A digital twin monitoring and regulation method for the domestication process of copepod bait organisms

Abstract

The present application provides a kind of digital twin monitoring and control method for domestication process of copepods bait organism, the method comprises: establishing micro-ecological, population dynamics coupled twin model, define monitoring parameter and control object uniform terminology system;Multi-source sensing data is collected in edge, clock synchronization, signal denoising and state estimation are carried out, and the joint estimation result of micro-ecological, population and water body state is obtained;Parameter online identification is executed based on state estimation, and safety constraint set such as ammonia nitrogen, dissolved oxygen, mortality and environmental change rate is constructed;The balance of reproduction efficiency improvement and system stability is realized by multi-objective optimization function;According to the solution result, the micro-ecological pulse, environmental gradual change cooperative control action set is generated, and when the set condition is satisfied, the domestication prescription is output and the process is terminated.Through digital twin modeling, rolling prediction control and self-evolution updating mechanism, intelligent, low stress and high stability control of copepod domestication process is realized.

Description

Technical Field

[0001] This invention relates to the field of digital twin monitoring technology, and in particular to a digital twin monitoring and control method for the domestication process of copepod prey organisms. Background Technology

[0002] Copepods are an important source of aquatic food organisms. Their domestication process is highly sensitive to environmental parameters such as salinity, temperature, light, and feed ratio. In addition, in the culture environment of copepods, the associated microbial community (including algae, bacteria and probiotics) plays an important role in maintaining the stability of the aquatic environment, promoting feed digestion and improving survival rate.

[0003] Existing research mainly focuses on adding probiotics or adjusting algal ratios, but these are still empirical operations and lack quantitative descriptions of the relationship between microbial community structure and copepod population dynamics. Especially during high-density cultivation and stress phases, the micro-ecosystem is highly susceptible to imbalance, leading to water quality fluctuations, ammonia nitrogen accumulation, and feeding disorders.

[0004] In recent years, digital twin and intelligent control technologies have been gradually applied to the field of aquaculture. However, most existing digital twin models only focus on the relationship between environmental parameters and individual growth parameters, failing to effectively incorporate microecological changes into the population dynamics model. Due to the complex feeding chains and competition relationships among copepods at different age levels, traditional single environmental control is difficult to capture ecological interaction effects, resulting in insufficient prediction accuracy of twin models and difficulty in promoting control strategies. Therefore, we propose a digital twin monitoring and regulation method for the domestication process of copepod prey organisms.

[0005] The information disclosed in the background section is only intended to enhance the understanding of the background of this disclosure, and therefore may include information that does not constitute prior art known to those skilled in the art. Summary of the Invention

[0006] The purpose of this invention is to address the shortcomings of existing technologies by providing a digital twin monitoring and control method for the domestication process of copepod prey organisms, thereby solving the technical problems mentioned in the background section.

[0007] To achieve the above objectives, the present invention provides the following technical solution:

[0008] A digital twin monitoring and control method for the domestication process of copepod prey organisms includes the following steps:

[0009] S1. Establish a coupled twin model of microecology and population dynamics for copepod domestication, and define a unified terminology system for monitoring parameters and control objects; generate reproducible standardized modeling configurations by defining variable ranges, interface formats and initialization strategies.

[0010] S2. Continuously collect multi-source monitoring data at the edge and perform time synchronization, missing data compensation and signal denoising. Input the coupled twin model to obtain the state estimation results and uncertainty index of micro-ecological state, population state and water state, and provide basic input for subsequent identification and control.

[0011] S3. Use the state estimation results to identify and robustly update the key parameters of the model online to minimize the error between the model prediction and the actual measurement. At the same time, construct a set of safety thresholds and water quality constraints based on real-time and historical data to form a control feasible region and freeze it synchronously with the model parameters.

[0012] S4. Establish a rolling predictive control model, set the prediction time window and control step size; construct a multi-objective optimization function with the goals of improving reproductive efficiency, reducing feeding fluctuations, suppressing stress mortality peaks and maintaining algal community stability, and adaptively adjust weights in combination with state uncertainty to form a dynamically solvable optimization solution configuration.

[0013] S5. Based on the optimized solution configuration, a set of coordinated control actions for micro-ecological pulses and gradual environmental changes is generated and sent to the edge execution end for real-time implementation. During the execution process, multi-channel anomaly detection is performed. When the prediction deviation or monitoring anomaly exceeds the set threshold, the safety backoff control law is automatically executed to maintain system stability and domestication continuity.

[0014] S6. Use execution feedback and monitoring data to perform closed-loop updates on the coupled twin model, train parameters and verify the effectiveness of the updates; the system comprehensively evaluates the convergence based on reproductive stability, survival rate stability, ecological balance and energy consumption indicators. When the comprehensive score reaches the set conditions, the domestication prescription is output and the process is terminated; otherwise, it returns to the rolling prediction stage for strategy fine-tuning and cyclic control.

[0015] S1 includes: establishing a coupled twin model of microecology and population dynamics for copepod domestication, wherein monitoring parameters are divided into microecological monitoring parameters, environmental monitoring parameters, and population monitoring parameters, and the controlled objects are divided into microecological pulses and environmental gradual changes.

[0016] The model input, output, and parameterized interface are defined based on the coupling relationship between the monitoring parameters and the controlled object; the model, interface, and data dictionary are unified by setting the variable value range, data format, and communication protocol.

[0017] When there is no historical data, default empirical parameters are used. When historical data is available, an initial parameter set is generated through statistical algorithms, and a model version baseline is established for subsequent data collection, identification, and control.

[0018] S2 includes: establishing a data acquisition link corresponding to the monitoring parameters at the edge execution end, and continuously collecting micro-ecology, environment and group status data through multi-source sensors;

[0019] Time synchronization algorithms are used to perform clock calibration, missing data compensation, and outlier correction for different sampling sources, and filtering, standardization, and cross-source alignment are performed on the returned data.

[0020] The processed data is input into the coupled twin model to obtain state estimation results of micro-ecological state, community state and water state, and the state estimation results and their uncertainty information are output to provide input for subsequent model identification and optimization control.

[0021] S3 includes: using the state estimation results obtained in step S2 to identify and dynamically update the key parameters in the coupled twin model online;

[0022] Robust fitting is achieved by minimizing the deviation between model predictions and measured values ​​and introducing a regularization term; at the same time, a set of safety thresholds and water quality constraints is constructed based on real-time and historical monitoring data.

[0023] The constraint set includes ammonia nitrogen concentration, dissolved oxygen content, mortality rate increment, and environmental change rate limits. After the parameters are updated, the model parameters and constraint set are frozen simultaneously to form a control feasible domain for subsequent rolling predictive control.

