A direct test loop digital twin system construction method of a power distribution device

Abstract

The application discloses a kind of direct test loop digital twin system construction methods of power distribution device, belong to high-voltage electrical apparatus test and digital twin field, method includes constructing test loop physical entity and digital model, and determine the electrical parameter and physical structure of each element in loop;Based on the domain of open source finite element software solution model, boundary condition, and calculate circuit breaker electric field by adaptive mesh;Test data are collected and fused with simulation data, and construct twin database;From twin database, select key characteristic parameters, combine the physical parameter values of all characteristic points in a certain breaking state into feature vector, and the breaking degree of circuit breaker corresponding to this state is taken as target value, to form complete training sample, construct GA-BP neural network model, and train neural network model.The application realizes the multi-physical field simulation, state prediction and operation optimization of test loop, realizes high-fidelity simulation, dynamic optimization and safety control of test process.

Description

Technical Field

[0001] This invention belongs to the field of high-voltage electrical testing and digital twin technology, and particularly relates to a method for constructing a direct test circuit digital twin system for a power distribution device. Background Technology

[0002] With rapid societal development and continuously rising electricity demand, the power grid capacity has expanded year by year, leading to a sustained increase in short-circuit current levels in the power system. The short-circuit currents in many substations have far exceeded the breaking capacity of switchgear, posing a significant threat to the safe and stable operation of the power grid. Therefore, the performance of switchgear and power transmission and transformation facilities such as transformers directly affects the safety of the power system. Conducting high-voltage, high-current tests on such equipment is of paramount importance.

[0003] Among numerous power transmission and transformation equipment, switchgear plays a crucial role in ensuring the stable operation of the power system and improving power quality. Typical switchgear in high-voltage, high-current circuits mainly includes disconnecting switches, load switches, and circuit breakers. Among these, circuit breakers, as the most critical and structurally complex type of high-voltage switchgear, are typically referred to as high-voltage circuit breakers in 3kV and above power systems. They are not only the last line of defense for the power system but also one of the core components ensuring the safe and stable operation of the power grid. Regardless of the power system's operating condition—no-load, loaded, or short-circuit fault—as long as the circuit breaker receives an operating command, it must reliably close or open the circuit. High-voltage circuit breakers are critical protection and control devices in the power system, and their short-circuit breaking capacity directly affects the safe and stable operation of the system. However, traditional direct short-circuit breaking tests rely on large test stations and complex circuits, resulting in high testing costs, long cycles, and safety risks.

[0004] Digital twin technology has seen rapid development in the power equipment field. By constructing high-fidelity models in virtual space, it enables real-time mapping, state prediction, and performance evaluation of physical equipment. Research on digital twins for high-voltage circuit breakers mainly focuses on single-physics simulations of electric, airflow, and temperature fields, lacking multi-physics coupling and dynamic characteristic simulations covering the entire short-circuit current breaking process. Therefore, there is an urgent need for a digital twin system for high-voltage AC circuit breaker short-circuit breaking test circuits that can integrate information from multiple fields such as electromagnetics, thermodynamics, and mechanics to achieve virtual-real synchronization, predictive control, and optimization of the test process. Summary of the Invention

[0005] Purpose of the invention: To address the core challenges in short-circuit breaking tests of high-voltage AC circuit breakers, such as insufficient accuracy of multi-physics coupling simulation, high test costs and risks, scarce failure data, difficulty in online model calibration, and low reliability of real-time safety decisions, this invention aims to provide a method for constructing a digital twin system for the direct test circuit of a power distribution device, which can be used to realize multi-physics simulation, state prediction, and operation optimization of the test circuit.

[0006] Technical solution: The present invention provides a method for constructing a digital twin system for a direct test circuit of a power distribution device, comprising the following steps:

[0007] Step 1: Construct the physical entity and digital model of the test circuit, and determine the electrical parameters and physical structure of each component in the circuit;

[0008] Step 2: Solve the domain and boundary conditions of the model using open-source finite element software, and calculate the circuit breaker electric field using adaptive mesh.

[0009] Step 3: Collect real experimental data and deeply integrate it with simulation data to construct a twin database;

[0010] Step 4: Select key feature parameters from the twin database, including electric field strength, airflow velocity, gas density, gas pressure and temperature. Combine the physical parameter values ​​of all feature points under a certain interruption state into a feature vector, and take the interruption degree of the circuit breaker corresponding to that state as the target value to form a complete training sample. Randomly divide all samples into training set and test set.

