A method for recommending suitable panax ginseng species and screening dominant strains based on linkage of environmental simulation and algorithm model

Abstract

The present application relates to the field of intelligent agriculture and digital breeding technology, especially to a method for recommending compatible varieties and screening dominant strains of Poria cocos based on linkage of environmental simulation and algorithm model, aiming at the problems of long period, low accuracy and low achievement transformation rate in traditional Poria cocos breeding, the present application constructs dynamic time sequence environmental parameter sequence by acquiring multi-dimensional environmental data of target planting area, synchronously collects dynamic growth characteristics of multiple varieties in whole growth period in multi-chamber parallel cultivation system, generates compatible variety recommendation and dominant trait fingerprint by using built-in environmental-physiological response coupling module, establishes two-way real-time linkage channel to realize closed-loop iterative optimization, and finally outputs target compatible variety, dominant strain and standardized cultivation promotion scheme. The present application greatly shortens the period of breeding compatible varieties, significantly improves the screening accuracy, generates planting guidance that can be directly applied, and can be extended to other medicinal fungus breeding compatible variety fields.

Description

Technical Field

[0001] This invention relates to the field of intelligent agriculture and digital breeding of suitable varieties, specifically a method for recommending suitable varieties of Poria cocos and screening dominant strains based on the linkage of environmental simulation and algorithm model. Background Technology

[0002] Poria cocos is a traditional, widely used medicinal fungus. Its dried sclerotia have extremely high medicinal and economic value and are widely used in medicine, food, health products, and other fields. With the rapid development of the health industry, the market demand for Poria cocos continues to grow, placing higher demands on its yield and quality. However, the current Poria cocos industry is still dominated by traditional cultivation methods, resulting in serious problems such as severe varietal degeneration, large yield fluctuations, and inconsistent quality, which severely restrict the sustainable development of the industry.

[0003] Traditional methods for breeding Poria cocos varieties and screening dominant strains primarily rely on field trials. This approach involves directly planting candidate varieties in the target area and screening based on final yield and appearance traits. However, due to limitations imposed by natural climate and seasonal variations, field trials typically require three to five years to complete a full breeding cycle, resulting in extremely low efficiency in selecting suitable varieties. Furthermore, the complex and variable field environment, with factors such as temperature, humidity, and soil conditions difficult to control, makes experimental results susceptible to random interference and exhibits poor repeatability. In addition, traditional screening methods focus only on traits observed at the final harvest stage, neglecting the dynamic coupling between environmental changes and physiological metabolism throughout the entire growth period of Poria cocos. This makes it impossible to accurately assess the environmental adaptability and stress resistance of varieties, leading to selected varieties often performing well in specific regions but experiencing significant yield and quality declines when promoted in other areas.

[0004] In recent years, with the development of smart agriculture and digital breeding technology for suitable varieties, environmental simulation and model-assisted breeding methods for suitable varieties have begun to be applied in the field of crop breeding. Simulating the environmental conditions of a target area using artificial climate chambers, combined with data analysis models, can shorten the breeding cycle and improve screening efficiency to some extent. However, the application of existing technologies in the breeding of Poria cocos still has significant shortcomings. First, as a typical saprophytic fungus, Poria cocos' growth and development process differs significantly from that of higher plants, exhibiting three distinct growth stages: mycelial stage, budding stage, and sclerotium stage. The environmental requirements for each stage vary greatly. Most existing environmental simulation systems are designed for higher plants and cannot accurately reproduce the dynamic environmental requirements of Poria cocos throughout its entire growth cycle, nor can they achieve a smooth transition of environmental parameters between different growth stages.

[0005] Secondly, most existing digital breeding models are static, capable of offline analysis based on historical data, and cannot achieve real-time interaction between the physical breeding process and the digital model. There is often a significant discrepancy between model predictions and actual growth conditions, and dynamic corrections based on actual growth feedback are not possible. Furthermore, existing models primarily focus on predicting single traits, lacking a comprehensive assessment of multi-dimensional core traits such as high yield, high resistance, and high stability, and are unable to extract quantitative trait thresholds that can guide actual breeding practices.

[0006] Furthermore, existing technologies generally suffer from a disconnect between variety selection and strain cultivation. Many superior strains selected in laboratories perform poorly in actual field planting, primarily because laboratory cultivation conditions differ significantly from actual field environments, and there is a lack of standardized cultivation and management programs for specific varieties and strains. This severe technological gap in the transformation of breeding results into actual production greatly limits the promotion and application of superior varieties.

[0007] In summary, existing Poria cocos breeding technologies suffer from problems such as long cycles, low efficiency, poor accuracy, and low conversion rates of research results, failing to meet the demands of the modern Poria cocos industry for rapid selection and standardized promotion of superior varieties. Therefore, there is an urgent need to develop a novel method for recommending Poria cocos varieties and screening superior strains that can accurately simulate the growth environment of Poria cocos, achieve two-way linkage between physical cultivation and digital models, and comprehensively evaluate multi-dimensional traits. Summary of the Invention

[0008] The purpose of this invention is to provide a method for recommending suitable varieties of Poria cocos and screening dominant strains based on the linkage of environmental simulation and algorithm model, so as to solve the problems mentioned in the background art.

[0009] To achieve the above objectives, the present invention provides the following technical solution: A method for recommending suitable varieties of Poria cocos and screening dominant strains based on the linkage of environmental simulation and algorithm model includes the following steps: S1. Obtain multidimensional environmental data covering the entire growth cycle of Poria cocos in the target planting area, and generate a structured target environmental feature set through time-series alignment, outlier filtering and principal component dimensionality reduction. S2. Input the target environmental feature set into the environmental simulation unit to construct a dynamic temporal environmental parameter sequence covering the mycelial stage, budding stage and sclerotium stage of Poria cocos; under the drive of the dynamic temporal environmental parameter sequence, conduct standardized parallel cultivation of multiple Poria cocos candidate varieties, collect phenotypic indicators and physiological metabolic indicators synchronously according to preset physiological nodes, and integrate to construct a dynamic growth characteristic temporal map of each variety. S3. Input the target environment feature set, the empirical breeding strategy feature vector generated by pre-training, and the time series map of dynamic growth characteristics into the variety adaptation network. The variety adaptation network has a built-in environment-physiological response coupling module. By calculating the phase difference between the environmental stress gradient and the physiological metabolic rate, it generates the adaptation variety recommendation vector that characterizes the regional adaptation priority of each variety, as well as the candidate strain dominant trait fingerprint map that characterizes the threshold of high yield, high resistance, and high stability core traits. S4. Establish a two-way real-time linkage channel between the simulation cultivation system and the algorithm model system. The channel includes a forward feature push link and a reverse decision control link. The forward feature push link inputs the dynamically updated time series map of growth characteristics into the variety adaptation network in real time for dynamic prediction. The reverse decision control link feeds back the recommended vector of the adapted varieties and the fingerprint map of the dominant traits of the candidate strains to the environmental simulation unit. The environmental simulation unit dynamically reduces the cultivation carriers of low-fit varieties based on the priority ranking of the recommended vector of the adapted varieties, and adjusts the stress intensity and nutrient ratio of the dynamic time series environmental parameter sequence in reverse through the model prediction control algorithm based on the core trait threshold of the fingerprint map of the dominant traits of the candidate strains. S5. At the cultivation stage node, the currently collected actual growth characteristic sequence is dynamically time-aligned with the benchmark sequence in the dynamic growth characteristic time series map, and a multidimensional growth deviation vector is calculated. When the Mahalanobis distance of the growth deviation vector exceeds the preset tolerance threshold, the online correction process of the model is triggered. The process generates an environmental compensation gradient based on the growth deviation vector, updates the mapping bias between the coupling weight of the variety adaptation network and the feature vector of the empirical cultivation strategy through the gradient compensation algorithm, and outputs the corrected adaptation variety recommendation vector and the corrected candidate strain dominant trait fingerprint map. S6. Drive the next round of parallel cultivation by the modified adaptive variety recommendation vector and the modified candidate strain dominant trait fingerprint spectrum through the reverse decision control link. Repeat steps S2 to S5 until the adaptability ranking information entropy of the adaptive variety recommendation vector stabilizes to the preset interval and the core trait coverage of the candidate strain dominant trait fingerprint spectrum converges to the target threshold, and generate a closed-loop iteration termination signal. S7. Responding to the closed-loop iteration termination signal, extract the Poria cocos variety with the highest priority in the recommended variety vector as the target suitable variety, match the measured strains of all core dimensions of the fingerprint spectrum of the candidate strains as the target dominant strains, integrate the environmental regulation parameter matrix and management strategy of the whole growth period, and generate a cultivation and promotion plan.

[0010] As can be seen from the technical solution provided by the present invention above, the method for recommending suitable varieties of Poria cocos and screening dominant strains based on the linkage of environmental simulation and algorithm model provided by the present invention has the following beneficial effects: This invention accurately reproduces the dynamic environmental evolution process of the entire growth cycle of Poria cocos in the target planting area through an environmental simulation unit. Combined with a multi-chamber standardized parallel cultivation system, it can simultaneously carry out synchronous cultivation experiments of dozens of candidate varieties indoors. With the help of an algorithm model-driven closed-loop iterative optimization mechanism, the entire process from initial screening of varieties to determination of dominant strains can be completed within three to six months, shortening the breeding cycle of suitable varieties by more than 80% and accelerating the promotion and application of superior Poria cocos varieties. This invention uses an in-situ multimodal sensing monitoring group to simultaneously collect phenotypic and physiological metabolic indicators during the growth process of Poria cocos, constructing a dynamic growth characteristic time series map. The variety adaptation network has a built-in environment-physiological response coupling module, which quantitatively characterizes the response speed and adaptability of varieties to environmental changes by calculating the phase difference between environmental stress gradient and physiological metabolic rate. Based on this, the generated adaptation variety recommendation vector and candidate strain dominant trait fingerprint map can comprehensively evaluate the performance of varieties and strains from three dimensions: high yield, high resistance, and high stability, making the screening results more scientific and accurate. This invention establishes a two-way real-time linkage channel between the simulation breeding system and the algorithm model system. Through the reverse decision-making control link, the breeding carriers of low-fit varieties can be dynamically reduced based on the variety suitability assessment results, and the limited breeding resources can be concentrated on high-potential candidate varieties. At the same time, based on the core trait thresholds of the fingerprint spectrum of the candidate strains' dominant traits, the model prediction control algorithm reversely adjusts the breeding environment parameters, so that the environmental conditions always evolve in a direction that is conducive to the expression of dominant traits. This avoids the waste of resources caused by blind environmental control and significantly reduces the consumption of human and material resources in the process of breeding suitable varieties. This invention, while screening out target suitable varieties and dominant strains, can integrate environmental regulation parameter matrix and agronomic management strategies throughout the entire growth period, generating a standardized management strategy set that includes microenvironment dynamic evolution curves, substrate water and fertilizer step replenishment procedures, and emergency intervention plans for abnormal working conditions. The cultivation and promotion plan is bound to the geographical coordinate attributes and climate adaptability rating tags of the target planting area, providing control parameters that can be directly imported into smart agricultural equipment and detailed field operation guidance. This solves the problem of technological disconnect in the process of transforming traditional breeding and suitable variety results, and can effectively guide farmers to carry out standardized planting, thereby improving the yield and quality stability of Poria cocos. The empirical strategy memory bank of this invention uses a temporal knowledge graph structure to store agronomic parameters and expert intervention rules in historical cultivation log areas. It can automatically update the knowledge graph after each cultivation experiment to continuously enrich prior knowledge. At the same time, the system's built-in online model correction process can update the coupling weights of the variety adaptation network and the mapping bias of the empirical cultivation strategy feature vector in real time according to the deviation between actual growth and model prediction. This enables the model to continuously adapt to different planting environments and variety characteristics, and continuously improve the accuracy and robustness of variety adaptability assessment and dominant strain screening. This invention adopts a modular design, and the core architecture of the simulation cultivation system and the algorithm model system has good scalability. By adjusting the parameter settings of the environmental simulation unit and the training data of the variety adaptation network, this method can quickly adapt to the needs of variety screening and dominant strain cultivation of other medicinal fungi such as Ganoderma lucidum, Gastrodia elata, and Polyporus umbellatus, providing a general technical solution for the digital breeding of adapted varieties in the entire medicinal fungi industry. Attached Figure Description

[0011] Figure 1 This is a schematic diagram of the steps of a method for recommending suitable varieties of Poria cocos and screening dominant strains based on the linkage of environmental simulation and algorithm model according to the present invention. Detailed Implementation

[0012] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention.

[0013] To better understand the above technical solutions, the following will provide a detailed explanation of the technical solutions in conjunction with the accompanying drawings and specific embodiments.

