Winter wheat yield prediction method and system fusing genetic constraints and attention model

Abstract

The present application relates to the field of agricultural information technology, and particularly relates to a winter wheat post-harvest phenotype prediction method and system fusing genetic constraints and attention models, the method comprising: S1, obtaining multi-source ecological time series data of a target plot and a historical same-period wheat phenotype observation sequence; S2, constructing a winter wheat growth model to simulate potential biomass accumulation and organ building dynamics at a daily step, and outputting intermediate variables; S3, encoding the multi-source ecological time series data into a high-dimensional feature tensor, inputting a pre-trained multi-modal model, and outputting spatio-temporal context features, the multi-modal model comprising a time series Transformer branch and a convolutional space branch; S4, predicting a predicted phenotype value of the target plot using the spatio-temporal context features and the intermediate variables, so that the predicted phenotype value conforms to basic physiological and ecological laws; and S5, calculating an uncertainty interval of the predicted phenotype value.

Description

Technical Field

[0001] This invention relates to the field of agricultural information technology, and in particular to a method and system for predicting the phenotypic traits of winter wheat by integrating genetic constraints and attention models. Background Technology

[0002] Winter wheat is a vital global food crop, and its stable yield is crucial for ensuring food security. Winter wheat phenotypes, such as plant height, biomass, and ultimate yield components, are direct indicators for measuring its growth status and assessing yield. Accurate and dynamic prediction of winter wheat phenotypes is of great significance for achieving refined field management, optimizing resource allocation, and addressing climate change, and is one of the core technologies for the development of modern smart agriculture.

[0003] Currently, the main technical means for predicting winter wheat phenotypes include: crop growth models based on physical processes, which use mathematical formulas to describe core physiological and ecological processes such as crop photosynthesis, water use, and nutrient cycling, and have strong mechanistic properties; and purely data-driven statistical or machine learning models, which use historically accumulated meteorological, remote sensing, and field observation data to learn the complex nonlinear relationship between input variables and output phenotypes through algorithms, and make predictions directly.

[0004] However, while physical process-based models have clear mechanisms, their numerous internal parameters are difficult to calibrate precisely, and they oversimplify the complex interactions between genes and the environment, resulting in limited prediction accuracy for specific locations. Pure data-driven models, on the other hand, heavily rely on the quantity and quality of training data. Their prediction process lacks clear physiological constraints, and when faced with extreme climate events not covered by the training data, they are prone to producing predictions that violate basic growth principles, exhibiting insufficient generalization ability and reliability. Summary of the Invention

[0005] This invention integrates environmental regulation coefficients and genetic constraints in phenotypic prediction, ensuring that the phenotypic prediction results conform to physiological and ecological laws while reflecting the influence of the environment, thus guaranteeing the accuracy and reliability of the prediction.

[0006] The technical solution proposed in this invention is: a method for predicting the phenotypic traits of winter wheat by integrating genetic constraints and attention models, the method comprising:

[0007] S1. Obtain multi-source ecological time-series data of the target plot and historical wheat phenotypic observation sequences for the same period;

[0008] S2. Construct a winter wheat growth model by inputting multi-source ecological time-series data into the winter wheat growth model, simulating potential biomass accumulation and organ formation dynamics with daily step size, and outputting intermediate variables.

[0009] S3. Encode the multi-source ecological time-series data into a high-dimensional feature tensor, input it into a pre-trained multimodal model, and output spatiotemporal context features. The multimodal model includes a temporal Transformer branch and a convolutional spatial branch.

[0010] S4. Utilize spatiotemporal context features and intermediate variables to predict the phenotypic values ​​of the target plots, so that the predicted phenotypic values ​​conform to basic physiological and ecological laws.

[0011] S5. Calculate and obtain the uncertainty interval of the predicted phenotypic value.

[0012] Preferably, the multi-source ecological time-series data includes daily meteorological data, soil moisture data, field management records, and remote sensing vegetation indices;

[0013] The daily meteorological data includes: temperature, precipitation, sunshine duration, and wind speed;

[0014] The soil moisture data includes: water content, electrical conductivity, and organic matter content of soil at different depths;

[0015] The field management records include: sowing date, fertilizer application amount, and irrigation date;

[0016] The remote sensing vegetation indices include: Normalized Difference Vegetation Index (NDVI), Enhanced Vegetation Index (EVI), and Chlorophyll Index (LCI).

[0017] The wheat phenotypic observation sequence includes: plant height, number of tillers per plant, leaf area index, aboveground biomass, spike length, number of grains per spike, and thousand-grain weight at harvest.