[0011] Step 5: Construct a GA-BP neural network model. Train and test the GA-BP neural network model using training and testing sets. Deploy the trained GA-BP neural network model into the digital twin system as a state recognition module.

[0012] Step 6: Construct a digital twin closed-loop framework that integrates a physical test circuit, a multi-physics simulation engine, a data management module, and a state identification module. Develop a graphical human-computer interaction interface to dynamically display the electrical wiring diagram and key parameters of the test circuit, as well as the dynamic cloud map distribution of the multi-physics field in the circuit breaker arc extinguishing chamber in real time.

[0013] Furthermore, step 1 specifically includes the following steps:

[0014] Step 1.1: Construct an electrical wiring model of the direct test circuit, including the impulse generator, main capacitor, adjustable reactor, closing switch, high-voltage AC circuit breaker under test, and measurement system, and determine the electrical parameters and physical structure of each component in the circuit; the electrical parameters include capacitance, inductance, resistance, and transient impedance of the generator;

[0015] Step 1.2: Establish a three-dimensional geometric model of the circuit breaker, describing the structure of the moving and stationary contacts, nozzles, air cylinders, and shielding cover; based on the physical mechanism, construct a multi-physics coupling mathematical model of the arc-extinguishing chamber;

[0016] The electrostatic field model employs the Laplace equation The spatial distribution of electric potential is described, and the electric field intensity is obtained by solving the problem using the finite element method. Where ε_r represents the dielectric constant, Indicates electric potential. This represents the vector differential operator, and grad represents the gradient.

[0017] The airflow and temperature fields are described based on the Navier-Stokes equations, including:

[0018]

[0019]

[0020]

[0021] Where ρ is the gas density, t represents time, V represents the velocity vector, p represents the gas pressure, τ is the viscous stress tensor, T represents the absolute temperature of the gas, and C P This represents the specific heat capacity at constant pressure, k represents the thermal conductivity of the gas, and Q is the source term, which includes the energy of the electric arc.

[0022] The turbulence model adopts the standard k-ε two-equation turbulence model, which accurately simulates the turbulence effect of high-speed, compressible airflow in the arc-extinguishing chamber by solving the transport equations of turbulent kinetic energy k and turbulent dissipation rate ε.

[0023] Furthermore, in step 2, the open-source finite element software includes FreeFem++ and OpenFOAM; FreeFem++ is responsible for the finite element calculation of the electrostatic field, while OpenFOAM is responsible for the coupled calculation of the airflow field and temperature field; the two transmit data in one direction through a data interface, that is, the electrostatic field calculation result can be used as a known condition to be input into the airflow field-temperature field calculation, so as to achieve efficient weak coupling solution.

[0024] Furthermore, step 2 specifically includes the following steps:

[0025] Step 2.1: Import the established two-dimensional or three-dimensional geometric model of the arc-extinguishing chamber into FreeFem++, and use its built-in Delaunay-Voronoi algorithm automatic mesh generator to perform finite element mesh generation;

[0026] Step 2.2: Import the arc-extinguishing chamber geometric model into OpenFOAM, use the snappyHexMesh mesh generation tool to automatically generate a high-quality hexahedral-dominated polyhedral mesh based on the geometric shape; perform local refinement in the key flow regions of the nozzle and contact gap, and solve the problem based on the OpenFOAM solver;

[0027] Step 2.3: Perform time-space alignment and integration of the electric field data output by FreeFem++ and the flow-temperature field data output by OpenFOAM; finally, output the complete dynamic distribution map of the internal multiphysics field of the circuit breaker under different opening states and the parameter history curve of the feature points, which constitutes a high-fidelity twin database for driving the state identification of the digital twin.

