Valve cold state early warning method and system based on digital twinning and counterfactual intervention

Abstract

The present application belongs to the technical field of valve cooling system state monitoring and testing, and relates to a valve cooling state early warning method and system based on digital twinning and counterfactual intervention. The present application simulates the running state of a physical valve cooling system by constructing a dynamic digital twin, applies a virtual disturbance signal to identify sensitive disturbances, and then generates a reversible reverse excitation signal to act on the actual system. By comparing the differences between the actual and virtual responses, the digital twin model is corrected to improve the prediction accuracy. The system uses the calibrated digital twin to simulate future state deduction and maintenance, generates predictive risk trends and recommended intervention strategies, and realizes accurate early warning and intelligent decision-making. The present application improves the test sensitivity and predictability, changes the correlation analysis to causal positioning, improves the accuracy and reliability of fault diagnosis, provides a quantitative basis and optimization strategy for equipment maintenance, and realizes the leap from state perception to intelligent decision-making.

Description

Technical Field

[0001] This invention belongs to the field of valve cooling system status monitoring and testing technology, and relates to a valve cooling status early warning method and system based on digital twin and counterfactual intervention. Background Technology

[0002] The valve cooling system is a core auxiliary device for HVDC converter valves. Its main function is to remove the heat generated by the converter valves and other power components during operation through circulating cooling media, ensuring that the converter valves operate within their allowable temperature range. This system typically consists of a main circulating pump, heat exchanger, filter, piping valves, and a series of sensors and control units. Its stable and reliable operation directly affects the safety of the entire power transmission project. Therefore, continuous condition testing and health assessment of the valve cooling system are crucial.

[0003] In existing technologies, the status monitoring and testing of valve cooling systems mainly rely on sensor threshold alarms and simple trend analysis. On-site, sensors for temperature, pressure, and flow are typically deployed at key points. When the monitored values ​​exceed pre-set safety thresholds, the system issues an alarm signal. Some systems also utilize historical operating data, analyzing the changing trends of various parameters through statistical methods, and providing alerts when the trends deviate abnormally. Furthermore, some preliminary digital solutions construct a visual interface for the system's operating status, aggregating and displaying real-time data from various sensors.

[0004] However, existing technical solutions have significant shortcomings in practical applications. Threshold-based alarms are a passive response mechanism, typically triggered only when a fault has already occurred or is about to occur, lacking early warning capabilities and failing to identify slow, gradual performance degradation. While statistical trend analysis methods possess some predictive ability, their analysis results are disconnected from the system's physical mechanisms, making it difficult to distinguish between normal operating condition fluctuations and true fault precursors, easily leading to false alarms or missed alarms, and failing to accurately pinpoint the root cause of the fault. Simple visualization solutions merely provide a "mirror image" of the equipment's current state, lacking in-depth analysis, diagnostic, and predictive capabilities. Summary of the Invention

[0005] In view of this, in order to solve the problems mentioned in the background technology, a valve cooling status early warning method and system based on digital twin and counterfactual intervention is proposed.

[0006] The objective of this invention can be achieved through the following technical solution: The first aspect of this invention provides a valve cooling status early warning method based on digital twin and counterfactual intervention, including: S1, acquiring real-time operating parameters and static model data of the physical valve cooling system.

[0007] S2. Based on real-time operating parameters and static model data, construct and drive a dynamic digital twin that operates synchronously with the physical valve cooling system.

[0008] S3. Apply a set of virtual perturbation signals to the dynamic digital twin to obtain a series of virtual system responses, and identify sensitive virtual perturbation signals from the virtual system responses that cause performance index changes to exceed the change threshold.

[0009] S4. Based on the sensitive virtual disturbance signal, generate a reversible reverse excitation signal that is used to counteract the effects of disturbance simulation in the physical system and whose operating parameters are limited.

[0010] S5. Apply the reversible reverse excitation signal to the controller of the physical valve cooling system and collect the actual system response corresponding to the physical valve cooling system.

[0011] S6. Compare the parts of the actual system response and the virtual system response that are related to the sensitive virtual disturbance signal, and generate system state deviation characteristics that quantify the difference between the two.

[0012] S7. Utilize the system state deviation characteristics to correct the parameters of the dynamic digital twin and generate a calibrated digital twin whose matching degree meets the convergence threshold.

[0013] S8. Perform future state projection and multi-scheme maintenance simulation on the calibrated digital twin to generate predictive risk trends and recommended intervention strategies.

[0014] S9. Combine predictive risk trends and recommended intervention strategies to generate early warning information.

[0015] The second aspect of the present invention provides a valve cooling status early warning system based on digital twin and counterfactual intervention, including: an operating parameter and model data acquisition module, which acquires the real-time operating parameters and static model data of the physical valve cooling system.

[0016] The dynamic digital twin construction driver module, based on real-time operating parameters and static model data, constructs and drives a dynamic digital twin that operates synchronously with the physical valve cooling system.

[0017] The sensitive virtual disturbance signal identification module applies a set of virtual disturbance signals to the dynamic digital twin to obtain a series of virtual system responses, and identifies sensitive virtual disturbance signals from the virtual system responses that cause performance index changes to exceed the change threshold.

[0018] The reversible reverse excitation signal generation module generates a reversible reverse excitation signal based on the sensitive virtual disturbance signal, which is used to counteract the effects of disturbance simulation in the physical system and has limited operating parameters.

[0019] The actual system response acquisition module applies the reversible reverse excitation signal to the controller of the physical valve cooling system and acquires the actual system response corresponding to the physical valve cooling system.

[0020] The system state deviation feature generation module compares the parts of the actual system response and the virtual system response that are related to the sensitive virtual disturbance signal, and generates system state deviation features that quantify the difference between the two.

[0021] The digital twin generation module uses system state deviation characteristics to correct the parameters of the dynamic digital twin and generate a calibrated digital twin with a matching degree that meets the convergence threshold.

[0022] The risk trend and intervention strategy generation module performs future state projection and multi-scenario maintenance simulation on the calibrated digital twin to generate predictive risk trends and recommended intervention strategies.

[0023] The early warning information generation module combines predictive risk trends with recommended intervention strategies to generate early warning information.

[0024] Compared with the prior art, the embodiments of the present invention have at least the following advantages or beneficial effects: (1) The present invention effectively realizes accurate early warning of latent faults in the system by constructing a closed-loop test mechanism of active detection and feedback calibration. This method does not rely on the explicit over-limit of parameters, but by applying small and safe physical stimuli and comparing the response differences between real and virtual models, it can discover the sub-health state of the equipment when the performance has not deteriorated significantly, thus significantly advancing the fault detection window and improving the sensitivity and predictability of the test.

[0025] (2) This invention provides a high-precision fault diagnosis capability based on causal inference. By identifying sensitive signals in virtual disturbances and designing physical world detection experiments based on these signals, this method establishes a clear logical chain from virtual hypotheses to physical verification. The generated system state deviation characteristics directly point to the root cause of equipment performance changes, transforming traditional correlation analysis into causal localization, thus improving the accuracy and reliability of fault diagnosis.

[0026] (3) This invention provides a quantitatively based optimization strategy for equipment maintenance by introducing adaptive model calibration and decision simulation. The system can self-correct the digital twin model based on the results of active detection, ensuring the accuracy of its prediction of the future behavior of physical entities. Based on this, the counterfactual intervention simulation can evaluate the long-term benefits of different maintenance measures, thereby outputting the optimal intervention plan, so that the test results can be directly transformed into high-value operation and maintenance decisions, realizing the leap from state perception to intelligent decision-making. Attached Figure Description

[0027] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0028] Figure 1 This is a schematic diagram of the method steps of the present invention.

[0029] Figure 2 This is a schematic diagram of the system structure connection of the present invention. Detailed Implementation

[0030] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0031] Please see Figure 1 The first aspect of the present invention provides a valve cooling status early warning method based on digital twin and counterfactual intervention, including: S1, acquiring real-time operating parameters and static model data of the physical valve cooling system.

