A deep-sea resource development digital twin risk early warning method

Abstract

This invention relates to the field of digital twin technology, specifically to a digital twin risk early warning method for deep-sea resource development, aiming to solve the problems of inconsistent time series of existing multi-source data, difficulty in effectively synchronizing real-time simulation and measured data, and lack of future risk prediction capabilities. The method includes the following steps: S100: Establishing a unified standard time grid; S200: Constructing real-time digital twin instances and ultra-real-time prediction instances; S300: Identifying risks based on future state sequences and generating a future risk timeline; S400: Synchronously displaying the current digital twin state, the future predicted state, and the future risk evolution process through a digital twin visualization interactive interface.

Description

Technical Field

[0001] This invention relates to the field of digital twin technology, specifically to a digital twin risk early warning method for deep-sea resource development. Background Technology

[0002] Deep-sea resource development, as a crucial technological means of acquiring seabed resources, typically comprises multiple subsystems, including a data acquisition system, a lifting system, a seabed relay station, a mother ship, and a communication network. Due to the characteristics of the deep-sea environment, such as high water pressure, complex seabed topography, strong ocean currents, and long-distance communication, resource development systems face challenges such as difficulty in acquiring multi-source heterogeneous data, challenges in directly observing equipment status, and complex system coupling relationships. To improve the observability and controllability of deep-sea resource development systems, digital twin technology is increasingly being applied to areas such as deep-sea equipment monitoring, operational analysis, and fault diagnosis.

[0003] Existing digital twin technology typically adopts a technical approach of "physical entity - data acquisition - virtual model - state mapping". By deploying sensing devices such as position sensors, attitude sensors, pressure sensors, flow sensors and electrical monitoring devices on physical equipment, real-time data on the equipment's operating status is collected and transmitted to the digital space through a communication network. The virtual model then maps, displays and analyzes the state of the physical entity.

[0004] For example, patent CN115267938A, "A Digital Twin Real-Time Monitoring System for Deep-Sea Mining Equipment," constructs a physical space layer composed of a sensor network, a data space layer for data storage and management, a central monitoring system, and a twin system layer. Through data interaction between the physical entity and the digital model, it achieves monitoring and visual management of the operational status of deep-sea mining equipment. To improve the consistency between the digital twin model and the real equipment, existing research further introduces state estimation and data fusion techniques. These methods typically employ algorithms such as Kalman filtering, extended Kalman filtering, unscented Kalman filtering, and particle filtering to fuse the system model's prediction results with real-time observation data, thereby reducing the impact of measurement noise and model errors. Another example is patent CN120705999A, "A Digital Twin Method for Predicting the Navigation Attitude of Unmanned Surface Vessels," which acquires real-time sensor fusion data from unmanned surface vessels, combines it with an improved unscented Kalman filtering algorithm to dynamically correct the digital twin model, and uses the corrected model to predict future navigation states.

[0005] With the development of digitization technology for complex equipment, co-simulation technology is increasingly being applied to the construction of digital twin systems. Existing research typically employs technical standards such as FMI or HLA to integrate kinematic, dynamic, structural, hydraulic, control, and environmental models, enabling collaborative computation across multiple domains. Co-simulation can describe the coupling relationships between different subsystems, improving the state analysis and prediction capabilities of complex systems. Currently, this approach has been applied in fields such as shipbuilding, aerospace, industrial equipment, and marine engineering, and is gradually expanding to deep-sea resource development systems.

[0006] Overall, existing technologies can achieve operational status monitoring, data fusion analysis, and multi-model joint simulation of deep-sea resource development systems. However, these technologies employ a single timescale operation mode, supporting only post-event alarms and lacking the ability to extrapolate future operating conditions, thus failing to predict potential risks in advance. They also rarely consider the differences in data sampling frequencies and model calculation step sizes among various subsystems in deep-sea resource development, compounded by communication latency and data out-of-order issues, making it difficult to align multi-source data in time, reducing data fusion and simulation accuracy. Furthermore, the fixed fidelity of prediction models, the high computational cost of multi-domain coupled simulations, and the low efficiency of long-term predictions make it impossible to balance prediction accuracy and real-time performance. Therefore, a digital twin risk early warning method for deep-sea resource development that can address these issues is urgently needed. Summary of the Invention

[0007] This invention proposes a digital twin risk early warning method for deep-sea resource development, aiming to solve the problems of inconsistent time series of existing multi-source data, difficulty in effectively synchronizing real-time simulation and measured data, and lack of future risk prediction capabilities.