[0024] S4 includes: establishing a rolling predictive control model based on the frozen set of model parameters and safety constraints, and setting the prediction time window and control step size to form a rolling solution cycle;

[0025] Construct a multi-objective optimization function with the goals of improving reproductive efficiency, reducing feeding fluctuations, suppressing stress mortality, and maintaining the stability of algal and bacterial communities;

[0026] By dynamically adjusting the target weights in conjunction with state uncertainty, an adaptive balance between efficiency and safety is achieved. Model constraints, actuator limitations, and nonlinear safety conditions after linearization are all injected into the optimization problem. Stable and real-time executable control commands are obtained through a two-level solution structure, providing configuration for subsequent collaborative control and edge execution.

[0027] S5 includes: generating a set of coordinated control actions for micro-ecological pulses and environmental gradual changes based on the optimization results of rolling predictive control, and sending the control actions to the edge execution end for periodic implementation through a standardized communication interface;

[0028] During execution, the system continuously collects monitoring parameters and performs multi-channel anomaly detection; when the prediction deviation or monitoring value exceeds the safety threshold, the safety backoff control law is automatically triggered.

[0029] The system maintains stability by reducing control amplitude, pausing execution, or switching to conservative mode; the execution end records control logs and monitoring feedback data in real time to form a closed-loop dataset for subsequent model updates.

[0030] S6 includes: after each control cycle, using execution logs and monitoring feedback data to perform a closed-loop update of the coupled twin model.

[0031] The model parameters are corrected and the prediction error is verified by retraining through a sliding window. When the error meets the threshold condition, the model update is confirmed to be effective; otherwise, it is rolled back to the previous version.

[0032] The system calculates a comprehensive convergence score based on reproductive stability, survival rate stability, ecological balance, and energy consumption indicators. When the convergence score continuously meets the set threshold, the system outputs the domestication prescription and terminates the process. If the convergence condition is not met, the system automatically adjusts the prediction window, weight parameters, and constraint thresholds and returns to the rolling prediction stage for strategy fine-tuning and cyclic control. At the same time, a multi-level version index of the model, constraints, and prescriptions is established to achieve full-process traceability.

[0033] The beneficial effects of this invention are as follows:

[0034] This invention introduces a three-dimensional coupling mechanism of microecology, community dynamics, and water state into the copepod domestication process for the first time, and constructs a digital twin model with time delay compensation capability. By mapping key variables such as community abundance, algal proportion, signal molecule intensity, ammonia nitrogen concentration, dissolved oxygen level, and reproduction rate into real-time acquisition monitoring parameters, the system can calculate the coupling state between the copepod community and the microecological environment at any time, and realize real-time dynamic monitoring with "virtual and real synchronization".

[0035] This invention employs a rolling predictive control (MPC) algorithm, using copepod net reproduction rate, feeding fluctuation, mortality rate, and algal-bacterial balance as multi-objective optimization functions. Through adaptive adjustment of uncertainty weights, it achieves dynamic prediction and optimal control of the steady state of the biological system. When external environmental disturbances (such as fluctuations in light, temperature, or dissolved oxygen) occur, the system can automatically adjust the control weights and step size, predicting and suppressing impending stress responses in advance, thus avoiding interruption of acclimatization.

[0036] This invention treats biological interventions (such as probiotic or algal supplementation) as "microecological pulses" and changes in the external physical environment (such as temperature, salinity, and light regulation) as "gradual environmental changes," and establishes a coupled control equation for both. The system generates the optimal combination of pulse dosage and environmental change rate through collaborative solving, achieving bidirectional coordinated regulation of the internal and external ecological environments of copepods.

[0037] This invention constructs a dual-channel safety protection system based on prediction deviation probability and statistical anomaly detection. When the monitoring signal is abnormal or the model prediction deviates beyond the set threshold, the system automatically triggers a safety backoff control law to dynamically reduce the amplitude of the control action or suspend execution. During this process, the system still maintains the key environmental variables within the safe range, thereby preventing the collapse of the aquatic ecosystem and the loss of copepod populations.

[0038] This invention introduces a closed-loop feedback and sliding window retraining mechanism, enabling the system to continuously optimize model parameters in multiple domestication cycles. After each iteration, the model automatically evaluates the prediction error and performs version rollback or parameter update, making it continuously approach the real ecological dynamics. This self-evolution process enables the model to gradually transform from an initial experience-based prediction into a high-precision, low-error personalized control system.

[0039] This invention establishes a three-layer index system of "model version, constraint version, and domestication prescription version". Each model update or parameter adjustment generates a unique version number and forms an anti-tampering verification code through hash signature. Parameter configuration, constraint setting, and control actions at any domestication stage can be traced and verified, ensuring the reliable operation of the entire digital twin system. Attached Figure Description

[0040] Figure 1 This is a schematic diagram of a digital twin monitoring and control method for the domestication process of copepod prey organisms according to the present invention. Detailed Implementation

[0041] 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.

[0042] Example: Figure 1 As shown, this embodiment provides a digital twin monitoring and control method for the domestication process of copepod prey organisms, including the following steps:

[0043] S1. Establish a coupled twin model of microecology and population dynamics for copepod domestication, and define a unified terminology system for monitoring parameters and control objects; generate reproducible standardized modeling configurations by defining variable ranges, interface formats and initialization strategies.

[0044] S2. Continuously collect multi-source monitoring data at the edge and perform time synchronization, missing data compensation and signal denoising. Input the coupled twin model to obtain the state estimation results and uncertainty index of micro-ecological state, population state and water state, and provide basic input for subsequent identification and control.

[0045] S3. Use the state estimation results to identify and robustly update the key parameters of the model online to minimize the error between the model prediction and the actual measurement. At the same time, construct a set of safety thresholds and water quality constraints based on real-time and historical data to form a control feasible region and freeze it synchronously with the model parameters.

[0046] S4. Establish a rolling predictive control model, set the prediction time window and control step size; construct a multi-objective optimization function with the goals of improving reproductive efficiency, reducing feeding fluctuations, suppressing stress mortality peaks and maintaining algal community stability, and adaptively adjust weights in combination with state uncertainty to form a dynamically solvable optimization solution configuration.

[0047] S5. Based on the optimized solution configuration, a set of coordinated control actions for micro-ecological pulses and gradual environmental changes is generated and sent to the edge execution end for real-time implementation. During the execution process, multi-channel anomaly detection is performed. When the prediction deviation or monitoring anomaly exceeds the set threshold, the safety backoff control law is automatically executed to maintain system stability and domestication continuity.

