In-vitro culture method for fertilized eggs of Xinxiu lobsters

Abstract

The invention relates to the technical field of breeding of Xinxiu lobsters, and discloses an in-vitro culture method for fertilized eggs of Xinxiu lobsters. According to the method, a multi-dimensional culture data set is formed by obtaining an image sequence of a fertilized egg development stage, culture solution component time sequence monitoring data and environmental factor fluctuation data. Analyzing the image sequence to obtain a developmental stage feature vector, extracting dynamic concentration gradient features from culture solution data, and generating a fused developmental feature set through cross-modal correlation fusion; calling a pre-trained embryonic development evaluation network model to generate a development risk score set, performing joint regulation and control analysis on the environmental factors and the culture solution data based on abnormal correlation degree distribution, generating a culture parameter optimization strategy, and further generating a dynamic regulation and control instruction to be fed back to culture equipment. According to the method, precise regulation and control of the fertilized egg culture process are achieved, the requirements of different development stages are met, and the scientificity and adaptability of in-vitro culture are improved.

Description

Technical Field

[0001] This invention relates to the field of ornate lobster farming technology, specifically to a method for in vitro culture of ornate lobster fertilized eggs. Background Technology

[0002] As a crustacean of significant economic value, the improvement of artificial breeding technology for the ornate lobster is crucial for the sustainable development of the industry. The cultivation of fertilized eggs is a key step in artificial breeding, directly affecting the survival rate of larvae and the quality of the population. Traditional in vitro culture methods rely heavily on empirical operations, adjusting culture conditions by manually observing changes in the morphology of fertilized eggs, which suffers from problems such as strong subjectivity and delayed response.

[0003] In their natural environment, the development of ornate spiny lobster fertilized eggs is influenced by a synergistic effect of multiple environmental factors, including water temperature, salinity, and dissolved oxygen. They are also highly sensitive to the nutrients and ion balance in the culture medium. The environmental requirements of fertilized eggs differ significantly at different developmental stages. For example, the cell division stage is more sensitive to fluctuations in dissolved oxygen, while the blastocyst stage has specific requirements for the rhythm of nutrient supply. Current culture techniques struggle to accurately capture these dynamic changes in demand, often leading to developmental abnormalities such as cell division arrest and embryonic malformations due to improper control of environmental parameters.

[0004] There is a lack of systematic methods for the dynamic monitoring and regulation of culture medium components. Traditional methods often rely on periodic sampling and testing, which is insufficient to reflect changes in concentration gradients in real time and cannot correlate changes in culture medium with the intrinsic relationship between embryonic development. When abnormalities occur, it is difficult to quickly pinpoint the influencing factors, resulting in insufficiently targeted regulatory measures. At the same time, the interaction between environmental factors and culture medium components is complex, and adjusting a single parameter may trigger a chain reaction, further increasing the difficulty of in vitro culture.

[0005] With the development of aquaculture technology, intelligent monitoring equipment is gradually being applied to the cultivation process. However, how to effectively integrate multi-source monitoring data and transform it into precise control strategies remains a problem to be solved. Current technologies often limit data processing to single-dimensional analysis, lacking cross-modal data correlation and fusion, making it difficult to comprehensively reflect the true state of embryonic development, resulting in insufficient scientific rigor and timeliness in control decisions. Therefore, establishing an in vitro culture method capable of dynamically sensing developmental status, accurately analyzing influencing factors, and generating optimized strategies in real time has become an important direction for upgrading artificial breeding technology of Orychomycosis var. spp. Summary of the Invention

[0006] The purpose of this invention is to provide a method for in vitro culture of fertilized eggs of the spiny lobster, so as to solve the problems mentioned in the background art.

[0007] To achieve the above objectives, the present invention provides a method for in vitro culture of fertilized eggs of the ornate lobster, the method comprising:

[0008] A multi-dimensional culture data set of the target fertilized egg batch is obtained, which includes image sequences of fertilized egg development stages, time-series monitoring data of culture medium components, and environmental factor fluctuation data.

[0009] The cell division pattern parsing processing is performed on the fertilized egg development stage image sequence to obtain the development stage feature vector. The dynamic concentration gradient extraction processing is performed on the culture medium component time-series monitoring data to obtain the culture medium change feature vector. The development stage feature vector and the culture medium change feature vector are subjected to cross-modal correlation fusion processing to generate a fused development feature set.

[0010] The pre-trained embryonic development assessment network model is invoked to perform developmental abnormality weight allocation processing on the fusion development feature set, generating a developmental risk score set corresponding to the batch of fertilized eggs, wherein the developmental risk score set includes the distribution of abnormal correlation between environmental factor fluctuation data and key nodes of each developmental stage;

[0011] Based on the abnormal correlation distribution, the environmental factor fluctuation data and the time-series monitoring data of culture medium composition are jointly regulated and analyzed to generate the culture parameter optimization strategy for the batch of fertilized eggs. The culture parameter optimization strategy is used to indicate the environmental compensation parameters and culture medium composition adjustment schemes for key developmental nodes.

[0012] Based on the culture parameter optimization strategy, dynamic control instructions for the culture environment are generated, and these instructions are fed back to the culture equipment control system to trigger real-time parameter calibration.

[0013] Preferably, the step of performing cell division pattern parsing processing on the fertilized egg development stage image sequence to obtain development stage feature vectors includes:

[0014] The fertilized egg development stage image sequence is segmented into developmental cycle stages to obtain multiple keyframe images of developmental stages;

[0015] The pre-trained cell structure segmentation model is invoked to encode cell division features in each keyframe image of the developmental stage, generating an initial developmental feature vector.

[0016] The initial developmental feature vector is subjected to embryological feature enhancement processing to obtain an enhanced developmental feature vector. The embryological feature enhancement processing includes the following steps: matching a set of morphological entities associated with the keyframe of the current developmental stage from a preset crustacean embryonic development knowledge base; inputting the set of morphological entities into the cell structure segmentation model for entity feature encoding processing to generate a set of entity feature vectors; and performing weighted fusion of the set of entity feature vectors with the initial developmental feature vector using an attention mechanism to generate the enhanced developmental feature vector.

[0017] The enhanced developmental feature vectors of each developmental stage keyframe are processed by temporal developmental stage encoding and concatenation to generate the developmental stage feature vector.