[0008] The technical solution of the present invention is as follows: A digital twin risk early warning method for deep-sea resource development includes the following steps: S100: Establish a unified standard time grid and map multi-source heterogeneous measured data to the unified standard time grid to achieve time alignment of multi-source heterogeneous data; S200: Construct real-time digital twin instances and ultra-real-time prediction instances. Real-time digital twin instances are used to synchronously reproduce the current physical system operating state, while ultra-real-time prediction instances are used to predict the future operating state of the system. Ultra-real-time prediction instances copy the latest state of real-time digital twin instances as the initial value for iteration and continuously predict the future state sequence. S300: Risk identification is performed based on future state sequences, and a future risk timeline is generated; S400: Simultaneously displays the current digital twin status, future predicted status, and future risk evolution process through a twin visualization interactive interface.

[0009] Furthermore, in step S100, the deep-sea data acquisition system and operations are acquired. The system collects heterogeneous operational data from various sources, including lifting systems, offshore platforms, and the marine environment. It performs time-series completion, interpolation resampling, and time alignment preprocessing on the raw data. Unified standard time grid is , In the formula, This is the initial time point. For the first There are discrete standard time nodes, with an interval of between adjacent time nodes. .

[0010] Furthermore, in step S200, each round of co-simulation iteration is based on the current reference grid time. The system drives the real-time digital twin instance to advance, calculates the simulation prediction value of the next simulation grid time, collects the corresponding measured data within the adaptive time delay buffer window, and completes the state update of the digital twin at the current time by performing state fusion or compensating state completion based on the integrity of the measured data.

[0011] Furthermore, the data queue is maintained in ascending order according to the unified standard time grid in step S100, the cache capacity is dynamically adjusted according to the real-time data traffic, and expired data earlier than the current computing node is cleaned up periodically; the operating system wall clock time when the measured data arrives at the platform is recorded. Calculate data transmission latency All delay samples are aggregated to form a delay set. Through the time delay statistical function Adaptive calculation of latency tolerance window duration ; The delay statistics function uses quantile statistics to extract the 95th percentile delay by sorting the delay set in ascending order. And combined with the safety margin factor Calculate the dynamic delay buffer window , The value range is 1.1 to 1.5; The number of valid delay samples during system initialization is insufficient compared to the preset number. At that time, the dynamic delay buffer window is calculated using the mean statistical method; Once the number of valid samples reaches the preset number, the system switches to quantile value statistics mode and dynamically updates the delay buffer window by re-statistically collecting the delay set according to the preset update cycle. When a single delay sample is detected to deviate from the current delay distribution, i.e., when it is greater than the sum of the average delay and three times the standard deviation, the sample is marked as an abnormal delay sample and is not directly included in the current buffer window calculation.

[0012] Furthermore, if the observation data is complete within the buffer window, an extended Kalman filter is used to fuse and correct the measured observations and simulated predictions, generating a complete digital twin state for the current moment. If there are still missing observations after the buffer window expires, a compensated state is constructed by weighting the corrected simulated predictions with the historical fused state from the previous moment using the extended Kalman filter. The compensated fused state is then used for the current moment. In the formula , The weighting coefficients corresponding to the predicted state and the historical fusion state are respectively, and satisfy the following conditions: The state update is completed by compensating for missing observations.

[0013] Furthermore, after completing the single-moment state fusion, the timing of the cache queue data is verified, and the simulation timing lag deviation is calculated. In the formula This is the standard grid time corresponding to the latest timestamp of the measured data in the queue. This refers to the grid time corresponding to the end of this simulation round, when a large number of timestamps remain in the cache queue later than [the specified time]. Subsequent measured data caused When the timing deviation continues to increase and exceeds the preset threshold, the system automatically switches to ultra-real-time catch-up mode; Acceleration coefficient in catch-up mode The time-series deviation solution function is calculated to satisfy the following conditions: The constraint interval is ,in The function for solving the acceleration coefficient, The system is preset with a maximum acceleration ratio, and the simulation duration covers the actual operating conditions in a single run. Keeping it unchanged, by reducing the time required for a single simulation calculation. Achieving ultra-real-time catch-up, the actual time consumption of a single operation meets the requirements. The system adjusts in real time according to... The reduction range is dynamically adjusted to lower the acceleration coefficient. Until the data backlog in the cache queue is cleared and the timing deviation returns to the normal range, then... k Automatically reset to 1, the system resumes standard real-time simulation operation mode.