[0032] It should be noted that this step aims to provide an absolutely reliable underlying device architecture benchmark and real-time state driving source for constructing a high-fidelity digital twin. To achieve accurate mapping from the physical system to the virtual space, the system must collect data from two dimensions: spatial structural attributes and dynamic temporal characteristics. The specific implementation process includes the following two sub-steps: 1) Acquisition and parsing of static model data: Static model data refers to the inherent geometric and physical boundary information of the valve cooling system itself, which does not change significantly during operation. In a specific embodiment of this invention, the system automatically extracts the required data from the engineering design drawing library, the product lifecycle management system, or the equipment factory data sheet through a data interface. Specifically, this includes: Spatial structural parameters: Extract geometric data containing the actual 3D topology of the equipment, including not only the inner diameter, wall thickness, length, bending radius, and elevation difference of each level of pipe, but also the impeller hydraulic model parameters of the main circulating pump, the flow channel cross-sectional area and plate assembly structural parameters of the heat exchanger, and the equivalent mesh count of the filter screen. Material thermophysical parameters: Extract the thermal conductivity, specific heat capacity, and density constants of the solid materials used in the equipment, such as stainless steel pipes and titanium plates of the heat exchanger; simultaneously, extract the basic physical property curves of the cooling medium, deionized water or water-alcohol mixture, at different reference temperatures, including density, dynamic viscosity, specific heat capacity, and thermal conductivity. After acquisition, the system parses and converts unstructured engineering drawings and tabular data into a standardized structural matrix that can be recognized by the digital twin engine, serving as the static skeleton for constructing the initial 3D multiphysics mesh model.

[0033] 2) Real-time Operating Parameter Acquisition and Preprocessing: Real-time operating parameters are dynamic time-series boundary conditions characterizing the current thermodynamic and hydrodynamic state of the valve cooling system. The system continuously acquires these parameters using a sensor network and data acquisition modules deployed at the physical valve cooling system site. Acquisition objects include: main inlet / outlet water temperature and branch circuit water temperature of the converter valve obtained through temperature sensors; absolute pressure of the main pump inlet / outlet and pressure difference across the heat exchanger and filter obtained through pressure transmitters; instantaneous flow rate of the main pipeline and branch pipelines obtained through flow meters; and real-time speed, operating current, power of the main pump frequency converter, and valve opening percentage of the electric regulating valve obtained through the electrical control cabinet. Communication and Transmission Protocol: After the analog signals from the field transmitters and the underlying PLC are digitized, the system uses a highly reliable industrial communication protocol to penetrate the industrial security gateway and pushes the data to the digital twin computing server in real time at a preset high-frequency sampling rate. Data Preprocessing: To eliminate electromagnetic interference and sensor zero drift at the site, the system incorporates a data cleaning algorithm after acquiring the real-time operating parameters. Anomaly spike noise is removed by using moving average filtering or Kalman filtering algorithms, and strict synchronization of sensor data at different locations is ensured at the same time section by using timestamp alignment technology.

[0034] S2. Based on real-time operating parameters and static model data, construct and drive a dynamic digital twin that operates synchronously with the physical valve cooling system.

[0035] In a specific embodiment of the present invention, constructing and driving a dynamic digital twin that operates synchronously with the physical valve cooling system includes: acquiring static model data that defines the spatial structure and material properties of the physical valve cooling system.

[0036] Based on static model data and the physical rules of fluid and heat transfer, an initial digital twin model describing the internal physical processes of the system is established.

[0037] Real-time operating parameters are input as boundary conditions into the initial digital twin model, and a dynamic digital twin is generated through continuous calibration using a state correction algorithm.

[0038] Specifically, the engineering objective of this process is to transform the physical entity of the valve cooling system into a computable virtual copy, providing a high-fidelity model foundation for subsequent state synchronization and early warning analysis. This process begins with acquiring the static data required for model construction. The system extracts the three-dimensional structural parameters of the physical valve cooling system from the engineering design database. These parameters precisely define the pipe's inner diameter, wall thickness, length, and bending radius, as well as spatial information such as the heat exchanger's flow channel structure and the pump impeller geometry. Simultaneously, the system retrieves material property parameters from equipment manuals and material libraries, including the density, specific heat capacity, thermal conductivity, and kinematic viscosity of the cooling medium, such as deionized water, at different temperatures, as well as physical constants such as the thermal conductivity of solid components like pipes and valve bodies.

[0039] Based on the acquired static data, the system initiates the engineering steps to construct an initial digital twin model. The engineering objective is to establish a mathematical model capable of solving the fluid flow and heat transfer processes within the system. The system generates a finite element mesh based on the three-dimensional structural parameters, discretizing the continuous fluid domain. Subsequently, material property parameters are assigned to the corresponding mesh elements, and pre-defined multiphysics coupling rules are loaded. These multiphysics coupling rules are essentially a set of Navier-Stokes equations and energy conservation equations, used to describe the transfer and exchange of momentum and heat within the system. Upon completion of this step, an initial digital twin model is formed that is structurally and physically identical to the physical device, but is not yet operational.

[0040] In a specific embodiment of the present invention, the initial digital twin model specifically refers to a high-fidelity three-dimensional numerical computing grid model that has been spatially discretized and assigned physical properties. Its essence is to transform the physical geometric structure and material thermal property parameters of the physical valve cooling system into a set of mathematical constraint boundaries and multiphysics field control equations that can be recognized by a computer through data mapping.

[0041] To achieve real-time state alignment between the virtual model and the physical system, thereby generating a dynamic digital twin with dynamic response capabilities, the system performs a model state synchronization operation. This operation inputs real-time operating parameters collected from the sensor network, such as inlet water temperature, outlet pressure, and main pump speed, as boundary conditions into the initial digital twin model. The model solver performs iterative calculations based on these real-time boundary conditions and internal physical rules. To ensure synchronization accuracy, a state correction algorithm is used in the synchronization process, the core of which can be expressed as: In this formula, This is the state vector of the model at the next time step, representing the complete distribution of the internal fields of the model, such as temperature and pressure. This is the state vector of the model at the current moment; The vector of real-time running parameters input at the current moment; function It represents the mathematical expression of the pre-defined multiphysics coupling rules, used to predict the natural evolution of the state; It is a vector of actual measured values ​​obtained from some key measuring points in the physical system; function From the complete state vector inside the model Extract the simulation values ​​corresponding to the physical measurement point locations; The correction gain matrix, obtained through calibration using historical data, is used to adjust the correction weights of the deviation between simulated and actual measured values ​​on the model state. Through continuous calculations of the formula, the state of the initial digital twin model is "pulled" towards the real state of the physical system in real time, ultimately generating a dynamic digital twin that accurately reflects the internal operating details of the physical system.

[0042] In a specific embodiment of the present invention, the function This represents a nonlinear state transition operator for multiphysics coupling in a valve-cooling system. Mathematically, it is a set of discretized numerical solutions to the Navier-Stokes equations and energy conservation equations under a finite volume method (FVM) grid. At the computational execution level, this function uses the full-physics state vector of the model at the current moment. As initial conditions, with real-time running parameters To solve the boundary conditions, by means of, etc. The pressure-velocity coupling algorithm performs numerical integration over a time step to accurately deduce the internal flow field and temperature field distribution of the system as it moves to the next moment under the drive of purely physical mechanisms, thereby achieving feedforward prediction calculation of the "natural evolution" of the state.

[0043] In a specific embodiment of the present invention, the gain matrix is ​​modified. Mathematically, it is a high-dimensional mapping operator used to solve the information fusion problem between the low-dimensional sparse sensor measurement space and the high-dimensional dense grid node physical field state space. Its adjustment mechanism for the correction weights is as follows: by calculating the actual measured values... The simulated observation values ​​of the model at the corresponding virtual measurement points The residuals between them, using a matrix The coefficient matrix values ​​are spatially inversely mapped to the residual and weighted by confidence level, i.e., when the sensor accuracy is higher than the model prediction... A larger weight is assigned to this deviation to strongly correct the internal flow field state; conversely, a smaller weight is assigned to suppress measurement noise interference. Regarding The acquisition of the matrix, 'obtained through historical data calibration', refers to the following: In the initial stage of system deployment, using historical data from the long-term normal operation of the collected physical equipment, offline statistical optimization algorithms such as maximum likelihood estimation or expectation maximization are used to quantitatively evaluate the sensor measurement noise covariance and the multiphysics model process noise covariance. Then, based on the aforementioned noise variance benchmark prior, the optimal gain matrix is ​​calculated or dynamically iterated to obtain the current optimal gain matrix. This ensures the mathematical convergence and physical reliability of the state correction process.