[0018] Preferably, the winter wheat growth model adopts a dry matter production framework based on radiation use efficiency (RUE) and introduces an accumulated temperature correction function. When the effective accumulated temperature deviates from the optimal range for the variety, the assimilation rate is reduced.

[0019] The ratio of effective soil moisture content in the root zone to actual evapotranspiration demand is defined as the water stress factor. The reference evapotranspiration is calculated using the Penman-Monteith formula for potential evapotranspiration, so as to dynamically adjust the stomatal conductance limitation coefficient.

[0020] Preferably, the temporal Transformer branch in the multimodal model uses non-uniform time-interval coding to adapt to situations where meteorological observations are missing or sampling is uneven;

[0021] The convolutional spatial branch employs dilated convolution to capture the synergistic effects of large-scale meteorological systems on wheat population structure.

[0022] Preferably, the step of encoding multi-source ecological time-series data into a high-dimensional feature tensor, inputting it into a pre-trained multimodal model, and outputting spatiotemporal context features includes:

[0023] Encode multi-source ecological time-series data into T×H×W×C tensors H and W represent the length and width of the grid for the target plot and the surrounding plots; C represents the number of channels for the variable, and T represents the time resolution.

[0024] For the first The H×W×C slices at each time point are denoted as follows: ;

[0025] use Obtain the global temporal feature vector and global spatial feature vector of the target plot;

[0026] Spatiotemporal context feature vectors are obtained by using global temporal feature vectors and global spatial feature vectors.

[0027] Preferably, the step of using spatiotemporal context features and intermediate variables to predict the predicted phenotypic values ​​of the target plot, so that the predicted phenotypic values ​​conform to basic physiological and ecological laws, includes:

[0028] Obtain the spatiotemporal context feature vector;

[0029] Construct a spliced ​​vector using all intermediate variables;

[0030] Obtain the spatiotemporal context feature vector, and input the concatenated vector and the spatiotemporal context feature vector into an affine parameter network for fusion to obtain the corrected feature vector. ;

[0031] Through a fully connected network ,Will Mapped to K scalar values, and the output constrained by the sigmoid activation function. The environmental response control vector is obtained from the interval; the environmental response control coefficient is a continuous value between 0 and 1.

[0032] The varietal-inherent parameters in the growth model were calibrated using wheat phenotypic observation sequences from the same historical period to obtain a baseline of genetic potential.

[0033] The predicted phenotypic value is corrected by the environmental response regulation vector to obtain the final predicted phenotypic value; the final predicted phenotypic value is the product of the genetic potential baseline and the environmental response regulation vector, so that the predicted phenotypic value conforms to basic physiological and ecological laws.

[0034] Preferably, the step of calibrating the variety-inherent parameters in the winter wheat growth model using wheat phenotypic observation sequences from the same historical period to obtain the genetic potential baseline includes:

[0035] Set the parameter set for the target variety ;

[0036] The maximum number of tillers per plant. For potential thousand-grain weight, Potential number of grains per ear;

[0037] The objective function is constructed by weighting the root mean square error of historical predicted phenotypic values ​​and the wheat phenotypic observation series of the same historical period as the optimization objective. :

[0038] ;

[0039] in, This represents the root mean square error of the leaf area index for that day. The root mean square error of plant height. The phenotypic value for the thousand-grain weight at maturity is [value]. This represents the observed weight of 1000 grains at maturity. The predicted phenotypic value for the number of grains per ear. This refers to the observed number of grains per ear. , , , These are the trait weights, all of which take a value of 1;

[0040] The initial value range of the intrinsic parameters is: The value range is 1.2-2.8 g / MJ; The value range is 0.4-0.8; The value range is 2-8; The value range is 35-55g. The value range is 25-55 grains / ear;

[0041] Using a gradient-free optimization algorithm The optimal parameter set is obtained by minimizing the range of values ​​for the intrinsic parameters within the objective function. ;

[0042] After calibration, the parameters in the optimal parameter set are the inherent parameters of the variety, and the inherent parameters are used to construct the genetic potential baseline.

[0043] Preferably, the calculation of the uncertainty interval for obtaining the predicted phenotypic value includes:

[0044] Based on Monte Carlo Dropout, output the variance of the predicted phenotypic values;

[0045] The uncertainty interval for the predicted phenotypic values ​​is constructed using the variance of the predicted phenotypic values.

[0046] A winter wheat successor representative type prediction system integrating genetic constraints and attention models, the system being used to execute the aforementioned winter wheat successor representative type prediction method integrating genetic constraints and attention models.