[0014] Furthermore, when the ultra-real-time prediction instance starts rolling prediction, it copies the latest fused state snapshot of the real-time digital twin instance as the initial value for iteration. Combining the system control input, the marine environmental disturbance input, and the selected simulation model, it iteratively extrapolates the future state, continuously iterating until the preset prediction duration is fully output. The corresponding sequence of all future states; once a single segment prediction is completed, the prediction instance immediately rereads the latest state snapshot of the real-time digital twin, starts a new round of complete duration simulation, and continuously updates the prediction results of future operating conditions.

[0015] Furthermore, the overall preset prediction duration Operating risk level Current computing resource reserves Three types of indicators are adaptively matched to simulation models with different fidelity levels. The model selection function is expressed as follows: ; The system has a built-in three-level differential fidelity model set. ,in A reduced-order multiphysics fully coupled model is adopted, which simultaneously considers the kinematics, hydraulic dynamics, structural stress and strain of deep-sea resource development equipment and the multiphysics coupling process of marine environmental disturbances, and reduces the order of high-frequency non-critical physical components. A method combining equipment kinematics and data-driven surrogate models is adopted. With equipment kinematics as the main body, the surrogate model is trained based on offline generated high-fidelity joint simulation samples to replace the multi-physics field solution of hydraulic system and structural response, while retaining the main dynamic evolution characteristics. A simplified rigid body dynamics model is adopted, which retains the macroscopic motion state of the equipment's position, attitude, and velocity, and eliminates hydraulic, structural, and multiphysics coupling calculations. The system sets the first prediction duration threshold. Second prediction duration threshold ,in Based on the prediction duration, a preliminary model is selected: when High-fidelity model used ,when Medium-fidelity model used ,when Low-fidelity model used ; Based on this, when the comprehensive risk level When the preset risk threshold is reached, the system improves the fidelity of the primary model, provided that the current computing resources are available. When the computational resource threshold is lower than the preset threshold, the fidelity of the model will be automatically reduced by one level while ensuring the real-time performance of the prediction.

[0016] Furthermore, in step S300, the current fused digital twin status is read to carry out real-time risk identification, and the future status time sequence is read to carry out future risk warning. Multiple risk indicators are constructed around structural stress, hydraulic load, floating body attitude, marine environment, and operational interference, and a comprehensive evaluation is carried out to form a comprehensive risk result at the current moment or future moment. Based on the comparison between the comprehensive risk result and the preset threshold, the risk status is divided into four levels: safe, attention, warning, and danger. When the input is the current status of the converged digital twin, the output includes the current risk level, risk type, and risk warning for real-time risk monitoring. When the input is a future predicted state time series, a future risk time series axis is generated, which records the risk level, risk type, evolution trend and duration range at each future moment, and outputs a risk curve to realize future risk early warning and early disposal.

[0017] Furthermore, in step S400, the twin visualization interactive interface supports free switching between three display modes: real-time operating condition monitoring, future state prediction, and dual-window synchronous comparison. The system includes a real-time operating condition interface that intuitively displays the real-time operating parameters and spatial status of each piece of equipment; a future simulation interface that dynamically displays the system's future time-series evolution curve and risk change trends; and a dual-window comparison interface that presents the equipment's current real operating condition and predicted future state in parallel. The real-time equipment status is represented by opaque virtual entities, and the predicted state is represented by transparent virtual entities. The system also includes time-series storage and global scheduling, which centrally stores all time-series data from sensing, simulation, prediction, risk, and historical data. It provides a unified data access interface, coordinates data interaction and task scheduling among modules, and ensures the stable and collaborative operation of the entire system with a globally unified time benchmark.

[0018] The beneficial effects of this invention are as follows:

[0019] 1. This invention constructs a unified time grid and delay buffer timing control system, combined with a parallel simulation architecture of real-time digital twin instances and ultra-real-time prediction instances, which solves the problems of asynchronous, out-of-order, and transmission delay of multi-source sensor data, and realizes the unification of real-time reproduction of on-site working conditions and advanced prediction of future states; at the same time, relying on state snapshot inheritance, rolling prediction, and multi-fidelity adaptive models, it balances prediction accuracy and computational efficiency.