[0014] like Figure 1 As shown, this embodiment of the invention provides a method for recommending suitable varieties of Poria cocos and screening dominant strains based on the linkage of environmental simulation and algorithm model, including the following steps: S1. Obtain multidimensional environmental data covering the entire growth cycle of Poria cocos in the target planting area, and generate a structured target environmental feature set through time-series alignment, outlier filtering and principal component dimensionality reduction. S2. Input the target environmental feature set into the environmental simulation unit to construct a dynamic temporal environmental parameter sequence covering the mycelial stage, budding stage and sclerotium stage of Poria cocos; under the drive of the dynamic temporal environmental parameter sequence, conduct standardized parallel cultivation of multiple Poria cocos candidate varieties, collect phenotypic indicators and physiological metabolic indicators synchronously according to preset physiological nodes, and integrate to construct a dynamic growth characteristic temporal map of each variety. S3. Input the target environment feature set, the empirical breeding strategy feature vector generated by pre-training, and the time series map of dynamic growth characteristics into the variety adaptation network. The variety adaptation network has a built-in environment-physiological response coupling module. By calculating the phase difference between the environmental stress gradient and the physiological metabolic rate, it generates the adaptation variety recommendation vector that characterizes the regional adaptation priority of each variety, as well as the candidate strain dominant trait fingerprint map that characterizes the threshold of high yield, high resistance, and high stability core traits. S4. Establish a two-way real-time linkage channel between the simulation cultivation system and the algorithm model system. The channel includes a forward feature push link and a reverse decision control link. The forward feature push link inputs the dynamically updated time series map of growth characteristics into the variety adaptation network in real time for dynamic prediction. The reverse decision control link feeds back the recommended vector of the adapted varieties and the fingerprint map of the dominant traits of the candidate strains to the environmental simulation unit. The environmental simulation unit dynamically reduces the cultivation carriers of low-fit varieties based on the priority ranking of the recommended vector of the adapted varieties, and adjusts the stress intensity and nutrient ratio of the dynamic time series environmental parameter sequence in reverse through the model prediction control algorithm based on the core trait threshold of the fingerprint map of the dominant traits of the candidate strains. S5. At the cultivation stage node, the currently collected actual growth characteristic sequence is dynamically time-aligned with the benchmark sequence in the dynamic growth characteristic time series map, and a multidimensional growth deviation vector is calculated. When the Mahalanobis distance of the growth deviation vector exceeds the preset tolerance threshold, the online correction process of the model is triggered. The process generates an environmental compensation gradient based on the growth deviation vector, updates the mapping bias between the coupling weight of the variety adaptation network and the feature vector of the empirical cultivation strategy through the gradient compensation algorithm, and outputs the corrected adaptation variety recommendation vector and the corrected candidate strain dominant trait fingerprint map. S6. Drive the next round of parallel cultivation by the modified adaptive variety recommendation vector and the modified candidate strain dominant trait fingerprint spectrum through the reverse decision control link. Repeat steps S2 to S5 until the adaptability ranking information entropy of the adaptive variety recommendation vector stabilizes to the preset interval and the core trait coverage of the candidate strain dominant trait fingerprint spectrum converges to the target threshold, and generate a closed-loop iteration termination signal. S7. Responding to the closed-loop iteration termination signal, extract the Poria cocos variety with the highest priority in the recommended variety vector as the target suitable variety, match the measured strains of all core dimensions of the fingerprint spectrum of the candidate strains as the target dominant strains, integrate the environmental regulation parameter matrix and management strategy of the whole growth period, and generate a cultivation and promotion plan.

[0015] In this embodiment, the core function of step S1 is to comprehensively acquire multi-dimensional environmental data of Poria cocos throughout its entire growth cycle in the target planting area. Through standardized processing procedures such as time-series synchronization, abnormal data cleaning, and feature dimensionality reduction, noise and redundant information in the original data are eliminated, and a structured dataset that can accurately characterize the regional environmental features is constructed, providing a reliable input foundation for subsequent environmental simulation and variety adaptation analysis. The detailed steps are as follows: Step S1-1: Collection of raw data on multidimensional environment throughout the entire growth cycle of Poria cocos By deploying an array of IoT sensor nodes and a regional agricultural meteorological database in the target planting area, multidimensional raw environmental data covering the mycelial, budding, and sclerotium stages of Poria cocos are collected. This multidimensional raw environmental data includes atmospheric meteorological parameters, soil physicochemical parameters, and forest microclimate parameters. Atmospheric meteorological parameters include air temperature, relative humidity, light intensity, wind speed, precipitation, and carbon dioxide concentration. Soil physicochemical parameters include soil temperature, soil moisture content, soil pH, soil organic matter content, soil nitrogen, phosphorus, and potassium content, and soil electrical conductivity. Forest microclimate parameters include forest light transmittance, forest wind speed, forest air humidity, and forest temperature gradient. The IoT sensor node array is evenly distributed in a grid pattern in the target planting area, with a node spacing of no more than 50 meters. Each node integrates multiple types of environmental sensors to achieve continuous real-time acquisition of the above parameters. The regional agricultural meteorological database provides historical meteorological data and medium- to long-term meteorological forecasts for the target planting area over the past three years to supplement the temporal coverage and spatial representativeness of the data collected by the sensor nodes. Step S1-2: Asynchronous time series alignment and environmental data matrix generation: Using a preset standard sampling period as a unified time reference axis, the asynchronous time series of each acquisition channel in the multidimensional environmental raw data are resampled and interpolated and fused. The preset standard sampling period is set according to the physiological response characteristics of Poria cocos at different growth stages: 1 hour for the mycelial stage, 30 minutes for the budding stage, and 1 hour for the sclerotium stage. For channels with sampling frequencies higher than the standard sampling period, downsampling is performed using the arithmetic mean method. For channels with sampling frequencies lower than the standard sampling period, upsampling is performed using the cubic spline interpolation method. The above processing eliminates the differences in device sampling frequencies and timestamp offsets, and outputs a time-synchronized environmental data matrix. The rows of the environmental data matrix correspond to the standard sampling time points, and the columns correspond to different environmental parameter dimensions. Each element represents the measured value of the corresponding environmental parameter at the corresponding time point. Step S1-3: Sliding window outlier detection and clean environment data sequence generation: A sliding window local statistical analysis is performed on the time-synchronized environmental data matrix, and outlier data points exceeding the normal fluctuation threshold are identified by combining dynamic confidence interval judgment rules. The sliding window size is set to 24 standard sampling periods: 24 hours for the mycelial stage, 12 hours for the budding stage, and 24 hours for the sclerotium stage. For the environmental parameter sequence within each sliding window, its mean and standard deviation are calculated, and a confidence interval is constructed as the mean plus or minus three times the standard deviation. Data points exceeding the confidence interval are identified as outlier data points. For the identified outlier data points, a smooth replacement is performed using the weighted mean of adjacent valid time-series data. The weight of the weighted mean is inversely proportional to the time distance between the outlier point and the valid data point; the closer the time distance, the greater the weight. The above processing generates a clean environment data sequence. Steps S1-4: Principal component dimensionality reduction and low-dimensional environmental feature vector generation: Calculate the variable covariance matrix of the clean environment data sequence; the elements of the covariance matrix represent the degree of linear correlation between any two environmental parameters; the formula for calculating the covariance matrix is: ,in, Let covariance be the characterization matrix. The number of time points in the clean environment data sequence. For the first An environmental parameter vector at each time point This represents the mean of the environmental parameter vector across all time points. The covariance representation matrix is ​​decomposed into eigenvalues ​​to obtain eigenvalues ​​and their corresponding eigenvectors. The eigenvalues ​​are then arranged in descending order, and the cumulative variance contribution rate is calculated. The formula for calculating the cumulative variance contribution rate is as follows: ,in, For the front The cumulative variance contribution rate of each principal component For the first 1 eigenvalue, The number of dimensions of the original environmental parameters; Extract principal component orthogonal bases whose cumulative variance contribution rate meets the preset retention threshold; the preset retention threshold is set to 95%; project and map high-dimensional environmental parameters to low-dimensional principal component space, remove multicollinear redundant features, and obtain dimensionality-reduced environmental feature vectors; the dimension of the dimensionality-reduced environmental feature vectors is equal to the number of principal components that meet the cumulative variance contribution rate threshold; Steps S1-5: Segmented Reorganization of Reproductive Stages and Generation of Structured Target Environmental Feature Sets: The dimensionality-reduced environmental feature vector is recombined into segments according to preset physiological nodes throughout the entire growth cycle of Poria cocos. The preset physiological nodes include the start point of the mycelial stage, the start point of the budding stage, the start point of the sclerotium stage, and the start point of the harvest stage. The time of each physiological node is predetermined based on the historical Poria cocos planting experience and meteorological conditions of the target planting area. The dimensionality-reduced environmental feature vector is divided into three segments corresponding to the mycelial stage, the budding stage, and the sclerotium stage. Each segment is bound to an environmental state label corresponding to the growth stage. The environmental state label includes the name of the growth stage, the duration of the stage, and the average environmental characteristics of the stage. The segmented sequences and their corresponding labels are encapsulated into a standardized matrix format to output a structured target environmental feature set. The structured target environmental feature set contains three sub-matrices, each corresponding to one of the three growth stages of Poria cocos. The rows of each sub-matrix correspond to the standard sampling time points within the growth stage, and the columns correspond to the dimensions of the dimensionality-reduced environmental features.

[0016] In this embodiment, the core function of step S2 is to accurately reproduce the dynamic environmental evolution process of Poria cocos throughout its entire growth cycle in the target planting area based on a structured target environmental feature set. Multiple groups of Poria cocos candidate varieties are simultaneously cultivated under standardized and controllable conditions. The growth phenotypes and physiological metabolic responses of each variety are continuously captured using multimodal sensing technology, constructing a dynamic growth characteristic time-series map that fully characterizes the environmental adaptability of the varieties. This provides core data support for subsequent variety suitability analysis and dominant strain screening. The detailed steps are as follows: Step S2-1: Generation of dynamic temporal environment parameter sequences based on reproductive stage mapping: The structured target environment feature set is loaded into the environment simulation unit, and the temporal boundaries and environmental state labels of each segment sequence in the target environment feature set are analyzed through the reproductive stage mapping model. The reproductive stage mapping model adopts a pre-trained long short-term memory network structure, which can identify the evolution patterns and transformation characteristics of environmental parameters in different reproductive stages. Using the cumulative environmental effect value as the physiological transition criterion, a dynamic temporal environmental parameter sequence is generated based on the differentiated physiological needs of the mycelial, budding, and sclerotium stages of Poria cocos. The formula for calculating the cumulative environmental effect value is as follows: ,in, From the start time up to the current time The cumulative environmental impact value at each stage, It is a comprehensive influence function of environmental parameters, which is composed of a weighted combination of parameters such as temperature, humidity, light, and nutrient supply; When the cumulative environmental effect value of a stage reaches the preset threshold of the corresponding growth stage, the stage transition of the environmental parameter sequence is automatically triggered. The dynamic time-series environmental parameter sequence includes four core dimensions: temperature and humidity control gradient, light and photoperiod, carbon dioxide concentration fluctuation and substrate water and fertilizer supply ratio. Each dimension is a numerical sequence that changes continuously over time, and the time resolution is consistent with the standard sampling period of the target environmental feature set. Step S2-2: Initialization and Environment-Driven Development of the Multi-Compartment Standardized Parallel Culture System The dynamic time-series environmental parameter sequence is injected into the multi-chamber standardized parallel cultivation system; the multi-chamber standardized parallel cultivation system is divided into standardized cultivation units that are physically isolated from each other, and the number of units is equal to the number of Poria cocos candidate varieties to be tested; each standardized cultivation unit is equipped with an independent temperature-controlled substrate tank and a micro-airflow circulation pipeline at the bottom, which can independently control the internal microenvironment parameters; Perform uniform initialization procedures; sterilize the substrate of all standardized cultivation units at high temperature and high pressure at 121 degrees Celsius for 120 minutes; after sterilization, cool the substrate to 25 degrees Celsius and maintain a constant temperature for 24 hours for pre-cultivation; use a quantitative inoculator to inoculate each standardized cultivation unit with an equal amount of Poria cocos inoculum, with an inoculation amount of 5 grams of mycelium per kilogram of substrate; The environmental simulation unit sends environmental control commands in real time through a digital-physical control interface, driving the microenvironment precision control array of each standardized cultivation unit to perform corresponding operations. The microenvironment precision control array includes a stepper temperature and humidity actuator, an adjustable spectral light source, a gas component ratio regulating valve, and a micro-pressure differential nutrient solution supply pump. Through a closed-loop feedback control mechanism, the actual microenvironment parameters of each standardized cultivation unit strictly follow the dynamic time-series environmental parameter sequence for continuous dynamic evolution, with control accuracy reaching ±0.1 degrees Celsius for temperature, ±1% for humidity, and ±50 ppm for carbon dioxide concentration. Step S2-3: Synchronous acquisition of multimodal growth indicators of preset physiological nodes: Within the driving cycle of the dynamic time-series environmental parameter sequence, the in-situ multimodal sensing monitoring group is activated to collect data using preset physiological nodes as synchronous trigger signals. The preset physiological nodes include once every 7 days during the mycelial stage, once every 3 days during the budding stage, once every 5 days during the sclerotium stage, and key time points for the transition of each growth stage. Phenotypic indicators of each Poria cocos candidate variety were captured non-contactly using a multispectral visual sensing array and a 3D structured light scanner. The multispectral visual sensing array acquired images in four bands: 450 nm, 550 nm, 650 nm, and 850 nm. The mycelial spread coverage and primordium differentiation density were calculated using image segmentation and feature extraction algorithms. The 3D structured light scanner acquired 3D point cloud data of the sclerotia, reconstructed the 3D morphological model of the sclerotia, and calculated morphological parameters such as sclerotia volume, surface area, and shape factor. The in-situ microfluidic sampling module was activated simultaneously to extract interstitial fluid. The in-situ microfluidic sampling module uses an embedded microfluidic percolation sampling needle, enabling quantitative extraction of interstitial fluid samples without disrupting the growth environment of Poria cocos. The extracted samples were automatically detected by an online biochemical analysis module to determine extracellular hydrolytic enzyme activity, characteristic triterpenoid metabolite concentration, and matrix respiration intensity. Extracellular hydrolytic enzyme activity included cellulase, hemicellulase, and ligninase activity; characteristic triterpenoid metabolites included pachymic acid, porous acid C, and dehydropachymic acid. All collected data are automatically stamped with precise timestamps to ensure that phenotypic and physiological metabolic indicators are fully synchronized over time. The collected data undergoes preliminary format conversion and caching through an edge data gateway to prepare for subsequent data fusion and map construction. Step S2-4: Multi-source data fusion and construction of dynamic growth feature time series map: The synchronously acquired phenotypic and physiological metabolic indicators were aligned in a multidimensional time series according to the acquisition timestamp to generate an initial multimodal growth feature matrix. Dynamic threshold normalization was then performed on the initial multimodal growth feature matrix to eliminate dimensional and numerical range differences between different indicators. The formula for dynamic threshold normalization is as follows: ,in, These are the normalized eigenvalues. These are the original eigenvalues. The mean of the feature values ​​within the sliding window. The standard deviation of the feature values ​​within the sliding window; the sliding window size is set to 5 consecutive acquisition time points; An anomaly trajectory removal algorithm is used to identify and remove abnormal data points. The anomaly trajectory removal algorithm is based on the isolated forest algorithm and can effectively identify abnormal data points that deviate significantly from the normal growth trajectory. For the identified abnormal data points, linear interpolation is used to fill and replace them. The cleaned multidimensional indicators were mapped to a standardized feature space with the number of cultivation days as the baseline axis and multimodal growth response as the dimension. The dimensions of the standardized feature space include mycelial spread coverage, primordium differentiation density, sclerotium volume, extracellular hydrolytic enzyme activity, concentration of characteristic triterpenoid metabolites, and substrate respiration intensity. A dynamic growth characteristic time series map is constructed to characterize the coupling relationship between the environmental response trajectory and metabolic dynamic evolution of each Poria cocos candidate variety throughout its entire growth period. The dynamic growth characteristic time series map is stored in a three-dimensional tensor structure, with the first dimension being the cultivation time point, the second dimension being the growth characteristic dimension, and the third dimension being the Poria cocos candidate variety number. The constructed dynamic growth characteristic time series map is encapsulated into a standardized time series data structure and output to the downstream algorithm model system.

