A digital-twin-based farm environment monitoring system
The farm environment monitoring system built using a digital twin model solves the problems of inaccurate control and poor robustness in farm environment monitoring, and achieves precise, efficient, and low-cost regulation of the farming environment, improving the system's adaptability and scientific decision-making.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- SHANDONG ACADEMY OF AGRICULTURAL SCIENCES
- Filing Date
- 2025-11-21
- Publication Date
- 2026-07-21
AI Technical Summary
Existing technologies lack the ability to predict and adapt to complex dynamic systems in monitoring the environment of aquaculture farms, resulting in inaccurate control, poor robustness, and difficulty in handling variable external disturbances and dynamic changes in aquaculture objects.
A digital twin model is used to construct a farm environment monitoring system. Through modules such as data acquisition, twin modeling, sequence generation, simulation, evaluation, control decision-making, and status monitoring, the system enables forward-looking projection and closed-loop correction of the farming environment, thereby optimizing control decisions.
It enables precise, efficient, and low-cost control of the aquaculture environment, improves the scientific nature and adaptability of the control, and can find the optimal balance among multiple objectives, taking into account stability, speed, and economy.
Smart Images
Figure CN121500768B_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of farm environmental monitoring and control technology, specifically a farm environmental monitoring system based on digital twins. Background Technology
[0002] Currently, in the field of environmental monitoring in aquaculture farms, related technologies mainly rely on real-time data collected by sensors for immediate response. When key water indicators deviate from preset thresholds, the control system passively triggers equipment to intervene. This approach is mostly based on simple logical thresholds, lacks forward-looking prediction of the system's future dynamics, and is also difficult to handle the dynamic changes brought about by volatile external disturbances and the aquaculture objects themselves. However, aquaculture water bodies are complex time-varying systems whose state evolution is affected by multiple nonlinear factors such as control time delay, biomass changes, and external environmental disturbances. Traditional control methods struggle to construct high-fidelity models to accurately describe these dynamic characteristics, especially when balancing multiple conflicting performance objectives such as control efficiency, operating costs, and water quality stability, lacking effective optimization tools. This results in low control accuracy and poor robustness in actual operation, and the model is prone to inaccuracy over time, making adaptive adjustment impossible. Therefore, there is an urgent need for a solution that can proactively extrapolate the complex dynamics of the aquaculture environment, optimize among multiple control objectives, and use actual monitoring data to perform closed-loop correction on the model, in order to solve the problems of inaccurate control, poor robustness, and lack of adaptability in existing technologies. Summary of the Invention
[0003] To address the aforementioned technical problems, this invention discloses a farm environment monitoring system based on digital twins. Specifically, the technical solution of this invention includes: The data acquisition module is used to collect multidimensional state data of aquaculture water bodies in real time. The multidimensional state data includes: key water body indicator dataset, biomass estimation dataset, environmental control equipment status dataset, and external environmental disturbance dataset. The twin modeling module is used to construct a digital twin model based on historical multidimensional state data and through system identification technology to describe the delay relationship between control inputs and state responses; The sequence generation module is used to generate multiple candidate multi-step control sequences in response to the deviation of key water indicators in the key water indicator dataset from the preset target steady-state range. The simulation module is used to perform prospective simulations of multiple candidate multi-step control sequences using a digital twin model, and obtain the corresponding multiple predicted state trajectories. The evaluation module is used to extract multi-objective performance indicators for multiple predicted state trajectories and calculate the comprehensive cost of each candidate multi-step control sequence based on the multi-objective performance indicators. The control decision module is used to determine the sequence with the minimum comprehensive cost value among multiple candidate multi-step control sequences as the optimal multi-step control sequence based on the comprehensive cost value. The state monitoring module is used to monitor the actual state trajectory during the physical execution of the optimal multi-step control sequence; The confidence calculation module is used to compare the actual state trajectory with the predicted state trajectory corresponding to the optimal multi-step control sequence and calculate the twin inference confidence. The calibration trigger module is used to trigger the twin modeling module to update the digital twin model using the monitored actual state trajectory when the twin inference confidence level is lower than the preset confidence level threshold; when the twin inference confidence level is greater than or equal to the preset confidence level threshold, the digital twin model remains unchanged.
