Power distribution device parameter setting system and method based on online learning and digital twinning

Abstract

The application discloses a power distribution device parameter setting system and method based on online learning and digital twinning, which is used for solving the problem of direct test loop multi-parameter collaborative setting. The system collects real-time operation data such as voltage, current and temperature of the actual test loop through a data acquisition module; the real-time data are transmitted to a digital twinning module through an Internet of Things cloud module, compared with the output of the digital twinning module, and dynamically corrected by using an adaptive algorithm; the test loop parameters after dimensionless processing are used as optimization variables, the differential equation output by the digital twinning module is used as a constraint condition, an optimization function with the minimum difference between the TRV characteristic parameters and the standard value as the target is constructed, parameter optimization and simulation verification are carried out, and the output through a control terminal realizes accurate setting of test parameters. The application can comprehensively simulate electrical, thermodynamic and electromagnetic characteristics under different test conditions, and greatly reduces the number and time of repeated debugging required by the traditional method.

Description

Technical Field

[0001] This invention relates to the field of testing technology for high-voltage electrical equipment in power systems, and in particular to a system and method for parameter setting of power distribution devices based on online learning and digital twins. Background Technology

[0002] With the continuous development of the economy and society, electricity demand is constantly rising, power grid capacity is continuously expanding, and the short-circuit current value of the power system is getting larger and larger. Currently, the short-circuit current of many substations has far exceeded the breaking capacity of switchgear, posing a serious threat to power grid safety. Against this backdrop, the performance of key power transmission and transformation equipment such as switches and transformers is directly related to the safe and stable operation of the power system. Therefore, conducting high-voltage and high-current tests on such equipment is of paramount importance.

[0003] Among various power transmission and transformation equipment, switchgear is a core component ensuring the stable operation of the power system and providing high-quality electrical energy. Common high-voltage switchgear mainly includes disconnecting switches, load switches, and circuit breakers. Among them, circuit breakers, as the most critical and complex equipment in the high-voltage switchgear family, are generally referred to as high-voltage circuit breakers in systems of 3kV and above. They are not only the final protective barrier of the power system but also a core component maintaining grid security. Regardless of whether the system is under no-load, load, or short-circuit fault conditions, the circuit breaker must reliably close or open the circuit upon receiving an operating command. Therefore, to ensure its rapid and reliable operation during system short circuits, short-circuit breaking tests on high-voltage circuit breakers are crucial.

[0004] To verify and assess the breaking performance of high-voltage AC circuit breakers, international standard IEC 62271-100 and Chinese standard GB / T 1984-2014 specify stringent short-circuit breaking test requirements. Direct testing, as the test method closest to actual power grid operating conditions, is widely used due to its realistic physical process and lack of complex equivalence theory. However, direct testing involves numerous test items. For example, switch short-circuit tests require multiple test items under different operating conditions, such as T10, T30, T60, T100a, and T100s. Each item requires adjustment of test circuit parameters. Currently, the parameter adjustment methods for test systems are mostly based on experimental experience, which cannot meet the needs of rapid parameter setting and accurate short-circuit current acquisition for short-circuit breaking capacity testing.

[0005] Therefore, a new method is needed to efficiently, accurately, and automatically complete the multi-parameter coordinated setting of the direct test circuit for short-circuit breaking of high-voltage AC circuit breakers, ensuring that the test waveform strictly conforms to the standard requirements and truly reflects the breaking capacity of the circuit breaker. Summary of the Invention

[0006] The problem to be solved by this invention is to provide a power distribution device parameter tuning system and method based on online learning and digital twins, which is used to solve the problem of multi-parameter coordinated tuning of direct test circuits.

[0007] The present invention adopts the following technical solution: a power distribution device parameter setting system based on online learning and digital twin, comprising: an actual test circuit, a data acquisition module, an Internet of Things cloud module, a digital twin module and a control terminal;

[0008] The actual test circuit includes: a voltage source, an equivalent inductor, a TRV frequency modulation device, and a high-voltage AC circuit breaker;

[0009] The data acquisition module collects voltage, current, temperature and electromagnetic interference data in real time during the test process through sensors installed at key nodes of the actual test circuit, and transmits the data to the Internet of Things cloud module.

[0010] The IoT cloud module receives real-time collected data and preprocesses the raw signal, automatically extracts key feature parameters from the collected waveform data, calculates feature values, and transmits the processed digital signal to the digital twin module.

