Oil-immersed power equipment risk assessment and regulation method based on digital twinning

Abstract

The application discloses a kind of oil-immersed power equipment risk assessment and regulation method based on digital twinning, to solve the technical problems that current monitoring system cannot form complete closed loop, lack online calibration and causal identification, cannot answer key decision problem.The method comprises: obtaining the current state snapshot of oil-immersed power equipment and multi-physical field model;Parameter calibration is carried out on the multi-physical field model using the current state snapshot, and a digital twin model is obtained;Based on the digital twin model, multi-physical field coupling solution is executed, and multi-physical field key indicators are obtained, and candidate action set is constructed according to multi-physical field key indicators, combined with comprehensive risk assessment;The optimal control scheme is determined by executing candidate action set through digital twin model, combined with multi-objective optimization decision;Control oil-immersed power equipment to execute optimal control scheme, and optimize digital twin model based on regulation result.

Description

Technical Field

[0001] This invention relates to the field of power simulation technology, and in particular to a method for risk assessment and control of oil-immersed power equipment based on digital twins, a device for risk assessment and control of oil-immersed power equipment based on digital twins, an electronic device, and a storage medium. Background Technology

[0002] Oil-immersed power equipment is a widely used type of electrical device in power systems. Its core technology is primarily based on the design and application of oil-immersed transformers. An oil-immersed transformer is a power device that uses transformer oil as both insulation and cooling medium, mainly used in voltage transformation, power transmission, and power distribution systems. With the rapid development of the power industry, transformer oil, due to its excellent insulation and heat dissipation properties, has gradually become an important material in transformer manufacturing.

[0003] Oil-immersed transformers operate in a complex environment with strong coupling of multiple physical fields including electricity, heat, and current, and are subjected to the combined effects of load fluctuations, ambient temperature changes, and insulation aging. Their internal hazards are diverse and often hidden, including hotspot overheating, partial discharge, insulation dampness, abnormal dissolved gases in the oil, sludge deposition, and mechanical loosening. With the high proportion of new energy sources connected to the grid, the randomness of operating conditions has significantly increased, and the risk chain of "short-term overload - delayed cooling response - material parameter drift" occurs more frequently.

[0004] Current field monitoring systems face numerous challenges, including sparse sensor locations, significant noise interference, inconsistent data standards, and inconsistent time scales, making it difficult to accurately reproduce the true field distribution in critical areas such as winding gaps and interlayer insulation. While the digital twin concept offers a new technological path for equipment status awareness, most systems in engineering practice remain in an open-loop "monitoring-assessment-alarm" stage, exhibiting three prominent bottlenecks: First, the system remains at the monitoring and early warning level, lacking standardized handling actions and implementation channels, failing to form a complete closed loop of "diagnosis-calibration-handling-verification." Second, model parameters heavily rely on initial calibration; as the moisture content of insulation materials changes, equipment ages, and operating conditions change, a continuous mismatch occurs between the model and the actual system, lacking effective online calibration and causal identification mechanisms. Simultaneously, the system lacks counterfactual reasoning and quantitative trade-off capabilities, unable to answer some key decision-making questions. Summary of the Invention

[0005] This invention provides a method for risk assessment and control of oil-immersed power equipment based on digital twins, a device for risk assessment and control of oil-immersed power equipment based on digital twins, an electronic device, and a storage medium, which are used to solve or partially solve the technical problems of current monitoring systems that cannot form a complete closed loop, lack online calibration and causal identification, and cannot answer key decision-making questions.

[0006] This invention provides a method for risk assessment and control of oil-immersed power equipment based on digital twins, the method comprising: Obtain a snapshot of the current state and a multiphysics model of oil-immersed power equipment; The current state snapshot is used to calibrate the parameters of the multiphysics model to obtain a digital twin model; Based on the digital twin model, multiphysics coupling solution is performed to obtain key multiphysics indicators, and a candidate action set is constructed based on the key multiphysics indicators and a comprehensive risk assessment. The candidate action set is executed through the digital twin model, and the optimal control scheme is determined by combining multi-objective optimization decision-making. The oil-immersed power equipment is controlled to execute the optimal control scheme, and the digital twin model is optimized based on the control results.

[0007] The present invention also provides a risk assessment and control device for oil-immersed power equipment based on digital twins, the device comprising: The data acquisition unit is used to acquire a current state snapshot and a multiphysics model of the oil-immersed power equipment. A parameter calibration unit is used to perform parameter calibration on the multiphysics model using the current state snapshot to obtain a digital twin model; The candidate action set construction unit is used to perform multi-physics coupling solution based on the digital twin model, obtain multi-physics key indicators, and construct a candidate action set based on the multi-physics key indicators and a comprehensive risk assessment. A multi-objective optimization decision-making unit is used to execute the candidate action set through the digital twin model and determine the optimal control scheme by combining multi-objective optimization decision-making. The optimal control scheme execution unit is used to control the oil-immersed power equipment to execute the optimal control scheme and optimize the digital twin model based on the control results.

[0008] The present invention also provides an electronic device, the device comprising a processor and a memory: The memory is used to store program code and transmit the program code to the processor; The processor is used to execute the risk assessment and control method for oil-immersed power equipment based on digital twins as described above, according to the instructions in the program code.

[0009] The present invention also provides a computer-readable storage medium for storing program code for executing the digital twin-based risk assessment and control method for oil-immersed power equipment as described in any of the preceding claims.

[0010] As can be seen from the above technical solutions, the present invention has the following advantages: This paper presents a method for risk assessment and control of oil-immersed power equipment based on digital twins. First, a snapshot of the current state of the oil-immersed power equipment and a multiphysics model are acquired. The current state snapshot is then used to calibrate the parameters of the multiphysics model, resulting in a digital twin model. This provides a dynamic mapping mechanism from physical state to model parameters, accurately mapping the operating state of the transformer entity to the adjustable interface of the boundary conditions and parameters of the multiphysics model in the digital twin. Next, multiphysics coupling solutions are performed based on the digital twin model to obtain key multiphysics indicators. Based on these indicators and a comprehensive risk assessment, a set of candidate actions is constructed. The candidate action set is then executed through the digital twin model, and the optimal control scheme is determined by multi-objective optimization decision-making. This provides a decision-making method based on virtual simulation experiments and multi-objective optimization. Within the digital twin, candidate control schemes are pre-executed, and quantitative evaluation and optimization are performed based on multi-objective optimization decision-making. The optimal control scheme is determined based on the pre-simulation results, and a complete decision-making process is established regarding whether to implement the scheme on the physical equipment. Finally, the oil-immersed power equipment is controlled to execute the optimal control scheme, and the digital twin model is optimized based on the control results. Thus, combined with the previous steps, a "prediction-measurement-perturbation" closed-loop self-learning system is provided. The accuracy of the model is evaluated by comparing the predicted and measured data, and the model is dynamically optimized based on the control results, enabling the digital twin to continuously self-optimize closed-loop learning and adaptive adjustment. Attached Figure Description

[0011] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art 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.

[0012] Figure 1 A flowchart illustrating the steps of a risk assessment and control method for oil-immersed power equipment based on digital twins; Figure 2 This is a schematic diagram of the overall process of a risk assessment and control method for oil-immersed power equipment based on digital twins; Figure 3 This is a structural block diagram of a risk assessment and control device for oil-immersed power equipment based on digital twins. Detailed Implementation

[0013] This invention provides a method for risk assessment and control of oil-immersed power equipment based on digital twins, a device for risk assessment and control of oil-immersed power equipment based on digital twins, an electronic device, and a storage medium, which are used to solve or partially solve the technical problems of current monitoring systems that cannot form a complete closed loop, lack online calibration and causal identification, and cannot answer key decision-making questions.

[0014] To make the objectives, features, and advantages of this invention more apparent and understandable, the technical solutions of the embodiments of this invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the embodiments described below are only some embodiments of this invention, and not all embodiments. Based on the embodiments of this invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this invention.

[0015] To enable those skilled in the art to better understand the technical solutions provided in the embodiments of the present invention, some of the technical features involved in the solutions are briefly described first: Oil-immersed power equipment refers to power equipment that uses transformer oil (or other insulating oil) as the insulating medium internally. Due to its excellent electrical insulation, heat dissipation, and arc-extinguishing capabilities, oil-immersed insulation is widely used in high-voltage applications in power systems. Oil-immersed power equipment includes oil-immersed power transformers, oil-immersed reactors, oil-immersed circuit breakers, oil-immersed instrument transformers, oil-immersed bushings, and oil-immersed power capacitors. The method provided in this embodiment uses the digital twin risk assessment of an oil-immersed transformer as an example for illustration; however, the method proposed in this invention is not only applicable to transformers but also to internal oil-immersed insulation in other similar equipment, and the overall approach is consistent.