[0020] 2. This invention is based on long-term time-series prediction output to quantitatively identify various operational risks and generate a risk timeline, realizing the transformation from post-event alarms to pre-event proactive warnings, and greatly improving the safety and risk management capabilities of deep-sea resource development operations.

[0021] 3. This invention presents the current real working condition and predicted future state of the equipment in parallel through a dual-window comparison interface, distinguishing between opaque and transparent virtual entities, intuitively displaying the equipment status and risk evolution trend, reducing the difficulty of judgment for operators, and improving the level of intelligent decision-making in on-site operations.

[0022] In summary, this invention can effectively improve the temporal consistency, state synchronization capability, future prediction capability, and risk early warning capability of digital twin systems for deep-sea resource development, and has good engineering application value. The overall technical effect of this invention comes from the dual-timescale joint simulation mechanism and the coordinated operation of each module. Attached Figure Description

[0023] Figure 1 This is a schematic diagram of the overall module relationships; Figure 2 This is a schematic diagram of the simulation timing control process; Figure 3 This is a flowchart illustrating the ultra-real-time prediction instance submodule. Figure 4 This is a schematic diagram of a multi-scale risk identification and early warning process. Detailed Implementation

[0024] The technical solutions of the present invention will be clearly and completely described below with reference to the embodiments of the present invention. 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 of ordinary skill in the art without creative effort are within the scope of protection of the present invention.

[0025] Example 1

[0026] like Figure 1-4 As shown, this embodiment provides a digital twin risk early warning method for deep-sea resource development based on dual-timescale co-simulation, including six functional modules: sensing and acquisition preprocessing, simulation time series control, dual-timescale co-simulation (including a real-time digital twin instance submodule and an ultra-real-time prediction instance submodule), multi-scale risk identification and early warning, twin visualization interaction, and time series storage and global scheduling. This embodiment uses the dual-timescale co-simulation mechanism as the core operating mechanism of the system, and all modules operate collaboratively around this mechanism.

[0027] In this embodiment, the sensing and acquisition preprocessing module acquires information about the deep-sea acquisition system and operations. The system collects heterogeneous operational data from various sources, including lifting systems, offshore platforms, and the marine environment. Due to differences in equipment sampling frequencies, communication, and clock references, the raw data commonly suffers from uneven sampling, timing shifts, and data asynchrony. This module performs timing completion, interpolation resampling, and time alignment preprocessing on the raw data, mapping all measured data to a unified global standard time grid. middle. In the formula: This is the initial time point; For the first Each is a discrete standard time node.

[0028] The simulation timing control module receives the output from the sensing acquisition and preprocessing module, which has been aligned to the standard time grid. Real-time measured data is synchronously recorded and transmitted to the platform's operating system wall clock time. Calculate data transmission latency The system continuously aggregates all delay samples to form a delay set. Then through the time delay statistical function Adaptive calculation of latency tolerance window duration This is used to distinguish between normal transmission delays and permanent data loss. The delay statistics function uses quantile statistics to calculate the dynamic delay buffer window: the system sorts the delay set in ascending order and extracts the 95th percentile delay. And combined with the safety margin factor The dynamic delay buffer window is calculated. .in, This is the safety margin factor, with a value ranging from 1.1 to 1.5, for example, 1.2 can be used.

[0029] When the number of valid delay samples during system initialization is less than the preset number When there are 100 groups (e.g., in the case of 100 groups), the system automatically calculates the dynamic latency buffer window using the mean statistical method. As the number of latency samples continues to increase, once the number of valid samples reaches a preset number, the system automatically switches to quantile statistical mode and recalculates the latency set according to a preset update cycle, dynamically updating the latency buffer window. For example, the dynamic latency buffer window is updated every 200-500 new latency samples. If a single latency sample is detected to significantly deviate from the current latency distribution, for example, exceeding the sum of the average latency and three times the standard deviation, then the sample is marked as an abnormal latency sample and is not directly included in the current buffer window calculation. After the dynamic latency buffer window is calculated, it serves as a unified timing reference for the dual-timescale co-simulation mechanism, providing a consistent time base for the state fusion of real-time digital twin instances, the state snapshot replication of ultra-real-time prediction instances, and rolling prediction. The system follows... The data queue is maintained in ascending order, the cache capacity is dynamically adjusted according to real-time data traffic, and expired data that is older than the current computing node is cleaned up periodically.