[0048] S6. Use execution feedback and monitoring data to perform closed-loop updates on the coupled twin model, train parameters and verify the effectiveness of the updates; the system comprehensively evaluates the convergence based on reproductive stability, survival rate stability, ecological balance and energy consumption indicators. When the comprehensive score reaches the set conditions, the domestication prescription is output and the process is terminated; otherwise, it returns to the rolling prediction stage for strategy fine-tuning and cyclic control.

[0049] S1 specifically includes the following sub-steps:

[0050] S110: Establishment of a list of terms and objects: Establish a unified terminology and scope for monitoring parameters and control objects;

[0051] The monitoring parameters include:

[0052] Microbial ecosystem monitoring parameters (community abundance, algal ratio, signal molecule intensity, ammonia nitrogen, dissolved oxygen, etc.);

[0053] Environmental monitoring parameters (salinity, temperature, light intensity, pH, etc.);

[0054] Population monitoring parameters (age class structure, survival rate, feeding rate, reproductive indicators, etc.).

[0055] The controlled objects are limited to "microecological pulses" and "gradual environmental changes." All terms should be used consistently in subsequent steps, without any ambiguous substitutions.

[0056] S120. Coupled Twin Model Structure Definition: A three-layer coupled twin model is established, comprising micro-ecology, community dynamics, and water state. Its core differential equation is:

[0057]

[0058] in The number density of the i-th ecological level (e.g., copepod larvae, adults, or algal communities) is the main state variable that changes over time. This represents the natural growth rate of the i-th population, ranging from -0.2 to 0.5, where a positive value indicates a natural increase in the number of individuals and a negative value indicates a natural decrease in the number of individuals. This represents the interaction coefficient matrix among various ecological populations in the system, where the element in the i-th row and j-th column is... This represents the strength of the influence of the i-th population on the j-th population;

[0059] when When <-0.1, it indicates that there is a competitive relationship between the two groups; when A value greater than 0.3 indicates the existence of a symbiotic relationship; a value less than or equal to -0.1 indicates a symbiotic relationship. When the value is ≤0.3, it indicates that the interaction between populations is in a neutral state, therefore the matrix... By all The composition comprehensively reflects the ecological balance characteristics of the system; The response coefficient represents the external control input and is used to quantify the impact of artificial controls (including light, dissolved oxygen, nutrient addition, etc.) on the ecological level. Its value ranges from 0 to 1. This represents the control input variable, which is an external operating signal that changes over time, such as a periodic feeding pulse, a gradual change in ambient temperature, or a light adjustment process. This represents the random disturbance term of the system, which follows a zero-mean Gaussian distribution. It is used to describe environmental fluctuations, measurement noise, or natural uncertainties.

[0060] The output of the above model includes state variables. Parameters such as algae-to-bacteria ratio, ammonia nitrogen concentration, and dissolved oxygen concentration can be used to calculate the predicted state variables for the next time step. .

[0061] The model's input consists of a set of monitored parameters, including real-time data collected on temperature, light intensity, salinity, nutrient concentration, and environmental disturbance signals. To ensure synchronization between the digital twin model and the actual ecosystem, the time step is... It is recommended to set the sampling interval to 1–3 minutes, with a sampling frequency of no less than 0.1 Hz (i.e., sampling once every 10 seconds), to ensure consistency and dynamic matching between monitoring data and model calculation results over time. Through the above steps, an ecological twin mathematical model that can be updated in real time is established, providing fundamental support for subsequent data collection, parameter identification, rolling prediction, and coordinated control.

[0062] S130, Interface and Parameterization Description: The controlled object is parameterized in the model as: Microecological Pulse: represented by the dosage or inhibitor amount (0-1 unit concentration) and time sequence (minute level);

[0063] Gradual environmental change: expressed as the magnitude of change (0-10%) and the rate of change (0-2% / h).

[0064] Data interaction between the edge execution terminal and the twin model uses a standardized interface:

[0065] Communication protocol: RESTful API or MQTT;

[0066] Data format: JSON, with corresponding fields including {parameter name, target value, execution time, unit, security identifier};

[0067] Return format: Uniform timestamp records and status receipts; interface parameters, field names and data units remain consistent throughout the entire process to ensure the callability and traceability of rolling prediction and control solution stages.

[0068] S140. Initialization and Version Baseline: When the system is running for the first time or when there is no historical data, use the default set of empirical parameters. As initial value;

[0069] If historical monitoring data exists, the model parameters are back-corrected using the least squares method or the maximum likelihood estimation method to calculate the optimal initial values. The optimization objective function is defined as follows:

[0070]

[0071] in This represents the number of the i-th ecological level predicted by the model at time t. Let represent the measured value of the i-th ecological level at time t. The sum of squares of parameter identification errors is used to minimize the difference between predicted and observed values. The parameter set obtained through the above optimization process is established as the initial parameter baseline and forms the model version baseline. To enable subsequent traceability and version management, the version numbers of the model, interface, and data dictionary adopt a unified naming format (e.g., "V1.0.0-date") to record the initial configuration status of the current twin model and serve as a version comparison benchmark in subsequent rolling predictions and model updates.

[0072] S150, Archiving and Recall Strategy: The aforementioned terminology system, model structure, parameter range, interface definition and baseline initial value are archived in a unified manner to generate a standardized modeling configuration that can be called by subsequent S2-S6. This configuration can be executed locally or deployed to the cloud or edge nodes via the network to ensure the reproducibility of the modeling layer and cross-device consistency.

[0073] S2 specifically includes the following sub-steps:

[0074] S210. Data Acquisition Link Configuration: Under the unified configuration framework established in step S150, the monitoring parameter acquisition link of the edge execution end is initialized and configured.

[0075] Each acquisition link corresponds to a sensor node, and a unique sensor ID and timestamp field are bound in the system configuration file to ensure the traceability and time uniqueness of the acquired data. The sampling frequency of each acquisition link is set according to the parameter change rate, and it is recommended to be no less than 0.1Hz (i.e., sampling once every 10 seconds), preferably at the 1Hz level, to meet the real-time monitoring requirements and improve the model update accuracy.

[0076] All acquisition nodes and the central clock achieve high-precision time calibration through the NTP protocol or IEEE-1588 Precision Clock Synchronization (PTP) method, ensuring that the system time error does not exceed 0.05s, thereby guaranteeing the consistency and alignment of various monitoring data in the time dimension.

[0077] The data output from the acquisition link adopts a standardized JSON structure, defined as:

[0078] {Timestamp, ParameterName, Value, Unit, SensorID}.

[0079] Among them, Timestamp represents the data acquisition time; ParameterName represents the name of the monitored parameter; Value represents the parameter value; Unit represents the unit of physical quantity; and SensorID represents the identifier of the acquisition sensor.