[0017] In this embodiment, the core function of step S3 is to integrate the environmental characteristics of the target area, historical breeding experience, and dynamic growth data of multiple varieties. Through the environment-physiological response coupling module built into the variety adaptation network, it quantifies the temporal matching relationship between environmental stress and variety physiological metabolism, generates a regional adaptation priority ranking for each variety, and extracts key thresholds for high-yield, high-resistance, and high-stability core traits. This provides a decision-making basis for subsequent screening of superior strains and optimization of environmental regulation. The detailed steps are as follows: Step S3-1: Generation of multimodal collaborative feature sequences: The target environmental feature set and the time series map of dynamic growth features are synchronously mapped with timestamps to establish a one-to-one correspondence between environmental parameters at each time point and the physiological response of the corresponding variety. The evolution trajectory of environmental parameters and the response trajectory of physiological indicators of each Poria cocos candidate variety in the mycelial stage, budding stage and sclerotium stage are extracted to form environmental trajectory matrix and physiological trajectory matrix respectively. The rows of the environmental trajectory matrix correspond to the time points and the columns correspond to the dimensions of environmental features after dimensionality reduction. The rows of the physiological trajectory matrix correspond to the time points and the columns correspond to the dimensions of phenotype and physiological metabolic features. The empirical breeding strategy feature vector generated during pre-training is used as a priori adjustment parameter and concatenated with the environmental trajectory matrix and physiological trajectory matrix in the feature space. The empirical breeding strategy feature vector contains vectorized representations of key agronomic parameters such as historical optimal planting density, water and fertilizer application cycle, and pest and disease control scheme. During the concatenation process, the empirical breeding strategy feature vector is expanded into a sequence with the same time length as the environmental trajectory matrix and physiological trajectory matrix to ensure that the feature dimensions are fully aligned on the time axis. After concatenation, a multimodal collaborative feature sequence is generated, which simultaneously contains environmental information, physiological response information, and historical experience information, and can comprehensively characterize the growth status of the variety under specific environment and breeding strategy. Step S3-2: Calculation of environmental-physiological phase lag characteristic sequence: The multimodal collaborative feature sequence is input into the environment-physiological response coupling module, and gradient decomposition and rate differentiation are performed on the multimodal collaborative feature sequence along the time dimension. First, the first-order differentiation calculation is performed on the time series of the environmental feature dimension to obtain the environmental stress gradient sequence characterizing the rate of change of external pressure. Then, the first-order differentiation calculation is performed on the time series of the physiological feature dimension to obtain the physiological metabolic rate sequence characterizing the rate of internal material transformation. The formula for calculating the physiological metabolic rate sequence is: ,in, for Physiological metabolic rate at any given time. for Cumulative values ​​of physiological characteristics at any given time; The formula for calculating the environmental stress gradient sequence is: ,in, for Environmental stress gradient at time t, for Cumulative values ​​of environmental parameters at any given time; The sliding window cross-correlation algorithm was used to calculate the temporal misalignment between the environmental stress gradient sequence and the physiological metabolic rate sequence at each physiological node. The sliding window size was set to 10 consecutive standard sampling periods, covering the time span of the main physiological responses of Poria cocos. For each sliding window, the cross-correlation value of the two sequences at different lag times was calculated. The lag time corresponding to the maximum cross-correlation value was the environmental-physiological phase lag time within that window. The formula for calculating the cross-correlation of sliding windows is: ,in, The environmental stress gradient sequence and the physiological metabolic rate sequence at lag time Interrelation between places To adjust the sliding window size, This is the window start time; The formula for calculating the environmental-physiological phase lag time is: ,in, This refers to the environmental-physiological phase lag time. The phase lag times calculated by all sliding windows are arranged in chronological order to generate an environmental-physiological phase lag characteristic sequence. This sequence can quantitatively characterize the response speed and adaptability of a variety to environmental changes. The shorter the phase lag time, the faster the variety responds to environmental changes and the stronger its adaptability. Step S3-3: Generation of recommended vectors for suitable varieties: The environmental-physiological phase lag feature sequence and the empirical breeding strategy feature vector are input in parallel into the hierarchical aggregation unit of the variety adaptation network. The hierarchical aggregation unit adopts a multilayer perceptron structure and uses a nonlinear activation function to evaluate the stress decay of the phase lag feature sequence. The stress decay evaluation is used to quantify the negative impact of phase lag on the growth performance of the variety. The longer the phase lag time, the greater the stress decay value and the lower the adaptation of the variety. A multi-dimensional scoring mapping is performed by superimposing the fit weights of empirical breeding strategies; the strategy fit weights are calculated based on the similarity between the variety and historically successful breeding cases, with higher similarity resulting in greater weights; the formula for calculating the comprehensive fit score is as follows: ,in, For the first The overall compatibility score of each variety This is the stress attenuation weighting coefficient. The decay rate constant is For the first The average phase lag time of each variety throughout its entire growth period This is the strategy fit weighting coefficient. For the first Individual variety fit score with empirical breeding strategies; The comprehensive suitability scores of all Poria cocos candidate varieties are sorted in descending order and then normalized so that all score values ​​are mapped to between 0 and 1. The normalized score sequence is the suitability recommendation vector that represents the regional suitability priority of each variety. The larger the element value in the suitability recommendation vector, the higher the regional suitability of the corresponding variety. Step S3-4: Generation of fingerprint profiles of dominant traits of candidate strains: Based on the ranking priority of the recommended vectors of suitable varieties, a subset of dynamic growth characteristic time series maps corresponding to the top 30% of highly suitable varieties is selected. This subset contains complete growth data of varieties that perform well in the target environment and can effectively represent the core trait characteristics of the dominant strains. The dynamic growth characteristic time series spectrum focuses on the overlapping period of the peak metabolic rate and the steady state of morphology and phenotype. The peak metabolic rate corresponds to the stage of most vigorous growth of the variety, when the material conversion efficiency of the variety is the highest. The steady state of morphology and phenotype corresponds to the stage of most stable growth of the variety, when the variety has the strongest resistance to stress. The overlapping period is the key period when the variety exhibits both high yield and high resistance characteristics. The overlapping time periods are projected onto the phase compensation safety domain in the environmental-physiological phase lag characteristic sequence. The phase compensation safety domain refers to the time period in which the phase lag time is within a preset range. During this time period, the variety can adapt to environmental changes through self-regulation and will not show obvious growth inhibition. Key indicators such as the critical value of extracellular hydrolytic enzyme activity, the upper limit of accumulation of characteristic triterpenoid metabolites, and the tolerance of sclerotium swelling morphology variation are extracted within the safety domain. The extracted key indicators are subjected to cross-dimensional threshold calibration and feature encoding; a minimum threshold is set for each indicator, and only strains that reach or exceed the threshold are considered to have the corresponding dominant trait; the calibrated thresholds are classified and encoded according to three dimensions: high yield, high resistance, and high stability, to generate a fingerprint map of the dominant trait of candidate strains that characterizes the core traits of high yield, high resistance, and high stability; the fingerprint map of the dominant trait of candidate strains adopts a binary encoding format, with each binary bit corresponding to a core trait threshold, 1 indicating that the threshold is reached and 0 indicating that the threshold is not reached.