[0016] Time-Varying Covariance Matrix: A matrix used to describe how the covariance relationship between random variables changes over time.

[0017] As an example, oil-immersed power equipment is a widely used type of electrical device in power systems. Its core technology is mainly based on the design and application of oil-immersed transformers. Oil-immersed transformers operate in a complex environment with strong coupling of multiple physical fields such as electricity, heat, and current for extended periods, and are subjected to the combined effects of load fluctuations, ambient temperature changes, and insulation aging. Their internal hazards are diverse and hidden, including hotspot overheating, partial discharge, insulation moisture, abnormal dissolved gases in the oil, sludge deposition, and mechanical loosening. With the high proportion of new energy sources connected to the grid, the randomness of operating conditions has significantly increased, and the risk chain of "short-term overload - delayed cooling response - material parameter drift" occurs more frequently.

[0018] Current field monitoring systems face numerous challenges, including sparse sensor locations, significant noise interference, inconsistent data standards, and inconsistent time scales, making it difficult to accurately reproduce the true field distribution in critical areas such as winding gaps and interlayer insulation. While the digital twin concept offers a new technological path for equipment status awareness, most systems in engineering practice remain in an open-loop "monitoring-assessment-alarm" stage, exhibiting three prominent bottlenecks: First, the system remains at the monitoring and early warning level, lacking standardized handling actions and implementation channels, failing to form a complete closed loop of "diagnosis-calibration-handling-verification." Second, model parameters heavily rely on initial calibration; as the moisture content of insulation materials changes, equipment ages, and operating conditions change, a continuous mismatch occurs between the model and the actual system, lacking effective online calibration and causal identification mechanisms. Simultaneously, the system lacks counterfactual reasoning and quantitative trade-off capabilities, unable to answer some key decision-making questions.

[0019] Further analysis of this invention reveals that current digital twin models are primarily used for state reproduction rather than control strategy simulation. Model parameters are typically static and cannot be dynamically mapped based on the real-time states of actuators such as fans, pumps, and valves. Furthermore, current decision-making processes are mainly based on simple "IF-THEN" rules, lacking in-depth simulation of the subsequent chain reactions of control actions and failing to quantitatively weigh multiple objectives. More importantly, the system lacks closed-loop verification and model calibration mechanisms after executing relevant control actions, preventing the digital twin from learning from each execution and causing model accuracy to decline over time.

[0020] Therefore, one of the core inventive points of this invention is to propose a risk assessment and control method for oil-immersed power equipment based on digital twins. Specifically, on the one hand, a dynamic mapping mechanism from physical state to model parameters is provided, which maps the operating state of the transformer entity (such as fan speed, oil pump speed, valve opening, tap position) to the adjustable interface of the boundary conditions and parameters (such as convective heat transfer coefficient, inlet velocity, branch impedance, and end potential) of the multiphysics model in the digital twin in real time and accurately. On the other hand, a decision-making method based on virtual experiment and multi-objective optimization is provided. In the digital twin, candidate control schemes are pre-executed, and quantitative evaluation and optimization are carried out using a multi-objective system consisting of risk reduction benefits, energy consumption costs, execution wear, smoothness, and system coupling effects. Finally, a complete decision-making process is made on whether to execute the scheme on the physical equipment based on the pre-simulation results. Based on this, a closed-loop self-learning system of "prediction-measurement-perturbation" is provided. By comparing predicted and measured data to evaluate the accuracy of the model, it actively initiates safe perturbation to estimate the true sensitivity, and then dynamically calibrates the model parameters and mapping relationship, enabling the digital twin to continuously optimize closed-loop learning and adaptive adjustment.

[0021] Reference Figure 1The diagram illustrates a flowchart of a risk assessment and control method for oil-immersed power equipment based on digital twins, as provided in an embodiment of the present invention. Specifically, it may include the following steps: Step 101: Obtain the current state snapshot and multiphysics model of the oil-immersed power equipment; In practical applications, it is necessary to obtain the current state snapshot and multiphysics model of oil-immersed power equipment (for ease of understanding, a transformer is used as an example in this embodiment of the invention) for subsequent risk assessment and control processes.

[0022] Specifically, the construction of a snapshot of the transformer's current state relies on the acquisition of panoramic data. Therefore, it is first necessary to construct a holographic sensing network covering the entire transformer state to collect electrical measurement data. The objects of electrical measurement data collection include, but are not limited to: electrical quantities (three-phase current, voltage, active / reactive power, power factor, tap position, etc.), thermal quantities (oil surface temperature, winding hot spot temperature, clamp and core temperature, ambient temperature and humidity, etc.), fluid and mechanical quantities (oil pump and fan status and speed, oil flow and differential pressure in key oil passages, valve opening, vibration and acoustic emission signals, etc.), and insulating chemical quantities (online monitoring data of dissolved gases (H2, CH4, C2, H2, C2H4, C2H6, etc.), trace water content, partial discharge count and amplitude, etc.).

[0023] To ensure data consistency in time, all data acquisition devices at the station use a precision time protocol for unified clock synchronization and align all acquired electrical measurement data to a unified millisecond-level timestamp sequence, with the maximum time deviation controlled within 0.1 seconds.

[0024] Subsequently, the system performs in-depth cleaning and anomaly identification on the raw collected electrical measurement data. Specifically, median filtering and exponential smoothing are used for routine data, while Kalman smoothing is used for high-frequency data. Spline interpolation is used to repair short-term missing data (less than 5 minutes), and long-term missing data is marked and reduced in weight in subsequent processes. At the same time, by combining statistical thresholds, interquartile ranges, and seasonal-trend decomposition residual methods, the system effectively identifies and removes abnormal data such as spikes, flat lines, and jitters caused by hardware anomalies.

[0025] After cleaning and standardizing the raw electrical force measurement data, the system dynamically assigns a quality label to each observation data point (each observation data point) and quantifies its measurement uncertainty. Specifically, when quantifying uncertainty, the system will comprehensively consider the following three dimensions: Sensor inherent accuracy: Based on the calibration error or minimum resolution provided by the equipment manufacturer, it serves as the lower limit of observation error.

[0026] Signal-to-noise ratio (SNR): This metric assesses the volatility and stability of data using methods such as frequency domain analysis, sliding window variance, or wavelet entropy.

[0027] Post-Cleaning Residuals: The residual distribution and missing information obtained through methods such as filtering, interpolation, and trend decomposition are used to characterize the degree of data quality degradation.

[0028] Based on the above-mentioned indicators, the system assigns three quality levels (High, Medium, Low) to each observation data point, and configures the corresponding observation error variance accordingly as follows: , in, The variance of the observation error; This represents the inherent accuracy variance of the sensor; To clean the variance of the residuals; Data missing rate; These are the weighting factors for the variance of the cleaning process; The missing value weighting coefficient; subscript Indicates the first Observation data; subscript express time.

[0029] Finally, the system constructs a dataset consisting of all observation data that has undergone quality assessment and uncertainty quantification into an observation vector. Its corresponding time-varying covariance matrix Together, they form a dynamic observation information matrix (dynamic observation information structure). This structure is encapsulated in a standardized format to form a data set called "Current Observation Snapshot," which serves as a key input module for the real-time operation of the digital twin system.

[0030] Specifically, the current state snapshot includes various observations with timestamp alignment, observation variance, quality labels, unit information, and source device identifiers, serving as the standard input interface for subsequent digital twin model data assimilation, state estimation, and control decisions.

[0031] In terms of time, this snapshot uses a unified full-site time base to align all measurements at the millisecond level. In terms of data content, this snapshot specifies the following core fields for each observation record: observation value (raw and cleaned values), and observation error variance. The data includes quality level labels (e.g., High / Medium / Low), physical units (e.g., °C, kPa, A), measurement type (temperature, current, flow rate, etc.), source device code, and interface identifier. These elements collectively form a data snapshot with time synchronization, structural consistency, and uncertainty annotation capabilities. This provides a unified, reliable, and traceable data interface for subsequent state estimation, model calibration, data assimilation (e.g., Ensemble Kalman Filter), risk identification, and control decisions within the twin.