[0028] Furthermore, step 3 specifically includes the following steps:

[0029] Step 3.1: In the physical short-circuit breaking test circuit, deploy a sensor network to collect data in real time; install high-voltage probes on both sides of the circuit breaker break to measure the recovery voltage and transient recovery voltage; for the current signal, use a Rogowski coil or current transformer to measure the waveform, amplitude, and zero-crossing characteristics of the breaking current; use a linear displacement sensor to monitor the stroke curve of the moving contact in real time, and obtain the opening distance, overtravel, and opening / closing speed to obtain the displacement signal; collect the current waveform of the opening / closing coil to analyze the mechanical state of the operating mechanism; install pressure and temperature sensors in the arc-extinguishing chamber of the circuit breaker to monitor the pressure and temperature changes of the gas during the breaking process.

[0030] Step 3.2: Apply a uniform, high-precision timestamp to all collected real-time data to ensure strict alignment of data from different sources on the timeline. Then, denoise and remove outliers from the original signal. Finally, normalize the data, scaling it to the [0, 1] interval. The normalization formula is as follows:

[0031]

[0032] In the formula, x i For the original data, x min and x max These are the minimum and maximum values ​​of the data segment, respectively.

[0033] Step 3.3: After data processing is completed, all data are integrated to build a twin database. The database driver module is updated and feedback is closed. After each new short-circuit breaking test, the newly collected data is added to the twin database.

[0034] Furthermore, in step 5, the GA-BP neural network model includes an input layer, a hidden layer, and an output layer. The number of nodes in the input layer is determined based on the total number of selected feature parameters. The hidden layer is set to 5 nodes and uses the Sigmoid activation function. The output layer has 1 node for outputting the predicted value of the opening distance percentage. The model training is divided into two stages: First, a genetic algorithm is used for global optimization. After binary encoding the network parameters, an initial population of 50 is generated. Individuals are selected based on fitness using the roulette wheel selection method. Single-point crossover is performed with a crossover probability of 0.8, and basic bit mutation is performed with a mutation probability of 0.09. After 100 generations of iterative evolution, the optimal initial parameters are obtained. Then, the error backpropagation algorithm is used for fine training. The learning rate is set to 0.01, and the network parameters are fine-tuned using the gradient descent method until the model converges.

[0035] The present invention also discloses a computer device, including a memory, a processor, and a computer program stored in the memory, wherein the processor executes the computer program to implement the steps of the method of the present invention.

[0036] The present invention also discloses a computer-readable storage medium having a computer program / instructions stored thereon, which, when executed by a processor, implements the steps of the method of the present invention.

[0037] The present invention also discloses a computer program product, including a computer program / instructions that, when executed by a processor, implement the steps of the method of the present invention.

[0038] Beneficial effects: Compared with the prior art, the present invention has the following significant advantages:

[0039] This invention innovatively constructs a unified digital twin model that fully couples "electromagnetism-thermal-fluidity-mechanical" systems, achieving for the first time a high-fidelity simulation of the entire dynamic process of short-circuit breaking, overcoming the challenges of strong nonlinear coupling and efficient solution in multi-physics fields. Based on this, a hybrid intelligent driving strategy of "mechanism model defining boundaries, data model for identification" is proposed. Rich virtual sensing data is generated through multi-physics simulation, driving a neural network optimized by a genetic algorithm for state identification. This effectively solves the pain points of scarce data and high noise in experiments, achieving a very high level of accuracy in predicting key state variables with high goodness of fit and low root mean square error. To ensure the long-term reliability of the model, this invention introduces an online collaborative calibration mechanism based on extended Kalman filtering and recursive least squares, enabling the digital twin to dynamically track device performance degradation and achieve adaptive evolution. This system successfully applies digital twins to the intelligent decision-making stage of experiments. Through massive "digital experiment" rehearsals, experimental parameters can be optimized and risks identified in advance. In practical applications, it can significantly reduce the number of high-cost physical experiments, significantly improving experimental safety and economy. Ultimately, by integrating uncertainty quantification technologies such as Monte Carlo Dropout, this system can not only make deterministic judgments, but also provide probabilistic confidence interval assessments for whether the interruption is successful or not. This achieves a leap from traditional simulation to reliable decision-making, providing a complete solution for intelligent testing and operation and maintenance of power equipment. Attached Figure Description

[0040] Figure 1 This is a schematic diagram of the multi-physics coupling principle of a circuit breaker.

[0041] Figure 2 A diagram showing the physical model of a vacuum circuit breaker and the characteristic points for monitoring electric field intensity.