[0044] S3. Apply a set of virtual perturbation signals to the dynamic digital twin to obtain a series of virtual system responses, and identify sensitive virtual perturbation signals from the virtual system responses that cause performance index changes to exceed the change threshold.

[0045] In a specific embodiment of the present invention, a set of virtual perturbation signals is applied within a dynamic digital twin to obtain a series of virtual system responses, and sensitive virtual perturbation signals that cause performance index changes to exceed a change threshold are identified from the virtual system responses, including: generating a set of virtual perturbation signals representing progressive performance degradation of the device.

[0046] Virtual perturbation signals are applied one by one to the dynamic digital twin, and a series of performance degradation curves characterizing the system's performance response are simulated and plotted.

[0047] Calculate the slope of the performance degradation curve, and identify the virtual disturbance signals corresponding to the segments where the slope exceeds a preset slope threshold as sensitive virtual disturbance signals.

[0048] Specifically, the engineering objective of this step is to proactively test the sensitivity of a dynamic digital twin to different types of potential faults in a completely secure virtual environment by simulating a series of minor, progressive fault symptoms, thereby identifying the risk factors that have the greatest impact on the overall system performance. This process begins by pre-setting a series of progressive degradation scenarios for the input parameters of the dynamic digital twin. These scenarios are constructed based on historical fault data of the valve cooling system and an expert experience database, covering typical fault modes such as the decrease in main circulation pump efficiency due to wear, the reduction in heat exchanger heat transfer coefficient due to fouling, and the increase in filter differential pressure due to blockage. At the data structure level, the virtual disturbance signals are constructed as arithmetic or geometric correction sequences for specific physical parameters, such as the fouling thermal resistance coefficient of the heat exchanger, the equivalent roughness of the pipes, and the volumetric efficiency of the pump, with each sequence corresponding to a type of degradation. The corresponding model parameters are modified incrementally or incrementally in very small steps, such as 0.1% to 0.5% per step, forming a set of continuous input variable parameters that can be successively substituted into the calculation by the digital twin solver, serving as a virtual disturbance signal for systematic testing.

[0049] Subsequently, the system simulates progressive degradation scenarios within a dynamic digital twin. The engineering objective is to acquire performance response data at each minor degradation step and visualize it as performance degradation curves. For each progressive degradation scenario, the system iteratively performs the following operations: at each step of the degradation sequence, a corresponding virtual perturbation signal is applied to the dynamic digital twin, the model is run until it reaches a new steady state, and key performance indicators of the system at this point are recorded, such as the maximum heat exchanger outlet temperature and the total system pressure drop. A series of performance degradation curves are plotted with the degradation amount of each scenario on the x-axis and the corresponding key performance indicators on the y-axis. These curves visually demonstrate the gradual decline in system performance as a specific failure mode deepens.

[0050] Finally, the system needs to automatically identify vulnerable points from numerous performance degradation curves. The engineering objective is to quantitatively assess and filter out risk sources that, once they occur, will cause a sharp deterioration in the system's state. The system performs differential calculations on each performance degradation curve to extract its slope variation characteristics. The slope represents the sensitivity of the system's performance to the disturbance input. To achieve this, the system uses the following formula to calculate local sensitivity. , .in, Locality sensitivity indicates how drastic the system's performance response to a specific disturbance is; and These are the key performance index values ​​of two adjacent sampling points on the performance degradation curve, which are directly obtained through model simulation output; and The degradation values ​​for the two corresponding virtual disturbance signals are obtained from a preset scenario definition. The system calculates the degradation of the entire curve. The value is then compared with a preset threshold set based on the system's safety margin. When the calculated... When the value exceeds the preset threshold, it indicates that the system's response in this deterioration section is extremely drastic, entering a nonlinear degradation stage. The system will trigger the virtual disturbance signal corresponding to this high-slope section, automatically identify and mark it as a sensitive virtual disturbance signal, and use it as a key input for subsequent physical system flaw detection.

[0051] In a specific embodiment of the present invention, the degradation value and In the computational script of the digital twin, these are defined as discrete control variables representing the key underlying physical characteristic parameters of the system. "Obtained from a pre-defined scenario" refers to the input sequence constructed based on expert experience using constant, small differential step sizes for the aforementioned physical parameters; while key performance index values... and These are thermodynamic or hydrodynamic scalar extreme values ​​directly related to the red line of safe system operation. The specific process of 'obtaining directly through model simulation output' refers to: obtaining specific degradation values... After being input as fixed physical properties or boundary conditions into the dynamic digital twin, it drives the internal Navier-Stokes and energy equation solvers to perform iterative calculations until the residuals of each physical field meet the numerical convergence criteria. Then, it automatically captures and derives unique scalar values ​​from specific virtual grid nodes that have been pre-mapped with the spatial coordinates of the physical sensors, thereby ensuring that the calculation of the local sensitivity slope has a strict causal deduction physical logic.

[0052] In a specific embodiment of the present invention, the typical value of the preset threshold based on the system safety margin in system temperature sensitivity detection is usually set as follows: This refers to the slope limit of the ratio of the change in performance indicators to the degree of degradation. The specific basis for its setting lies in the safe operating margin of the physical valve cooling system and the critical point at which the system enters the nonlinear deterioration stage. Its core logic is to identify the type of potential fault by analyzing the differential changes in the performance degradation curve. Heat exchanger scaling or a decrease in main pump efficiency will cause the system to cross the steady decay zone and enter the range of drastic temperature or pressure deterioration. In this way, the corresponding high-slope disturbance signal is locked as the key risk point affecting reliability, and the system's vulnerable links are quantitatively screened.

[0053] S4. Based on the sensitive virtual disturbance signal, generate a reversible reverse excitation signal that is used to counteract the effects of disturbance simulation in the physical system and whose operating parameters are limited.

[0054] In a specific embodiment of the present invention, a reversible reverse excitation signal is generated based on the sensitive virtual disturbance signal to counteract the effects of disturbance simulation in the physical system, and the action parameters are limited. This includes: parsing the device degradation type and degradation amount contained in the sensitive virtual disturbance signal.

[0055] For each type and amount of equipment degradation, a compensation target is calculated to offset its simulated impact.

[0056] Based on the compensation target, a reversible reverse excitation signal is generated that acts on the controller of the physical valve cooling system, and whose adjustment range and duration are both within the safe range.

[0057] Specifically, the engineering objective of this step is to safely transform the virtual risks identified in the previous stage—i.e., sensitive virtual disturbance signals—into physical operation commands that can be applied to the real physical system for active detection. This process aims to generate accurate test signals that elicit a measurable response from the system without affecting its normal operation. This process begins by parsing the sensitive virtual disturbance signal. The system needs to extract the simulated equipment performance degradation type and degradation amount from the signal's data structure. For example, if the sensitive virtual disturbance signal corresponds to a scenario of "main circulation pump efficiency decreasing by 3%," then the system will analyze the equipment performance degradation type as "main pump efficiency decline," with a degradation amount of 3%.

[0058] Based on the analyzed equipment performance degradation type and amount, the system will query a pre-set safety compensation strategy library to determine the compensation target. This safety compensation strategy library is a knowledge base established during the system debugging phase, storing safe and reversible control and adjustment schemes for different failure modes. The query operation uses the degradation type and amount as an index to match the corresponding compensation action and target value in the library. Continuing with the example of main pump efficiency degradation, the query result might determine the compensation target as "to offset the 3% efficiency loss by increasing the pump's input power without changing the outlet flow rate." This compensation target is qualitative, providing a clear engineering direction for generating the final signal.