[0032] Based on the preceding discussion, the process for constructing a current state snapshot of oil-immersed power equipment may include the following steps S01 to S05: Step S01: Collect electrical power measurement data covering the entire status of oil-immersed power equipment; Step S02: Perform unified time synchronization and cleaning quality control on the electrical force measurement data to obtain processed measurement data; Step S03: For each observation data in the processed measurement data, calculate the observation error variance of the observation data by quantifying the uncertainty in multiple dimensions; Step S04: Based on all observation error variances, and considering various observations with timestamp alignment and quality labels, construct a dynamic observation information matrix composed of observation vectors and time-varying covariance matrix; Step S05: Encapsulate the dynamic observation information matrix in a standardized format to form a snapshot of the current state of the oil-immersed power equipment.

[0033] The above outlines the process for constructing a snapshot of the transformer's current state. Next, we will introduce the construction of the transformer's multiphysics model. The steps involved in constructing the transformer's multiphysics model are crucial for building a high-fidelity, evolvable digital twin. It primarily relies on the transformer's design structure (design drawings), material properties, and physical laws to establish a basic multiphysics model encompassing all key components such as windings, core, insulating oil channels, radiators, and bushings. Through a well-defined and adjustable interface, real-time data is mapped to model parameters, ensuring synchronization between the twin and the physical entity.

[0034] Specifically, firstly, based on the transformer's design drawings, material properties, and physical laws, a multiphysics-based fundamental model is established, encompassing all key components such as windings, core, insulating oil channels, and radiators. Its internal heat sources consist of copper losses, iron losses, and additional losses, and a small number of correctable empirical coefficients (i.e., correctable coefficients) are reserved to provide an interface for subsequent data assimilation.

[0035] Among them, the correctable empirical coefficients may include conductor resistance temperature rise coefficient (used to correct copper loss deviation caused by changes in winding resistance with temperature), contact resistance correction coefficient (used to characterize the additional heating caused by poor micro-contact in parts such as leads and contacts), core loss adjustment coefficient (including hysteresis loss and eddy current loss correction terms, used to compensate for differences in magnetic flux density distribution in different structural regions), local heat source enhancement coefficient (used to improve the model's ability to respond to the heating of hot spots (such as winding ends and tap changers), heat dissipation efficiency correction coefficient (considering the impact of changes in the operating status of the cooling system and environmental factors on heat transfer capacity), and system energy residual compensation coefficient (used to dynamically correct the deviation between the model's total heat balance and actual measurements).

[0036] Building upon this foundation, the system operates on the core innovative aspect of this invention: real-time mapping of states to parameters. This mechanism precisely transforms the control signals and operating states of physical devices into calculable parameters and boundary conditions in a digital model. Specifically, the fan speed is mapped to the equivalent convective heat transfer coefficient of the radiator surface; the oil pump state is mapped to the inlet flow velocity and network pressure drop characteristics of the oil circulation; the valve opening is mapped to the local impedance coefficient in the fluid network; the tap position is synchronously mapped to the winding end potential distribution and loss distribution; the switching of cooling modes triggers the grouping update of preset heat transfer coefficients and flow velocities; and online degassing, dehydration, and other operations directly trigger the online refresh of the insulation medium parameter library.

[0037] Based on the preceding discussion, the process of constructing a multiphysics model for oil-immersed electrical equipment can include the following steps S11 to S12: Step S11: Based on the design structure, material properties, and physical laws of oil-immersed power equipment, construct a three-dimensional multiphysics model; Step S12: Obtain the control signals and operating status of the oil-immersed power equipment, and map the control signals and operating status into the calculation parameters and boundary conditions of the multiphysics basic model in real time to form the multiphysics model of the oil-immersed power equipment.

[0038] Step 102: Use the current state snapshot to calibrate the parameters of the multiphysics model to obtain a digital twin model; In this step, the system utilizes the current state snapshot generated in the preceding steps (especially observed data such as temperature and flow rate) and employs data assimilation algorithms such as Kalman filtering to perform parameter calibration (optimal estimation) on the pre-reserved correctable coefficients in the multiphysics model. This ensures that the twin's output and the measured data achieve optimal matching within the uncertainty range. The final output of this step is a digital twin instance (digital twin model) that has undergone real-time data calibration and whose parameters can be dynamically adjusted through a unified interface. This instance not only reflects the current physical state of the transformer but also possesses the ability to accurately and safely simulate the consequences of various control operations in virtual space, laying the foundation for subsequent risk assessment and decision optimization.

[0039] Specifically, for the parameter calibration process, in order to achieve dynamic consistency between the multiphysics model parameters and the on-site operating status, this embodiment of the invention adopts ensemble Kalman filtering as the core data assimilation method to perform optimal estimation and continuous correction of the reserved correctable coefficients.

[0040] First, the system jointly constructs an extended state vector by combining the multiphysics state variables (such as temperature field, flow velocity, and potential distribution) of the multiphysics model with correctable coefficients (identifiable parameters, such as conductor resistance temperature rise coefficient, iron loss adjustment factor, and local hot spot enhancement coefficient). At each time step, the system extracts actual observation data such as temperature, flow rate, and electric field from the current state snapshot, and performs forward propagation in the multiphysics model through multiple sets of simulation samples to generate a corresponding set of predicted observation data.

[0041] Next, based on each set of actual observation data and each set of predicted observation data, an error covariance matrix between the predicted and actual observations is constructed, and the Kalman gain is calculated. This gain is used to feed the predicted residuals back to the parameter space of the extended state vector to correct the parameters of the correctable coefficients of the extended state vector. The parameter update not only considers the error weights of the current observations but also makes full use of the covariance information between the observations, thereby achieving observation-driven multi-parameter linkage correction.

[0042] Following the forward propagation and parameter correction process described above, the correctable coefficients are continuously corrected until the Kalman gain of the error covariance matrix is ​​less than a preset gain threshold. The final parameter update results of the correctable coefficients under each set of simulation samples are then output.

[0043] Finally, by weighted averaging the parameter update results of multiple samples, the system obtains the current optimal parameter estimate. Based on the optimal parameter estimate, the correctable coefficients of the multiphysics model are updated to obtain a digital twin model, which makes its physical output approximate the measured state within the uncertainty range.

[0044] The parameter assimilation calibration method provided by this invention has the ability to handle nonlinear and multidimensional state-space problems, and is suitable for the dynamic updating of key parameters such as heat source distribution, heat exchange efficiency, and resistance changes during transformer operation.

[0045] Based on the foregoing discussion, in some embodiments, the multiphysics model corresponds to multiphysics state variables and correctable coefficients. Therefore, in a specific implementation, the process of using a current state snapshot to calibrate the parameters of the multiphysics model to obtain a digital twin model may include the following steps S21 to S28: Step S21: Construct an extended state vector based on the multiphysics state variables and correctable coefficients; Step S22: Extract actual observation data at multiple times under the multiphysics state variables from the current state snapshot to form multiple sets of simulation samples; Step S23: Forward propagation is performed in the multiphysics model using each group of simulation samples to generate corresponding predicted observation data; Step S24: Based on the actual observation data and the predicted observation data of each group, construct the error covariance matrix between the predicted observations and the actual observations; Step S25: Calculate the Kalman gain of the error covariance matrix and feed the Kalman gain as the prediction residual back to the extended state vector to correct the parameters of the correctable coefficients of the extended state vector. Step S26: Use each set of simulation samples to perform forward propagation again in the parameter-corrected multiphysics model until the Kalman gain of the error covariance matrix is ​​less than the preset gain threshold, and output the final parameter update results of the correctable coefficients under each set of simulation samples. Step S27: Perform a weighted average of the update results for each parameter to obtain the optimal parameter estimate; Step S28: Update the correctable coefficients of the multiphysics model based on the optimal parameter estimates to obtain the digital twin model.

[0046] Step 103: Perform multiphysics coupling solution based on the digital twin model to obtain key multiphysics indicators, and construct a candidate action set based on the key multiphysics indicators and comprehensive risk assessment. This step mainly uses a digital twin model to achieve rapid solution of the electro-thermal-fluid multiphysics field and extraction of key indicators. Based on the key indicators, a comprehensive risk assessment is conducted to generate various risk regions of the transformer and determine the risk-dominant type corresponding to each risk region. Furthermore, a candidate action set for control actions is constructed.