[0042] Figure 3 The diagrams show the electric field intensity distribution, where (a) is the electric field intensity distribution at characteristic point AB inside the vacuum circuit breaker arc-extinguishing chamber, and (b) is the electric field intensity distribution at characteristic point CD inside the vacuum circuit breaker arc-extinguishing chamber.

[0043] Figure 4 The diagram shows the distribution characteristics of physical quantities at sampling points AB as a function of location. In this diagram, (a), (b), (c), and (d) are the distribution characteristics of physical quantities such as velocity, pressure, temperature, and density at sampling points AB as a function of location, respectively.

[0044] Figure 5 The diagram shows the distribution characteristics of physical quantities at the CD sampling point as a function of location. (a), (b), (c), and (d) are the distribution characteristics of physical quantities such as velocity, pressure, temperature, and density at the CD sampling point as a function of location, respectively.

[0045] Figure 6 This is a flowchart of neural network prediction based on genetic algorithms.

[0046] Figure 7 A flowchart of multi-physics coupling for digital twin circuit breakers;

[0047] Figure 8 The following are the circuit breaker status identification results based on electric field intensity twin data using a BP neural network: (a) is a comparison of actual and predicted values ​​in the training set, and (b) is a comparison of actual and predicted values ​​in the test set.

[0048] Figure 9 The identification results of the multi-physics digital twin data of the circuit breaker are shown in (a) and (b) respectively. (a) is a comparison chart of the actual values ​​and predicted values ​​of the training set and (b) is a comparison chart of the actual values ​​and predicted values ​​of the test set.

[0049] Figure 10 This is a flowchart of the present invention. Detailed Implementation

[0050] The technical solution of the present invention will be further described below with reference to the accompanying drawings.

[0051] like Figure 10 As shown, the process of this invention includes the following steps:

[0052] Step 1, according to Figure 1 Please provide an explanation. Figure 1 This is a multi-physical coupling principle diagram of a high-voltage circuit breaker. As can be seen from the diagram, the high-voltage circuit breaker operates under the combined action of multiple physical fields such as electric field, airflow field, and temperature field, and is affected by complex physical and chemical processes.

[0053] Step 1.1: Establish the electrostatic field calculation model. Since the electric field inside the arc-extinguishing chamber of the circuit breaker satisfies the Laplace equation and boundary conditions, the finite element method is used to study the electrostatic field of the circuit breaker. The Laplace equation is:

[0054]

[0055] In the formula, ε_r is the dielectric constant. It represents the electrical potential.

[0056] Boundary conditions can be set to the moving contact potential. Static contact potential The boundaries of the remaining insulating components satisfy the second type of boundary conditions.

[0057] By solving the above equations, the space potential distribution can be obtained, and further... Calculate the electric field strength.

[0058] Step 1.2: Establish a mathematical model for the gas field and temperature field. The gas can be considered as a compressible, viscous, non-isothermal Newtonian fluid. Thus, its flow and heat transfer processes can be described by the following governing equations: the mass conservation equation, the momentum conservation equation, and the energy conservation equation.

[0059]

[0060]

[0061] In the formula, τ is the viscous stress tensor.

[0062]

[0063] Here, Q is the source term, which may contain the electric arc energy.

[0064] For the gas, the ideal gas law p=ρRT is used, relating pressure, density, and temperature. For the high-speed airflow within the arc-extinguishing chamber, the standard k-ε turbulence model is selected. This model simulates turbulence effects by solving two transport equations: the turbulent kinetic energy k and the turbulent dissipation rate ε.

[0065]

[0066]

[0067] in, The empirical constant takes the value of C. 1ε =1.44, C 2ε =1.92, Cμ=0.09, σ k =1, σ ε =1.3.

[0068] Finally, based on the model established in steps 1.1 and 1.2, a weakly coupled multiphysics solution is used. That is, the electrostatic field is solved independently first, and the resulting electric field distribution is used as a known condition; then the airflow field and temperature field are solved in a coupled manner.

[0069] Step 2 involves building a co-simulation platform using two open-source software programs: FreeFem++ and OpenFOAM. FreeFem++ is specifically responsible for the finite element calculation of the electrostatic field, while OpenFOAM is specifically responsible for the coupled calculation of the airflow and temperature fields. The two programs transmit data unidirectionally via a data interface, meaning that the electrostatic field calculation results can be used as known conditions input into the airflow and temperature field calculations to achieve efficient, weakly coupled solutions.