[0059] Finally, based on the determined compensation target, the system generates a reversible reverse excitation signal to achieve that target, with limited duration and amplitude. This signal is the actual control command applied to the physical system controller. Its amplitude and duration are strictly constrained within safe limits; for example, the amplitude is typically limited to 1% to 3% of the normal setpoint, and the duration is set to a short window of 30 to 120 seconds to ensure the system can immediately return to its original state after the test. The signal amplitude can be calculated using the following formula. .in, It is the adjustment amplitude of the final generated reversible reverse excitation signal, such as the increment of the inverter frequency; It is the degradation amount obtained from analysis; The compensation gain coefficients are obtained from the safety compensation strategy library and are specific to certain degradation types. These coefficients are predetermined through offline simulation or field calibration, and their dimensions and units ensure consistency in the physical meaning of both sides of the formula. The final output of this step is a directly executable reversible excitation signal containing specific adjustment parameters, amplitude, and duration.

[0060] In a specific embodiment of the present invention, the 'analysis' process of degradation amount refers to the steps of data feature extraction and dimensionless normalization of the unstructured sensitive virtual disturbance signal output by the digital twin engine at the algorithm level. At the data structure level, the sensitive virtual disturbance signal is encapsulated into a data dictionary containing 'device fault type labels' and 'absolute values ​​of specific physical field parameters that trigger mutations'. The system's analysis algorithm first reads the data dictionary according to the addressing engine, extracts the specific fault type to match the corresponding safety compensation strategy library; then, the algorithm extracts the absolute values ​​of the physical field parameters that trigger mutations, and retrieves the corresponding physical constants of the dynamic digital twin under the current health baseline state. By calculating the ratio of the absolute value of the difference between the two to the baseline constant, it converts it into a dimensionless relative degradation percentage. The relative degradation percentage is the degradation amount. This normalized analytical transformation mechanism not only eliminates the computational barriers between different physical dimensions, but also ensures that the results can be substituted into the subsequent linear compensation formula. At that time, it can accurately calculate the relative control adjustment range that is strictly proportional to the actual degree of degradation.

[0061] In a specific embodiment of the present invention, the compensation gain coefficient This refers to a conversion factor in the non-destructive detection strategy for "heat exchanger efficiency decline" faults, which linearly maps the amount of equipment degradation to the controller adjustment command. Its typical value is set to -3.33RPM / %; the value is based on offline simulation or historical real test. This coefficient is designed to ensure that the pump speed adjustment of -15RPM estimated based on the unmeasurable degree of heat exchanger degradation can be perceived in the actual operation of the physical system with a temperature response of about 0.3-0.5°C, but does not exceed the rated range of safe operation, which is usually within ±1%-3%. This enables minimally invasive and limited intervention in the system and is a calibration constant connecting latent degradation and explicit adjustment commands.

[0062] S5. Apply the reversible reverse excitation signal to the controller of the physical valve cooling system and collect the actual system response corresponding to the physical valve cooling system.

[0063] It should be noted that the purpose of this step is to bridge the physical gap between information technology (IT) and operational technology (OT), securely and accurately translating virtual intervention strategies calculated in the digital space into physical actions of underlying actuators in the physical world, and holographically recording the trajectory of physical field disturbances caused by these actions. To enable those skilled in the art to implement this step, the specific software and hardware interaction and implementation logic are as follows: First, at the level of signal transmission and hardware / software conversion architecture: The reversible reverse excitation signal generation module running on the computing server first encapsulates the calculated parameters into standardized industrial communication data packets. The system then sends these packets to the underlying controller, the programmable logic controller (PLC), at the valve cooling site via a unidirectional or highly encrypted industrial security isolation gateway supporting industrial standard communication protocols such as OPCUA. After parsing the data packet, the PLC's communication module maps its signature code and overwrites it into the corresponding register control word. Subsequently, the PLC replaces the original steady-state control signal with a control command containing disturbance offset, and outputs it to the specific physical actuator via a fieldbus or analog output card, thereby actually triggering micro-disturbance intervention in the physical loop.

[0064] Second, regarding the 'reversibility' of physical execution and the underlying security safeguard mechanism: To ensure the safety of core equipment such as converter valves, intervention actions are not only strictly constrained in terms of amplitude, but also incorporate a dual hardware-level foolproof protection mechanism based on time and threshold values ​​in the underlying control logic. When the PLC receives instructions, it not only receives the adjustment amplitude... At the same time, a specific time limit will be set. Write to the PLC's timer latch. Once the timer expires... If the signal is exhausted, or if any hard-connected sensor in the field detects a value approaching the first-level safety alarm threshold of the physical system during this period, the high-priority interrupt program at the PLC's underlying level will unconditionally take over control, immediately block interference commands from the external digital twin, and automatically revert the register control word back to the original steady-state setting value before this test. This forced state rollback, which can be automatically completed by the underlying hardware even in the event of a communication failure, constitutes the physical basis for the so-called "reversible" implementation of this invention.

[0065] Third, at the level of synchronous acquisition and data digitization of actual system responses: The acquisition of the actual system response corresponding to the physical valve cooling system does not involve obtaining static values ​​at a single moment. Instead, it involves acquiring high-density time-series data through a field sensor network at a sampling frequency higher than that of conventional steady-state monitoring. This acquisition window strictly covers three stages: the baseline steady-state period before the disturbance is triggered, the transient ramp-up period during the disturbance command, and the physical recovery period after the command is withdrawn. By aligning the timestamps of the Network Time Protocol (NTP), the system packages the action time points of the controlled equipment with the continuous change trajectories recorded by all sensors within the above three window periods into a multi-dimensional time-domain feature data matrix. This complete situational data, which includes delay time, change slope, and maximum offset, constitutes the actual system response used to characterize the true health status of the equipment.

[0066] S6. Compare the parts of the actual system response and the virtual system response that are related to the sensitive virtual disturbance signal, and generate system state deviation characteristics that quantify the difference between the two.

[0067] In a specific embodiment of the present invention, comparing the parts of the actual system response and the virtual system response that are related to the sensitive virtual disturbance signal, and generating system state deviation features that quantify the difference between the two, includes: extracting an actual response feature vector containing steady-state values ​​and dynamic characteristics from the actual system response.

[0068] Extract virtual response feature vectors from the virtual system response that have the same dimensions as the actual response feature vectors.

[0069] The weighted distance between the actual response feature vector and the virtual response feature vector is calculated to obtain the system state deviation feature.

[0070] Specifically, the engineering objective of this step is to accurately quantify the subtle differences between the predictions of the dynamic digital twin and the actual behavior of the physical system, and to transform these differences into structured information that can be used for model calibration. This process first requires converting the raw time-series response data into a standardized, comparable format, i.e., extracting the actual response feature vector from the actual system response and extracting the virtual response feature vector from the virtual system response corresponding to the sensitive virtual perturbation signal. In engineering terms, this means processing the time-series data such as temperature and pressure acquired during the application of the reverse excitation signal, calculating its steady-state values, response time, overshoot, peak value, and other key dynamic characteristics. Specifically, the response feature vector is defined as a multi-dimensional numerical array. ,in The difference in steady-state value before and after the excitation is applied. The time constant for the system response to reach 63.2% of its steady-state value. The percentage of dynamic peak overshoot. These characteristics, together, constitute the actual response feature vector, representing the settling time for the system to return to steady state. Similarly, the same time-domain feature extraction calculation is performed on the simulation results of the dynamic digital twin under virtual perturbations to generate a virtual response feature vector with strictly aligned dimensions.

[0071] Subsequently, the system performs a quantitative comparison, the engineering purpose of which is to calculate the degree of difference between the two feature vectors, obtaining a single index that can comprehensively evaluate the model's accuracy. The system uses weighted norm distance to calculate the difference between the actual response feature vector and the virtual response feature vector. Before calculating the difference metric, the system first uses historical benchmark data to perform dimensionless (or normalized) processing on each component of the actual and virtual response feature vectors to eliminate the interference of different physical dimensions (such as temperature, time, percentage) and numerical magnitude differences on the distance calculation results. The calculation formula can be expressed as follows: In this formula, The final calculated measure of difference is a dimensionless scalar value; It is the first in the actual response feature vector One component, such as the actual steady-state temperature value; It is the corresponding first element in the virtual response feature vector. Each component, such as the steady-state temperature value obtained from the simulation, is calculated through the preceding steps; It is the first The preset weighting coefficients for each component are used to adjust the importance of different feature components in the overall difference calculation. Their values ​​are set according to the degree of influence of the feature on system security.