[0080] Through the above structured design, it is ensured that data from different sources can be time-aligned, field-unified, and semantically compatible at the acquisition end, providing high-precision input data support for subsequent state estimation and twin model updates.

[0081] S220, Time Reference and Integrity Verification: After the collected data is transmitted back, the system performs a time consistency and integrity verification process to ensure that the multi-source data output by the sensor link maintains comparability and traceability in terms of time and physical quantity dimensions.

[0082] The specific process includes the following steps:

[0083] Clock drift correction: Time drift is corrected using the least mean square error method. Curve fitting and backcompensation are performed to correct minor offsets between the sensor and the central clock, ensuring time reference consistency.

[0084] Packet loss detection: If the sampling time interval exceeds twice the expected sampling period (i.e., sampling interval > 2 × set sampling period), the system automatically marks the time period as a missing point and records the abnormal label for subsequent interpolation compensation.

[0085] Interpolation compensation: For missing data, linear interpolation or cubic spline interpolation algorithms are used for smoothing and repair to ensure the continuity of the time series and the smoothness of signal changes.

[0086] Unit and accuracy verification: Before integrating data from different sources, the system standardizes and verifies the units of parameters to ensure that the physical units of similar monitoring parameters (such as dissolved oxygen, ammonia nitrogen, temperature, etc.) remain consistent (e.g., mg / L, ℃, etc.) and makes consistency adjustments to the numerical accuracy.

[0087] After the above steps, the monitoring time series data, which has undergone time base correction and integrity verification, is obtained and denoted as . This time series data is used in the subsequent state estimation and feature extraction stages.

[0088] S230. Preprocessing and Alignment: After completing time base calibration and integrity verification, the processed monitoring time series... Denoising and standardization processes are performed to eliminate dimensional differences and improve the fusion accuracy of multi-source data.

[0089] Its standardized expression is:

[0090]

[0091] in and These represent the mean and standard deviation of the data within the sliding window, respectively. To further suppress high-frequency noise, Kalman filtering or Savitzky-Golay smoothing filtering methods are used to smooth the signal.

[0092] After multi-source data aggregation, to ensure consistency in the time domain, time alignment is performed on the signals of each channel. The calculation formula is as follows:

[0093]

[0094] in This represents the time offset between sources. After S220 correction, the time offset of all channels within the system is typically less than or equal to 0.1s. Finally, the processed result is presented in a unified structure format. Mapped to the input of the copepod coupled twin model, it provides standardized, high-confidence input data for subsequent state estimation and uncertainty propagation.

[0095] S240, State Estimation and Uncertainty Assessment:

[0096] In this step, the time-aligned data processed by S230 is input into the copepod-like coupled twin model defined by S120. The system state estimate is obtained through model calculation, denoted as:

[0097]

[0098] in This represents the estimated copepod population density. Indicates the algae-to-bacteria ratio. Indicates dissolved oxygen level. This represents an estimated value for ammonia nitrogen concentration.

[0099] To assess the changes in confidence and uncertainty of the estimation results, the covariance propagation method is used for quantification, and its calculation formula is as follows:

[0100]

[0101] in The Jacobian matrix represents the model and is used to reflect the sensitivity of the output to input perturbations. The covariance matrix of the input data is used to characterize the degree of fluctuation in the monitoring data; This represents the process noise matrix, used to describe the dynamic uncertainties introduced in the model calculation.

[0102] Based on the covariance propagation results, the system output includes the state estimate and its 95% confidence interval, expressed as follows: ,in To estimate the standard deviation of the error, this uncertainty assessment result is used for subsequent online identification and dynamic weighting of safety constraints, providing statistical reliability support for subsequent rolling prediction and control optimization.

[0103] S250, Output and Transmission: [This section appears to be incomplete and requires further context.] , The data quality identifier, QualityFlag, is uniformly encapsulated as follows:

[0104] {Timestamp, StateEstimate, Uncertainty, QualityFlag};

[0105] Where Timestamp represents the timestamp of the state estimate; StateEstimate represents the state estimate value. Uncertainty indicates uncertainty information (including standard deviation and confidence interval range); QualityFlag is a data quality indicator used to mark the confidence level of the estimated value.

[0106] The data is transmitted to the S3 online identification module and the S4 rolling prediction setting module. During the transmission process, the parameter names and units are kept consistent with the S110 definition to ensure that the data flow has no semantic drift within the coupled twin system and to achieve closed-loop consistency of the "acquisition-estimation-control" link.

[0107] S3 specifically includes the following sub-steps:

[0108] S310, Online Identification of Coupling Parameters and Time Delays: State Estimation Results Based on S240 For the parameter set of the coupled twin model Online identification is performed to minimize the error between model predictions and measured values.

[0109] The identification objective function is defined as follows:

[0110]

[0111] in It is the number density of predictions for the i-th ecological level by the model at time k; It is the number density of the i-th ecological level as measured by the sensor; It is the length of the sliding window, used to limit the time range of the calculation error; It is a regularization coefficient (typically ranging from 0.01 to 0.1), used to balance prediction accuracy and parameter update smoothness, and to prevent parameters from oscillating excessively during the identification process; , These are the parameter vectors for the current time step and the previous time step, respectively, containing... , , , .

[0112] The parameter update law can be implemented using gradient descent or recursive least squares.

[0113]

[0114] or

[0115]

[0116] in The learning rate is (0.001-0.01). This is the gain matrix; used to adjust the parameter update magnitude and response speed. It represents the gradient of the objective function with respect to the parameters, reflecting the trend of error change in the current parameter direction; This is the actual measured output. The output is the model's predicted value; if the model contains a time delay term. Then update in each iteration , where h is the sampling period.

[0117] Through the above online identification process, adaptive updating of model parameters is achieved, enabling the copepod coupled twin model to maintain high fitting degree and high prediction accuracy under dynamic environmental conditions, laying a data foundation for subsequent safety constraint generation and optimized control.

[0118] S320. Robustness and Overfit Prevention: To prevent parameter divergence in small sample sizes or noisy environments, the following robust strategies are adopted:

[0119] Input-output standardization: Standardize the input and output variables to make their mean 0 and variance 1, in order to eliminate differences in dimensions and avoid high-order data dominating the model fitting process.

[0120] Introduction of confidence weights: A confidence weight term is introduced into the objective function for parameter identification. ,in This is the variance term of the uncertainty estimation result, used to suppress the impact of high uncertainty samples on model parameter updates.

[0121] Exponential moving average smoothing of parameter updates: To improve identification stability, the exponential moving average method is used to smooth parameter updates.

[0122]

[0123] in This is a smoothing coefficient, ranging from 0.8 to 0.95, used to balance the weights of the current identification results and historical estimation results.