[0070] Step 2.1 involves importing the two-dimensional or three-dimensional geometric model of the arc-extinguishing chamber established in Step 1 into FreeFem++, and using its built-in Delaunay-Voronoi algorithm automatic mesh generator to perform finite element mesh generation. For regions with drastic changes in electric field gradient, an adaptive function is called to refine the local mesh, effectively controlling the number of meshes while ensuring computational accuracy.

[0071] Based on the boundary conditions set in step 1, the moving contact potential φ=0, the stationary contact potential φ=0, and the stationary contact potential φ=U0 are explicitly specified in the FreeFem++ script, with the rest being insulating boundaries. Using the high-level programming language in FreeFem++, the variational form of the Laplace equation is defined, and a linear solver such as the conjugate gradient method is selected for numerical solution. After the solution is completed, the electric field intensity and potential distribution are directly visualized using FreeFem++'s post-processing capabilities.

[0072] Step 2.2: Import the arc-extinguishing chamber geometric model into OpenFOAM. Use the snappyHexMesh mesh generation tool to automatically generate a high-quality hexahedral-dominated polyhedral mesh based on the geometric shape. Local mesh refinement is applied to key flow regions such as the nozzle and contact gap to ensure accurate capture of high-speed, compressible flow. To simulate the gate opening process, dynamic meshing technology is enabled in OpenFOAM to accurately describe the motion of the moving contact and pressure chamber. Finally, a custom solver is written based on OpenFOAM's standard solver. This solver integrates the governing equations described in Step 1: the Navier-Stokes equations, the energy equation, the k-ε turbulence model, and the functional relationships between gas physical parameters and temperature.

[0073] Step 2.3 involves temporal and spatial alignment and integration of the electric field data output by FreeFem++ and the flow-temperature field data output by OpenFOAM. The final output includes a complete dynamic distribution map of the internal multiphysics field of the circuit breaker under different opening states, along with the parameter history curves of feature points, forming a high-fidelity twin database used to drive the state identification of the digital twin.

[0074] Step 3: By collecting real experimental data and deeply integrating it with simulation data, a twin database is constructed to drive the continuous evolution and precise mapping of the digital twin.

[0075] Step 3.1: Deploy a sensor network in the physical short-circuit breaking test circuit to collect data in real time. Install high-voltage probes on both sides of the circuit breaker break to measure the recovery voltage and transient recovery voltage. For current signals, use Rogowski coils or current transformers to accurately measure the waveform, amplitude, and zero-crossing characteristics of the breaking current. Use linear displacement sensors to monitor the stroke curve of the moving contact in real time, accurately obtaining the opening distance, overtravel, and opening / closing speed to obtain displacement signals. By collecting the current waveform of the opening / closing coils, the mechanical state of the operating mechanism can be analyzed. Install pressure and temperature sensors in the arc-extinguishing chamber of the circuit breaker to monitor the pressure and temperature changes of the gas during the breaking process.

[0076] Step 3.2: Apply a uniform, high-precision timestamp to all collected real-time data to ensure strict alignment of data from different sources on the timeline. Then, perform noise reduction and outlier removal on the original signal to improve data quality. Finally, normalize the data using the following formula to scale it to the [0, 1] interval, eliminating the influence of dimensions and preparing it for subsequent neural network training.

[0077]

[0078] In the formula, x i For the original data, x min and x max These are the minimum and maximum values ​​of the data segment, respectively.

[0079] After all the data has been processed, the data is integrated to build a structured circuit breaker digital twin historical database. Each data sample in the database contains a set of multi-physics field characteristic parameters consisting of electric field strength, airflow velocity, temperature, density, pressure, etc., at a specific time or at a specific opening distance, as well as the actual opening state of the circuit breaker at that time.

[0080] Step 3.3 involves updating and implementing closed-loop feedback for the database-driven module. After each new short-circuit breaking test, the newly collected data will be added to the database. The digital twin model can then use this new data for retraining and parameter fine-tuning, enabling the model to self-evolve and more accurately track performance changes caused by mechanical wear, electrical life consumption, etc., thus forming a closed-loop feedback optimization mechanism of "measurement-fusion-modeling-verification-update".