[0072] In a specific embodiment of the present invention, the weighting coefficients here... This refers to the safety importance adjustment factor assigned to the physical response characteristics of the valve cooling system in the multidimensional difference measurement formula. Its typical sum is set to 1.0, targeting the most critical steady-state temperature rise characteristic in the system. The adjustment is based on the degree of impact of this characteristic on system safety; in engineering implementation, this is not a subjective assumption but must rely on equipment reliability engineering algorithms. During system operation and maintenance or digital twin initialization, the built-in algorithm calls the Failure Mode and Effects Analysis (FMEA) knowledge base and the Analytic Hierarchy Process (AHP) model. The system establishes an algorithm program containing a judgment matrix through the AHP model: inputting the probability statistics of each physical parameter deviation leading to converter valve shutdown from the historical fault database, calculating the maximum eigenvalue and corresponding eigenvector of the judgment matrix, and passing a consistency check (…). After that, the components of the feature vector are extracted as security weights for each physical feature. Based on this, It possesses dynamic adaptive adjustment capabilities. The algorithm defines a safety margin penalty function based on the current operating condition boundary: when the current absolute operating temperature of the physical system is extremely close to the safety protection red line, the fault tolerance rate decreases drastically, and the system will automatically trigger the weight update logic, using the penalty function to exponentially amplify the weight of the temperature characteristic component, while correspondingly compressing the weights of other secondary dynamic characteristics. This adjustment strategy, combining static AHP analysis with a dynamic boundary penalty mechanism, ensures that the calculated system state deviates from the characteristics. It exhibits extreme sensitivity and amplification of minute virtual-to-realistic discrepancies in critical safety conditions. Finally, the system makes a decision based on the calculation results. Its engineering purpose is to determine whether the deviation between the current model and reality exceeds the normal range and to indicate the nature of the deviation. The system will calculate the difference measure. The system compares the difference with a preset tolerance range, which is typically set based on the statistical distribution obtained from multiple tests of the system under healthy conditions, and its value may be between 0.02 and 0.05. When the difference exceeds this tolerance range, the system determines that there are latent state changes in the physical system that were not captured by the model or that there is measurement bias in the sensors. At this point, the system further analyzes the individual components that make up the difference measure. The sign and magnitude of the deviation are used to infer the physical root cause of the deviation based on its combined characteristics, ultimately generating a structured system state deviation feature. This feature not only includes the overall magnitude of the deviation but also indicates the physical dimension of the deviation, such as "low heat exchange efficiency response, by about 5%", providing a clear target and direction for the next step of accurate model calibration.

[0073] In one specific embodiment of the present invention, the system state deviation characteristic refers to a structured diagnostic report with quantitative and physical meaning, such as "low heat exchange efficiency response, approximately 5%", which includes not only scalar values ​​reflecting the total difference between virtual and real responses. It also clearly points out the specific physical performance dimension to which the deviation points; its construction is based on the actual response feature vector. With virtual response feature vector The system analyzes the composition of the weighted norm distance calculation results between them. The system identifies the signs and deviation ratios of each component, and the measured temperature rise is 66.7% higher than the simulation. This maps the mathematical residuals back to the real physical sources of degradation, providing physically-guided deviation feedback for the next step of precise calibration of twin parameters. The system infers the physical sources of deviation based on pattern matching between the built-in "multi-physics residual sign matrix" and the "expert causal reasoning rule base." First, the system assigns the residuals of each dimension of feature components (...) The sign and magnitude of the error are extracted into a multi-dimensional state combination vector. Subsequently, the inference engine maps this vector to a specific fault mechanism fingerprint. Based on the strong coupling constraints of fluid mechanics and heat transfer, the system eliminates factors such as insufficient cooling water pump output, accurately deducing that the physical root cause of the deviation is uniquely "fouling in the heat exchanger's internal flow channels or filter blockage." Conversely, if only a single point of temperature changes abruptly while the flow rate and pressure drop residuals are close to zero, the system determines it as "sensor measurement drift" based on the law of conservation of energy. Finally, the system encapsulates this clear physical root cause node, the derived deviation direction, and the initial value of the iteration step size set based on the absolute value of the residual into a structured data dictionary that can be directly parsed by a computer. This serves as the precise navigation input for the next step of directional high-fidelity parameter calibration of the digital twin.

[0074] S7. Utilize the system state deviation characteristics to correct the parameters of the dynamic digital twin and generate a calibrated digital twin whose matching degree meets the convergence threshold.

[0075] In a specific embodiment of the present invention, the parameters of the dynamic digital twin are corrected by utilizing the system state deviation characteristics to generate a calibrated digital twin with a matching degree that meets the convergence threshold. This includes: determining the model parameters to be adjusted in the dynamic digital twin based on the deviation direction and magnitude indicated in the system state deviation characteristics.

[0076] The model parameters are adjusted through an iterative optimization algorithm to minimize the weighted distance represented by the system state deviation feature.

[0077] When the weighted distance is less than the convergence threshold, the currently adjusted model parameters are locked, and a calibrated digital twin with a matching degree that meets the convergence threshold is generated.

[0078] In a specific embodiment of the present invention, the convergence threshold in the calculation of system state deviation matching degree is typically set to 0.01, which is usually the dimensionless scalar limit after quantization. Its value is based on the strict definition of the "high-fidelity" alignment between the physical entity and the digital copy. This threshold is set more finely than the tolerance range of the previous deviation judgment (such as 0.05). The purpose is to ensure that the model parameter correction amount found by the iterative optimization algorithm can make the virtual response of the digital twin when simulating the same excitation signal substantially consistent with the actual system response of the physical system, thereby locking in a set of high-confidence operating parameters that best represent the potential health status of the current physical entity.

[0079] Specifically, the engineering objective of this step is to utilize the discrepancies between the model and reality discovered in the previous stage to perform a closed-loop, goal-oriented self-correction on the dynamic digital twin, thereby making its intrinsic parameters more closely approximate the implicit real state of the physical entity. This process first initiates a reverse adjustment operation based on the system state deviation characteristics. The goal of this operation is to find a set of model parameter corrections that minimize the discovered deviations. The system then performs an iterative optimization based on the direction and magnitude of the deviations indicated in the system state deviation characteristics, combined with model parameters associated with sensitive virtual perturbation signals. This adjustment process can be guided by the following formula. In this formula, These are the model parameter values ​​for the next iteration; These are the current model parameter values, such as main pump efficiency or heat exchanger fouling factor; It is the error value after the system state deviates from the characteristic quantization, which is directly obtained from the output of the previous step; It is a preset adjustment gain coefficient, the value of which is pre-calibrated through offline sensitivity analysis of the model, and determines the adjustment step size for each iteration. It is usually between 0.1 and 0.5. The unit of this coefficient ensures the uniformity of the dimensions on both sides of the formula.

[0080] After each adjustment of the model parameters, the system immediately enters the verification phase, the engineering purpose of which is to verify whether the parameter adjustment is effective. The system reruns the dynamic digital twin using the adjusted model parameters and resimulates the previously applied sensitive virtual perturbation signal, obtaining a new virtual response. The system then calculates the degree of matching between this new virtual response and the actual system response. The degree of matching here is calculated using the same difference metric as in the previous claim, but its purpose is for convergence determination.

[0081] This adjustment and verification process continues in a loop until an exit condition is met. In engineering terms, this exit condition is that the matching degree meets a preset accuracy requirement, which is a threshold with a stricter tolerance range than the previous stage, such as 0.01. When the system determines that the matching degree is less than this accuracy requirement, it indicates that the model's adjusted behavior is highly consistent with physical reality within the allowable error range. At this point, the optimization loop terminates, the system locks the current set of fully verified model parameters, and officially updates the dynamic digital twin equipped with these parameters into a calibrated digital twin, serving as the final model for subsequent accurate prediction and decision analysis.

[0082] S8. Perform future state projection and multi-scheme maintenance simulation on the calibrated digital twin to generate predictive risk trends and recommended intervention strategies.

[0083] In a specific embodiment of the present invention, future state extrapolation and multi-scheme maintenance simulation are performed on the calibrated digital twin to generate predictive risk trends and recommended intervention strategies, including: extrapolating predictive risk trends based on the current parameters of the calibrated digital twin and in conjunction with a degradation model describing the change of equipment performance over time.