[0047] For the rapid solution and key index extraction process of the electro-thermal-fluid multiphysics field, this process is the core computational link of the digital twin. Specifically, based on the model parameters and real-time boundary conditions provided by the aforementioned steps and after parameter assimilation and calibration, the control equations of the coupled electro-thermal and fluid physical fields are solved in virtual space, and finally the physical state of the entire domain (the distribution of physical fields in the entire domain) and the quantitative indicators (key performance indicators) used for risk assessment are output.

[0048] More specifically, solving multiphysics involves multiphysics governing equations and coupling mechanisms. In this embodiment of the invention, a transformer electro-thermal-fluid multiphysics considering space charge is established based on a space charge distribution model to obtain a more accurate electro-thermal-fluid multiphysics.

[0049] The electric field governing equations, which describe the influence of space charge distribution on the electric field and the transport behavior of charges under the influence of electric and flow fields, are composed of the electric potential equation and the charge transport equation, namely: , In the above formula, Electric potential is expressed in V (volt). The dielectric constant is expressed in F / m (farads per meter) and can be a constant or a spatial distribution function. This represents the space charge density, expressed in C / m³ (coulombs per cubic meter). This represents the divergence of the electric displacement vector field caused by the potential field, i.e., the source strength of the electric field; This represents the space charge transport current density, measured in A / m². Charge mobility is expressed in m² / (V·s) and characterizes the rate at which a charge responds in an electric field. This represents the mobile charge density, measured in C / m³. This represents the electric field, with its direction being the potential gradient direction. It represents the fluid velocity vector, which is the potential migration of electric charge along with fluid motion (such as oil flow), and its unit is m / s; The divergence of current density represents the net increase or decrease of charge in space; This represents the charge generation / recombination term, with units of A / m³.

[0050] The magneto-thermal-fluid coupling equations describing fluid motion (velocity, pressure) and its coupling effects with electric and temperature fields are composed of the continuity equation, the Navier-Stokes equations, and the temperature field governing equations, namely: , In this formula, Indicates fluid density (kg / m³); Indicates axial velocity (z direction, m / s); Indicates radial velocity (r direction, m / s); This indicates axial volume force (such as gravity or magnetic force, in N / m³). This indicates the driving force caused by the pressure gradient; Indicates dynamic viscosity (Pa·s); Indicates pressure (Pa); This indicates volumetric heat source terms (unit: W / m³), such as heat generation from copper or iron losses. This represents the specific heat capacity at constant pressure (J / (kg·K)). Represents the temperature field (K); It represents the thermal conductivity (W / (m·K)).

[0051] The mapping parameters output by the digital twin model are directly set as the boundary conditions for this solution. Specifically, for the thermal boundary, the convective heat transfer coefficient mapped by the fan speed is applied to the outer surface of the radiator; for the flow boundary, the inlet velocity mapped by the oil pump speed and the pipeline pressure drop are set as the inlet and outlet conditions of the flow field; for structural switching, the preset combination of heat transfer coefficient and flow velocity is switched with one click according to the cooling mode mapping.

[0052] Next, efficient numerical algorithms (such as the finite element method or the finite volume method) or proven reduced-order models are used to quickly solve the aforementioned equations while ensuring engineering accuracy. The output obtained from the solution is divided into two levels: one is the global physical field distribution, and the other is the key performance indicators.

[0053] The overall physical field distribution includes a three-dimensional temperature field (for identifying overheated regions), a three-dimensional velocity field / streamlines (for identifying low-velocity and dead zones), and a three-dimensional electric field intensity distribution (for identifying weak points in insulation). Key performance indicators include thermal performance indicators (hotspot temperatures and locations, maximum temperature differences between key components, and the proportion of low-velocity stagnant flow volume), flow performance indicators (pressure differences between inlet and outlet of key oil channels, and average flow velocity), electrical performance indicators (maximum electric field intensity on insulating components and its location), and confidence intervals (each key indicator is accompanied by a confidence interval calculated based on the uncertainty propagation of electrical force measurement data, such as a 95% confidence interval).

[0054] Based on the foregoing discussion, in some embodiments, the implementation process of performing multiphysics coupling solutions based on a digital twin model to obtain key multiphysics indicators may include the following steps S31 to S33: Step S31: Construct the electric field control equation and the magnetic-thermal-fluid coupling equation respectively, and combine the electric field control equation and the magnetic-thermal-fluid coupling equation to construct the electric-thermal-fluid multiphysics field equation considering space charge; Step S32: Use the mapping parameters output by the digital twin model as the boundary conditions to solve the electro-thermal-fluid multiphysics equations and obtain the global physical field distribution and key performance indicators. Step S33: Integrate the global physical field distribution and key performance indicators as key indicators for multiphysics.

[0055] The above outlines the rapid solution and key indicator extraction process for the electro-thermal-fluid multiphysics field. The following section covers risk profiling and regional early warning based on the indicator solution results.

[0056] The first step is the generation and clustering of risk areas. Specifically, potential risk areas are defined and identified on the 3D model of the digital twin. Based on transformer design drawings and prior knowledge of faults, high-risk structural areas are pre-divided in the model. These areas can generally be used to characterize concentrated or weak points in the electric field, thermal field, or flow field. For example, the ends, transposition points, and axial air passages of the winding area; the corners of the leads and grounding system, the vicinity of the equalizing ball, and the corners of the clamps; the edges of the core window, the tie plate, and the clamping parts of the core structure; the narrow oil passages of the fluid system, the blind pipes at the radiator inlet, and the bushing riser. This partitioning divides the entire complex model into several evaluation units with clear physical meaning. During the actual risk area identification, based on the global physical field distribution obtained from the aforementioned steps, the physical field data is scanned within each first-level partition to identify "red spots" that exceed preset thresholds (such as areas with temperatures exceeding 105℃, electric field strength exceeding 80% of the design value, and flow velocities below 0.01m / s). A connected component clustering algorithm is used to connect adjacent "red dots" in space to form continuous "risk patches" of varying shapes, i.e., risk regions. This step aggregates discrete risk points into meaningful "risk regions" and records their spatial location, volume, and shape.

[0057] In addition to defining and identifying risk areas, this invention also provides a complete multi-dimensional risk assessment index system for risk areas based on the key performance indicators obtained from the aforementioned steps. Specifically, for each identified risk area or structural partition, a set of quantitative risk indicators, including thermal risk indicators and electrical risk indicators, is calculated through multi-dimensional risk assessment.

[0058] The specific thermal risk indicators include: Temperature margin : Quantify the relative distance between the current temperature and the safety limit.

[0059] Temperature gradient intensity: the maximum temperature difference per unit spatial distance, reflecting the severity of thermal stress.

[0060] Lag-thermal coupling ratio: The volume ratio that simultaneously meets the low flow rate condition in the high-temperature region, identifying the "dual threat" region most susceptible to insulation aging.

[0061] The specific electrical risk indicators include: Electric field utilization rate This reflects the proportion of the current maximum field strength to the design allowable value.

[0062] Field distortion: By calculating the degree of deviation between the local field strength and the average field strength, it can identify whether there is a risk of partial discharge or breakdown.

[0063] Matching degree with partial discharge signal: The calculated high field strength region is spatiotemporally correlated with the location of the partial discharge pulse monitored online in the previous steps to verify the authenticity of the electrical risk.

[0064] Next, based on the aforementioned thermal and electrical risk indicators, index normalization and weighted fusion are performed. Specifically, the indicators with different dimensions are normalized to the [0,1] interval. Appropriate weights are assigned to thermal, electrical, and current risks according to the transformer's current operating conditions and maintenance strategy (manually set; this setting process is based on actual operating experience, equipment operating characteristics, and historical fault analysis results to ensure that each type of risk reflects a reasonable weight in the comprehensive score). A comprehensive risk score is obtained through weighted summation. The dominant risk type is identified for each medium-risk and above zone, such as "thermal-dominant - stagnation," "electrical-dominant - field distortion," and "thermal-electrical coupling."

[0065] The above covers risk profiling and regional early warning based on the indicator solution results. Building upon this, the automatic generation and initial screening of precise candidate actions will be explained next.

[0066] Using the risk areas and risk-dominant types output from the aforementioned steps as input, a structured decision-making and generation process is used to output a set of specific control actions that have undergone preliminary safety screening, in preparation for subsequent virtual simulation experiments within the digital twin.