[0004] Preferably, the twin modeling module constructs a digital twin model, including: Offline training and fitting are performed based on historical multidimensional state data using system identification technology. Determine the state transition matrix as biomass changes; Determine the control input matrix that varies with biomass; Determine the perturbation effect matrix as a function of biomass; Determine the system time delay vector as biomass changes; Based on the state transition matrix, control input matrix, disturbance influence matrix, and system time delay vector, a time delay response model is constructed to characterize the dynamic characteristics of biomass change and control time delay, serving as a digital twin model.
[0005] Preferably, the multi-objective performance indicators include: The predicted steady-state convergence time is the time it takes for the predicted state trajectory to first enter the target steady-state region and no longer deviate from it. Multi-objective performance metrics also include: Predictive control overshoot represents the maximum percentage by which the predicted state trajectory deviates from the upper or lower limit of the target steady-state range. Multi-objective performance metrics also include: The predicted unit steady-state maintenance cost is characterized as the ratio of the total cost consumed in executing the candidate multi-step control sequence to the predicted biomass gain.
[0006] Preferably, the evaluation module calculates the comprehensive cost value based on multi-objective performance indicators, including: By iterating through multiple candidate multi-step control sequences, the predicted steady-state convergence time, predicted control overshoot, and predicted unit steady-state maintenance cost are determined, and their respective maximum and minimum values are identified. The predicted steady-state convergence time is calculated using min-max normalization to obtain a normalized convergence time index. The predictive control overshoot is calculated using minimum-maximum normalization to obtain a normalized overshoot index. The minimum-maximum normalization process is used to calculate the normalized cost index from the predicted unit steady-state maintenance cost. Based on preset adjustable weight coefficients corresponding to three normalized indicators, the comprehensive cost is calculated using a weighted sum method.
[0007] Preferably, when the maximum value is equal to the minimum value, the corresponding normalization index is defined as zero.
[0008] Preferably, the confidence calculation module calculates the twin inference confidence, including: Calculate the mean of the actual state trajectory; Calculate the sum of squared residuals between the predicted state trajectory and the actual state trajectory; Calculate the sum of squares between the actual state trajectory and the mean of the actual state trajectory; The confidence level of twin inference is determined based on the coefficient of determination model by using the total sum of squares and the residual sum of squares.
[0009] Preferably, the correction triggering module triggers the twin modeling module to update the digital twin model, including: The optimal multi-step control sequence and the actual state trajectory monitored are added as new data points to the historical multi-dimensional state data. The twin modeling module is triggered to invoke the system identification technology and use the updated historical multidimensional state data to re-identify the parameters of the digital twin model.
[0010] Compared with the prior art, the present invention has the following beneficial effects: 1. This system achieves proactive optimization control. When water quality indicators deviate, the system can generate multiple candidate multi-step control sequences and use a digital twin model for forward-looking simulation. By evaluating the comprehensive cost-effectiveness of the predicted trajectory, the system can select the optimal sequence for execution, avoiding blind trial and error in the physical world. This enables precise, efficient, and low-cost regulation of the aquaculture environment, improving the scientific nature of decision-making.
[0011] 2. This system possesses adaptive correction capabilities. During control execution, the system continuously monitors the actual state trajectory and compares it with the model's predicted trajectory to calculate the twin inference confidence level. When the confidence level falls below a threshold, the system automatically triggers a model update, re-identifying parameters using the latest actual data. This ensures that the digital twin model continuously reflects the true state of the physical entity, guaranteeing the accuracy of long-term monitoring.
[0012] 3. This system employs a multi-objective comprehensive evaluation approach in its control decision-making. It not only considers the convergence time and overshoot of the control but also innovatively introduces a predictive unit steady-state maintenance cost index, achieving a balance between control effectiveness and the economic benefits of aquaculture. By weighted summing of the multi-objective performance indicators and calculating the comprehensive cost value, the final selected optimal control sequence balances stability, speed, and economy.