[0084] In the calibrated digital twin, a set of maintenance measures are simulated to generate the corresponding corrected risk trends.

[0085] By comparing the benefits of each revised risk trend with the implementation costs of each maintenance measure, a recommended intervention strategy including the target maintenance measures is generated.

[0086] Specifically, the engineering objective of this step is to utilize a precisely calibrated virtual model to perform forward-looking future state projections and multi-scenario decision simulations, thereby generating an optimal response strategy with quantitative evidence before the actual occurrence of a failure. This process first uses the current parameters of the calibrated digital twin to extrapolate its performance degradation curve to obtain predictive risk trends. The system uses calibrated equipment performance parameters, such as the current actual efficiency of the main pump, as initial values ​​at time zero and loads a degradation model based on the equipment's physical characteristics to predict its future changes. This extrapolation process can be expressed by the formula... To describe it. Among them, For the future Predictive performance parameters at each time step; The current parameters of the calibrated digital twin are obtained from the previous steps; To describe performance over time and operating conditions The degradation function changes, its form and coefficients based on device type and historical data presets. This is achieved by continuously calculating the degradation function at different future moments. The system calculates the value and inputs it into the twin for simulation. The system then plots a complete performance degradation curve. When the curve intersects with the preset safe operating threshold, the remaining effective lifespan of the system can be determined, forming a quantitative predictive risk trend.

[0087] It should be noted that the degenerate function At the underlying algorithm level, it is not an abstract concept, but rather a concrete entity represented as a nonlinear dynamic damage model with a condition-based acceleration factor. Its specific mathematical form is often set as an exponential decay function. ,in This represents the actual dynamic degradation rate. More importantly, it reflects the operating conditions. By introducing a quantified operating condition acceleration factor Compared with the base decline rate Direct mathematical multiplication coupling occurs (i.e.) This allows for the precise calculation of the rate at which equipment lifespan decays under harsh or mild operating conditions. Regarding its "form and coefficients being based on equipment type and historical data presets," this means that in the initial stages of system deployment, the algorithm accesses real-lifetime fault logs and historical operating condition test datasets of similar equipment. It then uses nonlinear least squares estimation or maximum likelihood estimation to iteratively optimize the residual loss function, accurately determining the optimal baseline degradation rate for that type of equipment. It is a constant matrix along with the thermal / mechanical stress sensitivity index, and is fixed as a factory preset value in the long life cycle simulation engine.

[0088] To further explain in detail the "operating conditions" in the above physical performance degradation model The specific composition of "and the "operating condition acceleration factor" The quantitative calculation method of "". In this invention, the operating conditions It is not a single abstract variable, but rather a set of parameters representing the multi-dimensional environmental and load pressures experienced by the physical valve cooling system during operation. Specifically, for the cooling system of the high-voltage direct current transmission converter valve, the operating conditions... It mainly includes the following three core physical state variable sets: heat load condition variables ( : Includes the average operating temperature of the main circulating medium (such as deionized water), the daily temperature fluctuation range (peak-valley difference), and the frequency of thermal shocks caused by frequent system start-ups and shutdowns. Mechanical stress condition variables ( ): Includes the long-term average operating frequency (speed percentage) of the main circulation pump, the working water pressure in the piping system, and the peak value of local high-frequency vibration acceleration caused by the pulsation of the pump outlet flow field. Water quality chemical operating condition variables ( This includes the real-time conductivity of the cooling medium, dissolved oxygen concentration, and particulate impurity content (typically indirectly characterized by the bypass filter pressure differential). Correspondingly, the quantified operating condition acceleration factor... It is a nonlinear aggregation function that transforms the aforementioned physical operating condition variables into dimensionless multipliers. Its physical meaning is: when the equipment is operating under its rated standard design conditions, When equipment is subjected to severe operating conditions such as overload, high temperature, or deteriorating water quality for an extended period, This indicates that the fatigue, wear, or scaling of components is accelerated; conversely, if the system is in a low-load dormant state for a long time, The decline is slowed down. At the underlying algorithm implementation level, The calculation is usually based on a combination of the Arrhenius model and the inverse power law model. Its multivariate aggregation calculation formula can be expressed as: in: and These are the real-time average operating absolute temperature and mechanical pressure extracted from the digital twin, respectively. and The standard rated temperature and rated pressure specified by the manufacturer when the equipment leaves the factory; The activation energy of the material Boltzmann's constant, The mechanical fatigue damage index (its value is derived from historical data regression calibration); A penalty function (such as an exponential function based on conductivity) is used to describe the nonlinear increase in scaling rate caused by water quality deterioration. The contribution weights of heat load, mechanical stress, and water quality chemical conditions to the overall degradation are respectively, and Through the aforementioned rigorous physicochemical empirical formulas, this system can accurately convert the chaotic macroscopic operating condition data (such as water temperature, rotational speed, and conductivity) collected by the sensors into a dynamically fluctuating multiplier. This, coupled with the basic decay rate, enables customized and high-precision extrapolation of the remaining lifespan of key components in the converter valve cooling system (such as water pump bearings and heat exchanger channels) in a "one-size-fits-all" manner.

[0089] In one specific embodiment of the present invention, the preset safe operating threshold is set to 2200 W / (m²) for the typical value of the heat transfer capacity of the valve cooling system in the extrapolation of predictive risk trends. 2·K); its value is based on the highest surface operating temperature that the physical system can allow, and the minimum heat dissipation performance requirement derived from the 80℃ safety alarm red line. When the total heat transfer coefficient value predicted by the digital twin loading device performance degradation model over time intersects with this threshold, it is predicted that the intersection point will be reached after 90 days, which means that the system will be unable to maintain safe cooling capacity at that point. Thus, the "remaining safe operating time (90 days)" of the physical device is accurately calculated, providing a clear time inflection point for early warning.

[0090] Next, the system simulates the corrective effects of various preset maintenance measures on the performance degradation curve within the calibrated digital twin. The engineering objective is to virtually execute different maintenance plans and assess their technical effectiveness in a risk-free manner. The system retrieves a series of feasible preset maintenance measures from a library of preset maintenance strategies, such as "cleaning the heat exchanger" and "replacing the pump impeller." For each measure, the system creates a "hypothesis" branch in the calibrated digital twin and modifies the model parameters directly related to that measure. For example, simulating "cleaning the heat exchanger" involves resetting its heat transfer coefficient parameter to the factory design value. Then, based on this corrected parameter, the system re-executes the aforementioned performance extrapolation process, generating a new, improved performance degradation curve. This set of new curves constitutes the corrective effect of each measure on the system's future performance.

[0091] Finally, the system generates a recommended intervention strategy by comparing the corrective effects and implementation costs of each preset maintenance measure. Its engineering purpose is to provide optimal decision-making for maintenance personnel through a comprehensive evaluation from both technical feasibility and economic perspectives. The system calculates the cost-effectiveness ratio for each maintenance measure using the following formula: .in, For the first The cost-effectiveness of each maintenance measure; It is the first The increase in the remaining effective lifespan of the system resulting from each maintenance measure is calculated by comparing the original and the corrected performance degradation curves. Is with the first The implementation costs of each maintenance measure, including spare parts costs and estimated downtime losses, are obtained from the operations and maintenance knowledge base. The system sorts all maintenance measures from highest to lowest cost-effectiveness and packages the highest-ranked maintenance measure, its expected corrective effect, and cost into a final recommended intervention strategy.

[0092] In one specific embodiment of the present invention, the increment of the system's remaining effective lifetime, "calculated by comparing the original and corrected performance degradation curves," is calculated using the following geometric and algebraic calculation process at the computer's underlying level: the system first uses the current moment... Starting from the time frame, the original degradation equation without any intervention is used. By using algebraic analytical expressions or numerical interpolation, the curve and the preset safe operating threshold constant can be determined. Precise time coordinates of the mathematical intersection Subsequently, the system extracts the simulation execution of the first... The modified post-degradation equation was regenerated after a maintenance measure. Using the same root-finding algorithm, the correction curve was found to fall below the safety threshold. New time coordinates Finally, the algorithm subtracts the x-coordinates of the two failed intersections on the time axis (i.e., This allows the abstract "curve comparison" to be precisely quantified as the actual extended safe operating time difference of the system due to the maintenance action.