[0030] Each round of co-simulation iteration is based on the current baseline grid time. The real-time digital twin instance submodule drives the instance to fix real-world operating conditions according to a unified time grid. Forward deduction ( (The interval between adjacent time nodes) is used to calculate the next simulation grid. Simulation predictions The system uses an adaptive buffer window. Collection within the scope The measured data corresponding to the time; if within the window time limit With complete real-time observation data, an extended Kalman filter is used to fuse and correct the measured observations and simulated predictions, generating... The complete digital twin state at all times, and Set this as the reference time for the next simulation; if there are still missing observations after the buffer window expires, then rely on the extended Kalman filter to... Corrected and merged with the historical state of the previous moment. Weighted construction of the compensated state. Current state after compensation and fusion. In the formula , The weighting coefficients corresponding to the predicted state and the historical fusion state are respectively, and satisfy the following conditions: The state update is completed by using a compensating state to fill in missing observations, thus avoiding interruption of the simulation process.

[0031] After completing the single-moment state fusion, the system verifies the timing of the data in the buffer queue and calculates the simulation timing lag deviation. In the formula This is the standard grid time corresponding to the latest timestamp of the measured data in the queue. This refers to the grid time corresponding to the end of this simulation round. When a large number of timestamps remain in the cache queue later than... When subsequent actual measurement data is obtained, it will directly cause As the timing deviation continues to increase, once it exceeds a preset threshold, the system automatically switches to ultra-real-time catch-up mode. The acceleration coefficient in catch-up mode... The time-series deviation solution function is calculated to satisfy the following conditions: Constraint interval ,in The function for solving the acceleration coefficient, The system is preset with the maximum acceleration factor. The simulation duration covers the actual operating conditions in a single run. Keeping it unchanged, by reducing the time required for a single simulation calculation. Achieving ultra-real-time catch-up, the actual time consumption of a single operation meets the requirements. The system adjusts in real time according to... The reduction range is dynamically adjusted to lower the acceleration coefficient. This continues until the data backlog in the cache queue is cleared and the timing deviation returns to the normal range, and then... Automatically reset to 1, the system resumes standard real-time simulation operation mode.

[0032] Example 2 Based on Example 1, the dual-timescale co-simulation module in this example includes a real-time digital twin instance submodule and an ultra-real-time prediction instance submodule.

[0033] The real-time digital twin instance submodule is used to reproduce the current physical system's operating state in real time. Its runtime sequence is synchronized with the actual operation sequence and is uniformly scheduled and executed by the simulation timing control module.

[0034] The ultra-real-time prediction instance submodule is used to infer the future operating status of the system and predict potential operating risks. With the support of the selected simulation model, it completes the working condition simulation at the maximum computing speed allowed by the model, and the overall computing speed is higher than that of the actual operation sequence.

[0035] When initiating rolling prediction, the prediction instance copies the latest fused state snapshot of the real-time digital twin instance as the initial value for iteration, and iteratively extrapolates the future state by combining system control input, marine environmental disturbance input, and the selected simulation model, continuing to iterate until the preset prediction duration is fully output. The corresponding sequence of all future states When a single segment Once all duration prediction calculations are completed, the prediction instance immediately rereads the latest state snapshot of the real-time digital twin, initiating a new round of complete duration extrapolation and continuously updating the future operating condition prediction results.