[0018] In this embodiment, the core function of step S4 is to establish a closed-loop interaction mechanism between the physical cultivation process and the digital algorithm model. This is achieved through a forward feature push link to enable real-time flow and dynamic prediction of growth data, and through a reverse decision-making and control link to achieve dynamic optimization of variety resources and precise adjustment of environmental parameters, forming a data-driven iterative optimization closed loop. This significantly improves the efficiency of variety selection and the targeted nature of environmental control. The detailed steps are as follows: Step S4-1: Building a two-way real-time linkage channel architecture: A two-way real-time linkage channel based on the industrial internet architecture is constructed. The channel includes two independent and collaborative data flow paths: a forward feature push link and a reverse decision control link. The forward feature push link is responsible for transmitting the real-time growth data collected by the simulated cultivation system to the algorithm model system, while the reverse decision control link is responsible for transmitting the decision instructions generated by the algorithm model system to the simulated cultivation system. The bidirectional real-time linkage channel adopts a layered architecture design, consisting of a hardware interface layer, a data transmission layer, a message middleware layer, and an application interface layer from bottom to top. The hardware interface layer provides a standard industrial bus interface for multi-chamber parallel cultivation systems and edge data gateways, supporting multiple industrial communication protocols such as Modbus, Profinet, and EtherCAT. The data transmission layer uses the TCP / IP protocol for data transmission, ensuring the reliability and integrity of data transmission. The message middleware layer adopts a publish-subscribe message bus architecture to achieve asynchronous data transmission and multi-node subscription. The application interface layer provides standardized API interfaces to achieve seamless integration between the algorithm model system and the simulation cultivation system. Step S4-2: Implementation of the forward feature push chain: The positive feature push link is built on a publish-subscribe message bus architecture, with the edge data gateway as the feature publishing node and the multimodal feature fusion engine as the subscription node. The edge data gateway aligns the phenotypic indicators and physiological metabolic indicators collected by the in-situ multimodal sensing and monitoring group with timestamps and encapsulates the format to generate a dynamically updated time series map of growth features. A sliding time window incremental synchronization mechanism is used for data transmission. The sliding time window size is set to 5 standard sampling periods. Each time, only the newly added feature data within the window is transmitted, instead of the full data, which effectively reduces network bandwidth consumption. The edge data gateway encapsulates the newly added feature data into feature increment packets, which contain three parts: timestamp information, feature dimension identifier, and feature value. Before data transmission, a data integrity check is performed. A cyclic redundancy check algorithm is used to calculate the checksum of the feature increment packet, and the checksum is appended to the end of the feature increment packet. After receiving the feature increment packet, the multimodal feature fusion engine recalculates the checksum and compares it with the appended checksum. If the checksums match, the data integrity is confirmed, the feature increment packet is merged into the dynamic growth feature time series map, and the online status update and dynamic prediction of the variety adaptation network are triggered. If the checksums do not match, a retransmission request is sent to the edge data gateway to request the retransmission of the feature increment packet. Step S4-3: Reduction of low-fitness breeding vectors in the reverse decision-making and control chain: The reverse decision control link is constructed based on hardware abstraction layer mapping and instruction queue scheduling mechanism. The model predictive control decision-maker serves as the instruction generation node, and the microenvironment precision control array and multi-chamber parallel cultivation chamber serve as the instruction execution node. The model predictive control decision-maker receives the recommended variety vectors output by the variety adaptation network and parses them into a variety priority ranking list. The variety priority ranking list is arranged from high to low according to the comprehensive adaptation score, and each list element contains the variety number and the corresponding adaptation priority score. A mapping relationship between variety numbers and the physical addresses of standardized breeding units is established through a hardware abstraction layer. The hardware abstraction layer shields the specific implementation details of the underlying hardware and provides a unified hardware control interface for upper-layer applications. Based on the variety priority ranking list, low-fitness varieties that need to be eliminated are identified. Typically, the bottom 20% of varieties are identified as low-fitness varieties, and their breeding carriers need to be eliminated to free up resources for the breeding of high-fitness varieties. The model predictive control decision-maker generates a carrier dormancy command and sends it to the multi-chamber parallel culture chamber through the hardware abstraction layer. The carrier dormancy command contains a list of physical addresses of the standardized culture units that need to be dormant. After receiving the carrier dormancy command, the multi-chamber parallel culture chamber dynamically disconnects the control loop of the stepper temperature and humidity actuator and the gas component proportion regulating valve of the corresponding standardized culture unit, stops environmental control and nutrient supply, and realizes the physical reduction of the low-fitness variety culture carrier. The reduced culture unit can be used to inoculate new high-fitness variety strains or as a spare unit for subsequent use. Step S4-4: Dynamic timing environmental parameter sequence adjustment of the reverse decision-making control link: The model predictive control decision-maker simultaneously receives the fingerprint profiles of the dominant traits of candidate strains output by the variety adaptation network and maps them to a multidimensional environmental constraint boundary. The multidimensional environmental constraint boundary includes the upper and lower limits of parameters such as temperature, humidity, light, carbon dioxide concentration and nutrient supply, as well as the coupling relationship between the parameters. The multidimensional environmental constraint boundary is input into the rolling time-domain optimizer of the model predictive control decision engine. Using the core trait threshold of the candidate strain's dominant trait fingerprint as the target reference trajectory, the dynamic time-series environmental parameter sequence correction values ​​for the next control cycle are calculated in real-time. The objective function of the model predictive control is: ,in, The objective function value, To predict the length of the time domain, for Always The predicted value output by the system at time [time]. for The target reference trajectory at any given time, for Changes in control quantity at any given time The output error weight matrix is... The weight matrix represents the changes in the control quantity. The rolling time-domain optimizer solves the above optimization problem in each control cycle to obtain the optimal control sequence; only the first element of the control sequence is applied to the system, and the above optimization process is repeated in the next control cycle to achieve rolling optimization control of environmental parameters. Based on the calculated dynamic time-series environmental parameter sequence correction values, the light intensity and spectral composition are adjusted by an adjustable spectral light source, the substrate water and fertilizer supply ratio is dynamically compensated by a micro-pressure differential nutrient solution supply pump, and the temperature, humidity and carbon dioxide concentration are adjusted by a step-type temperature and humidity actuator and a gas component ratio regulating valve. Through the above adjustments, the cultivation microenvironment evolves in a direction that is conducive to the expression of dominant traits, promoting the growth and metabolism of highly adaptable varieties. Step S4-5: Switching between bidirectional link timing synchronization and steady-state maintenance modes: The forward feature push link and the reverse decision control link maintain microsecond-level timing alignment through a high-precision hardware clock synchronization protocol; a precise time protocol is used to achieve clock synchronization of the entire system. The master clock is deployed on the server side of the algorithm model system, and the edge data gateway, micro-environment precision control array and in-situ multimodal sensing and monitoring group are all synchronized with the master clock as slave clocks; the clock synchronization accuracy is better than 1 microsecond, ensuring the timing consistency of data acquisition, transmission, processing and control command execution; The iterative convergence evaluation module monitors in real time the changes in the fitness ranking information entropy of the recommended vector of the suitable varieties and the core trait coverage of the fingerprint spectrum of the dominant traits of the candidate strains. When the information entropy fluctuation amplitude and the coverage residual value both meet the preset preliminary stability conditions, the iterative convergence evaluation module sends a control enable signal, and the two-way real-time linkage channel automatically switches to the steady-state maintenance mode. In steady-state maintenance mode, the forward feature push link continues to collect and transmit growth data, but only records data and monitors status, and no longer triggers dynamic prediction and parameter updates of the variety adaptation network; the reverse decision control link locks the current environmental control parameters and variety breeding configuration, stops dynamically reducing low-fitness varieties and adjusting environmental parameters, and maintains the stability of the breeding environment; the two-way real-time linkage channel remains in standby mode until it responds to the closed-loop iteration termination signal and outputs the final breeding and promotion plan.

[0019] In this embodiment, the core function of step S5 is to monitor the deviation between the actual cultivation process and the model prediction baseline in real time. Dynamic time warping technology is used to achieve precise alignment of different growth stages. Mahalanobis distance is used to quantify the joint deviation of multidimensional growth indicators. When the deviation exceeds the tolerance range, the online correction process of the model is triggered. The mapping relationship between network parameters and empirical strategies is updated through a gradient compensation algorithm to ensure the accuracy and robustness of variety suitability assessment and dominant trait screening. The detailed steps are as follows: Step S5-1: Dynamic time warping and alignment of the actual growth feature sequence with the reference sequence: The actual growth characteristic sequence captured in real time by the in-situ multimodal sensing monitoring group at the current cultivation stage node is obtained; the actual growth characteristic sequence contains phenotypic indicators and physiological metabolic indicators of the same dimension as the dynamic growth characteristic time series map, and the sampling time point is consistent with the preset physiological node. The dynamic time warping algorithm is used to construct the cumulative distance matrix between the actual growth feature sequence and the baseline sequence corresponding to the reproductive stage in the dynamic growth feature time series map; the formula for calculating the cumulative distance matrix is ​​as follows: ,in, For the actual sequence number The point and the baseline sequence The cumulative distance between points Let be the Euclidean distance between the two points. , , These are the cumulative distances between three adjacent locations; The optimal time-matching mapping relationship is searched by backtracking along the path of minimum cumulative distance. Starting from the bottom right element of the cumulative distance matrix, the process backtracks sequentially to the top left, selecting the element with the smallest cumulative distance among three adjacent positions as the next path point, until the top left element of the matrix is ​​reached. Through this path, a nonlinear time mapping relationship is established between the actual growth feature sequence and the baseline sequence, and the asynchronously sampled multidimensional indicators are forced to align to a unified physiological progress scale, generating a set of time-aligned feature pairs. Step S5-2: Calculation of multidimensional growth deviation vector and determination of Mahalanobis distance: The feature deviation is extracted by subtracting the set dimension by dimension based on the temporal alignment feature; for each feature dimension, the difference between the actual measured value and the benchmark value is calculated to obtain the feature deviation of that dimension. The feature deviations of each dimension are normalized and weighted according to the preset feature importance weights to construct a multidimensional growth deviation vector. The feature importance weights are preset according to the degree of influence of each indicator on the growth and quality of Poria cocos. The mycelial option is biased towards extracellular hydrolytic enzyme activity and mycelial spread coverage, the budding option is biased towards primordium differentiation density, and the sclerotium option is biased towards sclerotium volume and concentration of characteristic triterpenoid metabolites. The inverse matrix of the historical baseline environmental response covariance matrix is ​​used to perform a quadratic form operation with the multidimensional growth deviation vector to quantify the Mahalanobis distance scalar, which characterizes the joint deviation of the multidimensional indicators. The formula for calculating the Mahalanobis distance is: ,in, For Mahalanobis distance scalar, A multidimensional growth bias vector. The mean of the historical baseline deviation vector. The historical baseline environmental response covariance matrix; The Mahalanobis distance scalar is compared with a preset tolerance threshold for determination. The preset tolerance threshold is set based on the statistical analysis results of historical cultivation data, and is usually set to 3. If the Mahalanobis distance scalar is less than or equal to the preset tolerance threshold, the growth status is determined to be normal, no model correction is required, and the current cultivation process continues. If the Mahalanobis distance scalar is greater than the preset tolerance threshold, the growth status is determined to be abnormal, and the online model correction process is triggered. Step S5-3: Generation of multi-dimensional environmental compensation gradient instructions: Extract the abnormal dimension identifiers and their deflection directions from the multidimensional growth deviation vector that exceed the amplitude limit; the abnormal dimension identifiers correspond to specific indicators of growth characteristics, and the deflection directions are divided into positive deviation and negative deviation. Positive deviation indicates that the actual value is higher than the benchmark value, and negative deviation indicates that the actual value is lower than the benchmark value. The abnormal dimension identifiers are inversely mapped to the adjustable parameter space of the environmental simulation unit; a mapping relationship table between growth characteristic dimensions and environmental regulation parameters is established. This table is constructed based on the physiological response mechanism of Poria cocos and historical cultivation data. For example, low mycelial spread rate corresponds to temperature and humidity regulation parameters, and low extracellular hydrolytic enzyme activity corresponds to substrate nutrient ratio regulation parameters. The compensation intensity and regulatory polarity of each environmental regulation dimension are calculated by combining the deflection direction and the deviation amplitude. The regulatory polarity is determined according to the deflection direction. Negative deviation corresponds to positive regulation, that is, increasing the environmental parameter value, and positive deviation corresponds to negative regulation, that is, decreasing the environmental parameter value. The compensation intensity is proportional to the deviation amplitude. At the same time, the regulation sensitivity and safety range of environmental parameters are considered to avoid excessive regulation that leads to drastic fluctuations in the growth environment. Multidimensional environmental compensation gradient instructions are generated by splicing together. The multidimensional environmental compensation gradient instructions include the compensation amount and adjustment direction of all environmental parameters that need to be adjusted, such as temperature, humidity, light intensity, carbon dioxide concentration and nutrient supply. The instruction format is consistent with the dynamic time series environmental parameter sequence. Step S5-4: Variety adaptation network coupling weights and empirical policy mapping bias update: The multidimensional environmental compensation gradient command is input into the gradient compensation module of the variety adaptation network; the gradient compensation module calculates the weight update amount layer by layer in reverse along the forward propagation topology of the environmental-physiological response coupling module; A momentum decay strategy is used to iteratively correct the coupling weight matrix between hidden layer nodes; the momentum update calculation formula is: ; ; in, For the first The momentum vector of the next iteration. The momentum decay coefficient, The gradient of the loss function with respect to the weights. For the first The weight matrix after the next iteration The learning rate; Simultaneously, the multi-dimensional environmental compensation gradient instruction is projected onto the knowledge graph node space of the empirical strategy memory; the semantic similarity offset between the feature vector of the empirical cultivation strategy and the current environmental deviation is calculated; the semantic similarity is calculated using cosine similarity, which is obtained by comparing the environmental deviation features with the feature vectors of each node in the knowledge graph; The mapping bias parameters of the empirical cultivation strategy feature vector are dynamically recalibrated based on the semantic similarity offset. For empirical strategy nodes with high similarity, their mapping bias weights are increased, and for empirical strategy nodes with low similarity, their mapping bias weights are decreased, so that the empirical strategy can better adapt to the current growth environment deviation. Step S5-5: Corrected adaptation recommendation and output of dominant trait fingerprint spectrum: Load the corrected coupling weight matrix and the updated mapping bias parameters, reset the inference state of the variety adaptation network; re-input the current time-series aligned feature pair set and the target environment feature set into the corrected environment-physiological response coupling module and hierarchical aggregation unit. Perform forward inference operations to recalculate the comprehensive fit score and core trait thresholds for each variety; after bias compensation optimization, obtain the corrected fit variety recommendation vector and the corrected candidate strain dominant trait fingerprint spectrum. The revised recommended vectors for suitable varieties and the revised fingerprints of the dominant traits of candidate strains are packaged into a standardized correction data package. The standardized correction data package is then sent to the reverse decision-making and control link through a two-way real-time linkage channel to drive the dynamic adjustment and optimization of the next round of breeding process.