[0081] Step 4: First, based on Step 3, select key feature parameters from the twin historical database, including electric field strength, airflow velocity, gas density, gas pressure, and temperature. Combine the physical parameter values ​​of all feature points under a certain interruption state into a feature vector, and use the circuit breaker interruption degree corresponding to that state as the target value to form a complete training sample. Randomly divide all samples into a training set and a test set, with the training set accounting for 70% of the total samples and the test set accounting for 30%. Before model training, a normalization formula is used to preprocess the input feature data and target value, scaling them to the [0,1] interval to eliminate the influence of dimensions.

[0082] Next, a GA-BP neural network model is constructed, consisting of an input layer, hidden layers, and an output layer. The number of nodes in the input layer is determined based on the total number of selected feature parameters. The hidden layer has 5 nodes and uses the Sigmoid activation function. The output layer has 1 node to output the predicted opening percentage. Model training is divided into two stages: First, a genetic algorithm is used for global optimization. After binary encoding the network parameters, an initial population of 50 is generated. Individuals are selected based on fitness using a roulette wheel selection method, with single-point crossover at a crossover probability of 0.8 and basic bit mutation at a mutation probability of 0.09. After 100 generations of iterative evolution, the optimal initial parameters are obtained. Then, a backpropagation algorithm is used for fine training, with a learning rate of 0.01. The network parameters are fine-tuned using gradient descent until the model converges.

[0083] Finally, the trained GA-BP neural network model was deployed into the digital twin system as the core state identification module. During real-time prediction, the system inputs the acquired feature parameters into the model to output the predicted value of the current circuit breaker's open state. Validated on the test set, the model performs excellently across multiple evaluation metrics, with a goodness-of-fit R² reaching 1. The root mean square error, mean absolute percentage error, and mean absolute error are all maintained at extremely low levels, fully demonstrating the high accuracy and reliability of this state identification method.

[0084] Example

[0085] To illustrate the implementation of the present invention and verify its effectiveness, a complete digital twin system was constructed using a 550kVSF6 high-voltage circuit breaker as an example.

[0086] First, a three-dimensional geometric model was created based on the product drawings, including the arc-extinguishing chamber, moving and stationary contacts, nozzle, and compressor cylinder (its two-dimensional cross-sectional diagram is shown in Figure 1). Figure 2 (As shown). Subsequently, based on this geometric model, mathematical models of electrostatic field, airflow-temperature field and turbulence coupling as described in the background art were constructed.

[0087] The solution was obtained using a co-simulation platform combining FreeFem++ and OpenFOAM. The electric field intensity distribution calculated by FreeFem++ is shown below. Figure 3 As shown, where Figure 3 (a) and Figure 3 Figure (b) shows the changes in electric field intensity along lines AB and CD between the moving and stationary contacts, respectively. Analysis Figure 3 It can be seen that the electric field strength is most concentrated at the tip of the contact and the edge of the nozzle, which is consistent with the theoretical expectation and verifies the correctness of the electric field model. At the same time, the trend of the electric field strength decreasing with the increase of the opening distance provides a key basis for insulation design.

[0088] A physical test circuit is built at the high-voltage test station, and data such as interruption current, recovery voltage, contact displacement, gas chamber pressure and temperature are collected synchronously. The data are aligned with the simulation data in step 2 on the time axis to build a twin database.

[0089] Based on simulation results from OpenFOAM, we extracted the distribution of airflow parameters on the critical paths (AB and CD) inside the arc extinguishing chamber.

[0090] Figure 4 and Figure 5 The characteristic curves of airflow velocity, pressure, temperature, and density as a function of location are shown on sampling lines AB and CD, respectively. Figure 4 , Figure 5 Analysis shows that near the nozzle throat (corresponding to a specific location on the curve), the airflow velocity reaches supersonic speeds, pressure and temperature change drastically in this region, while the gas density decreases due to expansion. These distribution patterns are completely consistent with the physical mechanism of gas-blown arc extinguishing in SF6 circuit breakers, proving the validity of the multiphysics coupling model. Figure 7 The process shown can reproduce the complex fluid and heat transfer phenomena during the switching process with high fidelity.