[0018] Preferably, the step of performing dynamic concentration gradient extraction processing on the time-series monitoring data of the culture medium components to obtain a feature vector of culture medium changes includes:

[0019] The time-series monitoring data of the culture medium components were processed to detect the fluctuation range of key components, and abnormal fluctuation time windows in the dissolved oxygen, ion concentration and nutrient content data were identified.

[0020] Within the abnormal fluctuation time window, the original monitoring data is subjected to multi-scale sliding sampling to obtain multiple local concentration change segments;

[0021] The pre-trained fluid dynamics network model is invoked to perform local gradient feature extraction on each of the local concentration change segments, generating a local gradient feature vector;

[0022] Global temporal correlation aggregation is performed on the local gradient feature vectors of each of the local concentration change segments to generate the culture medium change feature vector.

[0023] Preferably, the step of performing cross-modal correlation and fusion processing on the developmental stage feature vector and the culture medium change feature vector to generate a fused developmental feature set includes:

[0024] The developmental stage feature vector is aligned with the developmental time dimension and mapped to the same monitoring time granularity as the culture medium change feature vector.

[0025] A development-culture medium cross-attention mechanism is constructed, and the cross-modal correlation matrix between the developmental features at each time point in the developmental stage feature vector and the concentration change features at the corresponding time point in the culture medium change feature vector is calculated.

[0026] Based on the cross-modal correlation matrix, the developmental stage feature vector and the culture medium change feature vector are subjected to bidirectional feature interaction processing to generate interactive developmental feature vector and interactive concentration feature vector;

[0027] The interaction development feature vector and the interaction concentration feature vector are subjected to gated fusion processing to generate the fused development feature set.

[0028] Preferably, the step of calling the pre-trained embryonic development assessment network model to perform developmental abnormality weight allocation processing on the fusion developmental feature set, generating a developmental risk score set corresponding to the batch of fertilized eggs, includes:

[0029] The set of fused developmental features is input into the feature filtering layer of the embryonic development assessment network model for redundant feature filtering to obtain the set of key developmental features after filtering.

[0030] The cross-feature generation layer of the embryonic development assessment network model is invoked to perform high-order feature combination processing on the key developmental feature set to generate a cross-developmental feature matrix.

[0031] The developmental risk sensitivity of the cross-developmental feature matrix is ​​calculated by the weight allocation layer of the embryonic development assessment network model, generating the risk sensitivity distribution of each developmental feature dimension.

[0032] Based on the risk sensitivity distribution, the cross-development feature matrix is ​​subjected to feature weighting and aggregation to generate the development risk score set.

[0033] Preferably, the step of jointly regulating and analyzing the environmental factor fluctuation data and the time-series monitoring data of culture medium components based on the abnormal correlation distribution to generate an optimization strategy for the culture parameters of the fertilized egg batch includes:

[0034] Based on the abnormal correlation distribution, abnormal correlation items exceeding the preset risk threshold are selected to generate a candidate control feature set;

[0035] For each abnormal correlation item in the candidate regulatory feature set, reverse parameter parsing is performed to determine its corresponding environmental factor deviation description and culture medium abnormal concentration description.

[0036] The pre-culture regulation model is invoked to perform joint optimization reasoning on the descriptions of environmental factor deviations and abnormal concentrations in the culture medium, generating a set of regulation paths.

[0037] The set of regulatory pathways is evaluated for effectiveness, and the target regulatory pathway with the highest effectiveness is selected as the optimization strategy for the cultivation parameters.

[0038] Preferably, the step of calling the pre-culture regulation model to perform joint optimization reasoning on the description of the environmental factor deviation and the description of the abnormal concentration of the culture medium to generate a set of regulation paths includes:

[0039] The environmental factor deviation description is input into the environmental parameter encoder of the pre-cultured regulation model for deviation type encoding processing to generate an environmental deviation feature vector.

[0040] The abnormal concentration description of the culture medium is input into the abnormal encoder of the culture medium in the pre-culture control model for concentration fluctuation pattern encoding processing to generate a concentration fluctuation feature vector.

[0041] A regulatory causal network of environment and culture medium is constructed, and the environmental deviation feature vector and concentration fluctuation feature vector are used as node features to input the regulatory causal network for multi-level regulatory inference processing;

[0042] The path generation layer of the regulatory causal network outputs the regulatory correlation path between the environmental deviation feature vector and the concentration fluctuation feature vector, thereby generating the regulatory path set.

[0043] Preferably, the step of generating dynamic control instructions for the culture environment based on the culture parameter optimization strategy includes:

[0044] Analysis of the key developmental node environmental compensation parameters and culture medium composition adjustment scheme in the culture parameter optimization strategy;

[0045] Match the set of historical regulatory records associated with the key developmental nodes from the historical culture database;

[0046] Based on the culture medium composition adjustment scheme, the set of historical regulation records is screened for effectiveness to obtain a set of effective regulation strategies.

[0047] The set of effective control strategies is optimized using multiple parameters to generate dynamic control instructions that meet the current embryonic development constraints.

[0048] Preferably, the step of performing multi-parameter optimization on the set of effective regulatory strategies to generate dynamic regulatory instructions that meet the current embryonic development constraints includes:

[0049] Construct a three-dimensional optimization target space that includes dissolved oxygen saturation, ion balance, and nutrient consumption rate;

[0050] Each effective control strategy is mapped to a coordinate point in the three-dimensional optimization target space;

[0051] A set of candidate control strategies for the Pareto front was selected using a non-dominated sorting algorithm.

[0052] The real-time fitness assessment model is invoked to dynamically score the candidate control strategy set based on the current operating status of the culture equipment.

[0053] The dynamic control instruction is generated by selecting the candidate control strategy with the highest dynamic score.

[0054] Preferably, after generating the dynamic control instruction by selecting the candidate control strategy with the highest dynamic score, the method further includes a precision calibration step for the dynamic control instruction:

[0055] The set of candidate regulation strategies is input into the accuracy calibration module, and the real-time environmental sensor data stream in the culture container is called to perform physical constraint verification processing on each candidate regulation strategy.

[0056] For candidate control strategies that fail the physical constraint verification, parameter boundary correction processing is performed to generate a set of correction strategies;

[0057] The matching degree between the regulatory strategies in the set of correction strategies and the key requirements of the current embryonic development stage is calculated by using a dynamic weight allocation algorithm.