[0067] The system maintains a structured action library containing all available control methods. Each control action explicitly defines its controlled object, physical effect, and adjustable range. The structured action library mainly includes the following common control actions: Heat dissipation enhancement: fan switching and speed regulation, oil pump speed regulation, and cooling mode switching.

[0068] Flow path optimization: valve opening adjustment, branch valve bypass.

[0069] Electrical regulation: tap position adjustment, reactive power compensation device switching and setting, voltage target optimization.

[0070] Insulation maintenance: Start-up and shutdown of online degassing devices and online dehydration devices.

[0071] Load management: In collaboration with upper-layer systems, it is recommended to optimize load distribution.

[0072] This invention embodiment achieves preliminary screening of control actions corresponding to risk areas based on principal-cause routing and action mapping. Specifically, based on the "primary contradiction label" (i.e., risk-dominant type) output from the aforementioned steps, the system intelligently routes the risk-dominant type to the action category with the highest relevance.

[0073] For heat-dominated / flow-dominated risks, priority is given to actions that enhance heat dissipation and optimize flow paths. For example, for "overheating at the winding end", priority is given to generating actions such as "increasing the oil pump speed to increase the oil flow rate", "increasing the speed of the corresponding fan unit", or "adjusting the valve to optimize the flow rate through the area".

[0074] For risks dominated by electricity, priority is given to routing to electrical regulation actions. For example, for "lead field distortion", priority is given to generating a combined action of "reducing the tap position to optimize the potential distribution" and "adjusting reactive power compensation to maintain bus voltage".

[0075] For media-dominated risks, priority is given to routing to insulation maintenance actions, supplemented by heat dissipation / flow support. For example, for "high moisture content", actions such as "starting the online dehydration device" and "increasing the oil pump speed to enhance oil circulation" are generated simultaneously.

[0076] Based on this, the system presets three operation templates for each individual action: mild, moderate, and intense. It also incorporates safety constraints such as slope limits and minimum interval times to ensure that the action amplitude and rhythm remain within safe limits. Subsequently, within the constrained variable domain, the system generates a limited number of control action combination strategies (each control action combination strategy can be considered as corresponding to an initial candidate action) through a combination of rule priority and spatial sampling. These initial combinations undergo rigorous conflict pruning, eliminating schemes with interlocking conflicts, excessive amplitude, or potential power quality issues, ultimately forming a deduplicated set of high-value candidate actions.

[0077] Specifically, regarding the generation of control action combination strategies and the conflict-reduction process for these initial combination strategies, the actions are first categorized into mandatory, optional, disabled, and mutually exclusive categories based on the risk-dominant type and equipment operating procedures. The priority, upper and lower limits of amplitude, maximum change slope, and minimum interval time for each action are then determined. A constrained variable domain is constructed only for actions that satisfy the above constraints. Using mild, moderate, and intensity levels as baselines, and combining methods such as weighted Latin hypercube or orthogonal sampling, a controlled set of control action combination strategies is generated in a multi-dimensional space.

[0078] For the obtained initial control action combination strategy, the system sequentially performs conflict pruning. Specifically, firstly, combinations containing logical mutual exclusion, electrical interlocking, or large reverse adjustments within a short period are eliminated using action interlocking rules. Next, combinations that may cause mechanical or thermal shocks are filtered out by comparing the maximum allowable adjustment range of the equipment, temperature rise gradient, and action frequency. Through a rapid power quality assessment model, schemes that may cause bus voltage exceeding limits, voltage imbalance, or excessive reactive power fluctuations are excluded. Finally, the remaining combinations are normalized and deduplicated, and similar strategies with approximate effects are merged. Only high-value candidate action sets that meet all safety constraints, are of moderate quantity, and have a significant mitigating effect on the dominant risk are retained as input for subsequent virtual simulation experiments and multi-objective optimization decision-making.

[0079] Each candidate solution that passes the initial screening includes a complete list of actions, the magnitude and execution rhythm of each action, the corresponding twin parameter mapping changes, the applicable working condition window, the constraint check results, and the rollback plan identifier, in preparation for subsequent virtual simulation experiments and effect evaluations within the digital twin.

[0080] Based on the foregoing discussion, the multiphysics key indicators proposed in the embodiments of the present invention may include the global physical field distribution and key performance indicators. Therefore, in some embodiments, the execution process for constructing a candidate action set based on the multiphysics key indicators and a comprehensive risk assessment may include the following steps S41 to S46: Step S41: Based on the global physical field distribution, identify the risk areas of the digital twin model using the connected component clustering algorithm, and generate at least one continuous risk area; Step S42: For each risk area, extract the target performance indicators corresponding to the risk area from the key performance indicators, and conduct a multi-dimensional risk assessment of the risk area based on the target performance indicators to obtain thermal risk indicators and electrical risk indicators. Step S43: Based on the thermal risk index and the electrical risk index, obtain a comprehensive risk index through weighted fusion, and determine the dominant risk type of the risk area based on the comprehensive risk index; Step S44: Based on the risk-driven type, select the most relevant control actions from the pre-built structured action library containing all available control measures through the main cause routing; Step S45: Based on the control actions of each risk area, generate multiple initial candidate actions with control action combination strategies within the restricted variable domain by combining rule priority and spatial sampling. Step S46: Perform action conflict pruning on multiple initial candidate actions to obtain at least one target candidate action, thereby forming a candidate action set.

[0081] Step 104: Execute the candidate action set through the digital twin model, and determine the optimal control scheme by combining multi-objective optimization decision-making. This step primarily involves virtual simulation experiments and multi-objective optimization decision-making within the transformer's digital twin. Specifically, it receives a structured set of candidate actions output from the preceding steps. By faithfully reproducing the entire execution process of each candidate scheme within the digital twin, it performs quantitative evaluation and comparison from multiple dimensions, including effect, cost, and risk, achieving a "priority-based" scientific decision-making. This is achieved by generating candidate actions for high-risk partitions, quantifying the "risk reduction-cost" process, and outputting the optimal combination, thus transforming risk assessment into executable actions. Partition-level counterfactual analysis and virtual experiments replace local re-simulation, achieving a computationally efficient counterfactual deduction closed loop of "rapid evaluation → security verification → execution → review." Furthermore, in conjunction with the preceding steps, this invention provides confidence intervals, chains of evidence, and boundary conditions for each control action, ensuring it meets engineering audit and compliance requirements, and achieving explainability and accountability.

[0082] First, for each candidate action, the system performs transient time-domain simulation in the digital twin strictly according to the preset rhythm of that action. By accurately simulating the rise, stabilization period, and pullback of each action, rather than simply comparing steady-state results, rhythm reproduction is achieved. Furthermore, for actions that may cause transient electrical shocks, such as tap adjustments, a brief decoupled observation window can be set to separately assess the potential impact of their switching transients on the electric field and protection system.

[0083] During the simulation, the system calculates and records a comprehensive multi-objective evaluation index system in real time: I. Risk reduction benefits as a core indicator. This corresponds to the decrease in the overall score of each risk area after quantitative implementation, or the decrease in key performance indicators (such as hotspot temperature, maximum field strength).

[0084] II. Energy Consumption Cost. This represents the additional cost. It corresponds to the time integral of the additional power consumption of auxiliary equipment such as fans and oil pumps during the calculation and execution process.

[0085] III. Wear and Tear. This indicates the wear and tear on the equipment. It includes indicators reflecting mechanical wear, such as the number of tap changes and the cumulative changes in valve and pump speeds.

[0086] IV. Smoothness and Reversibility of Action. This involves assessing the smoothness of the control process and whether the plan can be safely withdrawn after execution.

[0087] V. System Coupling Penalty. Used to evaluate the side effects of the proposed solution on the power grid, such as whether it causes bus voltage fluctuations, excessive harmonics, or circulating currents between parallel transformers.

[0088] In this embodiment of the invention, the optimal solution is found from numerous candidate solutions through multi-objective aggregation and Pareto screening. Specifically, the first step is weighted sum model design. Based on preset business preferences (such as safety first or energy efficiency priority), weights are assigned to each evaluation index to construct a comprehensive cost function. During actual calculation, based on the comprehensive cost function and the multi-objective evaluation indexes, the comprehensive score of each candidate solution is calculated and initially ranked.