[0091] Extract feature parameters from the twin database, according to Figure 6 The genetic algorithm optimization process shown is used to train the GA-BP neural network model. For comparison, a standard BP neural network model was also trained. The trained model was then used to predict the contact opening distance during the circuit breaker's breaking process.

[0092] Figure 8 The prediction results of a BP neural network based on a single electric field intensity feature are presented. Analysis Figure 8 As can be seen from (a) and (b) in the figure, although the predicted values ​​can roughly track the trend, there are visible biases on the test set, indicating that the ability of a single physical field information to represent the state is limited.

[0093] Figure 9The results of GA-BP neural network predictions based on multi-physics field features such as electric field, airflow, temperature, pressure, and density are presented. Comparative analysis is also provided. Figure 9 (a) and (b) in Figure 8 It is obvious that: Figure 9 The predicted curve almost perfectly matches the actual curve. Quantitative calculations show that the goodness-of-fit R² of the GA-BP model reaches 0.9993, and the root mean square error (RMSE) is only 0.15%. This fully demonstrates that by employing the multi-physics fusion features and GA-BP algorithm proposed in this invention, real-time identification of circuit breaker status with sub-millisecond precision can be achieved.

[0094] The above are merely preferred embodiments of the present invention, but do not limit the patent scope of the present invention. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions described in the foregoing specific embodiments, or make equivalent substitutions for some of the technical features. Any equivalent structures made using the content of the present invention specification and drawings, directly or indirectly applied to other related technical fields, are similarly within the patent protection scope of the present invention.

Claims

1. A method for constructing a digital twin system for a direct test loop of a power distribution device, characterized in that, Includes the following steps: Step 1: Construct the physical entity and digital model of the test circuit, and determine the electrical parameters and physical structure of each component in the circuit; Step 1 specifically includes the following steps: Step 1.1: Construct an electrical wiring model of the direct test circuit, including the impulse generator, main capacitor, adjustable reactor, closing switch, high-voltage AC circuit breaker under test, and measurement system, and determine the electrical parameters and physical structure of each component in the circuit; the electrical parameters include capacitance, inductance, resistance, and transient impedance of the generator; Step 1.2: Establish a three-dimensional geometric model of the circuit breaker, describing the structure of the moving and stationary contacts, nozzles, air cylinders, and shielding cover; based on the physical mechanism, construct a multi-physics coupling mathematical model of the arc-extinguishing chamber; The electrostatic field model employs the Laplace equation The spatial distribution of electric potential is described, and the electric field intensity is obtained by solving the problem using the finite element method. Where ε_r represents the dielectric constant, Indicates electric potential. This represents the vector differential operator, and grad represents the gradient. The airflow and temperature fields are described based on the Navier-Stokes equations, including: ； ； ； Where ρ represents gas density, t represents time, V represents velocity vector, p represents gas pressure, τ represents viscous stress tensor, and C p represents the specific heat capacity at constant pressure, k represents the thermal conductivity of the gas, and Q represents the source term, which includes the energy of the electric arc. The turbulence model adopts the standard k-ε two-equation turbulence model, which accurately simulates the turbulence effect of high-speed, compressible airflow in the arc-extinguishing chamber by solving the transport equations of turbulent kinetic energy k and turbulent dissipation rate ε. Step 2: Solve the domain and boundary conditions of the model using open-source finite element software, and calculate the circuit breaker electric field using adaptive mesh. Step 3: Collect real experimental data and deeply integrate it with simulation data to construct a twin database; Step 3 specifically includes the following steps: Step 3.1: In the physical short-circuit breaking test circuit, deploy a sensor network to collect data in real time; install high-voltage probes on both sides of the circuit breaker break to measure the recovery voltage and transient recovery voltage; for the current signal, use a Rogowski coil or current transformer to measure the waveform, amplitude, and zero-crossing characteristics of the breaking current; use a linear displacement sensor to monitor the stroke curve of the moving contact in real time, and obtain the opening distance, overtravel, and opening / closing speed to obtain the displacement signal; collect the current waveform of the opening / closing coil to analyze the mechanical state of the operating mechanism; install pressure and temperature sensors in the arc-extinguishing chamber of the circuit breaker to monitor the pressure and temperature changes of the gas during the breaking process. Step 3.2: Apply a uniform high-precision timestamp to all collected real-time data to ensure that data from different sources are aligned on the timeline. Then, denoise and remove outliers from the original signal. Finally, normalize the data and scale it to the [0, 1] interval. The normalization formula is as follows: ； In the formula, x i For the original data, x min and x max These are the minimum and maximum values ​​of the data segment, respectively. Step 3.3: After the data processing is completed, all data are integrated to build a twin database. The database driver module is updated and feedback is closed. After each new short-circuit breaking test, the newly collected data is added to the twin database. Step 4: Select key feature parameters from the twin database, including electric field strength, airflow velocity, gas density, gas pressure and temperature. Combine the physical parameter values ​​of all feature points under a certain interruption state into a feature vector, and take the interruption degree of the circuit breaker corresponding to that state as the target value to form a complete training sample. Randomly divide all samples into training set and test set. Step 5: Construct a GA-BP neural network model. Train and test the GA-BP neural network model using training and testing sets. Deploy the trained GA-BP neural network model into the digital twin system as a state recognition module. Step 6: Construct a digital twin closed-loop framework that integrates a physical test circuit, a multi-physics simulation engine, a data management module, and a state identification module. Develop a graphical human-computer interaction interface to dynamically display the electrical wiring diagram and key parameters of the test circuit, as well as the dynamic cloud map distribution of the multi-physics field in the circuit breaker arc extinguishing chamber in real time.