[0013] 4. The digital twin model of this system has high fidelity. Constructed using system identification technology, the model specifically considers the impact of biomass changes on state transitions and control inputs, and incorporates system time delay vectors. This enables the model to accurately characterize the dynamic characteristics brought about by biomass growth during aquaculture, as well as the delayed response of control inputs, thus improving the model's simulation accuracy for complex aquaculture physical processes. Attached Figure Description
[0014] The present invention will be further explained below with reference to the accompanying drawings and embodiments: Figure 1 This is a system structure diagram of the present invention. Detailed Implementation
[0015] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to specific embodiments.
[0016] Example 1: Please see Figure 1 A digital twin-based environmental monitoring system for livestock farms includes: The data acquisition module is used to collect multidimensional state data of aquaculture water bodies in real time. The multidimensional state data includes: key water body indicator dataset, biomass estimation dataset, environmental control equipment status dataset, and external environmental disturbance dataset. The twin modeling module is used to construct a digital twin model based on historical multidimensional state data and through system identification technology to describe the delay relationship between control inputs and state responses; The sequence generation module is used to generate multiple candidate multi-step control sequences in response to the deviation of key water indicators in the key water indicator dataset from the preset target steady-state range. The simulation module is used to perform prospective simulations of multiple candidate multi-step control sequences using a digital twin model, and obtain the corresponding multiple predicted state trajectories. The evaluation module is used to extract multi-objective performance indicators for multiple predicted state trajectories and calculate the comprehensive cost of each candidate multi-step control sequence based on the multi-objective performance indicators. The control decision module is used to determine the sequence with the minimum comprehensive cost value among multiple candidate multi-step control sequences as the optimal multi-step control sequence based on the comprehensive cost value. The state monitoring module is used to monitor the actual state trajectory during the physical execution of the optimal multi-step control sequence; The confidence calculation module is used to compare the actual state trajectory with the predicted state trajectory corresponding to the optimal multi-step control sequence and calculate the twin inference confidence. The calibration trigger module is used to trigger the twin modeling module to update the digital twin model using the monitored actual state trajectory when the twin inference confidence level is lower than the preset confidence level threshold; when the twin inference confidence level is greater than or equal to the preset confidence level threshold, the digital twin model remains unchanged. This invention provides a farm environment monitoring system based on digital twins. The system constructs a high-fidelity digital twin model for forward-looking simulation and combines it with actual monitoring data from the physical world to form a simulation-correction closed loop, thereby achieving accurate, robust, and adaptive monitoring and control of the farm environment. In this embodiment, the system includes a data acquisition module, the purpose of which is to provide the necessary multi-dimensional input data for subsequent twin modeling and real-time monitoring; this module is used to collect multi-dimensional state data of the aquaculture water body in real time; multi-dimensional state data in this invention specifically refers to a combination of multiple heterogeneous datasets, the source of which is various sensors and estimation models of the aquaculture farm, and its specific composition includes: Key water indicators dataset This refers to the core parameters characterizing the biochemical state of water bodies, which originate from water quality sensors. In this embodiment, such as dissolved oxygen... ammonia nitrogen pH value, temperature ; Biomass estimation dataset Biomass refers to parameters characterizing the overall size of farmed organisms such as fish and shrimp. These parameters are derived from biomass estimation models or periodically entered manually. These estimation models are, for example, based on key water quality indicators. and environmental control equipment status To estimate the dynamics of real-time biological growth; in this embodiment, such as stock density. Average weight ; Environmental control equipment status dataset This refers to a parameter characterizing the intensity of human intervention, which originates from the status readback of equipment controllers such as PLCs. In this embodiment, it could be the power of an aerator. Feeder speed ; External environmental disturbance dataset This refers to uncontrollable external factors affecting the aquaculture environment, which originate from weather stations or third-party weather data interfaces. In this embodiment, such as air pressure... Light intensity Meteorological data; The system includes a twin modeling module, the core purpose of which is to construct a mathematical model capable of reproducing the dynamic characteristics of a physical aquaculture farm with high fidelity. This model forms the basis for all subsequent simulations and