[0036] Example 3 Based on the above embodiments, this embodiment, in order to avoid the inability to update the initial prediction value in a timely manner due to the excessive computational time consumed by the high-fidelity model in long-duration prediction tasks, the system comprehensively presets the prediction duration. Operating risk level Current computing resource reserves Three types of indicators adaptively match simulation models of different fidelity. The system has a built-in set of three levels of differentiated fidelity models. .in, A reduced-order multiphysics fully coupled model is adopted, which simultaneously considers the kinematics, hydraulic dynamics, structural stress and strain of deep-sea resource development equipment and the multiphysics coupling process of marine environmental disturbance. While ensuring the integrity of the main physical mechanisms, the high-frequency non-critical physical components are reduced in order, which is suitable for short-time, high-risk and high-precision prediction scenarios. A method combining equipment kinematics and data-driven surrogate models is adopted. The equipment kinematics is the main body, and the surrogate model is trained based on a large number of offline high-fidelity joint simulation samples. The surrogate model is used to replace the multiphysics field solution with large computational load such as hydraulic system and structural response. Only the main dynamic evolution characteristics are retained, which takes into account both prediction accuracy and computational efficiency. It is suitable for medium prediction time and conventional risk scenarios. A simplified rigid body dynamics model is adopted, which only retains the macroscopic motion states such as equipment position, attitude, and velocity, and no longer performs hydraulic, structural and multiphysics coupling calculations. It completes long-term prediction with the lowest computational cost and is suitable for scenarios with long prediction time and limited computing resources.

[0037] When a prediction instance starts, the system determines the prediction duration. Current overall risk level and remaining computing resources Automatically select model fidelity; the model selection function is expressed as follows: The system sets a first prediction duration threshold. Second prediction duration threshold (in ), and initially select the model based on the prediction duration: when At that time, a high-fidelity model was used. ,when At that time, a medium-fidelity model was adopted. ;when At that time, a low-fidelity model was used. .

[0038] Based on this, the system further combines risk level and computing resources for dynamic adjustments. When the overall risk level... When the preset risk threshold is reached, the system increases the fidelity of the primary model; when the current computing resources are available... When the data is below a preset computing resource threshold, the model fidelity is automatically reduced by one level while ensuring real-time prediction, so as to achieve a dynamic balance between prediction accuracy, prediction efficiency and computing resources.

[0039] Example 4

[0040] Building upon the above embodiments, the multi-scale risk identification and early warning module in this embodiment performs unified risk identification and graded early warning for the current operating status and future predicted status of the deep-sea resource development system. On one hand, the module reads the current fused digital twin status output by the simulation time-series control module to conduct real-time risk identification; on the other hand, it reads the future state time-series output by the ultra-real-time prediction instance submodule to conduct future risk early warning. This module constructs multiple risk indicators around typical risk items such as structural stress, hydraulic load, floating body attitude, marine environment, and operational interference, and comprehensively evaluates these indicators to form a comprehensive risk result for the current or future time. Based on the comparison between the comprehensive risk result and a preset threshold, the system classifies the risk status into four levels: safe, concerning, warning, and dangerous. When the input is the current fused digital twin status, the module outputs the current risk level, risk type, and risk warning for real-time risk monitoring; when the input is the future predicted state time-series, the module generates a future risk time-series axis, recording the risk level, risk type, evolution trend, and duration interval for each future time, and outputs a risk curve for future risk early warning and proactive handling.

[0041] Example 5

[0042] Based on the above embodiments, the twin visualization interaction module in this embodiment receives output data from the simulation timing control module, the ultra-real-time prediction instance sub-module, and the multi-scale risk identification and early warning module to complete the twin rendering and human-computer interaction functions for the entire deep-sea resource development operation scenario. This module supports free switching between three display modes: real-time operating condition monitoring, future state simulation, and dual-window synchronous comparison. The real-time operating condition interface intuitively displays the real-time operating parameters and spatial status of each piece of equipment; the future simulation interface dynamically displays the system's future time-series evolution curve and risk change trends; the dual-window comparison interface presents the equipment's current real operating condition and predicted future state in parallel, using opaque virtual entities to represent the real-time equipment state and transparent virtual entities to represent the predicted state, facilitating operators' intuitive comparison of current operating conditions and potential evolution patterns, achieving visible risks and controllable operations.

[0043] The time series storage and global scheduling module centrally stores all time series data from sensing, simulation, prediction, risk, and historical data, provides a unified data access interface, coordinates data interaction and task scheduling among various modules, and ensures the stable and collaborative operation of the entire system with a globally unified time base.

[0044] The above alternative implementation methods do not depart from the technical concept of dual-timescale joint simulation, digital twin state synchronization, and future risk warning of the present invention, and all fall within the protection scope of the present invention.