[0124] Convergence criterion: If the following conditions are met in three consecutive iterations:

[0125]

[0126] This indicates that the identification process has converged. Let be the gradient vector of the objective function. This represents the Euclidean norm, used to measure the rate of change of adjacent gradients.

[0127] S330, Safety Threshold and Water Quality Constraint Set Construction: Based on the state estimation and historical data of S240, a set of safety thresholds and water quality constraints is constructed.

[0128]

[0129] in This indicates the concentration of ammonia nitrogen in the water (mg / L). This indicates the maximum permissible safe threshold for ammonia nitrogen; This indicates the dissolved oxygen concentration (mg / L). Indicates the minimum permissible value for dissolved oxygen; : Indicates the mortality rate of copepod populations; This represents the upper limit threshold for mortality in copepod populations; : Indicates the rate of change in water quality (such as the rate of change in turbidity or algae-to-bacteria ratio, in % / h); This represents the maximum permissible rate of change in water quality; typical values ​​are: =0.6mg / L =5.0mg / L =0.05、 =1% / h;

[0130] The constraint set is represented in linearized matrix form:

[0131]

[0132] in The control object vector includes a combination of microecological pulses and environmental gradual change parameters (such as light intensity, feeding rate, dissolved oxygen regulation, etc.). This is the constraint coefficient matrix, used to characterize the linear relationship between each variable and the safety constraints; The upper bound vector is the constraint vector, corresponding to the threshold of each safety boundary.

[0133] This matrix form facilitates real-time solutions in subsequent optimization steps (such as S340–S350) using linear programming or quadratic programming.

[0134] S340, Feasibility Domain Verification: Based on the updated... With constraint set Calculate the feasible domain of the controlled object in the current state:

[0135]

[0136] in This represents the system control object vector, which includes adjustable parameters of the ecosystem (such as light intensity, aeration rate, dissolved oxygen regulation, temperature regulation, etc.). This represents a nonlinear ecological constraint function, reflected in the parameters. Ecological response relationships, such as temperature-dissolved oxygen coupling, water pH change rate, or ammonia nitrogen metabolism rate; if feasible domain volume If the value falls below the threshold by 5% (indicating that the environment is approaching its limit), the constraint relaxation mechanism will be automatically triggered.

[0137]

[0138] in This is the upper bound vector of the relaxed constraints; To constrain the relaxation range, a typical value is 5% of the original constraint upper limit; this adjustment strategy ensures that the control optimization still has feasible solutions in the boundary region, maintaining the stability of the system solution.

[0139] S350, Identification-Constraint Result Freeze:

[0140] After completing parameter identification and feasible region verification, the updated parameter set will be... Safety constraint set and feasible region description Archive and freeze versions to ensure the consistency and traceability of the model in subsequent rolling control solutions.

[0141] Data archiving structure: The system archives the above results in a structured format as follows:

[0142]

[0143] The archive is automatically labeled with: Timestamp: records the archive timestamp, used for time series indexing; ModelVersion: used to identify the model baseline version corresponding to this identification-constraint result; the two together constitute the unique index key for subsequent control inputs.

[0144] Data input and transmission: The archived results are transmitted to the S4 module "Control Objective and Rolling Prediction Settings" as the sole input interface to ensure that the model parameters, constraint boundaries and feasible region descriptions used in the subsequent control solution stage are completely consistent.

[0145] Freeze Mechanism Explanation: Freezing means that within the current version lifecycle: parameter set No longer participating in real-time identification and updates; security constraints With feasible region The system is locked to prevent cross-cycle drift. If the system enters a new identification cycle, the frozen version is used as the baseline for differential comparison and version increment update. This freezing mechanism ensures the consistency and traceability of model parameters, constraints and control inputs in cross-stage iterations, and avoids control solution deviation caused by version drift.

[0146] By introducing a VersionID-Timestamp dual-index mechanism, the system can achieve time-series tracing and version management of the entire model-constraint-control chain, providing a long-term, stable, and verifiable digital twin control baseline for the domestication process of copepod prey organisms.

[0147] S4 specifically includes the following sub-steps:

[0148] S410, Scrolling Window and Step Size Setting: Based on the parameter and constraint set frozen in S350, set the time window and step size for the Model Predictive Control (MPC).

[0149] Prediction Time Window Typically, a time interval of 24-72 hours is used, corresponding to the control step size. =1-3 hours, both conditions are met. , where n is the prediction step number, which is used to define the number of steps the model rolls during prediction within a control cycle.

[0150] Within each rolling cycle, the model predicts the future state evolution:

[0151]

[0152] in The state estimate at time t+k includes parameters such as copepod population density, dissolved oxygen concentration, ammonia nitrogen level, and algae-to-bacteria ratio. To control object variables, such as aeration intensity, light power, feeding rate, nutrient salt addition amount, etc.; The set of model parameters obtained from the S350 freeze remains constant within the current rolling forecast period; The state transition function of the twin model is used to describe the dynamic changes of the state variables under the control input. To avoid cumulative calculation errors, a sliding window update mechanism is adopted: at the end of each cycle, the oldest time window is discarded and the latest observation data is added.

[0153] S420. Definition of Optimization Objective Function and Weight Adaptive Rule: The optimization objective of rolling predictive control is a multi-objective function that comprehensively balances population stability, reproductive efficiency, and ecological security.

[0154]

[0155] in Indicates net reproductive rate deviation; used to measure the degree of deviation of population growth from the target steady state; It indicates the variability of the feeding rate and reflects the stability of feeding behavior; This represents the predicted mortality rate; It represents the fluctuation of the algae-to-bacteria ratio and is used to measure the micro-ecological balance of water bodies; To control the intensity of the movement and comprehensively reflect the amplitude of microecological pulses and environmental regulation, , , , , This refers to the dynamic weighting factor for the corresponding indicator.

[0156] Weight adaptive rule:

[0157] Each weight Dynamically adjust based on state uncertainty:

[0158]

[0159] in Let represent the estimated variance of the i-th state index, reflecting the magnitude of the uncertainty of that state under the model prediction. Its calculation follows the uncertainty propagation formula of S240. The uncertainty variance of each state variable can be obtained from the matrix. diagonal elements Extracted; The adjustment coefficient is (0.1-0.5). When the uncertainty increases, the system automatically increases the weight of the steady-state index to prioritize safety and stability. This mechanism ensures that the optimization objective automatically balances the trade-off between "efficiency" and "safety" at different stages.