[0058] A priority ranking list is generated based on the matching degree calculation results, and the strategy at the top of the list is selected as the final execution scheme for the dynamic control command.

[0059] Compared with the prior art, the beneficial effects of the present invention are:

[0060] By integrating multi-dimensional culture data, comprehensive perception and dynamic control of the in vitro culture process of fertilized lobster eggs were achieved. The acquired image sequences of fertilized egg development stages, time-series monitoring data of culture medium components, and environmental factor fluctuation data provide a rich information foundation for analyzing the embryonic development status, breaking through the limitations of traditional culture methods that rely on single data or experience-based judgments.

[0061] By analyzing cell division patterns in image sequences of fertilized eggs at different developmental stages and enhancing features using a crustacean embryonic development knowledge base, morphological characteristics at different developmental stages can be accurately captured, reflecting subtle changes in embryonic development. This approach not only improves the accuracy of developmental feature extraction but also deeply integrates image information with embryological expertise, making the feature vectors more aligned with biological principles and providing a reliable basis for subsequent developmental assessment.

[0062] By extracting dynamic concentration gradients from time-series monitoring data of culture medium components and identifying anomalous fluctuation time windows through multi-scale analysis, local variations in dissolved oxygen, ion concentration, and nutrient content can be accurately captured. Combined with gradient features extracted using a fluid dynamics model, this approach reflects the dynamic process of changes in culture medium components, rather than static concentration values, providing a deeper perspective for understanding the relationship between culture medium and embryonic development.

[0063] Cross-modal correlation fusion processing organically combines developmental stage characteristics with culture medium change characteristics, establishing a dynamic correlation between the two through temporal alignment and cross-attention mechanisms. This fusion approach breaks down the barriers between different types of data, revealing the intrinsic connection between developmental state and environmental factors, and enabling the generated fused developmental feature set to more comprehensively reflect the overall state of the culture system.

[0064] The pre-trained embryonic development assessment network model, by processing the set of fusion developmental features, can quantify the abnormal correlation between different environmental factors and key nodes in the developmental stage, and identify risk factors affecting embryonic development. The resulting developmental risk score set provides clear targets for subsequent regulatory analysis, avoiding the blind application of regulatory measures.

[0065] The combined regulation and analysis approach integrates environmental factor fluctuation data and culture medium composition data into a unified analytical framework. Through inverse parameter analysis and causal network reasoning, the generated culture parameter optimization strategy can take into account the interactions of multiple factors, and the proposed environmental compensation parameters and culture medium composition adjustment schemes are more targeted. This multi-parameter synergistic regulation method can effectively cope with complex dynamic changes in the culture system and reduce the chain reaction problems caused by adjusting a single parameter.

[0066] The dynamic control command generation and feedback mechanism, optimized by combining historical control records and real-time equipment status, ensures the feasibility and adaptability of control measures. The precision calibration step further verifies the physical constraint compatibility of the control strategy, ensuring that the final execution plan meets both the needs of embryonic development and the actual operational capabilities of the culture equipment. Attached Figure Description

[0067] Figure 1 This is a schematic diagram illustrating the working principle of the in vitro culture method for fertilized lobster eggs described in this invention.

[0068] Figure 2 A flowchart for image sequence parsing and processing during the developmental stages of a fertilized egg;

[0069] Figure 3 A flowchart for cross-modal correlation fusion processing;

[0070] Figure 4 A flowchart generated for joint regulation analysis and processing and cultivation parameter optimization strategies;

[0071] Figure 5 A flowchart for generating instructions for dynamic environmental control. Detailed Implementation

[0072] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0073] Please see Figure 1 The present invention provides a method for in vitro culture of fertilized eggs of the spiny lobster, the method comprising:

[0074] A multi-dimensional culture data set of the target fertilized egg batch was obtained, which included image sequences of fertilized egg development stages, time-series monitoring data of culture medium components, and environmental factor fluctuation data.

[0075] Cell division pattern parsing is performed on image sequences of fertilized egg development stages to generate developmental stage feature vectors; dynamic concentration gradient extraction is performed on time-series monitoring data of culture medium components to generate culture medium change feature vectors. The developmental stage feature vectors and culture medium change feature vectors are then fused across modalities to output a fused developmental feature set.

[0076] The pre-trained embryonic development assessment network model is invoked to perform developmental abnormality weight allocation processing on the fusion development feature set, generating a developmental risk score set, which includes the distribution of abnormal correlation between environmental factor fluctuation data and key nodes of each developmental stage.

[0077] Based on the abnormal correlation distribution, joint regulation and analysis are performed on environmental factor fluctuation data and culture medium composition time-series monitoring data to output culture parameter optimization strategy. This strategy defines the environmental compensation parameters and culture medium composition adjustment schemes for key developmental nodes.

[0078] Based on the optimization strategy for culture parameters, dynamic control instructions for the culture environment are generated and fed back to the culture equipment control system to trigger real-time parameter calibration.

[0079] Example 1: See Figure 2 This embodiment performs the parsing of image sequences from the fertilized egg developmental stages and the feature extraction of time-series monitoring data on culture medium components. In the image processing workflow, the image sequences from the fertilized egg developmental stages are segmented into developmental cycle stages. This segmentation is based on embryonic morphology standards, automatically identifying morphological transition critical points in the time-series images according to the cell arrangement characteristics of key developmental events such as gastrulation and mesoderm differentiation. The segmentation output includes keyframe image sequences from multiple developmental stages, including the blastocyst stage, gastrulation stage, and nauplius stage, with each keyframe marked with a timestamp accurate to ±0.5 hours.

[0080] The pre-trained cell structure segmentation model is used to process keyframe images. This model employs a deep convolutional neural network architecture. Its encoder consists of a ResNet-50 backbone network, extracting global spatial features through a 5-level downsampling layer. The decoder uses a U-Net skip connection structure to recover local details of cell boundaries and yolk granule distribution. After performing a 512×512 pixel normalized crop on the input image, the model outputs a fusion result of the segmentation mask and the original image, generating a 128-dimensional initial developmental feature vector containing cell division direction, number of blastomeres, and yolk absorption rate.