deductions. This module is used for multi-dimensional state data based on historical data. , , , Using system identification techniques, a digital twin model is constructed from the dataset to describe the delay relationship between control inputs and state responses. The model It can reflect the impact of biomass changes on system dynamics and the time delay effect after the execution of control commands; The system includes a sequence generation module, the purpose of which is to proactively propose multiple possible correction strategies for selection when the system deviates from the desired state; this module is used to respond to... Key water indicators obtained from When the target steady-state range is deviated from, multiple candidate multi-step control sequences are generated. Target steady-state range This refers to the ideal range of permissible fluctuations for key indicators in aquaculture water, such as dissolved oxygen. to The range is determined in advance based on aquaculture process standards; Multi-step control sequence This refers to a series of pre-set time windows in the future. Inside Control instructions executed step-by-step within minutes, such as {Future 1-20 minutes: Aerator power 5kW; Future 21-60 minutes: Aerator power 3kW}; In this embodiment, the sequence generation module generates a set of candidate sequences covering different control intensities by applying random perturbations of different magnitudes or fine-tuning with heuristic algorithms to a baseline control sequence, such as maintaining the current power constant. ; The system includes a simulation module, the purpose of which is to use a digital twin model to conduct a sandbox simulation of the future and evaluate the potential consequences of different control strategies; this module is used to construct a digital twin model using the aforementioned twin modeling module. For the multi-step control sequence of multiple candidates Perform forward-looking simulations, that is, in the model Iteratively run in the middle to obtain the results for each... Corresponding multiple predicted state trajectories ; Describes the execution In the future, key water indicators Predicted evolution process within the time window; The system includes an evaluation module, the purpose of which is to select the optimal solution from multiple simulation results based on complex and even conflicting business objectives such as cost, efficiency, and stability; this module is used to extract the multiple predicted state trajectories. The multi-objective performance index is used to calculate the comprehensive cost of each candidate multi-step control sequence. ; The system includes a control decision module, the purpose of which is to ultimately determine and output the optimal execution strategy; this module is used to calculate the comprehensive cost value based on the aforementioned evaluation module. Among multiple candidate multi-step control sequences, the sequence with the minimum comprehensive cost is determined as the optimal multi-step control sequence. Optimal multi-step control sequence This refers to all candidate sequences Among them, it has the smallest The sequence of values, i.e. ;Should The system outputs data to the physical execution layer, such as the controller of an aerator or feeder; The system includes a state monitoring module, the purpose of which is to obtain real-world feedback to compare with the predictions of the twin model, providing the data foundation for closed-loop correction; this module is used to physically execute the optimal multi-step control sequence. During that period Within the time window, the actual state trajectory of the aquaculture water body is monitored through the water quality sensor in the data acquisition module. ; The system includes a confidence calculation module, the purpose of which is to quantitatively evaluate digital twin models. This module determines whether the model is inaccurate based on its fidelity to the current physical reality; it is used to compare the actual state trajectory. With this optimal multi-step control sequence Corresponding predicted state trajectory ,Should The simulation module is designed for Calculate the confidence level of the twin inference based on the specific simulation results generated. ; The system includes a calibration trigger module, the purpose of which is to build a closed-loop adaptive mechanism to ensure that the model remains consistent with reality in the long term and to prevent model drift; this module is used to adjust the confidence level of the twin inference. Below the preset confidence threshold At that time, the twin modeling module is triggered to update the digital twin model using the monitored actual state trajectory; Confidence threshold This refers to a lower bound used to judge whether a model is reliable, for example... Its value is determined based on empirical values that balance model sensitivity and stability, derived from extensive testing using historical data. The specific calibration method is as follows: Simulation backtesting is performed using historical datasets, and different... The relationship curve between model correction frequency and prediction error under various values was analyzed, and the inflection point value that significantly reduces prediction error without making the correction frequency too high was selected as the optimal threshold; simultaneously, when the confidence level of the twin inference... Greater than or equal to the preset confidence threshold When, it indicates the model The current fidelity is good, the system maintains the digital twin model unchanged, and no correction action is performed.