[0160] S430, Constraint and Actuator Limitation Injection: Injecting the linear constraint matrix formed by S330. Physical constraints at the edge execution end (dose cap, rate limit) are incorporated into the MPC optimization problem. If the system contains nonlinear safety constraints (such as temperature-dissolved oxygen coupling), a first-order linearization approximation is used:

[0161]

[0162] in This is the current linearization point (i.e., the previous iteration or the current state point). To constrain the gradient matrix with respect to the control variables, this equation performs a first-order Taylor expansion at the current point to achieve a linear approximation of the nonlinear constraints, so as to solve it in model predictive control optimization and ensure local feasibility and convergence.

[0163] It is added to the constraint set; to prevent the solution from failing due to an excessively small feasible region, a penalty slack variable is introduced. Add a penalty term to the objective function This ensures that the optimization problem is always solvable.

[0164] S440. Control Variable Parameterization and Solution Structure: Control Object Parameterized as:

[0165]

[0166] in The microecological pulse dose (representing the intensity and frequency of drug administration or aeration) corresponds to the ecological "pulse" control signal. It indicates the rate of change in salinity and regulates the osmotic pressure balance of water bodies; It represents the rate of temperature change and regulates metabolic rate and dissolved oxygen capacity; It represents the rate of change in light intensity, which affects phytoplankton photosynthesis and copepod activity.

[0167] The control optimization problem is formalized as follows:

[0168]

[0169] in For the multi-objective cost function of model predictive control, a two-stage optimization strategy is adopted in the solution structure: the first-stage fast prediction layer uses linear quadratic programming (QP) to obtain the initial solution of the control sequence; the second-stage refinement layer inputs the initial solution into a nonlinear solver (such as SQP or ADMM) for 2-3 iterations of refinement. This structure ensures both real-time control performance and nonlinear accuracy; and an adaptive uncertainty step size and energy consumption penalty weight can be introduced in the second-stage refinement to improve the overall stability and energy efficiency of the system.

[0170] S450, Solving for configuration freeze and output:

[0171] The final obtained control sequence and corresponding predicted status This is encapsulated as: {VersionID, ControlSequence, PredictedState, WeightSet, ConstraintSet, Timestamp}, where VersionID represents the model and algorithm version number; ControlSequence is the optimal control input sequence; PredictedState is the future predicted state sequence; WeightSet is the current adaptive weight distribution; ConstraintSet is the constraint matrix and slack variable set; and Timestamp is the execution timestamp. This encapsulation structure facilitates control command issuance, model version tracking, and subsequent performance optimization analysis. It is also passed to S5 "Collaborative Control Solving and Edge Execution"; simultaneously, solution logs and weight adjustment records are saved for subsequent tracking and performance optimization analysis.

[0172] S5 specifically includes the following sub-steps:

[0173] S510, Cooperative Control Action Solution: Based on the control sequence and predicted state frozen in S450, generate the set of cooperative control actions for the current rolling cycle:

[0174]

[0175] in: Indicates the dosage of microecological pulse administration; Indicates the rate of change in salinity; Indicates the rate of temperature change; This indicates the rate of change in light intensity.

[0176] To improve real-time performance, a hierarchical scheduling mechanism is adopted: The upper-level scheduler calculates the optimal motion trajectory for the next 24-72 hours; the lower-level executor scheduler executes fine-grained control in 5-15 minute increments. Before each cycle, the scheduler verifies the following through a constraint verification module:

[0177]

[0178] If verification fails, automatic downgrade optimization is triggered, adjusting only the variables within the security domain to maintain feasibility.

[0179] S520, Issuance and Execution Scheduling: Control commands are issued to the edge execution end through the same interface protocol (RESTful / MQTT) as S130.

[0180] Each instruction includes:

[0181] {CommandID, ParameterName, TargetValue, ExecutionTime, Checksum}

[0182] CommandID is used to identify a unique command, ParameterName is the name of the corresponding control parameter, TargetValue is the optimal control target value, ExecutionTime represents the planned execution timestamp, and Checksum is the instruction integrity check code.

[0183] After the edge execution end receives the data, it enters the two-layer scheduling system:

[0184] Real-Time Scheduler: Performs millisecond-level timing management based on the system clock to ensure synchronization of multiple actuators;

[0185] Fault Buffer: If network latency > 1s, the instruction is buffered and automatically resent three times; if it still fails, it enters a safe hold state (maintaining the action of the previous cycle by 1 step); the execution end sends back status feedback {CommandID, ActualValue, Deviation, StatusFlag} every 1 second for anomaly detection. ActualValue is the actual execution value; Deviation indicates the difference between the target value and the execution value; StatusFlag represents the execution status flag, used to indicate normal, retry, timeout, or failure status.

[0186] S530, Anomaly Detection and Safety Rollback Control Law: The anomaly detection module adopts a dual-channel structure.

[0187] Statistical Channel: Identifying Observation Anomalies Based on the Sliding Window Three Sigma (3σ) Rule;

[0188] Model channel: based on prediction error , These are actual observed values. These are model predictions; the probability of an anomaly is calculated as follows:

[0189]

[0190] in This is the model prediction uncertainty (obtained from the covariance propagation formula of S240), if >0.2 or three consecutive deviations If it does, it is considered an abnormal event.

[0191] Safety rollback control law:

[0192]

[0193] in It refers to the amount of ecological pulse addition after the regression (such as the addition of oxygen, nutrients, and other controlled variables). It is the rate of salinity change after the retreat. , It is the optimal target value obtained in the model predictive control stage (S510); If the anomaly persists for two steps, the system automatically enters conservative mode: fixing at the nearest safe point. At the same time, it sends an alarm to the upper-level controller.

[0194] S540, Execution Log and Feedback Collection: The execution end records the following information in real time:

[0195] {Timestamp, CommandID, ParameterName, TargetValue, ActualValue, Deviation, StatusFlag, ,ModeFlag}.

[0196] Where Timestamp is the time identifier; CommandID is the command number; ParameterName, TargetValue, and ActualValue represent the monitored parameter and its target value and measured value, respectively; Deviation is the execution deviation; StatusFlag is used to mark the execution status; and ModeFlag is used to indicate the current running mode.

[0197] Log storage is divided into two levels:

[0198] A short-term cache layer (1 hour) is used for rapid visualization and analysis;

[0199] The long-term archive layer (24 hours - 7 days) is automatically uploaded to the main control server for retraining of the twin model parameters; the log sampling rate is ≥1Hz and the time synchronization accuracy is ≤0.05s to ensure strict alignment with the S210-S240 sampling layer.