[0081] During embryological feature enhancement, relevant morphological entity sets are retrieved from a pre-constructed crustacean embryonic development knowledge base based on the timestamps and developmental stage labels corresponding to keyframes. Entities include the spatial coordinate set of appendage primordia, topological parameters of neural cord rudiments, geometric dimensions of the midgut cecum, and gradient data of yolk sac volume changes. The entity sets are input into the additional coding layer of the cell structure segmentation model, generating a 256-dimensional entity feature vector through three fully connected transformations. Subsequently, a multi-head attention mechanism is used to calculate the correlation matrix between the initial developmental feature vector and the entity feature vector: feature relevance scores are calculated using eight parallel attention heads, and the scores are standardized to generate feature fusion weights. Based on these weights, the two feature vectors are linearly combined to output a 256-dimensional enhanced developmental feature vector incorporating prior embryological knowledge.

[0082] The enhanced developmental feature vectors of each keyframe are arranged chronologically to form a temporal feature matrix, which is then temporally encoded using a bidirectional long short-term memory (LSTM) network. This network consists of a two-layer 64-unit LSTM structure: the forward layer captures the historical dependencies in embryonic morphological evolution, and the backward layer predicts future developmental trends. The network outputs a 512-dimensional developmental stage feature vector, concatenated from the temporal and feature dimensions, which comprehensively represents the developmental trajectory of a batch of embryos.

[0083] In the culture medium data processing workflow, key component fluctuation range detection was performed on time-series monitoring data such as dissolved oxygen, calcium ion concentration, and ammonium ion concentration. A sliding window-based statistical analysis method was adopted: a detection window of 10 minutes was defined, and the mean and standard deviation of each component data within each window were calculated; sampling points that continuously exceeded the standard deviation of the window were marked as abnormal fluctuation time windows, with three times the standard deviation of the window as the threshold. The start and end timestamps and the maximum fluctuation amplitude were recorded for each window.

[0084] Multi-scale sliding sampling is performed within the identified anomaly windows. Three time scales are set: 5 minutes, 30 minutes, and 120 minutes. The sampling range is expanded outwards from the center of the time window. Data slices are performed at each scale using a fixed step size: 30 seconds for the 5-minute scale, 5 minutes for the 30-minute scale, and 15 minutes for the 120-minute scale. Finally, each window outputs a set of local concentration change segments corresponding to the three scales.

[0085] A pre-trained hydrodynamic network model is used to process local concentration change segments. The model input dimension is adapted to the length of segments at multiple scales, and the core structure includes three sets of temporal convolutional modules and gated recurrent unit modules. The temporal convolutional modules use a dilated convolution structure with dilation coefficients of 1 / 3 / 5 to extract short-term concentration abrupt changes; the gated recurrent unit modules capture long-period diffusion effects and record the directional patterns of ion transport. The module outputs are aggregated by a feature fusion layer to generate a 128-dimensional local gradient feature vector describing the direction of local gradient changes.

[0086] Global temporal correlation aggregation is performed on the local gradient feature vectors generated by all windows. A graph neural network structure is constructed, with each local feature vector as a node, and edge connections are built based on the temporal overlap relationship between segments. Node information is passed through a three-layer graph convolution operation, and the representation weight of each node for the target time period is calculated. Finally, all feature vectors are summed according to their weights to generate a 256-dimensional culture medium change feature vector describing the abnormal evolution of culture medium concentration.

[0087] The entire implementation process completes the transformation from raw image sequences and monitoring data to high-dimensional feature vectors. The image processing subsystem extracts the dynamic changes in embryo morphology over time, while the culture medium analysis subsystem characterizes the fluctuations in chemical composition. The spatiotemporal parameters of the two types of feature vectors are strictly aligned, providing a matching feature structure basis for subsequent cross-modal analysis.

[0088] Example 2: See Figure 3 This embodiment performs cross-modal fusion of developmental features and culture medium features, as well as developmental risk assessment. During the feature alignment phase, the 512-dimensional developmental stage feature vectors are resampled to the same temporal resolution as the 256-dimensional culture medium change feature vectors using cubic spline interpolation, ensuring a one-to-one correspondence of feature vector pairs at each time point. The interpolation process strictly follows the biological intervals of the embryonic development timeline, preserving the synchronicity between developmental events and culture medium monitoring.

[0089] Constructing a developmental-culture medium cross-attention mechanism:

[0090] A. Using the feature components of the developmental stage feature vector at each time node as the query vector Q. dev ;

[0091] B. Using the characteristic components of the culture medium change characteristic vector at each time node as the key vector K cul Sum vector V cul ;

[0092] C. Calculate the cross-modal correlation matrix:

[0093] Where δ represents the hidden layer dimension of the feature vector (set to 64), and the soft maximum function normalizes along the row direction;

[0094] D. Multiply the incidence matrix Φ with the value vector:

[0095] Ψ cul =Φ·V cul

[0096] The bidirectional feature interaction processing is performed sequentially: the developmental feature vector is placed into input port A of the interaction layer, and the correlation matrix Φ is placed into port B. Interactive developmental feature generation is then performed: the features from port A and the matrix from port B are subjected to a Hadamard product, outputting a 512-dimensional interactive developmental feature V′. dev The feature vector of the culture medium change is placed into port C, and tensor shrinking is performed with port B to output a 256-dimensional interactive concentration feature V′. cul .

[0097] The gated fusion mechanism consists of a three-layer fully connected network: the first layer is a 128-unit linear transform layer, the second layer is a ReLU activation layer, and the third layer is a single-unit output layer with Sigmoid activation. The input data is the interactive feature V′. dev With V′ cul The vertically concatenated vectors output g ∈ [0,1]. The fusion operation is defined as:

[0098] V fusion =g·V′ dev +(1-g)·V′ cul

[0099] The final output is a 768-dimensional fusion developmental feature set. The feature dimensions are divided into 16 sub-vectors according to the key nodes of the developmental stage, and each sub-vector contains 48 features.

[0100] The feature selection layer of the embryonic development assessment network model employs two parallel structures: the first is a temporal convolutional module with a 1×1 kernel, using a 32-channel convolutional layer to extract cross-temporal feature correlations; the second is a gated linear unit, using a 62-dimensional weight vector to filter redundant features with importance scores below 0.65. The two outputs are merged in the concatenation layer into a 512-dimensional set of key developmental features.