[0020] In this embodiment, the core function of step S6 is to drive multiple rounds of cultivation and evaluation cycles based on the online correction results of the model. By continuously iterating and optimizing the variety cultivation configuration and environmental control parameters, the deviation between model prediction and actual growth is gradually reduced until the fitness ranking information entropy is stable and the coverage of core traits converges, ensuring that the finally selected suitable varieties and dominant strains have the highest reliability and stability. The detailed steps are as follows: Step S6-1: Initialize a new round of cultivation based on the corrected results: The revised recommended vectors of suitable varieties and the revised fingerprints of the dominant traits of candidate strains are injected into the environmental simulation unit through a reverse decision-making control link. The environmental simulation unit first analyzes the priority ranking of the recommended vectors of suitable varieties and resets the variety carrying configuration matrix of the multi-chamber parallel cultivation chamber according to the ranking results. For high-suitability varieties with high priority ranking, the number of their corresponding standardized cultivation units is increased. For low-suitability varieties with low priority ranking, their cultivation carriers are further reduced or cultivation is stopped. Meanwhile, the environmental simulation unit maps the core trait thresholds of the fingerprint spectrum of the dominant traits of candidate strains, reconstructs the environmental stress boundary and nutrient ratio gradient of the dynamic time-series environmental parameter sequence, and adjusts the temperature and humidity control range, photoperiod parameters, carbon dioxide concentration fluctuation range, and element ratio and application frequency of substrate water and fertilizer supply at each growth stage according to the core trait thresholds, so that the cultivation environment is more conducive to the full expression of dominant traits. Step S6-2: Generation of a new round of dynamic growth feature time series map: A new round of standardized parallel cultivation is triggered by the reconstructed environmental stress boundary and nutrient ratio gradient; the microenvironment precision control array drives each standardized cultivation unit to perform continuous dynamic environmental evolution, strictly following the reconstructed dynamic temporal environmental parameter sequence; at preset physiological nodes, the in-situ multimodal sensing monitoring group simultaneously captures the phenotypic and physiological metabolic indicators of Poria cocos varieties in each cultivation unit. The collected raw data is timestamped and formatted by the edge data gateway before being transmitted to the multimodal feature fusion engine. The multimodal feature fusion engine performs cross-source spatiotemporal registration, outlier removal, and data cleaning and fusion on the multi-source heterogeneous data to generate a new round of dynamic growth feature time series map. The new round of dynamic growth feature time series map contains complete growth and metabolic data of all surviving varieties in this round of cultivation, and the time resolution and data dimension are consistent with the initial map. Step S6-3: Fit evaluation and trait extraction for the current iteration round: The new dynamic growth characteristic time series map and the target environmental feature set are synchronously input into the variety adaptation network. The variety adaptation network first calculates the phase difference between the environmental stress gradient and the physiological metabolic rate of each variety during the current breeding process through the environmental-physiological response coupling module, and generates the updated environmental-physiological phase lag feature sequence. Then, the hierarchical aggregation unit combines the feature vector of the empirical breeding strategy to perform nonlinear mapping and multidimensional scoring on the phase lag feature sequence, and outputs the suitable variety recommendation vector for the current iteration round; at the same time, based on the sorting priority of the current suitable variety recommendation vector, the growth data of highly suitable varieties are screened, the core trait thresholds are extracted and labeled, and the fingerprint map of the dominant traits of the candidate strains in the current iteration round is generated. Step S6-4: Calculation of iterative convergence metric: The iterative convergence evaluation module captures the recommended vector of the suitable varieties and the fingerprint of the dominant traits of the candidate strains in the current iteration round, and calculates two core convergence indices respectively. The first indicator is the fitness ranking information entropy; fitness ranking information entropy is used to quantify the uniformity of the probability distribution of fitness priority scores for each variety. The lower the information entropy, the more stable the ranking. The formula for calculating fitness ranking information entropy is: ,in, Information entropy for adaptive sorting, This represents the number of currently surviving candidate varieties of Poria cocos. For the first The proportion of the fit priority score of each variety to the total score of all varieties; The second indicator is core trait coverage; core trait coverage is used to quantify the proportion of core trait dimensions that reach a preset threshold in the fingerprint of the dominant trait of candidate strains. Higher coverage indicates more complete dominant traits. The formula for calculating core trait coverage is: ,in, For core trait coverage, To meet the requirement of a certain number of feature dimensions for core trait thresholds, This represents the total number of all core dimensions. Step S6-5: Convergence State Determination and Iteration Control: The information entropy fluctuation amplitude is extracted by performing a sliding window variance calculation on the information entropy sequence of the current iteration and the information entropy sequence of consecutive historical iterations. The sliding window size is set to 5 consecutive iterations. The formula for calculating the information entropy fluctuation amplitude is as follows: ,in, The amplitude of information entropy fluctuation. For the first The adaptive sorting information entropy of each iteration round. The mean of the information entropy within the sliding window; Simultaneously, the coverage of the core trait in the current generation round is subtracted from the target threshold to extract the coverage residual value; the formula for calculating the coverage residual value is: ,in, This represents the coverage residual value. The preset target threshold for core trait coverage. This represents the coverage of core traits in the current iteration round. If the information entropy fluctuation amplitude does not fall into the preset stable range, or the coverage residual value is not lower than the allowable residual, the system is determined to have not reached the convergence state; the recommended vector of the adapted varieties and the fingerprint map of the dominant traits of the candidate strains in the current iteration are marked as the correction benchmark for the next iteration, and the closed-loop rolling execution of steps S2 to S5 is triggered through the two-way real-time linkage channel to drive the environmental simulation unit and the variety adaptation network to perform the next round of dynamic parameter optimization and physiological characteristic alignment; If the information entropy fluctuation amplitude falls within the preset stable range and the coverage residual value is lower than the allowable residual, the system is determined to have reached the convergence state; the preset stable range is usually set to 0 to 0.05, and the allowable residual is usually set to 0.02. Step S6-6: Termination of closed-loop iteration and system state locking: When the system reaches the convergence state, the iterative convergence evaluation module generates and sends a closed-loop iteration termination signal; this signal is simultaneously sent to the environmental simulation unit, the variety adaptation network, the model predictive control decision-maker, and the two-way real-time linkage channel. Upon receiving the termination signal, the environmental simulation unit locks all control parameters of the current dynamic time-series environmental parameter sequence and stops the dynamic adjustment of environmental parameters; the variety adaptation network and model prediction control decision-maker freeze all parameter update states and lock the current iteration final state of the adaptation variety recommendation vector ranking list and the candidate strain dominant trait fingerprint spectrum; the two-way real-time linkage channel closes the instruction update function of the reverse decision control link and only retains the data recording function of the forward feature push link. Once the system status is locked, all cultivation units maintain their current environmental conditions and variety configurations until the current growth stage of the cultivation process is completed, preparing for the final extraction of the target suitable variety and dominant strain.

[0021] In this embodiment, the core function of step S7 is to respond to the closed-loop iteration termination signal, extract the optimal combination of Poria cocos varieties and strains from the convergent and stable model output, integrate environmental control parameters and agronomic management experience throughout the entire growth period, and form a standardized and implementable cultivation and promotion plan to realize the transformation and application of digital breeding results to actual agricultural production; the detailed steps are as follows: Step S7-1: System final state locking and final state data packet output: In response to the closed-loop iteration termination signal, freeze all parameter update states of the variety adaptation network and model prediction control decision-maker, prohibiting any form of weight adjustment and strategy recalibration; lock the current iteration final state of the adaptation variety recommendation vector ranking list and the converged dynamic time series environment parameter sequence to ensure the uniqueness and immutability of all decision data; Extract the final nutrient ratio regulation matrix output by the model predictive control decision-maker, as well as the set of agronomic intervention rules in the experience strategy memory bank that best matches the current breeding process; integrate and encapsulate all the above final state data to generate a final state locked data package; the final state locked data package is stored in an encrypted and compressed format and contains four core modules: variety adaptability data, environmental regulation data, growth characteristic data, and strategy rule data. Each module is attached with version identifier and timestamp information. Step S7-2: Extraction of target suitable varieties and compliance verification: The matching variety recommendation vector ranking list in the final state locked data packet is parsed, and the unique variety identifier and matching priority score corresponding to the first element of the ranking are extracted. The unique variety identifier adopts the nationally unified Poria cocos germplasm resource coding system to ensure the traceability of variety identity. The regional Poria cocos germplasm resource database was used to verify the genetic stability and quarantine compliance of the extracted variety unique identifiers. The genetic stability verification was conducted by comparing the phenotypic and metabolic characteristics of the variety over three consecutive generations of planting to verify the consistency and stability of its genetic traits, requiring that the coefficient of variation of the main traits not exceed 5%. The quarantine compliance review was conducted in accordance with the relevant national agricultural plant quarantine standards to check whether the variety carries quarantine pests and whether it meets the quarantine requirements for cross-regional introduction. Once the review is approved, the unique identifier of the variety will be fixed as the identifier of the target suitable variety; if the review is not approved, the next variety in the recommended vector of suitable varieties will be selected in turn, and the above verification process will be repeated until a target suitable variety that meets the requirements is found. Step S7-3: Multidimensional constraint matching screening of target dominant strains: Based on the target adapted variety identifier, the core trait threshold set of the candidate strain dominant trait fingerprint spectrum after iterative convergence in the final state locked data package is extracted; the core trait threshold set includes the sclerotium yield threshold and triterpenoid content threshold of the high yield dimension, the disease resistance threshold and environmental stress tolerance threshold of the high resistance dimension, and the trait variation coefficient threshold of the high stability dimension. The core trait threshold set was mapped to the multidimensional phenotypic and metabolic characteristic data of the local experimental strain germplasm library. The local experimental strain germplasm library stores complete growth and metabolic data of all isolated and purified strains under the target adapted variety. The data dimensions are completely consistent with the fingerprint of the candidate strain's dominant traits. A multidimensional constraint matching algorithm was used to screen out experimental strain combinations that simultaneously met all core trait thresholds. The multidimensional constraint matching algorithm screened out strains with the smallest distance and all dimensions meeting the threshold requirements by calculating the Euclidean distance between the strain feature vector and the core trait threshold vector. The germplasm code of the screened strains was marked as the target dominant strain identifier. If there are multiple strains that meet the conditions, strains with clear genetic background and good passage stability were selected as the final target dominant strains. Step S7-4: Generation of a standardized management strategy set for the entire reproductive period: Based on the identification of the target dominant strain and the locked convergent dynamic time-series environmental parameter sequence, the final nutrient ratio regulation matrix output by the model predictive control decision-maker is extracted; the dynamic time-series environmental parameter sequence is segmented according to the physiological transition boundaries of the mycelial stage, budding stage and sclerotium stage of Poria cocos to obtain the microenvironment parameter regulation curves of each growth stage. The corresponding regional agronomic intervention rules and disease early warning benchmarks are associated with the experience strategy memory bank; the regional agronomic intervention rules include field operation specifications such as planting density control, timing of inter-row cultivation and weeding, and requirements for soil covering thickness; the disease early warning benchmarks include the occurrence conditions, early warning thresholds and control measures for common diseases at each growth stage. Integrating the above information, a standardized management strategy set is generated, which includes a microenvironment dynamic evolution curve, a substrate water and fertilizer tiered replenishment procedure, and an emergency intervention plan for abnormal operating conditions. The microenvironment dynamic evolution curve clarifies the specific values ​​of temperature, humidity, light, and carbon dioxide concentration for each growth stage. The substrate water and fertilizer tiered replenishment procedure specifies the application time, amount, and element ratio of water and fertilizer at different growth stages. The emergency intervention plan for abnormal operating conditions provides specific emergency handling procedures and measures for sudden situations such as extreme weather and outbreaks of pests and diseases. Step S7-5: Structured packaging and distribution of the cultivation and promotion plan: Based on the target compatible variety identifier, the target dominant strain identifier, and the standardized management strategy set, structured data is encapsulated and versioned; a unique scheme number and version number are assigned to each breeding and promotion scheme, and a scheme version management mechanism is established to facilitate subsequent updates and traceability; The geographic coordinates of the target planting area are linked to the climate adaptability rating label. The climate adaptability rating is calculated based on the historical meteorological data of the target planting area and the environmental requirements of the target suitable varieties, and is divided into four levels: highly suitable, suitable, basically suitable and unsuitable. The program generates a breeding and promotion plan that includes an optimal configuration list of varieties and strains, a full-cycle environmental control instruction sequence, and a standardized field operation guide. The optimal configuration list of varieties and strains clarifies the correspondence between target suitable varieties and target dominant strains, inoculation requirements, and strain preservation conditions. The full-cycle environmental control instruction sequence provides control parameters that can be directly imported into smart agricultural equipment. The standardized field operation guide details the entire process of operation from site preparation to harvesting and processing in a combination of text and images. The cultivation and promotion plan is distributed to the agricultural technology extension terminal through the data interaction interface. The agricultural technology extension terminal can fine-tune some parameters in the plan according to the actual production conditions, and feed back the fine-tuned execution data to the system to provide data support for subsequent model optimization and plan iteration.