[0017] Example 2 The twin modeling module constructs digital twin models, including: Offline training and fitting are performed based on historical multidimensional state data using system identification technology. Determine the state transition matrix as biomass changes; Determine the control input matrix that varies with biomass; Determine the perturbation effect matrix as a function of biomass; Determine the system time delay vector as biomass changes; Based on the state transition matrix, control input matrix, disturbance influence matrix, and system time delay vector, a time delay response model is constructed to characterize the dynamic characteristics of biomass change and control time delay, serving as a digital twin model. As a preferred implementation of the digital twin modeling module, this embodiment elaborates on its construction of digital twin models. The specific process; its underlying logic is to construct a system that can accurately reflect biomass. Changes such as increased system oxygen consumption due to biomass surge and control lag For example, the dissolved oxygen in the water is not a dynamic model that responds instantaneously after the aerator is turned on, which is crucial for high-fidelity simulation. The module is based on historical multidimensional state data , , , Offline training and fitting are performed using system identification technology; the purpose of this fitting process is to determine a series of core model parameters, specifically including: the state transition matrix as biomass changes. This parameter matrix characterizes biomass. Impact on water body self-purification and evolution rate yes The function reflects the nonlinearity of the system; the control input matrix varies with biomass. This parameter matrix characterizes biomass. The impact on control, such as oxygenation efficiency. Too The function; the perturbation effect matrix as a function of biomass. This parameter matrix characterizes external disturbances. The constant impact on water quality indicators; and the system time-delay vector as a function of biomass. This vector represents different control actions. Water body indicators The delay time between responses, this parameter characterizes the control action. Water body indicators The time delay vector between responses; Based on the aforementioned state transition matrix, control input matrix, disturbance influence matrix, and system time delay vector, this embodiment constructs a time-delay response model characterizing the dynamic characteristics of biomass change and control time delay, serving as the final digital twin model. ; The model Originating from a time-varying linear parameter A state-space model was developed, and adaptive improvements were made to address the time-delay characteristics of the aquaculture system. Its specific form is as follows: ; in, The next moment predicted by the twin model Vector of key water indicators; for The current water quality indicator vector at any given moment is derived from sensor data or data from the previous iteration. ; for Each time delay is applied and takes into account changes in biomass. The control vector originates from the sequence generation module. The introduction of this reflects the control time delay; for The external disturbance vector at time t is derived from the data acquisition module; The current biomass, its source is from Dataset acquisition; To follow biomass The changing state transition matrix originates from offline system identification and fitting; To follow biomass The changing control input matrix originates from offline system identification and fitting; To follow biomass The changing perturbation effect matrix originates from offline system identification and fitting; To follow biomass The changing system time delay vector originates from offline system identification and fitting; Fit the residual terms to the model; In application, this model Called by the simulation module, input candidate control sequence Current status Biomass and predicted disturbances To iteratively simulate future time windows Internal state evolution trajectory .