[0047] A computer-readable storage medium storing a computer program that is executed by a processor to implement the above-described method for predicting the phenotypic representation of winter wheat by incorporating genetic constraints and attention models.

[0048] The beneficial effects of this invention are:

[0049] 1. In this invention, a multimodal large model is used to effectively integrate multi-source heterogeneous data and can deeply capture the long-term temporal dependence and large-scale spatial synergistic influence in the growth process, thereby improving the accuracy of prediction.

[0050] 2. In this invention, the final predicted phenotypic value is the product of the genetic potential baseline and the environmental response regulation vector, reflecting the influence of the environment on the predicted phenotypic value and ensuring that the predicted phenotypic value does not exceed the genetic potential baseline, thus achieving genetic constraint. Furthermore, by quantifying the uncertainty range of the prediction results, the interpretability of the prediction results is improved. Attached Figure Description

[0051] Figure 1 This is a flowchart of the winter wheat phenotypic prediction method that integrates genetic constraints and attention models according to the present invention. Detailed Implementation

[0052] The following description is intended to disclose the present invention and enable those skilled in the art to implement it. The preferred embodiments described below are merely examples, and other obvious modifications will occur to those skilled in the art. The basic principles of the invention defined in the following description can be applied to other embodiments, modifications, improvements, equivalents, and other technical solutions that do not depart from the spirit and scope of the invention.

[0053] It is understood that the term "a" should be understood as "at least one" or "one or more", that is, in one embodiment, the number of an element can be one, while in another embodiment, the number of the element can be multiple, and the term "a" should not be understood as a limitation on the number.

[0054] refer to Figure 1 The technical solution provided by this invention is: a method for predicting the representative phenotype of winter wheat by integrating genetic constraints and attention models, comprising the following steps:

[0055] S1. Obtain multi-source ecological time-series data of the target plot and historical wheat phenotypic observation sequences for the same period.

[0056] Multi-source ecological time-series data includes daily meteorological data, soil moisture data, field management records, and remote sensing vegetation indices. The daily meteorological data includes temperature, precipitation, sunshine duration, and wind speed, acquired through meteorological stations deployed near the target site.

[0057] Soil moisture data includes: water content, electrical conductivity, and organic matter content of soil at different depths. This data is obtained through in-situ, continuous monitoring using soil sensor arrays buried at different depths in the target site, or through periodic soil sampling and analysis in a laboratory.

[0058] Field management records include: sowing date, fertilizer application rate, and irrigation date. These are obtained through experiments or planting logs.

[0059] Remote sensing vegetation indices include: Normalized Difference Vegetation Index (NDVI), Enhanced Vegetation Index (EVI), and Chlorophyll Index (LCI). These are obtained through remote sensing satellite or drone inspection images.

[0060] The formula for calculating the normalized vegetation index is: ,in, The normalized vegetation index is a calculated value, ranging from -1 to 1. The surface reflectance is in the near-infrared band. This represents the surface reflectance in the red light band.

[0061] The formula for calculating the enhanced vegetation index is: ,in, The calculated value represents the enhanced vegetation index; These are the surface reflectances in the near-infrared, red, and blue light bands, respectively. It is the gain factor; and This is the aerosol correction factor; These are soil conditioning factors. These coefficients are all dimensionless parameters, set according to the standard Moderate Resolution Imaging Spectroradiometer (MODIS) EVI algorithm, and are typically set as follows: .

[0062] The chlorophyll index is more sensitive to the chlorophyll content in leaves and is often used to indicate the nitrogen nutrition status of crops. Its calculation formula is: ,in, The calculated value representing the chlorophyll index; and The meaning is the same as above; Represents the surface reflectance of the red-edge band, which is sensitive to changes in chlorophyll concentration.

[0063] The wheat phenotypic observation sequences include: plant height, number of tillers per plant, leaf area index, aboveground biomass, spike length, number of grains per spike, and thousand-grain weight at harvest.

[0064] After preprocessing, all the data obtained above are aligned on the timeline to form a multi-source ecological time-series data set.