[0089] Secondly, Pareto optimal selection is employed. Simultaneously, a non-dominated sorting algorithm is used to find the Pareto optimal solution set. These solutions are characterized by the fact that they cannot improve on any one objective, nor worsen at least one other objective, thus providing users with multiple optimal trade-off options.

[0090] This invention also includes a minimum risk reduction threshold elimination mechanism. Specifically, a minimum risk reduction benefit threshold is set, and any scheme that fails to meet this requirement, regardless of how excellent its other indicators are, will be directly eliminated.

[0091] Through Pareto optimal selection, the final output is a decision recommendation package containing detailed predictive information. It includes the following: (1) Recommended scheme: the optimal control scheme corresponding to the best comprehensive score.

[0092] (2) Predicted trajectory: the predicted curves of the key performance indicators (such as hot spot temperature) of each scheme over time and their 95% confidence intervals.

[0093] (3) Comprehensive evaluation card: scores of each evaluation indicator and robustness evaluation results for each scheme.

[0094] (4) Implementation recommendations: specific implementation schedule and key monitoring points.

[0095] As discussed above, the candidate action set constructed in the embodiments of the present invention includes at least one target candidate action obtained after screening. Therefore, in some embodiments, the process of executing the candidate action set through a digital twin model and combining it with multi-objective optimization decision-making to determine the optimal control scheme may include the following steps S51 to S54: Step S51: For each target candidate action, perform transient time-domain simulation in the digital twin model according to the preset rhythm of the target candidate action, and record the multi-objective evaluation index of the target candidate action; Step S52: Obtain the comprehensive cost function constructed based on preset business preferences, and substitute the multi-objective evaluation index into the comprehensive cost function to calculate the comprehensive score of the target candidate action; Step S53: Combining Pareto optimal screening and minimum risk reduction elimination mechanisms, a non-dominated sorting algorithm is used to determine the optimal comprehensive score from multiple comprehensive scores; Step S54: Determine the target candidate action corresponding to the best comprehensive score as the optimal control scheme.

[0096] Step 105: Control the oil-immersed power equipment to execute the optimal control scheme, and optimize the digital twin model based on the control results.

[0097] This step is mainly based on the optimal control scheme obtained above. It involves issuing instructions and executing actions in a closed loop, and then optimizing the digital twin model in reverse based on the control results.

[0098] The system supports fully automatic and semi-automatic execution modes. In fully automatic mode, control commands are automatically issued in sequence. In semi-automatic mode, personnel confirmation is requested before critical operations. During execution, the system monitors key parameters such as hotspot temperature, maximum electric field strength, and oil flow status at high frequency, and compares the measured data with the predicted trajectory and confidence interval in the previous digital twin model virtual simulation experiment in real time.

[0099] In this embodiment of the invention, the key operations refer to the relevant categories of actions marked as moderate or intensity levels in the aforementioned action library and combination strategy that directly affect the key electrical-thermal-fluid indicators. In semi-automatic mode, these types of actions must be confirmed by technicians via a pop-up window before execution. Thus, based on the semi-automatic mode, relatively special control actions can be subject to secondary manual confirmation, ensuring accuracy before execution and preventing accidents.

[0100] This invention also incorporates an effect verification, causal calibration, and self-learning mechanism. Specifically, this mechanism is activated after the regulatory action is executed. It assesses the accuracy of the decision through prediction-actual reconciliation and fine-tunes the digital twin model and its mapping relationship using newly collected data, thus completing the transition from a single closed loop to continuous optimization.

[0101] The prediction-measurement reconciliation process involves comparing and verifying predicted and measured values ​​to identify discrepancies and analyze their causes. Its core purpose is to ensure the accuracy of predictions or to provide data support for subsequent decision-making. Specifically, the measured data within the execution window is precisely compared with the predicted trajectory of the digital twin model simulation experiment mentioned earlier, generating a quantitative evaluation report. More specifically, the measured key performance indicators (such as hotspot temperature and oil flow velocity) trajectories are first aligned point-by-point with the predicted trajectory and its confidence interval. Then, the accuracy is judged by the hit rate (the proportion of measured points falling within the predicted confidence interval). Based on preset standards (such as a hit rate > 85%), it is determined whether the prediction of this virtual simulation experiment "meets the standard."

[0102] If the hit rate fails to meet the preset standard, the system automatically determines that the prediction result of this virtual simulation experiment is "unsatisfactory" and triggers a joint correction process for the model and strategy. On the one hand, the digital twin model version and its control strategy corresponding to this experiment are marked as low confidence, prohibiting their direct inclusion in the online automatic control action library. At the same time, the control mode is downgraded from fully automatic to semi-automatic / recommended mode, for reference only by maintenance personnel. On the other hand, the aforementioned data assimilation and parameter re-evaluation calibration module is invoked to re-identify the correctable coefficients, boundary condition setpoints, etc., in the digital twin model. Only after the new prediction trajectory is verified by indicators such as hit rate under the same or similar working conditions can the automatic execution qualification of this type of strategy be restored.

[0103] Furthermore, this invention also incorporates a safety perturbation and empirical sensitivity estimation mechanism. Specifically, the various parameters of the transformer change continuously over time and with transformer aging. Therefore, the state-parameter mapping curve from "action to twin input" in the aforementioned steps is also necessarily constantly changing.

[0104] To more accurately calibrate the model, the system proactively initiates minimal exploratory operations under safe operating conditions. Specifically, within a time window where the risk level is "low" and the load is stable, minimal, short-duration exploratory adjustments are made to the pump speed, valves, or fans.

[0105] To address parameter drift over time and aging, the state-parameter mapping curve from "action to twin input" in the aforementioned steps (e.g., mapping the fan to the heat transfer coefficient) is used. The system is configured as an adaptive model and corrected by safe perturbations. In this case, the system performs only minimal, short-duration probing adjustments (e.g., [missing information]) within low-risk and boundary-stable time windows. The fan speed is set to ±1 level. The probe is immediately terminated and reverted to a conservative trajectory when any constraint approaches the threshold or an alarm occurs. During the perturbation period, the measurement sequence is first detrended and smoothed (Kalman smoothing), and then the empirical sensitivity is estimated (Jacobi approximation). ).

[0106] get Then, this is used as the correction direction for the "action-to-twin input" mapping. The parameterized mapping is updated in small steps using Recursive Least Squares (RLS), Normalized Least Mean Squares (NLMS), or spline or piecewise linear models with a forgetting factor to track aging. An example of the update formula is shown below. , where gain Simultaneously, physical consistency constraints are enforced to prevent numerical drift out of bounds. The entire "perturbation-response-correction" process is recorded as an atomic transaction, facilitating subsequent updates to the mapping parameters in the state-parameter mapping.

[0107] Through continuous operation of effect verification, causal calibration, and self-learning mechanisms, the digital twin evolves from a static model dependent on the initial design into a dynamic model that can evolve dynamically as equipment ages and the environment changes, thereby continuously improving the accuracy and reliability of its decisions.

[0108] This invention provides a method for risk assessment and control of oil-immersed power equipment based on digital twins. First, the relationship between "action → twin input" is transformed into an evolvable model, achieving adaptive mapping. In subsequent applications, model accuracy is continuously improved through perturbation adjustments. Minimal safety perturbations are implemented within low-risk windows, and the empirical Jacobian J is estimated in real-time, with the mapping updated in small steps under monotonicity and boundary constraints using RLS / NLMS / forgotten splines. This allows for long-term tracking of aging and seasonal drift, significantly reducing model deviation and ensuring that predictions and decisions consistently closely approximate real-world equipment. Second, through prior-based multi-objective decision-making, not only safety and economy are considered, but also the problem of on-site failure is avoided. All candidate actions are first transiently simulated within the twin according to a complete "rise → stabilize → fall" rhythm, comprehensively evaluating risk reduction, energy consumption, wear, and system coupling penalties. By employing weighted scoring and Pareto screening, and setting a minimum risk reduction threshold and a unit cost-benefit threshold, the optimal control scheme that can be directly executed is finally output. Simultaneously, a simple and auditable closed-loop self-learning method is used, with "confidence interval hit rate ≥ threshold" as the sole compliance criterion, to reconcile prediction and actual results. The atomic data of "context → action → prediction → actual result → conclusion → perturbation gain" is stored in a database, and re-assimilation and proxy retraining are performed daily / weekly or event-triggered, with shadow / canary / automatic rollback governance employed. The technical solution provided by this invention offers a unified evaluation caliber, traceable process, and an evaluation model where models and strategies can continuously evolve under controlled risk.