[0101] The cross-feature generation layer performs high-order combination processing: it divides the key developmental feature set into 8 feature blocks along the time dimension, each feature block containing 64-dimensional features. Tensor outer product operations are performed on adjacent feature blocks: the feature vector F of the i-th block is... i With the feature vector F of the (i+1)th block i+1 The Kronecker product is performed, outputting a 4096-dimensional second-order feature combination matrix. This layer generates a total of 7 sets of feature matrices, which are then reduced to a 256×256-dimensional cross-development feature matrix after max pooling.

[0102] The weight allocation layer comprises a feature sensitivity analysis network and a weight aggregation unit: the feature sensitivity analysis network employs a fully connected architecture with 128 hidden layers, taking as input the column vectors of the cross-development feature matrix and outputting the anomaly contribution score for each feature. The contribution scores are then converted into a risk sensitivity distribution using an exponential normalization function.

[0103]

[0104] Where c i The feature contribution score is represented by λ, which is a temperature coefficient (value 2.5). The weighted aggregation unit's receiver sensitivity distribution S i The feature matrix is ​​subjected to a row-weighted summation operation, outputting a set containing risk scores for 24 developmental nodes. This score set is stored as a two-dimensional tensor structure, with the first dimension representing the developmental stage index and the second dimension recording the correlation score of the corresponding environmental factor. Each score is standardized to the [0,1] interval, and a score threshold of 0.85 is set to indicate a high-risk state.

[0105] The entire processing flow transforms from multimodal feature fusion to developmental risk quantification. Feature alignment maintains consistency over time, a cross-attention model captures the implicit correlation between developmental state and culture environment, and a gating mechanism dynamically adjusts the fusion ratio of cross-modal information. The evaluation network reduces data noise interference through feature selection, high-order feature combinations reveal the potential coupling effects of developmental abnormalities, and sensitivity distribution objectively reflects the differences in response to environmental fluctuations at different embryonic developmental stages.

[0106] Example 3: See Figure 4 This embodiment executes the process of generating a culture parameter optimization strategy. Based on the developmental risk score set output by the embryonic development assessment model, this set records the correlation distribution between key nodes at each developmental stage and environmental factors. A preset mechanism scans all correlation data items in the score set, filtering out abnormal correlation items with values ​​exceeding a threshold of 0.85 to form a preliminary set. A topological sorting operation is performed on these abnormal items, arranging them hierarchically according to chronological order and their dependence on embryonic development stages, generating a structured candidate regulatory feature set.

[0107] Inverse parameter parsing is performed on each abnormal correlation item within the candidate regulatory feature set. This operation connects the environmental monitoring database and the culture medium record library to establish a dynamic index of environmental factor fluctuation data: raw sensor readings during the time period of the associated event are located by timestamp, and minute-level sampling sequences of temperature, salinity, and light intensity are extracted. The raw sequences are processed using a variational mode decomposition algorithm to separate the sudden fluctuation component caused by equipment failure, the periodic oscillation component generated by the environmental control system, and the environmental noise floor component. Fluctuation feature descriptors are constructed for each component: the sudden fluctuation component records the amplitude and duration; the periodic oscillation component records the frequency and phase shift; and the noise floor component records the statistical variance. The feature descriptors of the three components are combined to form a quantitative description of the environmental factor deviation.

[0108] Synchronous analysis of abnormal culture medium conditions: Within the same time window, raw data streams from the dissolved oxygen sensor, ion-selective electrode, and spectroscopic nutrient analyzer are read. Abnormal pattern matching is performed for each culture medium component: 32 typical abnormal waveform templates pre-stored in the database are invoked, including step-like sudden changes, exponential decay, and sinusoidal oscillation patterns. The similarity between real-time data and templates is calculated using a dynamic time warping algorithm, and templates with a similarity higher than 0.75 are selected as matching results. Combining the parameterized features of the matching templates with measured concentration values, a dynamic pattern description matrix of abnormal concentrations in the culture medium is output.

[0109] The pre-trained control model is invoked to process the quantitative description of environmental deviations and the description matrix of abnormal culture medium patterns. The environmental parameter encoder contains a parallel dual-channel network structure: the first channel uses a temporal convolutional network to process abrupt fluctuation components, extracting multi-scale mutation features through four layers of dilated convolutions; the second channel uses an autoregressive network to model periodic oscillation components, generating frequency response features. The dual-channel outputs are concatenated into a 128-dimensional intermediate vector at a fusion layer, and then compressed through a fully connected layer to generate a 32-dimensional environmental deviation feature vector. The culture medium anomaly encoder constructs a three-dimensional tensor input structure: the first dimension represents different component types, the second dimension records the time series, and the third dimension stores pattern parameters. A graph convolutional network is used to model the chemical correlations between components, and after updating node features through a message passing mechanism, a 32-dimensional concentration fluctuation feature vector is output.

[0110] A causal network for environment-culture medium regulation is constructed. This network contains two types of entities: environmental regulation actuators (temperature compensation units, light regulation modules, etc.) and culture medium regulation actuators (ion implantation pumps, aeration controllers, etc.). Edge weights are calculated statistically based on a historical regulation case database: when the frequency of response from culture medium regulator B after an action by environmental regulator A exceeds 75%, a causal edge from A to B is established, with weights representing conditional probability values. Environmental deviation feature vectors and concentration fluctuation feature vectors are mapped to corresponding node types. Multi-level regulation inference is performed through three-layer graph convolution propagation: the first layer calculates direct causal effects, the second layer evaluates indirect cascade effects, and the third layer aggregates global regulatory influences. The path generation layer traverses the network based on an improved heuristic search algorithm, retaining the top k paths with the highest sum of node activations from the starting point to the ending point, generating a set of paths containing the regulation action sequence.

[0111] Effectiveness evaluation of the set of control paths: A virtual culture environment was constructed in the digital twin system, loading the fusion development feature set of the current batch of fertilized eggs as the initial state. Simulated control operations were executed according to the parameter settings of each path, and the output in three dimensions was recorded: embryonic morphological development synchronization rate (based on image feature similarity), environmental parameter stability coefficient (based on the inverse of the variance of fluctuation), and resource consumption index (based on the energy and material costs of the control actions). A weighted aggregation method was used to calculate the comprehensive score of the paths: embryonic morphology weight 0.6, environmental stability weight 0.3, and resource consumption weight 0.1. The path with the highest comprehensive score was selected as the final culture parameter optimization strategy. This strategy clearly specifies the compensation parameter values ​​and component adjustment ranges required for specific developmental stage nodes. The entire implementation process achieved a closed-loop transformation from risk quantification to control strategy, and the strategy content is compatible with the instruction parsing format of the equipment control system.