2. The method for constructing a direct test loop digital twin system for a power distribution device according to claim 1, characterized in that, In step 2, the open-source finite element software includes FreeFem++ and OpenFOAM; FreeFem++ is responsible for the finite element calculation of the electrostatic field, while OpenFOAM is responsible for the coupled calculation of the airflow field and temperature field; the two transmit data in one direction through a data interface, that is, the electrostatic field calculation result can be used as a known condition to be input into the flow field-temperature field calculation, so as to achieve efficient weak coupling solution.

3. The method for constructing a direct test circuit digital twin system for a power distribution device according to claim 1, characterized in that, Step 2 specifically includes the following steps: Step 2.1: Import the established two-dimensional or three-dimensional geometric model of the arc-extinguishing chamber into FreeFem++, and use its built-in Delaunay-Voronoi algorithm automatic mesh generator to perform finite element mesh generation; Step 2.2: Import the arc-extinguishing chamber geometric model into OpenFOAM, use the snappyHexMesh mesh generation tool to automatically generate a high-quality hexahedral-dominated polyhedral mesh based on the geometric shape; perform local refinement in the key flow regions of the nozzle and contact gap, and solve the problem based on the OpenFOAM solver; Step 2.3: Perform time-space alignment and integration of the electric field data output by FreeFem++ and the flow-temperature field data output by OpenFOAM; finally, output the complete dynamic distribution map of the internal multiphysics field of the circuit breaker under different opening states and the parameter history curve of the feature points, which constitutes a high-fidelity twin database for driving the state identification of the digital twin.

4. The method for constructing a direct test loop digital twin system for a power distribution device according to claim 1, characterized in that, In step 5, the GA-BP neural network model includes an input layer, a hidden layer, and an output layer. The number of nodes in the input layer is determined based on the total number of selected feature parameters. The hidden layer has 5 nodes and uses the Sigmoid activation function. The output layer has 1 node to output the predicted value of the opening distance percentage. The model training is divided into two stages: First, a genetic algorithm is used for global optimization. After the network parameters are binary encoded, an initial population of 50 is generated. Individuals are selected based on fitness using the roulette wheel selection method. Single-point crossover is performed with a crossover probability of 0.8, and basic bit mutation is performed with a mutation probability of 0.

09. After 100 generations of iterative evolution, the optimal initial parameters are obtained. Then, the error backpropagation algorithm is used for fine training. The learning rate is set to 0.01, and the network parameters are fine-tuned using the gradient descent method until the model converges.

5. A computer device comprising a memory, a processor, and a computer program stored in the memory, characterized in that, The processor executes the computer program to implement the steps of the method of claim 1.

6. A computer-readable storage medium having a computer program / instructions stored thereon, characterized in that, When the computer program / instructions are executed by the processor, they implement the steps of the method of claim 1.

7. A computer program product comprising a computer program / instructions, characterized in that, When the computer program / instructions are executed by the processor, they implement the steps of the method of claim 1.

Images