[0093] S9. Combine predictive risk trends and recommended intervention strategies to generate early warning information.

[0094] In a specific embodiment of the present invention, combining predictive risk trends with recommended intervention strategies to generate early warning information includes: extracting the remaining safe operating time from the predictive risk trends and mapping it to a risk probability level.

[0095] The recommended intervention strategy is broken down into a series of specific maintenance actions with implementation priorities.

[0096] The remaining safe operating time, risk probability level, and specific maintenance actions are integrated to generate a structured early warning report as early warning information.

[0097] Specifically, the engineering objective of this step is to transform the complex analysis results generated in the preceding steps into a standardized information product that is intuitive, operable, and conforms to the work habits of operations and maintenance personnel, thereby ensuring the effective communication and execution of early warnings. This process first involves concretizing and quantifying the abstract predictive risk trends, converting them into remaining safe uptime and risk probability levels that are easily understood by operations and maintenance personnel. The system extracts the intersection of the performance degradation curve and the safety threshold from the predictive risk trends; the time corresponding to this intersection is the remaining safe uptime, typically in days or hours. Simultaneously, the system uses a pre-defined piecewise mapping function or lookup table method for risk conversion. For example, mapping rules are defined as follows: when the remaining time is less than 15 days, it is classified as "high risk (emergency shutdown maintenance)"; between 15 and 90 days, it is classified as "medium risk (monthly planned maintenance)"; and more than 90 days, it is classified as "low risk (normal monitoring)". This clearly maps the uptime to specific risk probability levels, providing a qualitative assessment of the risk severity.

[0098] Next, the system decomposes the recommended intervention strategy into an engineering approach. The engineering goal is to refine a high-level strategy recommendation into a series of specific, executable work instructions. The system breaks down the recommended intervention strategy, such as "prioritizing the main pump," into several maintenance actions arranged in a logical order. The system calls upon a pre-built maintenance procedure expert system based on a relational database, using the intervention strategy as the index key to query the action sub-table and automatically extract the underlying standard operating procedures (SOPs) subtasks, such as "applying for a shutdown window," "draining the primary circuit medium," and "disassembling the main pump head." For each maintenance action, the system correlates and extracts its corresponding expected effect from the knowledge base, such as "restore pump efficiency to over 95%," and a quantified implementation priority, which is directly derived from the cost-effectiveness ratio calculated previously.

[0099] Finally, the system integrates the processed information elements, aiming to generate a complete, logically clear, and key-focused final report, serving as the sole interface for user interaction. The system calls a preset report template, automatically filling in the designated areas with the remaining safe operating time, risk probability level, and recommended intervention strategies, including specific maintenance actions, expected effects, and implementation priorities. This process ensures the integrity and consistency of all critical information, ultimately outputting a standardized, structured early warning report. This report, presented in a visually appealing format, intuitively informs users "how much time is left," "how great the risk is," and "what should be done and why," thus completing the data-driven decision-making loop.

[0100] In a specific embodiment of the present invention, after generating the early warning information, the method further includes: obtaining the actual maintenance results after performing maintenance operations on the physical valve cooling system.

[0101] By combining the reversible reverse excitation signal, the actual system response, and the actual maintenance results, a set of event feedback data is formed.

[0102] Based on the event feedback data, the model parameters describing device performance degradation in the dynamic digital twin are iteratively optimized, and the compensation algorithm parameters used to calculate the reversible reverse excitation signal are updated.

[0103] Specifically, the engineering goal of this step is to build a complete self-evolving closed loop, enabling the early warning system to learn from each successful early warning and subsequent physical intervention, solidifying experiential knowledge into model capabilities, thereby achieving continuous self-improvement of system performance. After each early warning cycle is completed and the corresponding maintenance work is finished, the system first initiates a data archiving process, recording the complete parameters of the reversible reverse excitation signal, all time-series data of the actual system response it triggered, and the final implemented maintenance measures and results input by the operations and maintenance personnel. This result is the crucial "ground truth," such as "the gasket of heat exchanger No. 2 has been replaced, and the leakage rate has decreased from 0.5 ppm to zero per hour." This complete event chain, including "stimulus-response-result," is encapsulated into a timestamped structured data unit and formally marked as feedback data.

[0104] Next, the system inputs the feedback data into the iterative optimization process of the core model to update two key knowledge bases. The first is the optimization of the preset physical rule model. Its engineering purpose is to correct the parameters in the model that describe the long-term degradation patterns of equipment. The system uses actual maintenance results from the feedback data to verify or correct previous predictions. For example, if the system predicted a component's performance degradation rate of 2% per year, and the feedback data confirms that the component was replaced due to failure after only one year, the system will trigger a parameter correction algorithm. This algorithm can be represented as... In this formula, It is the new value of the degradation coefficient in the physical rule model to be optimized; It is its old value; It is the actual amount of degradation extracted and observed from the feedback data; This is the degradation amount previously predicted by the model; It is an adaptive learning rate with the inverse of time, ranging from 0.01 to 0.1, used to control the step size for each optimization.

[0105] Secondly, the system iteratively optimizes the security compensation strategy library using feedback data. The engineering goal is to improve the accuracy and effectiveness of the signals applied during future active detection. The system compares the relationship between the "reversible reverse excitation signal" and the "actual system response" in the current detection with the expected relationships stored in the strategy library. If a systematic deviation is found between the actual response and the expectation, such as the actual response consistently being 20% ​​weaker than expected, the system automatically adjusts the gain coefficients associated with that type of excitation signal in the strategy library, ensuring that the next generated excitation signal more accurately elicits the target response. Through continuous and automated optimization of these two core knowledge bases, the system achieves the ability to learn from experience, and its prediction accuracy and detection efficiency continuously improve with increasing runtime.

[0106] Reference Figure 2The second aspect of the present invention provides a valve cooling state early warning system based on digital twins and counterfactual intervention, comprising: an operating parameter and model data acquisition module, a dynamic digital twin construction driving module, a sensitive virtual disturbance signal identification module, a reversible reverse excitation signal generation module, an actual system response acquisition module, a system state deviation feature generation module, a digital twin generation module, a risk trend and intervention strategy generation module, and an early warning information generation module.

[0107] The operating parameter and model data acquisition module is connected to the dynamic digital twin construction driving module. The dynamic digital twin construction driving module is connected to the sensitive virtual disturbance signal identification module. The sensitive virtual disturbance signal identification module is connected to the reversible reverse excitation signal generation module. The reversible reverse excitation signal generation module is connected to the actual system response acquisition module. Both the sensitive virtual disturbance signal identification module and the actual system response acquisition module are connected to the system state deviation feature generation module. The system state deviation feature generation module is connected to the digital twin generation module. The digital twin generation module is connected to the risk trend and intervention strategy generation module. The risk trend and intervention strategy generation module is connected to the early warning information generation module.

[0108] The module for acquiring operating parameters and model data obtains real-time operating parameters and static model data of the physical valve cooling system.

[0109] The dynamic digital twin construction driver module, based on real-time operating parameters and static model data, constructs and drives a dynamic digital twin that operates synchronously with the physical valve cooling system.

[0110] The sensitive virtual disturbance signal identification module applies a set of virtual disturbance signals to the dynamic digital twin to obtain a series of virtual system responses, and identifies sensitive virtual disturbance signals from the virtual system responses that cause performance index changes to exceed the change threshold.

[0111] The reversible reverse excitation signal generation module generates a reversible reverse excitation signal based on the sensitive virtual disturbance signal, which is used to counteract the effects of disturbance simulation in the physical system and has limited operating parameters.

[0112] The actual system response acquisition module applies the reversible reverse excitation signal to the controller of the physical valve cooling system and acquires the actual system response corresponding to the physical valve cooling system.

[0113] The system state deviation feature generation module compares the parts of the actual system response and the virtual system response that are related to the sensitive virtual disturbance signal, and generates system state deviation features that quantify the difference between the two.