[0112] Example 4: See Figure 5 This embodiment executes the process of generating control paths and converting dynamic instructions. Taking the delayed gastrulation stage during the culture of fertilized eggs of Ormosia lobster as an example: when the embryonic development assessment model detects a risk score of 0.92 for the gastrulation stage (threshold 0.85), the pre-culture control model initiates the environmental parameter encoding process. Environmental monitoring data records show that abnormal temperature fluctuations occurred in the time interval [11:30-12:15]: the measured value reached a peak of 31℃ at 11:42 under the set value of 28℃. This deviation consists of two abnormal components: a step-like jump caused by equipment failure (a rise of 2.8℃ at 11:35) and a decaying wave generated by the oscillation of the refrigeration system (period 45 minutes / amplitude ±0.5℃). The environmental parameter encoder maps the step component to a feature vector [2.8,0,1] (amplitude / frequency / type identifier), and encodes the oscillation component as [0.5,45,2]. After fusion by a fully connected layer, a 32-dimensional environmental deviation feature vector is output.

[0113] Synchronous treatment of culture medium anomalies: Within the same time window, the dissolved oxygen sensor showed a sharp drop in concentration from 6.2 mg / L to 4.9 mg / L, while the ammonia nitrogen concentration rose to 0.25 ppm. The anomaly encoder identified this composite pattern as "oxygen-nitrogen inverse oscillation type," and the characteristic matrix is ​​shown in the following example:

[0114]

[0115] The culture medium anomaly encoder outputs a 32-dimensional concentration fluctuation feature vector, which includes mode type weights, change gradients, and cross-influence coefficients.

[0116] Constructing a causal network for regulation: The node library pre-stores 12 types of execution units, including:

[0117] Temperature compensation unit (T1 refrigeration valve / T2 heating rod)

[0118] Dissolved oxygen regulator (O1 aeration pump / O2 nitrogen replacement)

[0119] Ion balance module (N1 ammonium adsorbent / N2 nitrifying bacteria injection)

[0120] Edge weights are established based on the historical regulation knowledge graph: when T1 is activated, the probability of triggering O1 is 92%, and the probability of triggering N1 is 67%; the probability of O2 activation leading to an N2 response is 81%. After inputting the environmental deviation feature vector, graph convolutional inference shows:

[0121] First layer: T1 unit activation level 0.93 (stepwise jump in cooling response)

[0122] Second layer: T1→O1 edge weight 0.92 triggers aeration compensation (dissolved oxygen recovery).

[0123] Third layer: O1→N1 edge weight 0.57 weakly triggers ammonium adsorption intervention

[0124] Three regulatory paths are generated:

[0125] Path A: T1 (power 80%) → O1 (flow rate increased by 40%) → N1 (adsorbent 0.5 g / L) Path B: T1 (power 75%) + O2 (nitrogen injection 2 L / min) → N2 (bacterial agent 0.3 mL / L) Path C: T1 (power 85%) → O1 (flow rate 35%) + N1 (adsorbent 0.4 g / L) During the dynamic control instruction generation stage, 15 groups of gastrulation control records were matched from the historical database. The screening criteria were:

[0126] Dissolved oxygen adjustment range to match the current protocol (within ±40%).

[0127] Consistency of ammonia nitrogen treatment methods > 90%

[0128] Five effective strategies were retained after screening:

[0129] History ID Temperature compensation Dissolved oxygen regulation ammonia nitrogen treatment HG-20240703 Cooling power 82% Aeration +35% Adsorbent 0.48 g / L HG-20240715 Cooling power 78% Nitrogen purging rate: 1.8 L / min 0.32 mL / L of bacterial agent HG-20240628 Cooling power 85% Aeration +32% Adsorbent 0.52 g / L

[0130] Parameter optimization is achieved through three-dimensional target space mapping:

[0131] Dissolved oxygen saturation dimension: Path A at target space coordinates (0.86, 0.79, 0.62); Ion balance dimension: Path B at coordinates (0.82, 0.85, 0.71).

[0132] Nutrient consumption dimension: Path C coordinates (0.88, 0.76, 0.68). The NSGA-II algorithm screened the Pareto front, retaining paths A and C. Real-time fitness assessment showed that path C had higher compatibility with the current load state of the culture tank (control system response delay < 3 seconds). Therefore, path C was used to generate the final control command: "Execute T1 cooling power 85% + O1 aeration flow rate 35% + N1 adsorbent 0.4g / L injection during the time interval [12:20-13:00]". This command was transmitted to the culture equipment via the control bus, triggering the actuator real-time calibration operation.

[0133] Example 5: This example demonstrates the optimization and precision calibration of dynamic control commands. A set of candidate control strategies in a three-dimensional optimization target space, derived from Pareto front screening results, is used. A real-time fitness assessment model is invoked to process the candidate strategies: the model input is connected to the culture equipment's operational status monitoring bus, continuously receiving real-time parameters such as the temperature control unit's response delay in milliseconds, the remaining calibration time for the dissolved oxygen probe, and the percentage of available capacity of the ion addition pump. The evaluation logic is categorized by equipment type: for temperature compensation parameters, the model calculates the matching degree between the setpoint and the actuator's maximum adjustment rate, outputting an adjustment feasibility score; for culture medium control parameters, the model retrieves the calibration curves of the solution tank's remaining volume and the added reagent concentration, calculating a material compatibility score. Each candidate strategy receives a comprehensive score consisting of three parts: actuator response matching degree (40% weight), resource availability (35% weight), and operational conflict avoidance coefficient (25% weight). The strategy with the highest score is selected to generate the initial dynamic control command, in the form of a structured parameter group.