[0045] The above are merely preferred embodiments of the present invention and are 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 risk early warning method for deep-sea resource development, characterized in that, Includes the following steps: S100: Establish a unified standard time grid and map multi-source heterogeneous measured data to the unified standard time grid to achieve time alignment of multi-source heterogeneous data; S200: Construct real-time digital twin instances and ultra-real-time prediction instances. Real-time digital twin instances are used to synchronously reproduce the current physical system operating state, while ultra-real-time prediction instances are used to predict the future operating state of the system. Ultra-real-time prediction instances copy the latest state of real-time digital twin instances as the initial value for iteration and continuously predict the future state sequence. S300: Risk identification is performed based on future state sequences, and a future risk timeline is generated; S400: Simultaneously displays the current digital twin status, future predicted status, and future risk evolution process through a twin visualization interactive interface.

2. The digital twin risk early warning method for deep-sea resource development according to claim 1, characterized in that, In step S100, the deep-sea data acquisition system and operations are obtained. The system collects heterogeneous operational data from various sources, including lifting systems, offshore platforms, and the marine environment. It performs time-series completion, interpolation resampling, and time alignment preprocessing on the raw data. Unified standard time grid is , In the formula, This is the initial time point. ( ) is the first There are discrete standard time nodes, with an interval of between adjacent time nodes. .

3. The digital twin risk early warning method for deep-sea resource development according to claim 1, characterized in that, In step S200, each round of co-simulation iteration is based on the current reference grid time. The system drives the real-time digital twin instance to advance, calculates the simulation prediction value of the next simulation grid time, collects the corresponding measured data within the adaptive time delay buffer window, and completes the state update of the digital twin at the current time by performing state fusion or compensating state completion based on the integrity of the measured data.

4. The digital twin risk early warning method for deep-sea resource development according to claim 3, characterized in that, The data queue is maintained in ascending order according to the unified standard time grid in step S100. The cache capacity is dynamically adjusted according to the real-time data flow, and expired data earlier than the current computing node is cleaned up periodically. The operating system wall clock time when the measured data arrives at the platform is recorded. Calculate data transmission latency All delay samples are aggregated to form a delay set. Through the time delay statistical function Adaptive calculation of latency tolerance window duration ; The delay statistics function uses quantile statistics to extract the 95th percentile delay by sorting the delay set in ascending order. And combined with the safety margin coefficient Calculate the dynamic delay buffer window , The value range is 1.1 to 1.5; The number of valid delay samples during system initialization is insufficient compared to the preset number. At that time, the dynamic delay buffer window is calculated using the mean statistical method; Once the number of valid samples reaches the preset number, the system switches to quantile value statistics mode and dynamically updates the delay buffer window by re-statistically collecting the delay set according to the preset update cycle. When a single delay sample is detected to deviate from the current delay distribution, i.e., when it is greater than the sum of the average delay and three times the standard deviation, the sample is marked as an abnormal delay sample and is not directly included in the current buffer window calculation.

5. The digital twin risk early warning method for deep-sea resource development according to claim 3, characterized in that, If the observation data is complete within the buffer window, an extended Kalman filter is used to fuse and correct the measured observations and simulation predictions, generating a complete digital twin state for the current moment. If there are still missing observations after the buffer window expires, a compensated state is constructed by weighting the corrected simulation predictions with the historical fused state from the previous moment using the extended Kalman filter. The compensated fused state is then generated for the current moment. In the formula , The weighting coefficients corresponding to the predicted state and the historical fusion state are respectively, and satisfy the following conditions: The state update is completed by compensating for missing observations.

6. The digital twin risk early warning method for deep-sea resource development according to claim 3, characterized in that, After completing the single-moment state fusion, the timing of the data in the buffer queue is verified, and the simulation timing lag deviation is calculated. In the formula This is the standard grid time corresponding to the latest timestamp of the measured data in the queue. This refers to the grid time corresponding to the end of this simulation round, when a large number of timestamps remain in the cache queue later than [the specified time]. Subsequent measured data caused When the timing deviation continues to increase and exceeds the preset threshold, the system automatically switches to ultra-real-time catch-up mode; Acceleration coefficient in catch-up mode The time-series deviation solution function is calculated to satisfy the following conditions: The constraint interval is ,in The function for solving the acceleration coefficient, The system is preset with a maximum acceleration ratio, and the simulation duration covers the actual operating conditions in a single run. Keeping it unchanged, by reducing the time required for a single simulation calculation. Achieving ultra-real-time catch-up, the actual time consumption of a single operation meets the requirements. The system is based on real-time data. The reduction range is dynamically adjusted to lower the acceleration coefficient. Until the data accumulation in the cache queue is eliminated and the timing deviation returns to the normal range, k is automatically reset to 1, and the system resumes the standard real-time simulation operation mode.