[0018] Example 3 Multi-objective performance metrics include: The predicted steady-state convergence time is the time it takes for the predicted state trajectory to first enter the target steady-state region and no longer deviate from it. Multi-objective performance metrics also include: Predictive control overshoot represents the maximum percentage by which the predicted state trajectory deviates from the upper or lower limit of the target steady-state range. Multi-objective performance metrics also include: The predicted unit steady-state maintenance cost is characterized as the ratio of the total cost consumed in executing the candidate multi-step control sequence to the predicted biomass gain. As a preferred implementation of multi-objective performance indicators in the evaluation module, this embodiment defines the following set of indicators that must be derived from the predicted state trajectory to resolve the inherent conflicts between control objectives such as energy consumption, water quality, and growth rate. Key performance indicators extracted from: Predicting steady-state convergence time This index characterizes the predicted state trajectory of the k-th candidate sequence. First entry into the target steady-state region And within the preset time window The remaining time during which the deviation does not occur; its calculation method is as follows: ,in This is the start time of the simulation. This is the first time the trajectory has entered. And the point at which they will not leave again; this indicator is used to assess the speed of control; Multi-objective performance metrics also include: predictive control overshoot. This index characterizes the predicted state trajectory. In the time window Internal deviation from the target steady-state range upper limit or lower limit The maximum percentage; this indicator is used to assess the stability or smoothness of the control, and excessive overshoot may cause stress to the cultured organisms; Multi-objective performance metrics also include: predicted unit steady-state maintenance cost This indicator characterizes the execution of the candidate's multi-step control sequence. The total cost consumed, such as electricity Predicted biomass gain during this period The ratio; its calculation method is as follows: ,in Depend on That is, the power is calculated by integrating it over time; Then it is based on an independent biological growth model, according to The predicted water quality trajectory is The impact on biological growth within a time window is estimated, and this indicator is used to evaluate the economics or benefits of control.
[0019] Example 4 The evaluation module calculates the overall cost based on multiple performance indicators, including: By iterating through multiple candidate multi-step control sequences, the predicted steady-state convergence time, predicted control overshoot, and predicted unit steady-state maintenance cost are determined, and their respective maximum and minimum values are identified. The predicted steady-state convergence time is calculated using min-max normalization to obtain a normalized convergence time index. The predictive control overshoot is calculated using minimum-maximum normalization to obtain a normalized overshoot index. The minimum-maximum normalization process is used to calculate the normalized cost index from the predicted unit steady-state maintenance cost. Based on the preset adjustable weight coefficients corresponding to the three normalized indicators, the comprehensive cost is calculated using the weighted sum method. This embodiment, based on embodiment 3, specifically illustrates how the evaluation module uses this multi-objective performance index. , , To calculate the comprehensive cost Detailed steps; To further clarify its computational logic: through normalization, the problem is solved. time, percentage and The difference in physical dimensions between cost and benefit makes them additive; and through weighted summation, a quantitative evaluation standard is provided. To balance conflicting objectives; Multi-step control sequence for traversing multiple candidates in the current batch The corresponding predicted steady-state convergence time Predictive control overshoot and the predicted unit steady-state maintenance cost Determine their respective maximum values in the current batch. , , and minimum value , , ; Minimum-maximum normalization is used to map all indices to a dimensionless interval of 0 to 1; the predicted steady-state convergence time is then calculated. The normalized convergence time index was calculated. Its calculation is The predictive control overshoot is processed using minimum-maximum normalization. The normalized overshoot index is calculated. Its calculation is The min-max normalization process is used to predict the unit steady-state maintenance cost. The normalized cost index was calculated. Its calculation is ; Based on the three normalization indices respectively , , Preset adjustable weight coefficient , , The final comprehensive cost is calculated using the weighted sum method. ; This comprehensive value It is a dimensionless scalar used to quantify the overall performance of the k-th sequence, and its calculation formula is: ; in, For the first The comprehensive cost of each candidate sequence; the smaller the value, the better the comprehensive performance. , , These are the three normalized performance indices calculated using the formulas above; , , For each corresponding , , The adjustable weighting coefficients are dynamically set based on the aquaculture strategy. For example, operators can select preset weight combinations, such as emergency combinations, based on current production targets through a human-computer interaction interface. Economic combination Or through a set of preset rules, such as Cold wave disturbance detected in China Automatically triggers adjustments, and .
[0020] Example 5 When the maximum value equals the minimum value, the corresponding normalization index is defined as zero; This embodiment, based on embodiment 4, employs a minimum-maximum normalization processing-based exception handling mechanism; in calculation... , , At that time, there is a possibility that the denominator is zero; Specifically, when, for example Maximum value and, for example When the minimum values are equal, this indicates that, in this batch of simulations, all candidate sequences have the same metric, for example... All are exactly the same; at this point, if the original formula is used for calculation, it will lead to a calculation error where the denominator is zero. To avoid computational crashes, this embodiment sets the corresponding normalization index as follows: Defined as zero; normalized overshoot metric and normalized cost indicators The same exception handling logic is used.