[0114] The digital twin generation module uses system state deviation characteristics to correct the parameters of the dynamic digital twin and generate a calibrated digital twin with a matching degree that meets the convergence threshold.

[0115] The risk trend and intervention strategy generation module performs future state projection and multi-scenario maintenance simulation on the calibrated digital twin to generate predictive risk trends and recommended intervention strategies.

[0116] The early warning information generation module combines predictive risk trends with recommended intervention strategies to generate early warning information.

[0117] The above content is merely an example and illustration of the concept of the present invention. Those skilled in the art can make various modifications or additions to the specific embodiments described, or use similar methods to replace them, as long as they do not deviate from the concept of the invention or exceed the scope defined by the present invention, and all such modifications and additions should fall within the protection scope of the present invention.

Claims

1. A valve cooling status early warning method based on digital twin and counterfactual intervention, characterized in that, include: S1. Obtain the real-time operating parameters and static model data of the physical valve cooling system; S2. Based on real-time operating parameters and static model data, construct and drive a dynamic digital twin that operates synchronously with the physical valve cooling system; S3. Apply a set of virtual perturbation signals to the dynamic digital twin to obtain a series of virtual system responses, and identify from the virtual system responses sensitive virtual perturbation signals that cause performance index changes to exceed the change threshold. S4. Based on the sensitive virtual disturbance signal, generate a reversible reverse excitation signal that is used to counteract the effects of disturbance simulation in the physical system and whose operating parameters are limited. S5. Apply the reversible reverse excitation signal to the controller of the physical valve cooling system and collect the actual system response corresponding to the physical valve cooling system. S6. Compare the parts of the actual system response and the virtual system response that are related to the sensitive virtual disturbance signal, and generate system state deviation characteristics that quantify the difference between the two. S7. Utilize the system state deviation characteristics to correct the parameters of the dynamic digital twin and generate a calibrated digital twin with a matching degree that meets the convergence threshold. S8. Perform future state projection and multi-scheme maintenance simulation on the calibrated digital twin to generate predictive risk trends and recommended intervention strategies; S9. Combine predictive risk trends and recommended intervention strategies to generate early warning information.

2. The valve cooling status early warning method based on digital twin and counterfactual intervention according to claim 1, characterized in that, The construction and driving of a dynamic digital twin that operates synchronously with the physical valve cooling system includes: Obtain static model data defining the spatial structure and material properties of the physical valve cooling system; Based on static model data and the physical rules of fluid and heat transfer, an initial digital twin model describing the internal physical processes of the system is established. Real-time operating parameters are input as boundary conditions into the initial digital twin model, and a dynamic digital twin is generated through continuous calibration using a state correction algorithm.

3. The valve cooling status early warning method based on digital twin and counterfactual intervention according to claim 1, characterized in that, The process of applying a set of virtual perturbation signals within a dynamic digital twin to obtain a series of virtual system responses, and identifying sensitive virtual perturbation signals from the virtual system responses that cause performance index changes to exceed a change threshold, includes: Generate a set of virtual disturbance signals representing the progressive performance degradation of the device; Virtual perturbation signals are applied one by one to the dynamic digital twin, and a series of performance degradation curves characterizing the system performance response are simulated and plotted. Calculate the slope of the performance degradation curve, and identify the virtual disturbance signals corresponding to the segments where the slope exceeds a preset slope threshold as sensitive virtual disturbance signals.

4. The valve cooling status early warning method based on digital twin and counterfactual intervention according to claim 1, characterized in that, The step of generating a reversible reverse excitation signal based on the sensitive virtual disturbance signal, which is used to counteract the effects of disturbance simulation in the physical system and has limited operating parameters, includes: Analyze the types and amounts of device degradation contained in sensitive virtual disturbance signals; For each type and amount of equipment degradation, a compensation target is calculated to offset its simulated impact. Based on the compensation target, a reversible reverse excitation signal is generated that acts on the controller of the physical valve cooling system, and whose adjustment range and duration are both within the safe range.

5. The valve cooling status early warning method based on digital twin and counterfactual intervention according to claim 1, characterized in that, The comparison of the portions of the actual system response and the virtual system response related to the sensitive virtual disturbance signal generates system state deviation features that quantify the difference between the two, including: Extract the actual response feature vector containing steady-state values ​​and dynamic characteristics from the actual system response; Extract virtual response feature vectors from the virtual system response that have the same dimensions as the actual response feature vectors; The weighted distance between the actual response feature vector and the virtual response feature vector is calculated to obtain the system state deviation feature.

6. The valve cooling status early warning method based on digital twin and counterfactual intervention according to claim 1, characterized in that, The step of using system state deviation characteristics to correct the parameters of the dynamic digital twin and generating a calibrated digital twin with a matching degree that meets the convergence threshold includes: Based on the deviation direction and magnitude indicated in the system state deviation characteristics, determine the model parameters to be adjusted in the dynamic digital twin; The model parameters are adjusted through an iterative optimization algorithm to minimize the weighted distance represented by the system state deviation feature; When the weighted distance is less than the convergence threshold, the currently adjusted model parameters are locked, and a calibrated digital twin with a matching degree that meets the convergence threshold is generated.

7. The valve cooling status early warning method based on digital twin and counterfactual intervention according to claim 1, characterized in that, The process of performing future state projections and multi-scenario maintenance simulations on the calibrated digital twin to generate predictive risk trends and recommended intervention strategies includes: Based on the current parameters of the calibrated digital twin, and combined with a degradation model describing the change of device performance over time, predictive risk trends are extrapolated. In the calibrated digital twin, a set of maintenance measures are simulated and executed to generate the corresponding corrected risk trends; By comparing the benefits of each revised risk trend with the implementation costs of each maintenance measure, a recommended intervention strategy including the target maintenance measures is generated.

8. The valve cooling status early warning method based on digital twin and counterfactual intervention according to claim 1, characterized in that, The combination of predictive risk trends and recommended intervention strategies generates early warning information, including: Extract the remaining safe operating time from the predicted risk trends and map it to a risk probability level; The recommended intervention strategy is broken down into a series of specific maintenance actions with implementation priorities; The remaining safe operating time, risk probability level, and specific maintenance actions are integrated to generate a structured early warning report as early warning information.

9. The valve cooling status early warning method based on digital twin and counterfactual intervention according to claim 8, characterized in that, After generating the warning information, it also includes: Obtain the actual maintenance results after performing maintenance operations on the physical valve cooling system; By combining reversible reverse excitation signals, actual system responses, and actual maintenance results, a set of event feedback data is formed. Based on the event feedback data, the model parameters describing device performance degradation in the dynamic digital twin are iteratively optimized, and the compensation algorithm parameters used to calculate the reversible reverse excitation signal are updated.

10. A valve cooling status early warning system based on digital twin and counterfactual intervention, characterized in that, include: The module for acquiring operating parameters and model data obtains real-time operating parameters and static model data of the physical valve cooling system. The dynamic digital twin construction driver module, based on real-time operating parameters and static model data, constructs and drives a dynamic digital twin that operates synchronously with the physical valve cooling system. The sensitive virtual disturbance signal identification module applies a set of virtual disturbance signals to the dynamic digital twin to obtain a series of virtual system responses, and identifies the sensitive virtual disturbance signals from the virtual system responses that cause the performance index to change beyond the change threshold. The reversible reverse excitation signal generation module generates a reversible reverse excitation signal based on the sensitive virtual disturbance signal, which is used to counteract the effects of disturbance simulation in the physical system and has limited operating parameters. The actual system response acquisition module applies the reversible reverse excitation signal to the controller of the physical valve cooling system and acquires the actual system response corresponding to the physical valve cooling system. The system state deviation feature generation module compares the parts of the actual system response and the virtual system response that are related to the sensitive virtual disturbance signal, and generates system state deviation features that quantify the difference between the two. The digital twin generation module uses system state deviation characteristics to correct the parameters of the dynamic digital twin and generate a calibrated digital twin with a matching degree that meets the convergence threshold. The risk trend and intervention strategy generation module performs future state projection and multi-scheme maintenance simulation on the calibrated digital twin to generate predictive risk trends and recommended intervention strategies. The early warning information generation module combines predictive risk trends with recommended intervention strategies to generate early warning information.

Images