[0109] For better explanation, refer to Figure 2This diagram illustrates the overall process of a risk assessment and control method for oil-immersed power equipment based on digital twins, as provided in an embodiment of the present invention. It should be noted that this embodiment only provides a brief overview of the general process of risk assessment and control for oil-immersed power equipment based on digital twins. The specific implementation process of each step can be understood by referring to the relevant content in the foregoing embodiments, and will not be elaborated upon here. It is understood that the present invention does not impose any limitations on this.

[0110] Step 201: Collect power measurement data covering the entire state of oil-immersed power equipment, perform unified time synchronization and cleaning quality control on the power measurement data, and construct a snapshot of the current state of the oil-immersed power equipment considering multi-dimensional quantitative uncertainties; Step 202: Based on the design structure, material properties and physical laws of oil-immersed power equipment, construct a multiphysics basic model, and map the control signals and operating status of oil-immersed power equipment into the calculation parameters and boundary conditions of the multiphysics basic model in real time to form a multiphysics model; Step 203: Combining Kalman filtering, the correctable coefficients of the multiphysics model are assimilated and calibrated using the current state snapshot to obtain a digital twin model; Step 204: Perform multiphysics coupling solution based on digital twin model to obtain key multiphysics indicators, and construct candidate action set based on key multiphysics indicators and comprehensive risk assessment; Step 205: Execute the candidate action set through the digital twin model, and combine it with multi-objective optimization decision-making based on Pareto optimal screening and minimum risk reduction line elimination mechanism to determine the optimal control scheme; Step 206: Control the oil-immersed power equipment to execute the optimal control scheme, and based on the control results, combine the prediction-measurement reconciliation and the sensitivity of the safety perturbation estimation to optimize the digital twin model in real time.

[0111] Reference Figure 3 The diagram illustrates a structural block diagram of a risk assessment and control device for oil-immersed power equipment based on digital twins, provided by an embodiment of the present invention. Specifically, it may include: Data acquisition unit 301 is used to acquire the current state snapshot and multiphysics model of oil-immersed power equipment; The parameter calibration unit 302 is used to perform parameter calibration on the multiphysics model using the current state snapshot to obtain a digital twin model; The candidate action set construction unit 303 is used to perform multi-physics coupling solution based on the digital twin model, obtain multi-physics key indicators, and construct a candidate action set based on the multi-physics key indicators and a comprehensive risk assessment. The multi-objective optimization decision unit 304 is used to execute the candidate action set through the digital twin model and determine the optimal control scheme by combining multi-objective optimization decision. The optimal control scheme execution unit 305 is used to control the oil-immersed power equipment to execute the optimal control scheme and optimize the digital twin model based on the control results.

[0112] In one optional embodiment, the multiphysics model corresponds to multiphysics state variables and correctable coefficients; the parameter calibration unit 302 includes: An extended state vector construction unit is used to construct an extended state vector based on the multiphysics state variables and the correctable coefficients. The actual observation data extraction unit is used to extract actual observation data at multiple times under the multi-physics state variables from the current state snapshot, and form multiple sets of simulation samples. The forward propagation unit is used to propagate forward through each group of simulation samples in the multiphysics model to generate corresponding prediction observation data. The error covariance matrix construction unit is used to construct the error covariance matrix between the predicted observations and the actual observations based on the actual observation data of each group and the predicted observation data of each group. The parameter correction unit is used to calculate the Kalman gain of the error covariance matrix and feed the Kalman gain as the prediction residual back to the extended state vector to correct the correctable coefficients of the extended state vector. The forward propagation re-execution unit is used to re-perform forward propagation in the parameter-corrected multiphysics model using each set of simulation samples until the Kalman gain of the error covariance matrix is ​​less than a preset gain threshold, and outputs the final parameter update results of the correctable coefficients under each set of simulation samples. The optimal parameter estimation calculation unit is used to perform a weighted average of the update results of each parameter to obtain the optimal parameter estimate. The parameter update unit is used to update the correctable coefficients of the multiphysics model according to the optimal parameter estimate to obtain a digital twin model.

[0113] In one optional embodiment, the candidate action set construction unit 303 includes: The multiphysics equation construction unit is used to construct the electric field control equation and the magnetic-thermal-fluid coupling equation respectively, and to combine the electric field control equation and the magnetic-thermal-fluid coupling equation to construct the electric-thermal-fluid multiphysics equation considering space charge. The multiphysics equation solving unit is used to solve the electro-thermal-fluid multiphysics equation by taking the mapping parameters output by the digital twin model as the boundary conditions, and obtaining the global physical field distribution and key performance indicators. The indicator integration unit is used to integrate the global physical field distribution and the key performance indicators as key indicators of multiphysics.

[0114] In one optional embodiment, the multiphysics key indicators include global physics field distribution and key performance indicators; the candidate action set construction unit 303 includes: The risk area identification unit is used to identify the risk areas of the digital twin model based on the global physical field distribution and combined with the connected component clustering algorithm, and generate at least one continuous risk area. A multi-dimensional risk assessment unit is used to extract the target performance index corresponding to the risk area from the key performance index for each risk area, and to perform a multi-dimensional risk assessment on the risk area based on the target performance index to obtain thermal risk index and electrical risk index. The risk-dominant type determination unit is used to obtain a comprehensive risk index by weighted fusion based on the thermal risk index and the electrical risk index, and to determine the risk-dominant type of the risk area based on the comprehensive risk index; The main cause routing unit for regulatory actions is used to select the most relevant regulatory actions from a pre-built structured action library containing all available regulatory means based on the risk-dominant type through main cause routing. The initial candidate action generation unit is used to generate multiple initial candidate actions with a combination strategy of control actions within a restricted variable domain by combining rule priority and spatial sampling, based on the control actions of each of the risk regions. The action conflict trimming unit is used to trim the multiple initial candidate actions to obtain at least one target candidate action, so as to form a candidate action set.

[0115] In one optional embodiment, the candidate action set includes at least one target candidate action obtained after screening; the multi-objective optimization decision unit 304 includes: The transient time-domain simulation execution unit is used to perform transient time-domain simulation in the digital twin model for each of the target candidate actions according to the preset rhythm of the target candidate actions, and record the multi-objective evaluation index of the target candidate actions; The comprehensive score calculation unit is used to obtain a comprehensive cost function constructed according to preset business preferences, and to substitute the multi-objective evaluation index into the comprehensive cost function to calculate the comprehensive score of the target candidate action; The optimal comprehensive score screening unit is used to combine Pareto optimal screening and the lowest risk reduction line elimination mechanism, and to determine the optimal comprehensive score from multiple comprehensive scores using a non-dominated sorting algorithm. The optimal control scheme determination unit is used to determine the target candidate action corresponding to the optimal comprehensive score as the optimal control scheme.

[0116] In one optional embodiment, the apparatus further includes a current state snapshot construction unit; the current state snapshot construction unit includes: An electrical power measurement data acquisition unit is used to collect electrical power measurement data covering the entire state of the oil-immersed power equipment; The data preprocessing unit is used to perform unified time synchronization and cleaning quality control on the electrical force measurement data to obtain processed measurement data. The observation error variance calculation unit is used to calculate the observation error variance of each observation data in the processed measurement data by multi-dimensional quantification of uncertainty. The dynamic observation information matrix construction unit is used to construct a dynamic observation information matrix composed of observation vectors and time-varying covariance matrix based on all the observation error variances, while also considering various observation values ​​and quality labels with timestamp alignment. The current state snapshot forming unit is used to encapsulate the dynamic observation information matrix in a standardized format to form a current state snapshot of the oil-immersed power equipment.

[0117] In one optional embodiment, the apparatus further includes a multiphysics model building unit; the multiphysics model building unit includes: The multiphysics basic model building unit is used to build a three-dimensional multiphysics basic model based on the design structure, material properties and physical laws of the oil-immersed power equipment. The state mapping unit is used to acquire the control signals and operating status of the oil-immersed power equipment, and to map the control signals and operating status into the calculation parameters and boundary conditions of the multiphysics basic model in real time, thereby forming the multiphysics model of the oil-immersed power equipment.

[0118] As the device embodiment is basically similar to the method embodiment, it is described in a relatively simple way. For relevant details, please refer to the description of the method embodiment above.