[0200] S550, Execution Cycle Closure: After each rolling cycle ends, the system executes a closed-loop update process: summarizing the execution logs and status feedback from S540;

[0201] Calculate control effectiveness indicators:

[0202]

[0203] in It is the target control sequence predicted by the MPC model; It is the actual control quantity sequence at the execution end; when If no abnormal rollback is triggered, the cycle is considered to have executed successfully; the successful and abnormal cycle data are marked and archived. The results are passed to the S6 model closed-loop update module for retraining the coupled Siamese model. CycleID represents the cycle number, and AbnFlag represents the exception flag, which is used to identify the cycle execution status (0 for normal, 1 for minor exception, and 2 for safe maintenance).

[0204] S6 specifically includes the following sub-steps:

[0205] S610, Model Closed-Loop Update: After each rolling cycle, read the execution log output by S550. and state feedback data; a sliding window retraining mechanism is used to update the parameter set of the coupled twin model. :

[0206]

[0207] in The dynamic learning rate is 0.001-0.01. The momentum term is (0.8-0.95). The objective function defined for S310, The gradient of the objective function defined by S310 (usually the least squares or Bayesian estimation error term).

[0208] Training dataset Composed of the latest N cycles:

[0209]

[0210] in This represents the observed state vector at the current moment; it includes multi-dimensional environmental variables such as salinity, temperature, dissolved oxygen, light intensity, and microbial ratio. Represents the historical state vector from n periods ago; represents the initial state of the model's sliding window; This indicates the control input being executed at the current moment; including control variables such as the amount of microecological pulse dosage and the light adjustment rate. This represents the historical execution input from n cycles ago; corresponding to the historical control action sequence; n is the sliding window length; its value ranges from 5 to 10, used to balance the training sample size and real-time performance; after updating, execution verification is performed: if the average prediction error... If the model update is successful, it is confirmed to be valid; otherwise, it will be automatically rolled back to its previous state. The version is marked "Update Rejected".

[0211] S620. Evaluation of domestication convergence conditions: The system calculates the following convergence indices within each window period:

[0212] Reproductive stability index

[0213]

[0214] in This represents the reproductive rate value observed in the i-th instance, used to reflect the reproductive intensity of the system at the i-th sampling point; The average reproductive rate is expressed as follows: This value reflects the average reproductive level within the sample period; This indicates the number of observations, i.e., the sample size, and is used to characterize the size of a statistical sample. The standard deviation of the reproductive rate measures the fluctuation range of the reproductive process; the smaller the value, the more stable the reproductive process. .

[0215] Survival rate stability index:

[0216]

[0217] in The standard deviation of survival rate is used to measure the range of fluctuation in survival rate among different observed samples. The average survival rate represents the average survival level of the ecosystem within the statistical period. The coefficient of variation is the survival rate; the smaller the value, the more stable the individual's survival process.

[0218] Ecological balance indicators:

[0219]

[0220] in The (ecological balance deviation coefficient) reflects the community balance between algae and bacteria in the system. This index represents the difference between the current algae-to-bacteria ratio and the ideal algae-to-bacteria ratio. The relative deviation between algae and bacteria indicates that the system is in a good ecological balance when the deviation is less than 10%. The algae-bacteria ratio represents the ratio of the biomass or density of algae to bacteria in the system and is a key parameter for energy and material cycling in the ecosystem.

[0221] Control energy consumption indicators:

[0222]

[0223] in This represents the ecological control input signal at time t, which typically corresponds to the microecological pulse dose or the environmental regulation amplitude. This indicates the length of the time window for energy consumption calculation, i.e., the total number of steps in one control cycle; The upper limit for energy consumption is generally set based on experimental experience, and is usually taken as 0.5-1 (unit concentration squared). This is the actual average energy consumption indicator; the smaller the value, the higher the control efficiency.

[0224] The comprehensive scoring function is defined as follows:

[0225]

[0226] like If the condition is maintained for three consecutive cycles, the system is considered to have met the conditions for domestication and convergence.

[0227] S630, Prescription Output and Process Termination: When the convergence condition is met, the system outputs a "training prescription," including:

[0228]

[0229]

[0230] in This represents the final parameter set after model retraining (S610) converges, containing all model parameters used for prediction and control; This indicates the final optimized set of safety constraints, which is the result of integrating ecological constraints and actuator limitations in steps S330–S430. This represents the sequence of control actions obtained through the S440–S510 co-optimization solution, i.e., the optimal control trajectory; The convergence verification metric (see S620) is used to quantify the overall stability of the training and control system; the Timestamp records the timestamp of the result generation for version tracking and subsequent backtracking, ensuring the consistency of parameters, constraints and control strategies at different stages.

[0231] The prescription is automatically packaged into JSON and uploaded to the main control server, while simultaneously freezing the current model version (VersionID=Vfinal). The prescription includes a microecological pulse dose table, an environmental gradual change rate table, and corresponding safety thresholds, which can directly guide subsequent mass production cultivation.

[0232] S640, Strategy Fine-tuning and Loop Entry: If If the value is less than 0.9 or any single indicator exceeds the limit, a strategy fine-tuning will be triggered.

[0233] Adjust the rolling forecast window (±12 hours);

[0234] Adjusting weights in S420 This increases the weighting of safety and stability by 10%.

[0235] Adjusting the S330 constraint set The threshold is ±5%;

[0236] Mark the adjustment result as "Cycle+1" and return to S410 to solve again. This loop can be executed up to 30 times. If it still fails to converge, an anomaly report is generated for manual intervention.

[0237] S650, Version Management and Audit Traceability Mechanism: To ensure the traceability and verifiability of the system, a multi-layered version control system is established:

[0238] Model Version Layer, each time Update and generate a unique VersionID (Vn.mx); archive fields ConstraintVersion layer; generated for each constraint adjustment. Version number Cn.yx; Prescription Version layer; The taming prescription archive for converged output is Pn.zx.

[0239] Audit Index Table: Establish a unified mapping table:

[0240] AuditIndex={VersionIDmodel,VersionIDconstraint,VersionIDprescription,Timestamp,Author};

[0241] Where VersionIDmodel is the model version number, representing the version identifier of the currently used model parameter set; VersionIDconstraint is the constraint version number, representing the version identifier of the system safety boundary and constraint condition set; VersionIDprescription is the prescription version number, representing the version identifier of the generated control instructions or optimized prescriptions; and Author is the creator, representing the entity that generated this version record.

[0242] A checkcode is generated using SHA-256 hash signature to prevent data tampering. Through this three-layer version control plus audit index mechanism, a closed-loop, verifiable, and traceable data system is achieved throughout the entire process.

[0243] The above formulas are all dimensionless calculations. The formulas are derived from software simulations based on a large amount of collected data to obtain the most recent real-world results. The preset parameters and thresholds in the formulas are set by those skilled in the art according to the actual situation.

