CN-121997729-A - Digital twin growth simulation and optimization method for safflower seedling breeding

CN121997729ACN 121997729 ACN121997729 ACN 121997729ACN-121997729-A

Abstract

The invention relates to the technical field of digital twin and agricultural information, and discloses a digital twin growth simulation and optimization method for safflower seedling breeding, which comprises the steps of constructing an individualized digital twin body fusing genotype, phenotype and multidimensional environment perception data; and generating an accurate cultivation regulation and control instruction based on multi-objective optimization of biomass, medicinal ingredient content and water efficiency. The system comprises a data acquisition unit, a twin body construction unit, an environment fusion unit, a hybrid modeling unit, a visualization unit, an optimization decision and instruction execution unit. The invention obviously improves the yield, quality and resource utilization efficiency of safflower seedling breeding through closed-loop intelligent regulation and control.

Inventors

XU LANJIE
FENG JUNPING
LIU YAZHOU
ZHENG DONGFANG
LIU FENGMING
SUN YUSHU
LIANG HUIZHEN
WU XIAOHUI
YU YONGLIANG
An Sufang
CHEN PEIYU
DONG WEI
TAN ZHENGWEI
SU XIAOYU

Assignees

河南省农业科学院中药材研究所

Dates

Publication Date: 20260508
Application Date: 20260115

Claims (10)

1. The digital twin growth simulation and optimization method for safflower seedling breeding is characterized by comprising the following steps: acquiring individual genotype information, phenotype initial parameters and multidimensional real-time sensing data of a cultivation environment where safflower seedlings are located; constructing an individualized three-dimensional geometric form framework of the safflower seedling based on the individual genotype information and the phenotype initial parameters, and initializing physiological state variables of the framework; performing space-time alignment and fusion processing on the multidimensional real-time sensing data and the historical environment time sequence data to form a unified environment driving field data set; establishing a mechanism submodel set comprising core physiological processes such as photosynthesis, respiratory metabolism, water transpiration, nutrient absorption and distribution, organ establishment and the like, wherein each submodel adopts a differential equation set to describe a dynamic evolution rule; constructing a data driving correction module based on a depth time sequence neural network, wherein the module takes an environmental driving field data set and a physiological state variable as input and outputs real-time compensation quantity of prediction deviation of a mechanism sub-model; The output of the mechanism sub-model set and the compensation quantity of the data driving correction module are subjected to weighted fusion, and the updated physiological state variable and three-dimensional geometric form parameter of the safflower seedling at the next time step are generated; Dynamically reconstructing a digital twin body visual model of the safflower seedling based on the updated three-dimensional geometric form parameters, and synchronously updating an internal state database of the model; setting a multi-objective optimization problem of taking the maximization of biomass accumulation rate, the achievement of the standard of the effective medicinal ingredient content threshold and the optimal water resource utilization efficiency as an objective function; Adopting a genetic algorithm based on non-dominant ranking to iteratively optimize 4 decision variables including irrigation quantity, fertilization proportion, illumination intensity regulation and control interval and temperature and humidity set point; and converting the decision variable sequence obtained by optimization into an executable cultivation regulation instruction, and issuing the cultivation regulation instruction to a physical planting unit through an Internet of things execution terminal.
2. The digital twin growth simulation and optimization method for safflower seedling propagation of claim 1, wherein the obtaining of the idiotype information of safflower seedlings comprises: Carrying out whole genome resequencing on the safflower seed sample through a high-throughput sequencing platform to obtain a single nucleotide polymorphism locus map, and extracting molecular marker combinations related to key enzyme coding genes of plant height, branch number, flowering phase, glandular hair density and hydroxy safflower yellow A synthesis pathway.
3. The digital twin growth simulation and optimization method for safflower seedling propagation according to claim 2, wherein the phenotypic initial parameters comprise thousand seed weight, radicle length, cotyledon expansion angle, initial moisture content, and the multidimensional real-time perception data comprises total solar irradiance above canopy, photosynthetic effective radiant flux density, air temperature, air relative humidity, carbon dioxide concentration, soil volume moisture content, soil conductivity, soil temperature.
4. The method for simulating and optimizing digital twin growth for propagation of safflower seedlings according to claim 3, wherein said constructing an individualized three-dimensional geometric skeleton of safflower seedlings comprises: based on an L system fractal algorithm, generating an initial topological structure according to an initial branching series, an internode length and a leaf inclination angle distribution function; carrying out parameterized modeling on the blade profile by adopting spherical harmonic functions; and mapping the discrete organ measurement point cloud data to a continuous curved surface model by using a radial basis function interpolation method.
5. The method for simulating and optimizing digital twin growth for safflower seedling propagation according to claim 4, wherein the environmental driving field dataset has a temporal resolution of 10 minutes and a spatial resolution of a single plant scale, and the fusion processing comprises filling the missing data by cubic spline interpolation, and removing the abnormal mutation points by sliding window median filtering.
6. The digital twin growth simulation and optimization method for safflower seedling propagation according to claim 5, wherein a photosynthesis sub-model in the mechanism sub-model set adopts a Farquhar biochemical model frame, the input is intercellular carbon dioxide concentration, mesophyll conductivity and maximum carboxylation rate, the output is net photosynthetic rate, the moisture transpiration sub-model is based on Penman-Monteth equation, a pore conductivity dynamic feedback term is introduced, and a nutrient absorption sub-model adopts Michaelis-Menten kinetic equation to couple a root system distribution density function with a soil nutrient diffusion coefficient.
7. The method for simulating and optimizing digital twin growth for safflower seedling propagation according to claim 6, wherein the deep timing neural network is a two-way gated cyclic unit network, the number of hidden layer nodes is 128, the length of input sequence is 72 time steps, and the output is a first derivative correction term of each physiological state variable, the network uses a historical field observation data set for supervision training in an off-line stage, and the network receives real-time data stream in a sliding window mode and outputs an instant compensation quantity in an on-line stage.
8. The method of claim 7, wherein the weighted fusion is performed by using an adaptive weight distribution strategy, the weight coefficient is dynamically adjusted according to the variance of the residual error of the mechanism model, when the residual error variance exceeds a preset threshold value of 0.5, the weight of the data-driven correction module is increased to 0.7, otherwise, the weight is maintained at 0.3.
9. The method for simulating and optimizing digital twin growth for safflower seedling propagation according to claim 8, wherein the objective function of the multi-objective optimization problem is defined by a first objective function being an increase of dry matter on the upper part in unit time, a second objective function being that the mass fraction of hydroxysafflor yellow A in petals is not less than 1.5%, a third objective function being that the amount of irrigation water consumed per kilogram of dry matter is not more than 8 liters, and constraints including that the soil moisture content is not less than 60% and not more than 90% of the field water holding capacity, and the daily average air temperature is between 15 ℃ and 28 ℃.
10. The digital twin growth simulation and optimization method for safflower seedling propagation according to claim 9, wherein the population scale of the non-dominant ranking genetic algorithm is 200, the crossover probability is 0.9, the mutation probability is 0.1, the maximum evolution algebra is 500 generations, the coding mode of the decision variable is real number coding, the irrigation value range is 0 to 10 liters per day, the mass fraction of nitrogen element in the nitrogen-phosphorus-potassium fertilization ratio is 0.1 to 0.5%, the lower limit of the illumination intensity regulation interval is not lower than 400 micromoles per square meter per second, and the difference between the daytime temperature set value and the night temperature set value in the temperature and humidity set point is not lower than 6 ℃.