[0022] In this embodiment, the simulated cultivation system is the core carrier of the physical cultivation process in this invention. It is responsible for reproducing the dynamic environment of the target planting area under standardized and controllable conditions, performing parallel cultivation of multiple varieties, and collecting multimodal data of Poria cocos growth in real time. The system adopts a modular design, and the components interact with each other and transmit control commands through a standard industrial bus. The system has the characteristics of high precision, high stability, and high parallelism. The simulated cultivation system specifically includes four core components: a multi-chamber parallel cultivation chamber, a microenvironment precision control array, an in-situ multimodal sensing and monitoring group, and an edge data gateway. Multi-chamber parallel culture chamber: The multi-chamber parallel cultivation chamber is the physical space carrier for cultivating Poria cocos varieties. Its interior is divided into standardized cultivation units that are physically isolated from each other. The number of standardized cultivation units can be flexibly expanded according to the number of Poria cocos candidate varieties to be tested. The effective cultivation volume of a single unit is 0.5 cubic meters, which can meet the growth needs of Poria cocos throughout its entire growth cycle. Each standardized cultivation unit is equipped with an independent temperature-controlled substrate tank and a micro-airflow circulation pipeline at its bottom. The independent temperature-controlled substrate tank is made of stainless steel and has built-in heating and cooling elements, enabling precise control of the substrate temperature, with a temperature control range of 10 degrees Celsius to 40 degrees Celsius. The micro-airflow circulation pipeline is evenly arranged along the inner wall of the cultivation unit, achieving uniform air circulation through a low-speed fan to avoid excessive local temperature and humidity differences. The multi-chamber parallel cultivation chamber has a unified initialization operation capability, enabling synchronous substrate sterilization, isothermal pre-cultivation, and equal-volume inoculation of all standardized cultivation units. Substrate sterilization adopts high-temperature and high-pressure steam sterilization, with a sterilization temperature of up to 121 degrees Celsius and a sterilization pressure of 0.1 MPa. During the isothermal pre-cultivation stage, the substrate temperature of all cultivation units can be uniformly stabilized at the preset value, with a temperature uniformity better than ±0.2 degrees Celsius. Equal-volume inoculation is achieved through an automated inoculation robot, with an inoculation accuracy of ±0.1 grams, ensuring that the initial inoculation conditions of each cultivation unit are completely consistent. Microenvironment precision control array: The microenvironment precision control array is coupled one-to-one with each standardized cultivation unit, serving as the actuator for precise control of the cultivation microenvironment. The array comprises four core actuators: a stepping temperature and humidity actuator, an adjustable spectral light source, a gas component proportioning valve, and a micro-pressure differential nutrient solution supply pump. The stepping temperature and humidity actuator employs a high-precision PID control algorithm, achieving a temperature control accuracy of ±0.1 degrees Celsius and a humidity control accuracy of ±1%. The adjustable spectral light source can output any spectral combination within the range of 400 nm to 700 nm, with an illumination intensity adjustment range of 0 lux to 10,000 lux and a photoperiod accuracy down to the minute level. The gas component proportioning valve precisely controls the carbon dioxide concentration within the cultivation unit, with an adjustment range of 300 ppm to 2000 ppm and a control accuracy of ±50 ppm. The micro-pressure differential nutrient solution supply pump uses a peristaltic pump structure, enabling precise quantitative supply of nutrient solution with a supply accuracy of ±0.1 ml. The microenvironment precision control array communicates with the environment simulation unit through a digital-physical control interface, and can receive the dynamic time-series environmental parameter sequence issued by the environment simulation unit in real time. The array integrates a closed-loop feedback control mechanism, which monitors the actual environmental parameters in the cultivation unit in real time through built-in micro sensors, compares them with the target parameters, and automatically adjusts the output state of the actuators, so that the actual cultivation microenvironment strictly follows the dynamic time-series environmental parameter sequence to continuously and dynamically evolve. In-situ multimodal sensing monitoring group: The in-situ multimodal sensing monitoring group is embedded in the gap between the inner wall and the matrix of each standardized cultivation unit and is the core sensing device for acquiring Poria cocos growth data. The monitoring group integrates four types of sensing elements: multispectral imaging probe, three-dimensional morphology scanner, microfluidic permeation sampling needle and respiratory quotient detection probe, which can realize non-contact, in-situ synchronous acquisition of Poria cocos phenotypic indicators and physiological metabolic indicators. The multispectral imaging probe can simultaneously acquire images in four bands: 450 nm, 550 nm, 650 nm, and 850 nm, with an image resolution of 1920×1080 pixels and a frame rate of 1 frame per second. Through image segmentation and feature extraction algorithms, it can automatically calculate phenotypic indicators such as hyphal spread coverage and primordium differentiation density. The 3D morphology scanner uses structured light scanning technology to acquire high-precision 3D point cloud data of Poria cocos sclerotia, with a point cloud density of up to 1 million points per square centimeter and a reconstruction accuracy of ±0.1 mm. It can accurately measure morphological parameters such as sclerotia volume, surface area, and shape factor. The microfluidic percolation sampling needle adopts an embedded design, enabling quantitative extraction of interstitial fluid samples without damaging the Poria cocos growth environment, with a single sampling volume ranging from 10 μL to 100 μL. The respiratory quotient detection probe can monitor changes in oxygen and carbon dioxide concentrations within the cultivation unit in real time, calculate the substrate respiration intensity, and reflect the metabolic activity of Poria cocos. The in-situ multimodal sensing and monitoring group is uniformly controlled by the edge data gateway, which can automatically trigger data acquisition at preset physiological nodes to ensure that all acquired data has a unified time reference. Edge data gateway: The edge data gateway is electrically connected to the in-situ multimodal sensing and monitoring group, serving as the data processing and transmission hub for the simulated cultivation system. The gateway incorporates a time-series buffer and a multi-protocol conversion module, enabling simultaneous access to data streams from multiple sensing and monitoring groups and supporting various communication protocols such as Modbus, RS485, and Ethernet. The time-series buffer module employs a ring buffer structure to cache the most recent 72 hours of collected data, preventing data loss. The multi-protocol conversion module can convert heterogeneous data output from different sensing elements into a unified, standardized format. The edge data gateway is responsible for the initial processing of the collected raw data, including timestamp alignment, preliminary screening of outliers, and data format encapsulation. The gateway performs multi-dimensional time-series alignment of the synchronously acquired phenotypic and physiological metabolic indicators according to the collection timestamps, generating a dynamically updated time-series map of growth characteristics. The dynamic growth characteristic time-series map is stored using a standardized time-series data structure, which can be directly read and processed by the downstream algorithm model system. The edge data gateway communicates with the multimodal feature fusion engine of the algorithm model system through a standard industrial bus to achieve real-time data transmission.

[0023] In this embodiment, the algorithm model system is the decision-making core of the invention, responsible for processing the multimodal data collected by the simulated breeding system and executing core algorithm tasks such as variety suitability assessment, dominant trait extraction, environmental regulation decision-making, and iterative convergence determination. The system adopts a layered architecture design, with each module interacting with data through standardized interfaces, and features high real-time performance, high accuracy, and self-learning and adaptive characteristics. The algorithm model system specifically includes five core components: a multimodal feature fusion engine, a variety suitability network, an experience strategy memory, a model prediction and control decision-maker, and an iterative convergence evaluation module. Multimodal feature fusion engine: The multimodal feature fusion engine communicates directly with the edge data gateway, serving as the data entry point for the algorithm model system. The engine incorporates a feature alignment and dimensionality reduction mapping unit, capable of receiving the dynamically grown feature time-series map output from the edge data gateway and performing cross-source spatiotemporal registration with the target environment feature set. The feature alignment unit employs a dynamic time warping algorithm, aligning environmental data and grown data with different sampling frequencies and time starting points to a unified time reference axis. The dimensionality reduction mapping unit uses principal component analysis, enabling dimensionality reduction processing of high-dimensional multimodal data, eliminating redundant features, and retaining core information. The multimodal feature fusion engine stitches together the aligned and dimensionality-reduced environmental and growth features in the feature space to generate a standardized collaborative input tensor. The standardized collaborative input tensor is the input data of the variety adaptation network, which contains environmental information, physiological response information and phenotypic information during the growth process of Poria cocos, and can comprehensively characterize the growth status of the variety under a specific environment. The engine has real-time data processing capabilities and can complete data processing and tensor generation within 100 milliseconds after receiving a new feature increment package, meeting the real-time requirements of the system. Variety matching network: The variety adaptation network, deployed downstream of the multimodal feature fusion engine, is the core algorithm module of this invention. The network adopts a deep learning architecture, with a built-in environment-physiological response coupling module and a hierarchical aggregation unit. The environment-physiological response coupling module can perform gradient decomposition and rate differentiation processing on the input multimodal collaborative feature sequence along the time dimension, separating the environmental stress gradient sequence and the physiological metabolic rate sequence, and calculating the phase difference between the two through a sliding window cross-correlation algorithm to generate an environment-physiological phase lag feature sequence. The hierarchical aggregation unit adopts a multilayer perceptron structure, which can perform nonlinear mapping and multidimensional scoring on the phase lag feature sequence, and generate a comprehensive adaptation score for each variety by combining the feature vector of empirical breeding strategy. The output of the variety adaptation network consists of two parts. The first part is a recommended vector of adapted varieties that characterizes the regional adaptation priority of each variety. Each element in the vector corresponds to the adaptation score of a candidate Poria cocos variety. The higher the score, the better the regional adaptability of the variety. The second part is a fingerprint map of the dominant traits of candidate strains that characterize the threshold of core traits such as high yield, high resistance, and high stability. The map includes key trait thresholds such as the critical value of extracellular hydrolytic enzyme activity, the upper limit of accumulation of characteristic triterpenoid metabolites, and the tolerance of sclerotium swelling morphology variation. The variety adaptation network has online reasoning capabilities and can dynamically adjust the output results based on the real-time updated dynamic growth characteristic time series map. Experience strategy memory bank: The experience strategy memory bank uses a temporal knowledge graph structure to store historical cultivation logs, regional agronomic parameters, and expert intervention rules, serving as the prior knowledge source for this invention. The knowledge graph nodes include variety nodes, environment nodes, strategy nodes, and trait nodes, with edges representing their relationships. The historical cultivation log stores complete data from all past cultivation experiments, including environmental parameters, growth data, variety information, and cultivation results. The regional agronomic parameters store soil characteristics, climate features, and traditional cultivation experience from different planting areas. The expert intervention rules store the experience and knowledge summarized by agricultural experts during the cultivation of Poria cocos regarding pest and disease control, water and fertilizer management, and environmental regulation. The experience strategy memory bank can dynamically provide prior adjustment parameters and strategy fit weights during the variety adaptation network inference process. When processing data from new target planting areas, the memory bank can retrieve similar historical breeding cases based on regional characteristics, extract corresponding empirical breeding strategy feature vectors, and input them as prior information into the variety adaptation network to improve the accuracy of evaluation results. The memory bank has self-learning capabilities and can automatically update the results of each breeding experiment into the knowledge graph, continuously enriching and optimizing prior knowledge. Model predictive control decision maker: The model predictive control decision-maker, cascaded with the output of the variety adaptation network, is the core module for realizing reverse decision-making and control. The decision-maker has a built-in rolling time-domain optimizer and constraint solver, which can analyze the recommended vector of adapted varieties and the fingerprint spectrum of the dominant traits of candidate strains output by the variety adaptation network to generate corresponding control commands. The rolling time-domain optimizer adopts the model predictive control algorithm, which can predict the system state in the future multiple control cycles with the core trait threshold as the target reference trajectory and solve for the optimal control quantity sequence. The constraint solver can consider the physical constraints of environmental parameters and the physiological constraints of Poria cocos growth to ensure that the generated control commands are safe and feasible. The output of the model predictive control decision-maker includes two types of control commands. The first type is the variety priority control command, which is used to dynamically reduce the breeding vectors of low-fit varieties. The decision-maker parses the suitable variety recommendation vector into a variety priority ranking list, and issues a vector dormancy command to the multi-chamber parallel cultivation chamber through the hardware abstraction layer, cutting off the control loop of the low-priority cultivation unit. The second type is the environmental regulation command, which is used to dynamically adjust the cultivation microenvironment. The decision-maker maps the fingerprint of the candidate strain's dominant traits into a multi-dimensional environmental constraint boundary, inverts and calculates the correction value of the dynamic time-series environmental parameter sequence, and executes the corresponding regulation operation through the microenvironment precision control array. Iterative convergence evaluation module: The iterative convergence evaluation module communicates with the model prediction and control decision-maker to monitor the iterative convergence status of the system. The module has a built-in information entropy calculation unit and coverage determiner, which can calculate the two core convergence indicators, fitness ranking information entropy and core trait coverage, in real time. The information entropy calculation unit can calculate fitness ranking information entropy based on the fitness score distribution of each variety in the fitness variety recommendation vector, and quantify the stability of the ranking. The coverage determiner can count the number of core trait dimensions that reach the threshold in the fingerprint spectrum of the dominant traits of candidate strains, calculate the core trait coverage, and quantify the completeness of the dominant traits. The iterative convergence evaluation module compares the calculated convergence index with a preset threshold. When the fluctuation amplitude of the fitness ranking information entropy falls within a preset stable range, and the residual value of the core trait coverage is lower than the allowable residual, the system is determined to have reached convergence, and a closed-loop iteration termination signal is generated and sent. The termination signal is simultaneously sent to all modules of the system, locking the system state and terminating the iteration process. If the system has not reached convergence, the module will trigger the next round of iteration, driving the simulated breeding system and the algorithm model system to continue optimization. Although embodiments of the present invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions, and variations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A method for recommending suitable varieties of Poria cocos and screening dominant strains based on the linkage of environmental simulation and algorithm model, characterized in that: Includes the following steps: S1. Obtain multidimensional environmental data covering the entire growth cycle of Poria cocos in the target planting area, and generate a structured target environmental feature set through time-series alignment, outlier filtering and principal component dimensionality reduction. S2. Input the target environmental feature set into the environmental simulation unit to construct a dynamic temporal environmental parameter sequence covering the mycelial stage, budding stage and sclerotium stage of Poria cocos; under the drive of the dynamic temporal environmental parameter sequence, conduct standardized parallel cultivation of multiple Poria cocos candidate varieties, collect phenotypic indicators and physiological metabolic indicators synchronously according to preset physiological nodes, and integrate to construct a dynamic growth characteristic temporal map of each variety. S3. Input the target environment feature set, the empirical breeding strategy feature vector generated by pre-training, and the time series map of dynamic growth characteristics into the variety adaptation network. The variety adaptation network has a built-in environment-physiological response coupling module. By calculating the phase difference between the environmental stress gradient and the physiological metabolic rate, it generates the adaptation variety recommendation vector that characterizes the regional adaptation priority of each variety, as well as the candidate strain dominant trait fingerprint map that characterizes the threshold of high yield, high resistance, and high stability core traits. S4. Establish a two-way real-time linkage channel between the simulation cultivation system and the algorithm model system. The channel includes a forward feature push link and a reverse decision control link. The forward feature push link inputs the dynamically updated time series map of growth characteristics into the variety adaptation network in real time for dynamic prediction. The reverse decision control link feeds back the recommended vector of the adapted varieties and the fingerprint map of the dominant traits of the candidate strains to the environmental simulation unit. The environmental simulation unit dynamically reduces the cultivation carriers of low-fit varieties based on the priority ranking of the recommended vector of the adapted varieties, and adjusts the stress intensity and nutrient ratio of the dynamic time series environmental parameter sequence in reverse through the model prediction control algorithm based on the core trait threshold of the fingerprint map of the dominant traits of the candidate strains. S5. At the cultivation stage node, the currently collected actual growth characteristic sequence is dynamically time-aligned with the benchmark sequence in the dynamic growth characteristic time series map, and a multidimensional growth deviation vector is calculated. When the Mahalanobis distance of the growth deviation vector exceeds the preset tolerance threshold, the online correction process of the model is triggered. The process generates an environmental compensation gradient based on the growth deviation vector, updates the mapping bias between the coupling weight of the variety adaptation network and the feature vector of the empirical cultivation strategy through the gradient compensation algorithm, and outputs the corrected adaptation variety recommendation vector and the corrected candidate strain dominant trait fingerprint map. S6. Drive the next round of parallel cultivation by the modified adaptive variety recommendation vector and the modified candidate strain dominant trait fingerprint spectrum through the reverse decision control link. Repeat steps S2 to S5 until the adaptability ranking information entropy of the adaptive variety recommendation vector stabilizes to the preset interval and the core trait coverage of the candidate strain dominant trait fingerprint spectrum converges to the target threshold, and generate a closed-loop iteration termination signal. S7. Responding to the closed-loop iteration termination signal, extract the Poria cocos variety with the highest priority in the recommended variety vector as the target suitable variety, match the measured strains of all core dimensions of the fingerprint spectrum of the candidate strains as the target dominant strains, integrate the environmental regulation parameter matrix and management strategy of the whole growth period, and generate a cultivation and promotion plan.