[0244] Those skilled in the art will recognize that the modules and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0245] In conclusion, the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A digital twin monitoring and control method for the domestication process of copepod prey organisms, characterized in that, Includes the following steps: S1. Establish a coupled twin model of microecology and population dynamics for copepod domestication, and define a unified terminology system for monitoring parameters and control objects; generate reproducible standardized modeling configurations by defining variable ranges, interface formats and initialization strategies. S2. Continuously collect multi-source monitoring data at the edge and perform time synchronization, missing data compensation and signal denoising. Input the coupled twin model to obtain the state estimation results and uncertainty index of micro-ecological state, population state and water state, and provide basic input for subsequent identification and control. S3. Use the state estimation results to identify and robustly update the key parameters of the model online to minimize the error between the model prediction and the actual measurement. At the same time, construct a set of safety thresholds and water quality constraints based on real-time and historical data to form a control feasible region and freeze it synchronously with the model parameters. S4. Establish a rolling predictive control model, set the prediction time window and control step size; construct a multi-objective optimization function with the goals of improving reproductive efficiency, reducing feeding fluctuations, suppressing stress mortality peaks and maintaining algal community stability, and adaptively adjust weights in combination with state uncertainty to form a dynamically solvable optimization solution configuration. S5. Based on the optimized solution configuration, generate a set of coordinated control actions for micro-ecological pulses and gradual environmental changes, and send them to the edge execution end for real-time implementation; during the execution process, perform multi-channel anomaly detection, and when the prediction deviation or monitoring anomaly exceeds the set threshold, automatically execute the safety backoff control law to maintain system stability and domestication continuity.

2. The digital twin monitoring and control method for the domestication process of copepod prey organisms according to claim 1, characterized in that, It also includes S6, which uses execution feedback and monitoring data to perform closed-loop updates on the coupled twin model, trains parameters and verifies the effectiveness of the updates; the system comprehensively evaluates the convergence based on reproductive stability, survival rate stability, ecological balance and energy consumption indicators. When the comprehensive score reaches the set conditions, it outputs the domestication prescription and terminates the process; otherwise, it returns to the rolling prediction stage for strategy fine-tuning and cyclic control.

3. The digital twin monitoring and control method for the domestication process of copepod prey organisms according to claim 1, characterized in that, S1 includes: A coupled twin model of microecology and population dynamics for copepod domestication was established. Monitoring parameters were categorized into microecological monitoring parameters, environmental monitoring parameters, and population monitoring parameters. Controlled objects were classified into microecological pulses and environmental gradual changes. The model input, output, and parameterized interface are defined based on the coupling relationship between the monitoring parameters and the controlled object; the model, interface, and data dictionary are unified by setting the variable value range, data format, and communication protocol.

4. The digital twin monitoring and control method for the domestication process of copepod prey organisms according to claim 3, characterized in that, S1 also includes: When there is no historical data, default empirical parameters are used. When historical data is available, an initial parameter set is generated through statistical algorithms, and a model version baseline is established for subsequent data collection, identification, and control.

5. The digital twin monitoring and control method for the domestication process of copepod prey organisms according to claim 1, characterized in that, S2 specifically includes: Establish a data acquisition link corresponding to the monitoring parameters at the edge execution end, and continuously collect micro-ecology, environment and group status data through multi-source sensors; Time synchronization algorithms are used to perform clock calibration, missing data compensation, and outlier correction for different sampling sources, and filtering, standardization, and cross-source alignment are performed on the returned data. The processed data is input into the coupled twin model to obtain state estimation results of micro-ecological state, community state and water state, and the state estimation results and their uncertainty information are output to provide input for subsequent model identification and optimization control.

6. The digital twin monitoring and control method for the domestication process of copepod prey organisms according to claim 1, characterized in that, S3 specifically includes: Using the state estimation results obtained in step S2, key parameters in the coupled twin model are identified and dynamically updated online. Robust fitting is achieved by minimizing the deviation between model predictions and measured values ​​and introducing a regularization term. Simultaneously, a set of safety thresholds and water quality constraints is constructed based on real-time and historical monitoring data; The constraint set includes ammonia nitrogen concentration, dissolved oxygen content, mortality rate increment, and environmental change rate limits. After the parameters are updated, the model parameters and constraint set are frozen simultaneously to form a control feasible domain for subsequent rolling predictive control.

7. The digital twin monitoring and control method for the domestication process of copepod prey organisms according to claim 1, characterized in that, S4 specifically includes: Based on the frozen set of model parameters and safety constraints, a rolling predictive control model is established, and the prediction time window and control step size are set to form a rolling solution cycle. Construct a multi-objective optimization function with the goals of improving reproductive efficiency, reducing feeding fluctuations, suppressing stress mortality, and maintaining the stability of algal and bacterial communities; By dynamically adjusting the target weights in conjunction with state uncertainty, an adaptive balance between efficiency and safety is achieved, and model constraints, actuator limitations, and nonlinear safety conditions after linearization are all injected into the optimization problem.

8. The digital twin monitoring and control method for the domestication process of copepod prey organisms according to claim 7, characterized in that, S4 also includes: Stable and real-time executable control commands are obtained through a two-level solution structure, providing configuration for subsequent collaborative control and edge execution.

9. The digital twin monitoring and control method for the domestication process of copepod prey organisms according to claim 1, characterized in that, S5 include: Based on the optimization results of rolling predictive control, a set of coordinated control actions for micro-ecological pulses and environmental gradual changes is generated, and the control actions are sent to the edge execution end for periodic implementation through a standardized communication interface; During execution, the system continuously collects monitoring parameters and performs multi-channel anomaly detection; when the prediction deviation or monitoring value exceeds the safety threshold, the safety backoff control law is automatically triggered. The system maintains stability by reducing control amplitude, pausing execution, or switching to conservative mode; the execution end records control logs and monitoring feedback data in real time to form a closed-loop dataset for subsequent model updates.

10. A digital twin monitoring and control method for the domestication process of copepod prey organisms according to claim 2, characterized in that, S6 include: After each control cycle, the coupled twin model is updated in a closed loop using execution logs and monitoring feedback data. The model parameters are corrected and the prediction error is verified by retraining through a sliding window. When the error meets the threshold condition, the model update is confirmed to be effective; otherwise, it is rolled back to the previous version. The system calculates a comprehensive convergence score based on reproductive stability, survival rate stability, ecological balance, and energy consumption indicators. When the convergence score continuously meets the set threshold, the training prescription is output and the process is terminated. If the convergence condition is not met, the prediction window, weight parameters and constraint thresholds are automatically adjusted and the process returns to the rolling prediction stage for strategy fine-tuning and cyclic control. At the same time, a multi-level version index of the model, constraints and prescriptions is established to achieve full process traceability.

Images