[0021] Example 6 The confidence calculation module calculates the confidence level of the twin inference, including: Calculate the mean of the actual state trajectory; Calculate the sum of squared residuals between the predicted state trajectory and the actual state trajectory; Calculate the sum of squares between the actual state trajectory and the mean of the actual state trajectory; The confidence level of twin inference is determined based on the coefficient of determination model by using the total sum of squares and the residual sum of squares. As a preferred implementation of the confidence calculation module, this embodiment elaborates on its calculation of twin inference confidence. The specific process; Its underlying logic is to use the standard coefficient of determination in statistics as a goodness-of-fit index to quantify the twin model. Fidelity of controlling the behavior of physical entities; It serves as the trigger for the deduction-correction mechanism; Calculation within the time window Internal actual state trajectory The mean, denoted as ; It represents The average level of actual water quality indicators during the period; Calculate the predicted state trajectory Compared with the actual state trajectory The sum of squared residuals between them; It measures the total error between the model's predicted values and the actual values, i.e. ; Calculate the actual state trajectory Its actual state trajectory mean The total sum of squares between them; It measures the total volatility of the actual data itself, i.e. ; Through the total sum of squares With the sum of squares of the residuals Based on the coefficient of determination model, the confidence level of the twin inference was determined. ; The confidence level of the twin inference The calculation formula is: ; in, The confidence score for twin inference is dimensionless. The closer its value is to 1, the better the model fits the actual trajectory and the higher the fidelity. The state vectors predicted by the twin model originate from the simulation module's... Simulation results; This is the actual state vector monitored by the sensor, and its source is the state monitoring module; In order to be in The mean of the internal actual state vector is given by Calculated; The L2 norm square of the vector is the square of the Euclidean distance, ensuring that the numerator and denominator have consistent dimensions.
[0022] Example 7 The calibration trigger module triggers the twin modeling module to update the digital twin model, including: The optimal multi-step control sequence and the actual state trajectory monitored are added as new data points to the historical multi-dimensional state data. The twin modeling module is triggered to call the system identification technology and use the updated historical multidimensional state data to re-identify the parameters of the digital twin model; This embodiment, based on embodiment 6, corrects the trigger module in... When the condition is met, i.e. when the model is inaccurate, the twin modeling module is triggered to update the specific implementation of the digital twin model; when When triggered, the system identifies the most recent one. Data within the cycle, i.e., data with significant deviations between model predictions and actual results, has high correction value; the system will use the optimal multi-step control sequence that has just been executed. and the actual status trajectory monitored Together with the period corresponding and Data is added as a new data point or a set of new data points to the historical multidimensional state data. , , , In this process, an updated and complete dataset is formed; The system triggers the twin modeling module to invoke system identification technology, utilizing this updated historical multidimensional state data to analyze the digital twin model. The parameters are , , , The system performs online updates or re-identification; after re-identification, it further invokes the model verification process, using a set of boundary input sequences containing maxima, minima, and zero values to verify the updated model. Conduct test simulations to check whether its predicted trajectory consistently remains within the preset physical constraint range, such as dissolved oxygen. Furthermore, the value is less than the saturation value, and only models that have passed verification are deployed for online control.
[0023] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention.