2. The method for recommending suitable varieties of Poria cocos and screening dominant strains based on the linkage of environmental simulation and algorithm model according to claim 1, characterized in that: Multidimensional environmental data covering the entire growth cycle of Poria cocos in the target planting area were obtained. After time-series alignment, outlier filtering, and principal component dimensionality reduction, a structured target environmental feature set was generated; specifically including: By deploying an array of IoT sensor nodes and a regional agricultural meteorological database in the target planting area, multidimensional raw environmental data covering the mycelial stage, budding stage and sclerotium stage of Poria cocos are collected. The multidimensional raw environmental data includes atmospheric meteorological parameters, soil physicochemical parameters and forest microclimate parameters. Using a preset standard sampling period as a unified time reference axis, the asynchronous time series of each acquisition channel in the multidimensional environmental raw data is resampled and interpolated and fused to eliminate the differences in device sampling frequency and timestamp offset, and output a time-synchronized environmental data matrix. Sliding window local statistics are performed on the time-synchronized environmental data matrix. Outlier data points exceeding the normal fluctuation threshold are identified by combining dynamic confidence interval judgment rules. The weighted average of adjacent time-series valid data is used for smooth replacement to generate a clean environment data sequence. Calculate the variable covariance representation matrix of the clean environment data sequence, extract the principal component orthogonal basis whose cumulative variance contribution rate meets the preset retention threshold, project and map the high-dimensional environmental parameters to the low-dimensional principal component space, remove multicollinear redundant features, and obtain the dimensionality-reduced environmental feature vector. The dimensionality-reduced environmental feature vector is segmented and recombined according to the preset physiological nodes of the entire growth cycle of Poria cocos, bound to the environmental state labels of the corresponding growth stage, encapsulated into a standardized matrix format, and output a structured target environmental feature set.

3. The method for recommending suitable varieties of Poria cocos and screening dominant strains based on the linkage of environmental simulation and algorithm model according to claim 1, characterized in that: The target environmental feature set is input into the environmental simulation unit to construct a dynamic temporal environmental parameter sequence covering the mycelial stage, budding stage, and sclerotium stage of Poria cocos. Driven by the dynamic temporal environmental parameter sequence, multiple groups of Poria cocos candidate varieties are cultured in a standardized parallel manner, and phenotypic and physiological metabolic indicators are collected synchronously according to preset physiological nodes. The dynamic growth characteristic time-series map of each variety is then integrated and constructed. Specifically, this includes: The structured target environmental feature set is loaded into the environmental simulation unit. The temporal boundary and environmental state label of each segment sequence in the target environmental feature set are analyzed through the reproductive stage mapping model. The cumulative environmental effect value of the stage is used as the physiological conversion criterion. According to the differentiated physiological needs of the mycelial stage, budding stage and sclerotium stage of Poria cocos, a dynamic temporal environmental parameter sequence including temperature and humidity control gradient, light photoperiod, carbon dioxide concentration fluctuation and substrate water and fertilizer supply ratio is generated. The dynamic temporal environmental parameter sequence is injected into the multi-chamber standardized parallel cultivation system. Each independent cultivation chamber carries a single Poria cocos candidate variety and performs unified substrate sterilization, constant temperature pre-cultivation and equal amount of inoculation initialization operations. The environmental simulation unit sends environmental control commands in real time through the digital-physical control interface, driving the actuators of each chamber to make the actual cultivation microenvironment strictly follow the dynamic temporal environmental parameter sequence for continuous dynamic evolution. Within the driving cycle of the dynamic time-series environmental parameter sequence, with preset physiological nodes as synchronous trigger signals, the mycelial spread coverage, primordium differentiation density, and sclerotium swelling morphological phenotypic indicators of each Poria cocos candidate variety are captured non-contactly using a multispectral visual sensing array and a three-dimensional structured light scanner. Simultaneously, the in-situ microfluidic sampling module is activated to extract interstitial fluid from the matrix. The extracellular hydrolytic enzyme activity, characteristic triterpenoid metabolite concentration, and matrix respiration intensity physiological metabolic indicators are measured by the online biochemical analysis module. The phenotypic and physiological metabolic indicators acquired synchronously were aligned in a multidimensional time series according to the collection timestamp. Dynamic threshold normalization and outlier removal algorithms were used to clean and fuse the multi-source heterogeneous data. The cleaned multidimensional indicators were mapped to a standardized feature space with the number of cultivation days as the baseline axis and multimodal growth response as the dimension. A dynamic growth characteristic time series map representing the coupling relationship between the environmental response trajectory and metabolic dynamic evolution of each Poria cocos candidate variety throughout the entire growth period was constructed and encapsulated into a standardized time series data structure for output.

4. The method for recommending suitable varieties of Poria cocos and screening dominant strains based on the linkage of environmental simulation and algorithm model according to claim 1, characterized in that: The target environmental feature set, the feature vector of the pre-trained empirical breeding strategy, and the time series map of dynamic growth characteristics are input into the variety adaptation network. The variety adaptation network has a built-in environment-physiological response coupling module. By calculating the phase difference between the environmental stress gradient and the physiological metabolic rate, it generates a recommended variety vector characterizing the regional adaptation priority of each variety, as well as a fingerprint map of the dominant traits of candidate strains characterizing the thresholds of high yield, high resistance, and high stability core traits. Specifically, this includes: The target environmental feature set and the time series map of dynamic growth features are synchronously mapped with timestamps. The evolution trajectory of environmental parameters and the response trajectory of physiological indicators of each Poria cocos candidate variety at each growth stage are extracted. The feature vector of empirical cultivation strategy is used as a priori adjustment parameter and spliced ​​with the evolution trajectory and response trajectory in the feature space to generate a multimodal collaborative feature sequence. The multimodal collaborative feature sequence is input into the environment-physiological response coupling module. Gradient decomposition and rate differentiation are performed on the multimodal collaborative feature sequence along the time dimension to separate the environmental stress gradient sequence representing changes in external pressure and the physiological metabolic rate sequence representing internal material transformation. The sliding window cross-correlation algorithm is used to calculate the time misalignment value of the environmental stress gradient sequence and the physiological metabolic rate sequence at each physiological node, and the environment-physiological phase lag feature sequence is generated quantitatively. The environmental-physiological phase lag feature sequence and the empirical breeding strategy feature vector are input in parallel into the hierarchical aggregation unit of the variety adaptation network. The phase lag feature sequence is evaluated for stress decay through a nonlinear activation function. The strategy fit weight is superimposed to perform multi-dimensional scoring mapping. The results are sorted in descending order and normalized and encoded to output the recommended variety vector that represents the regional adaptation priority of each variety. Based on the ranking priority of the recommended vectors for suitable varieties, a subset of dynamic growth characteristic time series maps corresponding to highly suitable varieties is selected. The overlapping period between the peak metabolic rate and the steady state of morphological phenotype in the subset of dynamic growth characteristic time series maps is located. The overlapping period is projected onto the phase compensation safety domain in the environmental-physiological phase lag characteristic sequence. The critical value of extracellular hydrolase activity, the upper limit of accumulation of characteristic triterpenoid metabolites, and the tolerance of sclerotium swelling morphological variation are extracted within the safety domain. Cross-dimensional threshold calibration and feature encoding are performed to generate fingerprint maps of the dominant traits of candidate strains that characterize the core traits of high yield, high resistance, and high stability.

5. The method for recommending suitable varieties of Poria cocos and screening dominant strains based on the linkage of environmental simulation and algorithm model according to claim 1, characterized in that: The simulated cultivation system specifically includes: a multi-chamber parallel cultivation chamber, a microenvironment precision control array, an in-situ multimodal sensing and monitoring group, and an edge data gateway; The multi-chamber parallel cultivation chamber is divided into standardized cultivation units that are physically isolated from each other. Each cultivation unit is equipped with an independent temperature-controlled substrate tank and a micro-airflow circulation pipeline at the bottom, which is used to carry a single Poria cocos candidate variety and perform uniform sterilization, constant temperature pre-cultivation and equal amount of inoculation. The microenvironment precision control array is coupled one-to-one with each standardized cultivation unit, including a stepping temperature and humidity actuator, an adjustable spectral light source, a gas component ratio regulating valve, and a micro-pressure differential nutrient solution supply pump. It is used to receive the dynamic time-series environmental parameter sequence issued by the environmental simulation unit and drive the actuator to make the cultivation microenvironment continuously and dynamically evolve according to the sequence instructions. The in-situ multimodal sensing and monitoring group is embedded in the gap between the inner wall and the matrix of each standardized culture unit, and integrates a multispectral imaging probe, a three-dimensional morphology scanner, a microfluidic percolation sampling needle and a respiratory quotient detection probe, which are used to simultaneously capture phenotypic indicators and physiological metabolic indicators when triggered by preset physiological nodes. The edge data gateway is electrically connected to the in-situ multimodal sensing and monitoring group. It has a built-in time-series cache and multi-protocol conversion module, which is used to align and encapsulate the captured multi-source heterogeneous indicators with timestamps, generate a dynamically updated time-series map of growth features, and output it downstream through a standard industrial bus.

6. The method for recommending suitable varieties of Poria cocos and screening dominant strains based on the linkage of environmental simulation and algorithm model according to claim 5, characterized in that: The algorithm model system specifically includes: a multimodal feature fusion engine, a variety adaptation network, an empirical strategy memory, a model prediction control decision-maker, and an iterative convergence evaluation module; The multimodal feature fusion engine is directly connected to the edge data gateway and has a built-in feature alignment and dimensionality reduction mapping unit. It is used to receive dynamically grown feature time series maps and perform cross-source spatiotemporal registration with the target environment feature set, and output a standardized collaborative input tensor. The variety adaptation network is deployed downstream of the multimodal feature fusion engine. It has a built-in environment-physiological response coupling module and hierarchical aggregation unit. It is used to receive standardized collaborative input tensors and empirical breeding strategy feature vectors from the empirical strategy memory bank, perform phase difference calculation between environmental stress gradient and physiological metabolic rate, and output adaptation variety recommendation vector and candidate strain dominant trait fingerprint map. The experience strategy memory bank uses a temporal knowledge graph structure to store historical breeding logs, regional agronomic parameters and expert intervention rules, which are used to dynamically provide prior adjustment parameters and strategy fit weights during variety adaptation network reasoning. The model predictive control decision-maker is cascaded with the output of the variety adaptation network, and has a built-in rolling time-domain optimizer and constraint solver. It is used to analyze the priority ranking of the adaptation variety recommendation vector and the core trait threshold of the candidate strain's dominant trait fingerprint spectrum, and generate environmental compensation gradient instructions and nutrient ratio regulation matrix. The iterative convergence evaluation module communicates with the model prediction control decision-maker and has a built-in information entropy calculation unit and coverage arbiter. It is used to monitor the fluctuation range of fitness ranking information entropy and the convergence status of core trait coverage in real time, and generate a closed-loop iteration termination signal when the preset convergence conditions are met.