[0134] The accuracy calibration module performs physical constraint verification on the initial commands: loading real-time data streams from multiple sensors within the culture vessel, including multi-point temperature probe arrays, dissolved oxygen electrode sampling sequences, and conductivity sensor network feedback data. The verification process is executed in three levels: the first level constrains and verifies the physical limits of the equipment; for example, if the temperature adjustment command exceeds the equipment's maximum gradient setting value, the parameter exceeding the limit is marked. The second level verifies the coupling relationship of environmental parameters; when there is a hardware conflict in the resources required for two control operations (such as simultaneous full-load operation of the cooling pump and heating rod), logical conflict items are marked. The third level verifies biological boundary conditions; if an ion concentration adjustment command is detected to exceed the osmotic pressure tolerance threshold of the spiny lobster embryo, a warning mark is generated. During the verification, each violation triggers an anomaly flag and records the original parameter coordinates.

[0135] For strategies that fail validation, parameter boundary correction is performed: a three-dimensional boundary hyperplane constraint model is established in the parameter space of the control variables. When the parameter set is identified as physically infeasible, an optimization equation set is constructed in the objective function space. The decision variables include the deviation of the parameters to be corrected from the physical limits of the equipment. The constraints are: the temperature adjustment rate does not exceed the gradient limit of ±5℃ / min; the single adjustment range of dissolved oxygen concentration is kept within the threshold of ±2mg / L; any ion concentration adjustment follows the isotonic rule. The correction algorithm adopts the iterative projection method: starting from the initial parameter point, it is gradually adjusted along the gradient direction of the feasible region until all parameters satisfy the hyperplane constraint conditions. In each iteration, the Euclidean distance between the parameter point and the corrected parameter is calculated as the loss function value. When the change of the loss function is less than 0.01 for three consecutive iterations, the final corrected strategy set is output.

[0136] A dynamic weighting algorithm matches the modified strategy with embryo requirements: a feature vector of key requirements for the current batch of fertilized eggs is established, obtained through analysis of the embryo development feature set. The requirement dimensions include: blastocyst cleavage synchronization acceleration coefficient (0-1 scale), expected gradient value of gastrulation rate (unit: micrometer / hour), and percentage of nauplius appendage differentiation meeting the target. The matching process calculates the cosine of the multidimensional angle between the changes in environmental parameters generated by the modified strategy and the requirement vector, and weights them according to the biological priority of each dimension: blastocyst parameters weighted at 0.7, gastrulation option weighted at 0.9, and nauplius option weighted at 0.8. Each strategy yields a final matching score output value.

[0137] Priority ranking is implemented in the feature space: a three-dimensional demand-regulation correlation mapping coordinate system is constructed, with the first axis representing the morphological development achievement rate, the second axis representing physiological metabolic stability, and the third axis representing developmental temporal synchronicity. The matching scores of each correction strategy are converted into spatial coordinate values, and a density clustering algorithm is used to identify coordinate clustering regions. For each clustering region, a representative strategy is selected at its geometric center, and ranked according to its spatial distance from the ideal demand point (coordinate origin). When multiple strategies are at the same distance level, their resource consumption entropy values ​​(based on the product of equipment energy consumption and material consumption of the regulation action) are additionally compared. The ranking results generate a command priority sequence, and the strategy at the top of the sequence is selected as the final dynamic regulation command scheme. This scheme is converted into an equipment control protocol format, such as: setting the refrigeration unit to linearly adjust from the initial value V1 to the target value V2 at a slope S within the time interval T; instructing the aeration controller to start the flow rate Q at time interval T+Δt and continue for D minutes; synchronously triggering the adsorbent addition module to inject nano-adsorbent particles of mass M within the time window T'. The commands are transmitted to the execution unit of the cultivation equipment via an industrial bus, and the execution log is returned to the system to form a closed-loop regulation system.

[0138] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus.

[0139] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations 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 in vitro culture of fertilized eggs of the spiny lobster, characterized in that, The method includes: A multi-dimensional culture data set of the target fertilized egg batch is obtained, which includes image sequences of fertilized egg development stages, time-series monitoring data of culture medium components, and environmental factor fluctuation data. The cell division pattern parsing processing is performed on the fertilized egg development stage image sequence to obtain the development stage feature vector. The dynamic concentration gradient extraction processing is performed on the culture medium component time-series monitoring data to obtain the culture medium change feature vector. The development stage feature vector and the culture medium change feature vector are subjected to cross-modal correlation fusion processing to generate a fused development feature set. The pre-trained embryonic development assessment network model is invoked to perform developmental abnormality weight allocation processing on the fusion development feature set, generating a developmental risk score set corresponding to the batch of fertilized eggs, wherein the developmental risk score set includes the distribution of abnormal correlation between environmental factor fluctuation data and key nodes of each developmental stage; Based on the abnormal correlation distribution, the environmental factor fluctuation data and the time-series monitoring data of culture medium composition are jointly regulated and analyzed to generate the culture parameter optimization strategy for the batch of fertilized eggs. The culture parameter optimization strategy is used to indicate the environmental compensation parameters and culture medium composition adjustment schemes for key developmental nodes. Based on the culture parameter optimization strategy, dynamic control instructions for the culture environment are generated, and these instructions are fed back to the culture equipment control system to trigger real-time parameter calibration.

2. The method for in vitro culture of fertilized lobster eggs according to claim 1, characterized in that, The cell division pattern parsing process is performed on the image sequence of the fertilized egg development stage to obtain the development stage feature vector, including: The fertilized egg development stage image sequence is segmented into developmental cycle stages to obtain multiple keyframe images of developmental stages; The pre-trained cell structure segmentation model is invoked to encode cell division features in each keyframe image of the developmental stage, generating an initial developmental feature vector. The initial developmental feature vector is subjected to embryological feature enhancement processing to obtain an enhanced developmental feature vector. The embryological feature enhancement processing includes the following steps: matching a set of morphological entities associated with the keyframe of the current developmental stage from a preset crustacean embryonic development knowledge base; inputting the set of morphological entities into the cell structure segmentation model for entity feature encoding processing to generate a set of entity feature vectors; and performing weighted fusion of the set of entity feature vectors with the initial developmental feature vector using an attention mechanism to generate the enhanced developmental feature vector. The enhanced developmental feature vectors of each developmental stage keyframe are processed by temporal developmental stage encoding and concatenation to generate the developmental stage feature vector.