[0065] S2. Construct a winter wheat growth model by inputting multi-source ecological time-series data into the model. Simulate potential biomass accumulation and organ formation dynamics using daily step sizes, and output intermediate variables. This includes the following steps:

[0066] A winter wheat growth model was constructed using a dry matter production framework based on radiation use efficiency (RUE), and an accumulated temperature correction function was introduced. When the effective accumulated temperature deviates from the optimal range for the variety, the assimilation rate was lowered; specifically:

[0067] Daily potential dry matter increment The calculation formula is:

[0068] ,in The potential radiation utilization efficiency (g / MJ) of a variety under optimal temperature and light conditions is an inherent genetic parameter of the variety. The instantaneous limiting factor of the daily average temperature on photosynthesis is still the aforementioned three-point piecewise linear function. It is an accumulated temperature correction function that reflects whether the accumulated temperature is appropriate throughout the entire reproductive process. It is a downregulation factor on a seasonal or developmental stage scale, rather than an instantaneous response to daily temperature. The formula for calculating the photosynthetically active radiation (MJ / m² / day) intercepted by the canopy is as follows: . The extinction coefficient (0.6-0.7 for winter wheat) is used. Total solar radiation of the day, This represents the leaf area index for that day.

[0069] in, ; The average daily temperature The growth base point temperature is 0°C. The optimal growth temperature is 20℃. This is the upper limit temperature for growth (35℃).

[0070] Accumulated temperature correction function This term describes how, when the actual accumulated heat conditions at a certain growth stage deviate significantly from the historical adaptation range of a variety, even if the instantaneous temperature is suitable, the crop assimilation efficiency will decrease due to the disruption of the overall growth rhythm.

[0071] ,in The actual cumulative effective accumulated temperature from sowing to the current day ( ). It represents a day (24 hours).

[0072] Daily effective accumulated temperature ,in, The base temperature for development (0℃ for winter wheat). .in, This represents the total number of days from sowing to the current day.

[0073] The optimal accumulated temperature for the variety at the current developmental stage ( It is a developmental stage. Linear functions:

[0074] , The optimal total accumulated temperature for the entire growth period of the variety (variety parameter). This represents the current developmental stage (value 0-1).

[0075] This is a tolerance parameter for the suitable range of accumulated temperature, indicating the degree to which the accumulated temperature is allowed to deviate from the optimum value. Variety parameters, such as 150~250 .

[0076] When the actual accumulated temperature Approaching the theoretical optimum hour, No further downward adjustment will be applied. If a prolonged period of low temperatures in spring causes a significant lag in accumulated temperature, or if abnormally high temperatures cause a substantial lead in accumulated temperature, then... The assimilation rate is reduced by decaying in the form of a Gaussian function, thus simulating the decrease in photosynthetic efficiency caused by the disruption of reproductive rhythm.

[0077] The ratio of effective soil moisture content in the root zone to actual evapotranspiration demand is defined as the water stress factor. A reference evapotranspiration rate is calculated using the Penman-Monteith formula for potential evapotranspiration, thereby dynamically adjusting the stomatal conductance limitation coefficient. Specifically:

[0078] The effective water content of the root zone soil is defined as the total water content within the effective root depth of the crop, between the field capacity and the permanent wilting point. ,in, This represents the current average volumetric water content in the root zone, derived from soil moisture data or model water balance simulations. is the wilting coefficient (volume water content at the permanent wilting point), and is a soil property constant. The current effective root depth (mm) increases with the developmental stage, gradually deepening from the initial depth at sowing to the maximum root depth.

[0079] The daily water balance update is based on the formula mentioned above:

[0080] , For precipitation, For irrigation volume, This represents the actual evaporation on that day. For deep penetration amount This occurs when the water content exceeds the field capacity. All parameters are in millimeters (mm).

[0081] The reference evapotranspiration was calculated using the Penman-Monteith formula. (mm / day) represents the evapotranspiration rate of a fully irrigated dwarf grass canopy, which is entirely determined by meteorological conditions.

[0082] ; Net radiation, Soil heat flux (on a daily scale, it can be ignored, i.e.) ), The slope of the saturated water vapor pressure-temperature curve. This is the hygrometer constant. The wind speed at a height of 2 meters The saturated vapor pressure, Actual water vapor pressure

[0083] Crop potential evapotranspiration Depend on Multiply by crop coefficient get. ,in, The coefficient represents the crop's evapotranspiration capacity under ideal moisture conditions, and varies with the developmental stage.

[0084] The water stress factor is calculated based on the ratio of effective water content to evapotranspiration demand, including:

[0085] Define the water supply satisfaction index .

[0086] This index indicates how many days' worth of atmospheric evapotranspiration can be met under current available soil water reserves. The unit is days; a higher value indicates more abundant moisture.

[0087] Water stress factors Using this index as the independent variable, a piecewise function is employed, namely:

[0088] ;

[0089] in, The critical moisture requirement is defined as the number of days required, depending on the variety or soil-related parameters. In this example, 3 days can be used. When soil water storage can meet the potential evapotranspiration for more than this number of days, it is considered that there is no stress.