Claims
1. A farm environmental monitoring system based on digital twins, characterized in that, include: The data acquisition module is used to collect multidimensional state data of aquaculture water bodies in real time. The multidimensional state data includes: key water body indicator dataset, biomass estimation dataset, environmental control equipment status dataset, and external environmental disturbance dataset. The twin modeling module is used to construct a digital twin model based on historical multidimensional state data and through system identification technology to describe the delay relationship between control inputs and state responses; The sequence generation module is used to generate multiple candidate multi-step control sequences in response to the deviation of key water indicators in the key water indicator dataset from the preset target steady-state range. The simulation module is used to perform prospective simulations of multiple candidate multi-step control sequences using a digital twin model, and obtain the corresponding multiple predicted state trajectories. The evaluation module is used to extract multi-objective performance indicators for multiple predicted state trajectories and calculate the comprehensive cost of each candidate multi-step control sequence based on the multi-objective performance indicators. The control decision module is used to determine the sequence with the minimum comprehensive cost value among multiple candidate multi-step control sequences as the optimal multi-step control sequence based on the comprehensive cost value. The state monitoring module is used to monitor the actual state trajectory during the physical execution of the optimal multi-step control sequence; The confidence calculation module is used to compare the actual state trajectory with the predicted state trajectory corresponding to the optimal multi-step control sequence and calculate the confidence of the twin inference. The calibration trigger module is used to trigger the twin modeling module to update the digital twin model using the monitored actual state trajectory when the twin inference confidence level is lower than the preset confidence level threshold; when the twin inference confidence level is greater than or equal to the preset confidence level threshold, the digital twin model remains unchanged.
2. The farm environment monitoring system based on digital twins according to claim 1, characterized in that, The twin modeling module constructs a digital twin model, including: Offline training and fitting are performed based on historical multidimensional state data using system identification technology. Determine the state transition matrix as biomass changes; Determine the control input matrix that varies with biomass; Determine the perturbation effect matrix as a function of biomass; Determine the system time delay vector as biomass changes; Based on the state transition matrix, control input matrix, disturbance influence matrix, and system time delay vector, a time delay response model is constructed to characterize the dynamic characteristics of biomass change and control time delay, serving as a digital twin model.
3. The farm environment monitoring system based on digital twins according to claim 1, characterized in that, The multi-objective performance metrics include: The predicted steady-state convergence time is the time it takes for the predicted state trajectory to first enter the target steady-state interval and no longer deviate from it. Multi-objective performance metrics also include: Predictive control overshoot represents the maximum percentage by which the predicted state trajectory deviates from the upper or lower limit of the target steady-state range. Multi-objective performance metrics also include: The predicted unit steady-state maintenance cost is characterized as the ratio of the total cost consumed in executing the candidate multi-step control sequence to the predicted biomass gain.
4. The farm environment monitoring system based on digital twins according to claim 3, characterized in that, The evaluation module calculates a comprehensive cost value based on multi-objective performance indicators, including: By iterating through multiple candidate multi-step control sequences, the predicted steady-state convergence time, the predicted control overshoot, and the predicted unit steady-state maintenance cost are determined, and their respective maximum and minimum values are identified. The predicted steady-state convergence time is calculated using min-max normalization to obtain a normalized convergence time index. The predictive control overshoot is calculated using minimum-maximum normalization to obtain a normalized overshoot index. The minimum-maximum normalization process is used to calculate the normalized cost index from the predicted unit steady-state maintenance cost. Based on preset adjustable weight coefficients corresponding to three normalized indicators, the comprehensive cost is calculated using a weighted sum method.
5. A farm environment monitoring system based on digital twins according to claim 4, characterized in that, When the maximum value equals the minimum value, the corresponding normalization index is defined as zero.
6. The farm environment monitoring system based on digital twin according to claim 1, characterized in that, The confidence calculation module calculates the twin inference confidence, including: Calculate the mean of the actual state trajectory; Calculate the sum of squared residuals between the predicted state trajectory and the actual state trajectory; Calculate the sum of squares between the actual state trajectory and the mean of the actual state trajectory; The confidence level of twin inference is determined based on the coefficient of determination model by using the total sum of squares and the residual sum of squares.
7. A farm environmental monitoring system based on digital twins according to claim 1 or 6, characterized in that, The correction trigger module triggers the twin modeling module to update the digital twin model, including: The optimal multi-step control sequence and the actual state trajectory monitored are added as new data points to the historical multi-dimensional state data. The twin modeling module is triggered to invoke the system identification technology and use the updated historical multidimensional state data to re-identify the parameters of the digital twin model.