7. The method for recommending suitable varieties of Poria cocos and screening dominant strains based on the linkage of environmental simulation and algorithm model according to claim 6, characterized in that: A two-way real-time linkage channel is established between the simulation cultivation system and the algorithm model system. This channel includes a forward feature push link and a reverse decision-making and control link. The forward feature push link inputs the dynamically updated time-series growth characteristic maps into the variety adaptation network in real time for dynamic prediction. The reverse decision-making and control link feeds back the recommended vectors of adapted varieties and the fingerprint maps of the dominant traits of candidate strains to the environmental simulation unit. The environmental simulation unit dynamically reduces the cultivation carriers of low-fitness varieties based on the priority ranking of the recommended vectors of adapted varieties, and, based on the core trait thresholds of the fingerprint maps of the dominant traits of candidate strains, uses a model prediction and control algorithm to reversely adjust the stress intensity and nutrient ratio of the dynamic time-series environmental parameter sequences. Specifically, this includes: The positive feature push link is built on a publish-subscribe message bus architecture. The edge data gateway serves as the feature publishing node, and the multimodal feature fusion engine serves as the subscription node. It adopts a sliding time window incremental synchronization mechanism to encapsulate the newly collected phenotypic and physiological metabolic indicators of each standardized breeding unit into feature increment packages. After data integrity verification, the packages are pushed to the variety adaptation network in real time, triggering the online status update of the environment-physiological response coupling module and the rolling correction of the dynamic growth feature time series map. The reverse decision control link is constructed based on hardware abstraction layer mapping and instruction queue scheduling mechanism. The model predictive control decision-maker is used as the instruction generation node, and the microenvironment precision control array is used as the instruction execution node. The suitable variety recommendation vector is parsed into a variety priority ranking list. The carrier dormancy instruction is issued to the multi-chamber parallel cultivation chamber through the hardware abstraction layer. The control loop of the stepping temperature and humidity actuator and gas component ratio regulating valve of the standardized cultivation unit corresponding to the low priority ranking is dynamically cut off, so as to realize the physical reduction of the low-fitness variety cultivation carrier. Meanwhile, the reverse decision-making control link maps the fingerprint of the candidate strain’s dominant traits into a multidimensional environmental constraint boundary, inputs it into the rolling time-domain optimizer of the model predictive control decision-maker, uses the core trait threshold as the target reference trajectory, and calculates the dynamic time-series environmental parameter sequence correction value for the next control cycle in real time. Through an adjustable spectral light source and a micro-differential nutrient solution supply pump, it performs stress intensity gradient decay and dynamic compensation of nutrient ratio, thus completing the reverse decision-making control. The forward feature push link and the reverse decision control link maintain microsecond-level timing alignment through a high-precision hardware clock synchronization protocol. After the iterative convergence evaluation module issues a control enable signal, the bidirectional real-time linkage channel automatically switches to steady-state maintenance mode, locks the current environmental control parameters and variety cultivation configuration, until it responds to the closed-loop iteration termination signal and outputs the final cultivation and promotion plan.

8. The method for recommending suitable varieties of Poria cocos and screening dominant strains based on the linkage of environmental simulation and algorithm model according to claim 1, characterized in that: At each stage of the cultivation process, the currently collected actual growth characteristic sequence is dynamically time-aligned with the baseline sequence in the dynamic growth characteristic time series map, and a multidimensional growth deviation vector is calculated. When the Mahalanobis distance of the growth deviation vector exceeds a preset tolerance threshold, an online model correction process is triggered. This process generates an environmental compensation gradient based on the growth deviation vector, updates the mapping bias between the coupling weights of the variety adaptation network and the feature vector of the empirical cultivation strategy through a gradient compensation algorithm, and outputs the corrected adaptation variety recommendation vector and the corrected fingerprint map of the dominant traits of the candidate strains. Specifically, this includes: The actual growth feature sequence captured in real time by the in-situ multimodal sensing monitoring group at the current cultivation stage node is obtained. The dynamic time warping algorithm is called to construct the cumulative distance matrix between the actual growth feature sequence and the corresponding fertility stage benchmark sequence in the dynamic growth feature time series map. The optimal time matching mapping relationship is searched by backtracking along the minimum cumulative distance path. The asynchronously sampled multidimensional indicators are forced to be aligned to a unified physiological progress scale, and a time-aligned feature pair set is generated. Based on the temporal alignment feature, the feature offset is extracted by subtracting the set dimension by dimension. The offset is then normalized and weighted according to the preset feature importance weights to construct a multidimensional growth deviation vector. The inverse matrix of the historical benchmark environmental response covariance matrix is ​​called and the multidimensional growth deviation vector is used to perform a quadratic operation to quantify the Mahalanobis distance scalar that represents the joint deviation of the multidimensional indicators. The Mahalanobis distance scalar is compared with the preset tolerance threshold. If the result is that the tolerance threshold is exceeded, the abnormal dimension identifier and its deflection direction in the multidimensional growth deviation vector are extracted. The abnormal dimension identifier is inversely mapped to the adjustable parameter space of the environmental simulation unit. The compensation intensity and adjustment polarity of each environmental control dimension are calculated by combining the deflection direction and deviation amplitude, and multidimensional environmental compensation gradient instructions are spliced ​​together. The multidimensional environmental compensation gradient command is input into the gradient compensation module of the variety adaptation network. The weight update amount is calculated layer by layer in reverse along the forward propagation topology of the environment-physiological response coupling module. The coupling weight matrix between each hidden layer node is iteratively corrected using a momentum decay strategy. Simultaneously, the multidimensional environmental compensation gradient command is projected onto the knowledge graph node space of the empirical strategy memory bank. The semantic similarity offset between the empirical breeding strategy feature vector and the current environmental deviation is calculated. The mapping bias parameters are dynamically recalibrated to complete the online correction of the model. Load the corrected coupling weight matrix and the updated mapping bias parameters, reset the inference state of the variety adaptation network, re-input the current time-series aligned feature pair set and the target environment feature set into the corrected environment-physiological response coupling module and hierarchical aggregation unit, perform forward inference operation, output the corrected adaptation variety recommendation vector and the corrected candidate strain dominant trait fingerprint spectrum optimized by bias compensation, and encapsulate them into a standardized correction data package and send it to the reverse decision control link.

9. The method for recommending suitable varieties of Poria cocos and screening dominant strains based on the linkage of environmental simulation and algorithm model according to claim 1, characterized in that: The revised suitable variety recommendation vector and the revised candidate strain dominant trait fingerprint are used to drive the next round of parallel breeding through a reverse decision-making control link. Steps S2 to S5 are repeated until the fitness ranking information entropy of the suitable variety recommendation vector stabilizes to a preset interval, and the core trait coverage of the candidate strain dominant trait fingerprint converges to the target threshold, generating a closed-loop iteration termination signal; specifically including: The modified recommended vectors for suitable varieties and the modified fingerprint maps of the dominant traits of candidate strains are injected into the environmental simulation unit through the reverse decision-making control link. The environmental simulation unit analyzes the priority ranking of the recommended vectors, resets the variety carrying configuration matrix of the multi-chamber parallel cultivation chamber, and maps the core trait thresholds of the fingerprint maps to reconstruct the environmental stress boundary and nutrient ratio gradient of the dynamic time-series environmental parameter sequence. Based on the reconstructed environmental stress boundary and nutrient ratio gradient triggering a new round of standardized parallel cultivation, the microenvironment precision control array drives each standardized cultivation unit to perform continuous dynamic environmental evolution, and simultaneously captures phenotypic indicators and physiological metabolic indicators at preset physiological nodes. After cross-source spatiotemporal registration and cleaning fusion through a multimodal feature fusion engine, a new round of dynamic growth feature time series map is generated. The new round of dynamic growth characteristic time series map and target environmental feature set are synchronously input into the variety adaptation network. The phase difference calculation of the environment-physiological response coupling module and the nonlinear mapping reasoning of the hierarchical aggregation unit are performed to output the adaptation variety recommendation vector and the candidate strain dominant trait fingerprint map of the current iteration. The iterative convergence evaluation module captures the recommended vector of suitable varieties in the current iteration round, extracts the probability distribution of the suitability priority score of each variety in the vector, and calculates and generates the suitability ranking information entropy of the current iteration round; at the same time, it analyzes the fingerprint map of the dominant traits of the candidate strains in the current iteration round, counts the proportion of the number of feature dimensions that meet the core trait threshold to the total number of core dimensions, and calculates and generates the core trait coverage of the current iteration round. The information entropy fluctuation amplitude is extracted by performing a sliding window variance calculation on the information entropy sequence of the current iteration and the information entropy sequence of the continuous historical iterations; the coverage residual value is extracted by subtracting the core trait coverage of the current iteration from the target threshold. If the information entropy fluctuation amplitude does not fall into the preset stable range, or the coverage residual value is not lower than the allowable residual, the recommended vector of the adapted varieties and the fingerprint map of the dominant traits of the candidate strains in the current iteration are marked as the correction benchmark for the next iteration. The closed-loop rolling execution of steps S2 to S5 is triggered through the two-way real-time linkage channel, driving the environmental simulation unit and the variety adaptation network to perform the next round of dynamic parameter optimization and physiological characteristic alignment. If the information entropy fluctuation amplitude falls within the preset stable range and the coverage residual value is lower than the allowable residual, then the information entropy of the suitability ranking of the suitable variety recommendation vector is determined to be stable to the preset range, and the core trait coverage of the candidate strain's dominant trait fingerprint converges to the target threshold. The iterative convergence evaluation module generates and issues a closed-loop iteration termination signal, locks the regulation parameters of the current dynamic time series environmental parameter sequence and the variety configuration matrix of the multi-chamber parallel cultivation chamber, and terminates the dynamic adjustment command update of the reverse decision-making control link.

10. The method for recommending suitable varieties of Poria cocos and screening dominant strains based on the linkage of environmental simulation and algorithm model according to claim 1, characterized in that: In response to the closed-loop iteration termination signal, the highest-priority Poria cocos variety in the recommended variety vector is extracted as the target suitable variety. Tested strains from all core dimensions of the candidate strains' dominant trait fingerprints are matched as the target dominant strains. The entire growth period environmental regulation parameter matrix and management strategies are integrated to generate a cultivation and promotion plan; specifically including: In response to the closed-loop iteration termination signal, the parameter update state of the variety adaptation network and the model prediction control decision-maker is frozen, the adaptation variety recommendation vector sorting list and the converged dynamic time series environment parameter sequence of the current iteration final state are locked, and the final state locking data packet is output. Based on the matching variety recommendation vector sorting list in the final state locked data packet, the first element of the sorting is parsed to extract the corresponding variety unique identifier and matching priority score. The regional Poria cocos germplasm resource database is called to perform genetic stability verification and quarantine compliance review. After the review is passed, the variety unique identifier is solidified as the target matching variety identifier. Based on the target adapted variety identifier, the core trait threshold set of the candidate strain dominant trait fingerprint spectrum after iterative convergence and labeling in the final state locked data package is extracted. The core trait threshold set is then mapped in feature space with the multidimensional phenotypic-metabolic feature data of the local experimental strain germplasm library. A multidimensional constraint matching algorithm is used to screen out experimental strain combinations that simultaneously meet the high yield threshold, high resistance threshold, and high stability threshold, and their germplasm codes are marked as the target dominant strain identifier. Based on the identification of the target dominant strain and the locked convergent dynamic time-series environmental parameter sequence, the final nutrient ratio regulation matrix output by the model predictive control decision-maker is extracted. The dynamic time-series environmental parameter sequence is segmented into time periods according to the physiological transition boundaries of the mycelial stage, budding stage and sclerotium stage of Poria cocos. The corresponding regional agronomic intervention rules and disease early warning benchmarks in the experience strategy memory bank are associated to generate a standardized management strategy set that includes the dynamic evolution curve of the microenvironment, the substrate water and fertilizer step-by-step replenishment procedure and the emergency intervention plan for abnormal working conditions. Based on the target compatible variety identifier, the target dominant strain identifier, and the standardized management strategy set, structured data is encapsulated and versioned, and the geographical coordinate attributes and climate adaptability rating tags of the target planting area are bound together. A cultivation and promotion plan is generated, which includes a list of optimal configurations for varieties and strains, a full-cycle environmental regulation instruction sequence, and standardized field operation guidelines. The plan is then distributed to agricultural technology extension terminals for execution through a data interaction interface.

Images