[0090] when Below When the stress increases linearly, this is equivalent to directly using the ratio. It is still based on the ratio of effective water content to actual evaporation demand, but it introduces the concept of time buffer, which is more physiologically meaningful than a simple water volume ratio.

[0091] By calibrating the variety-specific parameters in the winter wheat growth model using wheat phenotypic observation sequences from the same historical period, a baseline of genetic potential was obtained. Specifically:

[0092] Set the parameter set for the target variety ;

[0093] This represents the maximum number of tillers per plant (used to limit tillering simulation). Potential thousand-grain weight (g / 1000 grains, used as the upper limit of final grain weight). This represents the potential number of grains per ear (individual grains per ear).

[0094] The objective function is constructed by weighting the sum of the root mean square error (RMSE) of the predicted phenotypic values ​​and the wheat phenotypic observation sequences of the same historical period as the optimization objective:

[0095] ; This represents the root mean square error of the leaf area index for that day. The root mean square error of plant height. The phenotypic value for the thousand-grain weight at maturity is [value]. The values ​​represent the thousand-grain weight at maturity (obtained from wheat phenotypic observation sequences). The predicted phenotypic value for the number of grains per ear. This represents the observed number of grains per ear. , , , These are the trait weights, which can all be set to 1 or adjusted according to the measurement reliability.

[0096] The initial value range of the intrinsic parameters is: The value range is 1.2-2.8 g / MJ; The value range is 0.4-0.8; The value range is 2-8; The value range is 35-55g. The value range is 25-55 grains / ear.

[0097] Gradient-free optimization algorithms (such as the Nelder-Mead simplex method) are used in The range of values ​​for the intrinsic parameters is minimized by the objective function. The optimal parameter set is obtained. .

[0098] After calibration, the parameters in the optimal parameter set are the inherent parameters of the variety.

[0099] The parameters that determine the maximum production potential are extracted and used as the genetic potential baseline. For example, the baseline for the maximum tillering number is... Potential particle weight baseline is The baseline for potential grain count per ear is The potential biomass accumulation curve can be obtained by running the model under stress-free conditions using these parameters.

[0100] The actual evaporation on that day Water use efficiency Light energy utilization rate , as an intermediate variable for output.

[0101] S3. Encode multi-source ecological time-series data into a high-dimensional feature tensor, input it into a pre-trained multimodal model, and output spatiotemporal context features. The multimodal model includes a temporal Transformer branch and a convolutional spatial branch. The temporal Transformer branch uses non-uniform time-interval encoding to adapt to situations where meteorological observations are missing or sampling is uneven. The convolutional spatial branch uses dilated convolution to capture the synergistic effects of large-scale meteorological systems (such as persistent drought or continuous rain) on wheat population structure. Specifically, this includes the following steps:

[0102] Using daily meteorological data, soil moisture data, field management records, and remote sensing vegetation index data from multi-source ecological time series data, a variable set was constructed that includes daily meteorological variables, soil moisture variables, field management variables, and remote sensing vegetation variables. Each variable includes C channels.

[0103] Daily meteorological variables and remote sensing index variables are covered by a grid covering the target area and surrounding areas, with a grid resolution of 500 meters, consisting of H×W grids.

[0104] Soil moisture variables and field management variables are broadcast spatially as a network identical to that of daily meteorological variables and remote sensing index variables, comprising H×W grids;

[0105] Multi-source ecological time-series data are then organized as a tensor of T×H×W×C. Where T is the time resolution, i.e., the time from seeding to the prediction endpoint; C is the number of channels for the variable;

[0106] For the first The H×W×C slices at each time point are denoted as follows: , representing the Spatial distribution of all variables in the sky.

[0107] Z-score standardization is performed on each channel individually to make the mean 0 and the standard deviation 1, thus eliminating the difference in dimensions.

[0108] First, map the C-dimensional vector at each time step to... The latent space is used to obtain the time mapping vector. ,in . For the first Tianwei A C-dimensional vector; This is the bias vector.

[0109] Non-uniform time interval encoding is performed, including:

[0110] Adopting based on actual number of days after sowing Continuous-time encoding replaces fixed-location encoding (due to potential missing observations or uneven sampling).

[0111] Time-coded vector Generated by sine functions of different frequencies:

[0112] ; This is a dimensional index. The encoding adapts to the actual time interval.

[0113] Spatial location encoding includes:

[0114] To distinguish different spatial locations, 2D learnable positional encoding is added. For each spatial location Generate time series units ; Pick or .