7. The digital twin risk early warning method for deep-sea resource development according to claim 1, characterized in that, When the ultra-real-time prediction instance starts rolling prediction, it copies the latest fused state snapshot of the real-time digital twin instance as the initial value for iteration. Combining system control input, marine environmental disturbance input, and the selected simulation model, it iteratively extrapolates the future state and continues iterating until a complete output is generated for the preset prediction duration. The corresponding sequence of all future states; once a single segment prediction is completed, the prediction instance immediately rereads the latest state snapshot of the real-time digital twin, starts a new round of complete duration simulation, and continuously updates the prediction results of future operating conditions.

8. The digital twin risk early warning method for deep-sea resource development according to claim 7, characterized in that, Comprehensive preset prediction duration Operating risk level Current computing resource reserves Three types of indicators adaptively match simulation models with different fidelity levels. The model selection function is expressed as follows: ; The system has a built-in three-level differential fidelity model set. ,in A reduced-order multiphysics fully coupled model is adopted, which simultaneously considers the kinematics, hydraulic dynamics, structural stress and strain of deep-sea resource development equipment and the multiphysics coupling process of marine environmental disturbances, and reduces the order of high-frequency non-critical physical components. A method combining equipment kinematics and data-driven surrogate models is adopted. With equipment kinematics as the main body, the surrogate model is trained based on offline generated high-fidelity joint simulation samples to replace the multi-physics field solution of hydraulic system and structural response, while retaining the main dynamic evolution characteristics. A simplified rigid body dynamics model is adopted, which retains the macroscopic motion state of the equipment's position, attitude, and velocity, and eliminates hydraulic, structural, and multiphysics coupling calculations. The system sets the first prediction duration threshold. Second prediction duration threshold ,in Based on the prediction duration, a preliminary model is selected: when High-fidelity model used ,when Medium-fidelity model used ,when Low-fidelity model used ; Based on this, when the comprehensive risk level When the preset risk threshold is reached, the system increases the fidelity of the primary model, provided that the current computing resources are available. When the computational resource threshold is lower than the preset threshold, the fidelity of the model will be automatically reduced by one level while ensuring the real-time performance of the prediction.

9. The digital twin risk early warning method for deep-sea resource development according to claim 1, characterized in that, In step S300, the current fused digital twin status is read to carry out real-time risk identification, and the future status time sequence is read to carry out future risk warning. Multiple risk indicators are constructed around structural stress, hydraulic load, floating body attitude, marine environment, and operational interference, and a comprehensive evaluation is carried out to form a comprehensive risk result at the current moment or future moment. Based on the comparison between the comprehensive risk result and the preset threshold, the risk status is divided into four levels: safe, attention, warning, and danger. When the input is the current status of the converged digital twin, the output includes the current risk level, risk type, and risk warning for real-time risk monitoring. When the input is a future predicted state time series, a future risk time series axis is generated, which records the risk level, risk type, evolution trend and duration range at each future moment, and outputs a risk curve to realize future risk early warning and early disposal.

10. The digital twin risk early warning method for deep-sea resource development according to claim 1, characterized in that, In step S400, the twin visualization interactive interface supports free switching between three display modes: real-time operating condition monitoring, future state prediction, and dual-window synchronous comparison. The system includes a real-time operating condition interface that intuitively displays the real-time operating parameters and spatial status of each piece of equipment; a future simulation interface that dynamically displays the system's future time-series evolution curve and risk change trends; and a dual-window comparison interface that presents the equipment's current real operating condition and predicted future state in parallel. The real-time equipment status is represented by opaque virtual entities, and the predicted state is represented by transparent virtual entities. The system also includes time-series storage and global scheduling, which centrally stores all time-series data from sensing, simulation, prediction, risk, and historical data. It provides a unified data access interface, coordinates data interaction and task scheduling among modules, and ensures the stable and collaborative operation of the entire system with a globally unified time benchmark.

Images