3. The method for in vitro culture of fertilized lobster eggs according to claim 1, characterized in that, The dynamic concentration gradient extraction process is performed on the time-series monitoring data of the culture medium components to obtain a feature vector of culture medium changes, including: The time-series monitoring data of the culture medium components were processed to detect the fluctuation range of key components, and abnormal fluctuation time windows in the dissolved oxygen, ion concentration and nutrient content data were identified. Within the abnormal fluctuation time window, the original monitoring data is subjected to multi-scale sliding sampling to obtain multiple local concentration change segments; The pre-trained fluid dynamics network model is invoked to perform local gradient feature extraction on each of the local concentration change segments, generating a local gradient feature vector; Global temporal correlation aggregation is performed on the local gradient feature vectors of each of the local concentration change segments to generate the culture medium change feature vector.

4. The method for in vitro culture of fertilized lobster eggs according to claim 1, characterized in that, The step of performing cross-modal correlation and fusion processing on the developmental stage feature vector and the culture medium change feature vector to generate a fused developmental feature set includes: The developmental stage feature vector is aligned with the developmental time dimension and mapped to the same monitoring time granularity as the culture medium change feature vector. A development-culture medium cross-attention mechanism is constructed, and the cross-modal correlation matrix between the developmental features at each time point in the developmental stage feature vector and the concentration change features at the corresponding time point in the culture medium change feature vector is calculated. Based on the cross-modal correlation matrix, the developmental stage feature vector and the culture medium change feature vector are subjected to bidirectional feature interaction processing to generate interactive developmental feature vector and interactive concentration feature vector; The interaction development feature vector and the interaction concentration feature vector are subjected to gated fusion processing to generate the fused development feature set.

5. The method for in vitro culture of fertilized lobster eggs according to claim 1, characterized in that, The pre-trained embryonic development assessment network model is invoked to perform developmental abnormality weight allocation processing on the fusion developmental feature set, generating a developmental risk score set corresponding to the batch of fertilized eggs, including: The set of fused developmental features is input into the feature filtering layer of the embryonic development assessment network model for redundant feature filtering to obtain the set of key developmental features after filtering. The cross-feature generation layer of the embryonic development assessment network model is invoked to perform high-order feature combination processing on the key developmental feature set to generate a cross-developmental feature matrix. The developmental risk sensitivity of the cross-developmental feature matrix is ​​calculated by the weight allocation layer of the embryonic development assessment network model, generating the risk sensitivity distribution of each developmental feature dimension. Based on the risk sensitivity distribution, the cross-development feature matrix is ​​subjected to feature weighting and aggregation to generate the development risk score set.

6. The method for in vitro culture of fertilized lobster eggs according to claim 1, characterized in that, The method of jointly regulating and analyzing the environmental factor fluctuation data and the time-series monitoring data of culture medium components based on the abnormal correlation distribution to generate the culture parameter optimization strategy for the batch of fertilized eggs includes: Based on the abnormal correlation distribution, abnormal correlation items exceeding the preset risk threshold are selected to generate a candidate control feature set; For each abnormal correlation item in the candidate regulatory feature set, reverse parameter parsing is performed to determine its corresponding environmental factor deviation description and culture medium abnormal concentration description. The pre-culture regulation model is invoked to perform joint optimization reasoning on the descriptions of environmental factor deviations and abnormal concentrations in the culture medium, generating a set of regulation paths. The set of regulatory pathways is evaluated for effectiveness, and the target regulatory pathway with the highest effectiveness is selected as the optimization strategy for the cultivation parameters.

7. The method for in vitro culture of fertilized lobster eggs according to claim 6, characterized in that, The method of calling the pre-culture regulation model to perform joint optimization reasoning on the description of the environmental factor deviation and the description of the abnormal concentration of the culture medium generates a set of regulation paths, including: The environmental factor deviation description is input into the environmental parameter encoder of the pre-cultured regulation model for deviation type encoding processing to generate an environmental deviation feature vector. The abnormal concentration description of the culture medium is input into the abnormal encoder of the culture medium in the pre-culture control model for concentration fluctuation pattern encoding processing to generate a concentration fluctuation feature vector. A regulatory causal network of environment and culture medium is constructed, and the environmental deviation feature vector and concentration fluctuation feature vector are used as node features to input the regulatory causal network for multi-level regulatory inference processing; The path generation layer of the regulatory causal network outputs the regulatory correlation path between the environmental deviation feature vector and the concentration fluctuation feature vector, thereby generating the regulatory path set.

8. The method for in vitro culture of fertilized lobster eggs according to claim 1, characterized in that, The step of generating dynamic control instructions for the culture environment based on the culture parameter optimization strategy includes: Analysis of the key developmental node environmental compensation parameters and culture medium composition adjustment scheme in the culture parameter optimization strategy; Match the set of historical regulatory records associated with the key developmental nodes from the historical culture database; Based on the culture medium composition adjustment scheme, the set of historical regulation records is screened for effectiveness to obtain a set of effective regulation strategies. The set of effective control strategies is optimized using multiple parameters to generate dynamic control instructions that meet the current embryonic development constraints.

9. The method for in vitro culture of fertilized lobster eggs according to claim 8, characterized in that, The step of performing multi-parameter optimization on the set of effective regulatory strategies to generate dynamic regulatory instructions that meet the current embryonic development constraints includes: Construct a three-dimensional optimization target space that includes dissolved oxygen saturation, ion balance, and nutrient consumption rate; Each effective control strategy is mapped to a coordinate point in the three-dimensional optimization target space; A set of candidate control strategies for the Pareto front was selected using a non-dominated sorting algorithm. The real-time fitness assessment model is invoked to dynamically score the candidate control strategy set based on the current operating status of the culture equipment. The dynamic control instruction is generated by selecting the candidate control strategy with the highest dynamic score.

10. The method for in vitro culture of fertilized lobster eggs according to claim 9, characterized in that, After generating the dynamic control command by selecting the candidate control strategy with the highest dynamic score, the process further includes a precision calibration step for the dynamic control command: The set of candidate regulation strategies is input into the accuracy calibration module, and the real-time environmental sensor data stream in the culture container is called to perform physical constraint verification processing on each candidate regulation strategy. For candidate control strategies that fail the physical constraint verification, parameter boundary correction processing is performed to generate a set of correction strategies; The matching degree between the regulatory strategies in the set of correction strategies and the key demand characteristics of the current embryonic development stage is calculated by using a dynamic weight allocation algorithm. A priority ranking list is generated based on the matching degree calculation results, and the strategy at the top of the list is selected as the final execution scheme for the dynamic control command.

Images