[0115] Sequence of all positions The same lightweight Transformer encoder (e.g., 2 layers, based on a multi-head self-attention mechanism) is shared and processed in parallel to obtain temporal feature maps. .

[0116] in, This indicates a lightweight Transformer encoder.

[0117] For each After several dilated convolutional layers and downsampling, a spatial feature map is obtained. .

[0118] For each Perform global average pooling to obtain the spatial context vector. All time steps Time-average pooling is used to obtain the global spatial feature vector. ,Right now: ;

[0119] Spatiotemporal feature fusion includes:

[0120] Global average pooling is performed on the temporal feature map to obtain the global temporal feature vector. The global temporal feature vector and the global spatial feature vector are concatenated and fused through a linear layer to obtain the final spatiotemporal context feature vector. ;in This represents vector concatenation. We fuse weights and biases for linear layers.

[0121] S4. Utilize spatiotemporal context features and intermediate variables to predict the phenotypic values ​​of the target plot, ensuring that the predicted phenotypic values ​​conform to basic physiological and ecological laws; this includes the following steps:

[0122] Obtain the spatiotemporal context feature vector, and then input the concatenated vector and the spatiotemporal context feature vector into an affine parameter network for fusion to obtain the corrected feature vector. , For fusion bias.

[0123] in, Scaling factor , The intermediate variable encoding vector is obtained after the intermediate variable is encoded by the gated recurrent unit (GUR). Here is the weight matrix of the affine parameter network. For the bias of the affine parameter network, This is the activation function.

[0124] Through a fully connected network ,Will The mapping is performed on K scalar values ​​(K being the predicted number of phenotypic traits), and then activated by an activation function. The function's output is constrained to the interval (0,1). This can be formally expressed as:

[0125] Environmental response control vector ; This is the output layer weight matrix. For the fusion weight matrix of the gated layer, For gating layer fusion bias. The activation function of the gated layer. This is the output layer bias.

[0126] , For the first The environmental response regulation coefficient of an individual trait reflects the overall degree of restriction of the trait by the current ecological conditions (1 for no restriction, 0 for complete inhibition).

[0127] For each trait, its final predicted phenotypic value ,in For the first The genetic potential baseline for individual traits is directly determined by the inherent parameters of the variety. The potential baseline for plant height determines the maximum plant height of the variety; the potential baseline for tillering and number per plant determines the maximum potential number of tillers; the potential baseline for grain number determines the potential number of grains per ear; and the potential baseline for 1000-grain weight determines the potential 1000-grain weight. These baseline values ​​are achievable without any environmental stress, meaning that the genetic constraints for predicting phenotypes are achieved through the genetic potential baseline.

[0128] Final output phenotypic prediction vector This allows for direct comparison and training with observed values. Through the above steps, the mapping from multi-source ecological data to environmental regulation coefficients is achieved.

[0129] S5. Calculate and obtain the uncertainty interval for the predicted phenotypic value. The specific steps are as follows:

[0130] Based on Monte Carlo Dropout, the variance of the predicted phenotypic values ​​is output; the uncertainty interval of the predicted phenotypic values ​​is constructed using the variance of the predicted phenotypic values; specifically:

[0131] After the fully connected layers of the temporal Transformer branch and the convolutional spatial branch, a Dropout regularization layer is added.

[0132] Multi-source ecological time-series data from multiple time points are acquired. The temporal Transformer branch and the convolutional spatial branch are repeatedly propagated independently M times. During each forward propagation, the Dropout regularization layer randomly discards neurons with a probability of 0.2 to obtain different multimodal models.

[0133] Collect the predicted phenotypic values ​​output by each different multimodal model, and calculate the standard deviation of the predicted phenotypic values ​​at different time points. and mean ;

[0134] Then for each time node, the first The uncertainty interval for the predicted phenotypic values ​​(not the final predicted phenotypic values) of a personality trait is: ,in, It is the critical value of the standard normal distribution, that is, the value corresponding to the 95% confidence interval.

[0135] The present invention also provides a winter wheat successor representative type prediction system that integrates genetic constraints and attention models, the system being used to execute the winter wheat successor representative type prediction method that integrates genetic constraints and attention models.

[0136] The present invention also provides a computer-readable storage medium storing a computer program that is executed by a processor to implement the above-described method for predicting the phenotypic traits of winter wheat by incorporating genetic constraints and attention models.