[0119] This invention also provides an electronic device, which includes a processor and a memory: The memory is used to store program code and transfer the program code to the processor; The processor is used to execute the risk assessment and control method for oil-immersed power equipment based on digital twins according to the instructions in the program code of any embodiment of the present invention.

[0120] This invention also provides a computer-readable storage medium for storing program code, which is used to execute the digital twin-based risk assessment and control method for oil-immersed power equipment according to any embodiment of this invention.

[0121] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working processes of the systems, devices, and units described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here.

[0122] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, data stored, data displayed, etc.) involved in this invention are all information and data authorized by the user or fully authorized by all parties. Furthermore, the collection, use and processing of related data must comply with the relevant laws, regulations and standards of the relevant countries and regions, and corresponding operation entry points are provided for users to choose to authorize or refuse.

[0123] In the embodiments provided by this invention, it should be understood that the disclosed systems, apparatuses, and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be an indirect coupling or communication connection between devices or units through some interfaces, and may be electrical, mechanical, or other forms.

[0124] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0125] Furthermore, the functional units in the various embodiments of the present invention can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit.

[0126] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0127] The above-described embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit it. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for risk assessment and control of oil-immersed power equipment based on digital twins, characterized in that, include: Obtain a snapshot of the current state and a multiphysics model of oil-immersed power equipment; The current state snapshot is used to calibrate the parameters of the multiphysics model to obtain a digital twin model; Based on the digital twin model, multiphysics coupling solution is performed to obtain key multiphysics indicators, and a candidate action set is constructed based on the key multiphysics indicators and a comprehensive risk assessment. The candidate action set is executed through the digital twin model, and the optimal control scheme is determined by combining multi-objective optimization decision-making. The oil-immersed power equipment is controlled to execute the optimal control scheme, and the digital twin model is optimized based on the control results.

2. The method for risk assessment and control of oil-immersed power equipment based on digital twins according to claim 1, characterized in that, The multiphysics model corresponds to multiphysics state variables and correctable coefficients; the step of using the current state snapshot to calibrate the parameters of the multiphysics model to obtain a digital twin model includes: Based on the multiphysics state variables and the correctable coefficients, an extended state vector is constructed; The actual observation data at multiple times under the multiphysics state variables are extracted from the current state snapshot to form multiple sets of simulation samples; Each set of simulation samples is forward-propagated in the multiphysics model to generate corresponding predicted observation data. Based on the actual observation data and the predicted observation data of each group, construct the error covariance matrix between the predicted observations and the actual observations. Calculate the Kalman gain of the error covariance matrix and feed the Kalman gain as the prediction residual back to the extended state vector to correct the parameters of the correctable coefficients of the extended state vector; The simulation samples from each group are used to perform forward propagation again in the multiphysics model after parameter correction until the Kalman gain of the error covariance matrix is ​​less than a preset gain threshold. The final parameter update results of the correctable coefficients under each group of simulation samples are then output. The optimal parameter estimate is obtained by weighted averaging of the update results of each parameter. The correctable coefficients of the multiphysics model are updated based on the optimal parameter estimates to obtain a digital twin model.

3. The method for risk assessment and control of oil-immersed power equipment based on digital twins according to claim 1, characterized in that, The multiphysics coupling solution is performed based on the digital twin model to obtain key multiphysics indicators, including: The electric field control equation and the magnetic-thermal-fluid coupling equation are constructed separately, and the electric field control equation and the magnetic-thermal-fluid coupling equation are combined to construct the electric-thermal-fluid multiphysics field equation considering space charge. The mapping parameters output by the digital twin model are used as the boundary conditions to solve the electro-thermal-fluid multiphysics equations, thereby obtaining the global physical field distribution and key performance indicators. The global physical field distribution and the key performance indicators are integrated as key indicators for multiphysics.

4. The method for risk assessment and control of oil-immersed power equipment based on digital twins according to claim 1, characterized in that, The key indicators of the multiphysics field include the global physical field distribution and key performance indicators; the construction of a candidate action set based on the key indicators of the multiphysics field and a comprehensive risk assessment includes: Based on the global physical field distribution, and combined with the connected component clustering algorithm, the risk regions of the digital twin model are identified, and at least one continuous risk region is generated. For each of the aforementioned risk areas, the target performance indicators corresponding to the risk areas are extracted from the key performance indicators, and a multi-dimensional risk assessment is performed on the risk areas based on the target performance indicators to obtain thermal risk indicators and electrical risk indicators. Based on the thermal risk index and the electrical risk index, a comprehensive risk index is obtained through weighted fusion, and based on the comprehensive risk index, the dominant risk type of the risk area is determined; Based on the aforementioned risk-dominant type, the most relevant control actions are selected from a pre-built structured action library containing all available control measures through cause-based routing. Based on the control actions of each of the aforementioned risk regions, multiple initial candidate actions with control action combination strategies are generated within the restricted variable domain by combining rule priority and spatial sampling. Action conflict pruning is performed on the multiple initial candidate actions to obtain at least one target candidate action, thus forming a candidate action set.

5. The method for risk assessment and control of oil-immersed power equipment based on digital twins according to claim 1, characterized in that, The candidate action set includes at least one target candidate action obtained after screening; The step of executing the candidate action set through the digital twin model and determining the optimal control scheme by combining multi-objective optimization decision-making includes: For each of the target candidate actions, a transient time-domain simulation is performed in the digital twin model according to the preset rhythm of the target candidate action, and the multi-objective evaluation index of the target candidate action is recorded. Obtain a comprehensive cost function constructed based on preset business preferences, and substitute the multi-objective evaluation index into the comprehensive cost function to calculate the comprehensive score of the target candidate action; Combining Pareto optimal screening and minimum risk reduction line elimination mechanisms, a non-dominated sorting algorithm is used to determine the optimal comprehensive score from multiple comprehensive scores. The target candidate action corresponding to the optimal comprehensive score is determined as the optimal control scheme.

6. The method for risk assessment and control of oil-immersed power equipment based on digital twins according to any one of claims 1 to 5, characterized in that, The process of constructing the current state snapshot includes: Collect electrical power measurement data covering the entire state of the oil-immersed power equipment; The electrical force measurement data is uniformly synchronized and cleaned for quality control to obtain processed measurement data; For each observation data in the processed measurement data, the observation error variance of the observation data is calculated by multi-dimensional quantification of uncertainty. Based on all the observation error variances, and taking into account various timestamp-aligned observations and quality labels, a dynamic observation information matrix is ​​constructed, which is composed of the observation vector and the time-varying covariance matrix. The dynamic observation information matrix is ​​encapsulated in a standardized format to form a snapshot of the current state of the oil-immersed power equipment.

7. The method for risk assessment and control of oil-immersed power equipment based on digital twins according to any one of claims 1 to 5, characterized in that, The construction process of the multiphysics model includes: Based on the design structure, material properties and physical laws of the oil-immersed power equipment, a three-dimensional multiphysics model is constructed. The control signals and operating status of the oil-immersed power equipment are acquired, and the control signals and operating status are mapped in real time to the calculation parameters and boundary conditions of the multiphysics basic model to form the multiphysics model of the oil-immersed power equipment.

8. A risk assessment and control device for oil-immersed power equipment based on digital twins, characterized in that, include: The data acquisition unit is used to acquire a current state snapshot and a multiphysics model of the oil-immersed power equipment. A parameter calibration unit is used to perform parameter calibration on the multiphysics model using the current state snapshot to obtain a digital twin model; The candidate action set construction unit is used to perform multi-physics coupling solution based on the digital twin model, obtain multi-physics key indicators, and construct a candidate action set based on the multi-physics key indicators and a comprehensive risk assessment. A multi-objective optimization decision-making unit is used to execute the candidate action set through the digital twin model and determine the optimal control scheme by combining multi-objective optimization decision-making. The optimal control scheme execution unit is used to control the oil-immersed power equipment to execute the optimal control scheme and optimize the digital twin model based on the control results.

9. An electronic device, characterized in that, The device includes a processor and a memory: The memory is used to store program code and transmit the program code to the processor; The processor is used to execute the risk assessment and control method for oil-immersed power equipment based on digital twins as described in any one of claims 1-7, according to the instructions in the program code.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium is used to store program code for executing the risk assessment and control method for oil-immersed power equipment based on digital twins as described in any one of claims 1-7.

Images