Description

Digital twin growth simulation and optimization method for safflower seedling breeding Technical Field The invention belongs to the technical field of digital twin and agricultural information, and particularly relates to a digital twin growth simulation and optimization method for safflower seedling breeding. Background Along with the rapid development of precise agriculture and intelligent breeding technology, the role of crop growth simulation in seedling breeding, variety breeding and cultivation management is increasingly critical. The traditional crop growth model is mostly based on an empirical formula or a simplified mechanism equation, and relies on group average parameters to macroscopically describe physiological processes such as photosynthesis, respiration, material distribution and the like. Although the method has a certain prediction capability on the field scale, the dynamic response characteristics of individual plants under genotype difference, micro-environment fluctuation and agronomic management intervention are difficult to be described. Especially in the seedling breeding of medical cash crops such as safflower, the single plant phenotype has strong plasticity, sensitive growth period and obvious influence of coupling of heated light and water and fertilizer, and the traditional model lacks fine analysis on a genotype-environment-management ternary interaction mechanism, so that systematic deviation exists between a simulation result and an actual growth track, and high-precision breeding decision and individual breeding regulation and control are difficult to support. The digital twin technology provides a new paradigm for crop growth modeling, and the core of the digital twin technology is to construct a virtual-real mapping, real-time interaction and dynamic evolution high-fidelity virtual individual. Aiming at the safflower seedling breeding scene, a digital twin body capable of fusing a physiological mechanism and data driving is required to be established so as to realize accurate crossing from population average to single plant. The key of the direction is how to keep the inherent constraint of the plant physiological process and have the self-adaptive learning capability on the complex nonlinear interaction relationship under the condition of limited observation data. While the pure mechanism model has interpretability, the parameters are fixed, the structure is stiff, and the model is difficult to adapt to individual variation and environmental disturbance, while the pure data driving model (such as a deep neural network) has strong fitting capability, but is easy to fall into overfitting, has poor generalization performance under the condition of a small sample, and lacks physiological rationality constraint, so that the prediction result violates the biological rule. More importantly, the current method generally neglects the high sensitivity of safflower seedlings to micro-environmental disturbance at the early development stage, and can not reveal implicit interaction rules (such as the nonlinear enhancement effect of specific genotypes on nitrogen response under low-temperature weak light), so that the optimization of a breeding scheme lacks scientific basis. Disclosure of Invention The invention provides a digital twin growth simulation and optimization method for safflower seedling breeding, which is used for realizing full life cycle high-fidelity simulation of safflower seedlings in a complex dynamic environment by constructing an individualized safflower plant digital twin body fused with multi-source heterogeneous data and combining a mixed modeling mechanism of a physical mechanism model and a data driving model, and generating an accurate cultivation regulation strategy based on a simulation result by performing multi-objective collaborative optimization, thereby solving the technical problems that a traditional crop growth model depends on an empirical formula and is difficult to accurately simulate individual differences and complex environment interactions. The invention provides a digital twin growth simulation and optimization method for safflower seedling breeding, which comprises the following steps: acquiring individual genotype information, phenotype initial parameters and multidimensional real-time sensing data of a cultivation environment where safflower seedlings are located; constructing an individualized three-dimensional geometric form framework of the safflower seedling based on the individual genotype information and the phenotype initial parameters, and initializing physiological state variables of the framework; performing space-time alignment and fusion processing on the multidimensional real-time sensing data and the historical environment time sequence data to form a unified environment driving field data set; establishing a mechanism submodel set comprising core physiological processes such as photosynthesis, respiratory metabolism, water