

8. A planting optimization system based on crop yield estimation in saline and alkaline regions for implementing the method of planting optimization based on crop yield estimation in saline and alkaline regions according to any one of claims 1 to 7, characterized in that: The system comprises a data acquisition and fusion processing module, a soil water and salt transport dynamic prediction module, a crop salt tolerance modeling module, a transport salt tolerance coupling modeling module, and a planting optimization module. 9.The system for crop yield estimation and planting optimization based on saline-alkali region of claim 8, wherein: The data acquisition and fusion processing module is configured to perform data acquisition and fusion processing to obtain a standardized fusion data set and send the standardized fusion data set to the soil water and salt transport dynamic prediction module and the crop salt tolerance modeling module. The soil water and salt transport dynamic prediction module is configured to perform soil water and salt transport dynamic prediction to obtain water and salt transport dynamic data and send the water and salt transport dynamic data to the transport salt tolerance coupling modeling module. The crop salt tolerance modeling module is configured to perform crop salt tolerance modeling to obtain a crop salt tolerance response model and send the crop salt tolerance response model to the transport salt tolerance coupling modeling module. The transport salt tolerance coupling modeling module is configured to perform transport salt tolerance coupling modeling to obtain a coupling prediction model and send the coupling prediction model to the planting optimization module. The planting optimization module is configured to perform crop yield estimation and planting optimization to obtain regional yield estimation data and planting optimization strategies.

Images