[0137] The processes described above with reference to the flowcharts in the embodiments disclosed in this invention can be implemented as computer software programs. The embodiments disclosed in this invention include a computer program product comprising a computer program carried on a computer-readable medium, the computer program containing program code for performing the methods shown in the flowcharts. In such embodiments, the computer program can be downloaded and installed from a network via a communication component, and / or installed from a removable medium. When the computer program is executed by a central processing unit (CPU), it performs the functions defined in the methods of this application. It should be noted that the computer-readable medium described above in this application can be a computer-readable signal medium or a computer-readable storage medium, or any combination of the two. Computer-readable storage media can be, for example, but not limited to, electrical, magnetic, optical, electromagnetic, infrared, or semiconductor systems, apparatuses, or devices, or any combination thereof. More specific examples of computer-readable storage media may include, but are not limited to: electrical connections having one or more wire segments, portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fiber, portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination thereof. In this application, a computer-readable storage medium can be any tangible medium containing or storing a program that can be used by or in connection with an instruction execution system, apparatus, or device. In this application, a computer-readable signal medium may include a data signal propagated in baseband or as part of a carrier wave, carrying computer-readable program code. Such propagated data signals can take various forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination thereof. A computer-readable signal medium can also be any computer-readable medium other than a computer-readable storage medium, which can send, propagate, or transmit a program for use by or in connection with an instruction execution system, apparatus, or device. The program code contained on a computer-readable medium may be transmitted using any suitable medium, including but not limited to: wireless segments, wire segments, optical fibers, RF, etc., or any suitable combination thereof.

[0138] The flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of systems, methods, and computer program products according to various embodiments of the present invention. In this regard, each block in a flowchart or block diagram may represent a module, segment, or portion of code containing one or more executable instructions for implementing a specified logical function. It should also be noted that in some alternative implementations, the functions indicated in the blocks may occur in a different order than those indicated in the drawings. For example, two consecutively indicated blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each block in the block diagrams and / or flowcharts, and combinations of blocks in the block diagrams and / or flowcharts, may be implemented using a dedicated hardware-based system that performs the specified function or operation, or using a combination of dedicated hardware and computer instructions.

[0139] Those skilled in the art should understand that the embodiments of the present invention described above and shown in the accompanying drawings are merely examples and do not limit the present invention. The purpose of the present invention has been fully and effectively achieved. The functions and structural principles of the present invention have been shown and explained in the embodiments. Without departing from the principles described, the implementation of the present invention may have any changes or modifications.

Claims

1. A method for predicting the phenotypic traits of winter wheat by integrating genetic constraints and attention models, characterized in that, The method includes: S1. Obtain multi-source ecological time-series data of the target plot and historical wheat phenotypic observation sequences for the same period; S2. Construct a winter wheat growth model by inputting multi-source ecological time-series data into the winter wheat growth model, simulating potential biomass accumulation and organ formation dynamics with daily step size, and outputting intermediate variables. S3. Encode the multi-source ecological time-series data into a high-dimensional feature tensor, input it into a pre-trained multimodal model, and output spatiotemporal context features. The multimodal model includes a temporal Transformer branch and a convolutional spatial branch. S4. Utilize spatiotemporal context features and intermediate variables to predict the phenotypic values ​​of the target plots, so that the predicted phenotypic values ​​conform to basic physiological and ecological laws. S5. Calculate and obtain the uncertainty interval of the predicted phenotypic value.

2. The method for predicting the successive representative traits of winter wheat by integrating genetic constraints and attention models according to claim 1, characterized in that, The multi-source ecological time-series data includes daily meteorological data, soil moisture data, field management records, and remote sensing vegetation indices; The daily meteorological data includes: temperature, precipitation, sunshine duration, and wind speed; The soil moisture data includes: water content, electrical conductivity, and organic matter content of soil at different depths; The field management records include: sowing date, fertilizer application amount, and irrigation date; The remote sensing vegetation indices include: Normalized Difference Vegetation Index (NDVI), Enhanced Vegetation Index (EVI), and Chlorophyll Index (LCI). The wheat phenotypic observation sequence includes: plant height, number of tillers per plant, leaf area index, aboveground biomass, spike length, number of grains per spike, and thousand-grain weight at harvest.

3. The method for predicting the phenotypic traits of winter wheat by integrating genetic constraints and attention models according to claim 2, characterized in that, The winter wheat growth model adopts a dry matter production framework based on radiation use efficiency (RUE) and introduces an accumulated temperature correction function. When the effective accumulated temperature deviates from the optimal range for the variety, the assimilation rate is reduced. The ratio of effective soil moisture content in the root zone to actual evapotranspiration demand is defined as the water stress factor. The reference evapotranspiration is calculated using the Penman-Monteith formula for potential evapotranspiration, so as to dynamically adjust the stomatal conductance limitation coefficient.

4. The method for predicting the representative phenotype of winter wheat by integrating genetic constraints and attention models according to claim 3, characterized in that, The temporal Transformer branch in the multimodal model uses non-uniform time-interval coding to adapt to situations where meteorological observations are missing or sampling is uneven. The convolutional spatial branch employs dilated convolution to capture the synergistic effects of large-scale meteorological systems on wheat population structure.

5. The method for predicting the successive representative traits of winter wheat by integrating genetic constraints and attention models according to claim 4, characterized in that, The process of encoding multi-source ecological time-series data into a high-dimensional feature tensor, inputting it into a pre-trained multimodal model, and outputting spatiotemporal context features includes: Encode multi-source ecological time-series data into T×H×W×C tensors H and W represent the length and width of the grid for the target plot and the surrounding plots; C represents the number of channels for the variable, and T represents the time resolution. For the first The H×W×C slices at each time point are denoted as... ; use Obtain the global temporal feature vector and global spatial feature vector of the target plot; Spatiotemporal context feature vectors are obtained by using global temporal feature vectors and global spatial feature vectors.

6. The method for predicting the successive representative traits of winter wheat by integrating genetic constraints and attention models according to claim 5, characterized in that, The method of using spatiotemporal context features and intermediate variables to predict the phenotypic values ​​of the target plot, so that the predicted phenotypic values ​​conform to basic physiological and ecological laws, includes: Obtain the spatiotemporal context feature vector; Construct a spliced ​​vector using all intermediate variables; Obtain the spatiotemporal context feature vector, then input the concatenated vector and the spatiotemporal context feature vector into an affine parameter network for fusion to obtain the corrected feature vector. ; Through a fully connected network ,Will Mapped to K scalar values, and the output constrained by the sigmoid activation function. The environmental response control vector is obtained from the interval; the environmental response control coefficient is a continuous value between 0 and 1. The varietal-inherent parameters in the growth model were calibrated using wheat phenotypic observation sequences from the same historical period to obtain a baseline of genetic potential. The predicted phenotypic value is corrected by the environmental response regulation vector to obtain the final predicted phenotypic value; the final predicted phenotypic value is the product of the genetic potential baseline and the environmental response regulation vector, so that the predicted phenotypic value conforms to basic physiological and ecological laws.

7. The method for predicting the successor phenotypic traits of winter wheat by integrating genetic constraints and attention models according to claim 6, characterized in that, The process of calibrating the variety-specific parameters in the winter wheat growth model using wheat phenotypic observation sequences from the same historical period to obtain the genetic potential baseline includes: Set the parameter set for the target variety ; The maximum number of tillers per plant. For potential thousand-grain weight, Potential number of grains per ear; The objective function is constructed by weighting the root mean square error of historical predicted phenotypic values ​​and the wheat phenotypic observation series of the same historical period as the optimization objective. : ； in, This represents the root mean square error of the leaf area index for that day. The root mean square error of plant height. The phenotypic value for the thousand-grain weight at maturity is [value]. This represents the observed weight of 1000 grains at maturity. The predicted phenotypic value for the number of grains per ear. This refers to the observed number of grains per ear. , , , These are the trait weights, all of which take a value of 1; The initial value range of the intrinsic parameters is: The value range is 1.2-2.8 g / MJ; The value range is 0.4-0.8; The value range is 2-8; The value range is 35-55g. The value range is 25-55 grains / ear; Using a gradient-free optimization algorithm The optimal parameter set is obtained by minimizing the range of values ​​for the intrinsic parameters within the objective function. ; After calibration, the parameters in the optimal parameter set are the inherent parameters of the variety, and the inherent parameters are used to construct the genetic potential baseline.

8. The method for predicting the successive representative traits of winter wheat by integrating genetic constraints and attention models according to claim 7, characterized in that, The calculation of the uncertainty interval for obtaining the predicted phenotypic value includes: Based on Monte Carlo Dropout, output the variance of the predicted phenotypic values; The uncertainty interval for the predicted phenotypic values ​​is constructed using the variance of the predicted phenotypic values.

9. A winter wheat successor phenotypic prediction system integrating genetic constraints and attention models, characterized in that, The system is used to execute the winter wheat post-representative phenotypic prediction method according to any one of claims 1-8, which integrates genetic constraints and attention models.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program that is executed by a processor to implement the method for predicting the post-representative phenotype of winter wheat according to any one of claims 1-8, which integrates genetic constraints and attention models.

Images