[0011] The digital twin module receives digital signals from the IoT cloud module, constructs a digital twin model, performs simulation calculations, simulates the corresponding TRV waveform, compares and analyzes the real-time collected data with the model output features, dynamically corrects the model parameters using a particle swarm optimization algorithm, uses a BP neural network model online learning and update mechanism for prediction, and verifies the effect of the optimized parameters.

[0012] The control terminal receives the prediction results from the digital twin module, performs diagnosis and displays the results visually, establishes an optimization history archive, and records the input conditions, optimization process and output results for each optimization.

[0013] As a preferred embodiment, the actual test circuit includes a voltage source, an equivalent inductance, a TRV frequency modulation device, and a high-voltage AC circuit breaker.

[0014] As a preferred embodiment, the data acquisition module deploys a sensor array at key nodes of the actual test circuit, including: voltage sensors at both ends of the circuit breaker, loop current sensors, reactor temperature sensors, and system frequency sensors, and ensures the time consistency of the data from each sensor through system clock synchronization.

[0015] As a preferred embodiment, the IoT cloud module processes data as follows:

[0016] The system receives real-time acquired physical and static parameters, including: transient recovery voltage waveform, loop current characteristics, component temperature changes, and system oscillation frequency; and the static parameters include: parameters of stationary contact, moving contact, nozzle, and shield.

[0017] The acquired raw signals are preprocessed, including signal amplification, noise filtering, AD conversion, and data packaging.

[0018] Key feature parameters are automatically extracted from the acquired waveform data, including: peak voltage, rise time, peak time, oscillation frequency, and decay characteristics.

[0019] The sliding window algorithm is used to calculate feature values ​​in real time, and a feature change trend model is established by combining historical data.

[0020] The processed digital signal is transmitted to the digital twin model via a high-speed communication interface.

[0021] As a preferred embodiment, the digital twin model includes an electrical parameter model, a thermodynamic model, and an electromagnetic compatibility model, used to establish multi-physics coupling that includes electrical characteristics, thermodynamic effects, and electromagnetic interference.

[0022] As a preferred embodiment, the digital twin module processes data as follows:

[0023] Receive real-time data and perform multi-scenario simulation calculations, including: normal start-up, reignition, and different load conditions, simulate the corresponding TRV waveforms, and perform sensitivity analysis and parameter impact assessment.

[0024] Establish a comprehensive deviation evaluation index system and compare the characteristics of actual measurement data with simulation prediction results;

[0025] For deviations or changes in system characteristics, the BP neural network model is used for online learning and updating to make predictions. Based on historical optimization data and system response characteristics, the data features are corrected, the structural parameters and learning rate of the digital twin model are adjusted, and the changed parameters are output and sent to the actual system through the feedback control channel to realize the dynamic adjustment of the digital twin model.

[0026] Based on the adjusted digital twin model, the actual test loop parameters are dimensionless, and a mapping relationship between the dimensionless parameters and the TRV characteristic parameters is established. Using the dimensionless parameters as optimization variables and the differential equations output by the digital twin model as constraints, an optimization function is constructed with the goal of minimizing the differences between the TRV amplitude coefficient, time delay parameters and standard values.

[0027] The optimization algorithm is run in the digital twin model to correct the data characteristics. The optimal combination of dimensionless parameters is obtained through multiple iterations and converted into actual inductance, capacitance and resistance values.

[0028] The optimized system is validated by checking the rationality of parameters, evaluating simulation accuracy, and analyzing stability. An optimization history archive is established to record the input conditions, optimization process, and output results for each optimization.

[0029] As a preferred option, a mapping relationship between dimensionless parameters and TRV characteristic parameters is established, expressed as:

[0030] ;

[0031] in, p norm The value of the dimensionless parameter. ω (1) , ω (2) This is the weight matrix. σ It is the ReLU activation function. b (1) , b (2) For bias, F TRV This represents the approximate relationship between the dimensionless parameter and the TRV parameter.

[0032] As a preferred option, an optimization function is constructed with the objective of minimizing the differences between the TRV amplitude coefficient, time delay parameter, and standard values, expressed as:

[0033] ;

[0034] in, p This represents the dimensionless parameter vector to be optimized. p 0 This represents the initial dimensionless parameter vector. Indicates the peak voltage of TRV. This is the standard value for TRV peak voltage. Peak time, This is the standard value for the delay parameter. α 1 and α 2 These are the weighting coefficients. λ ∈ (0, 1), J It is the minimum optimization function.

[0035] As a preferred embodiment, the optimization algorithm is the adaptive momentum gradient descent algorithm, and the iterative process is as follows:

[0036] ;

[0037] in, Let be the gradient descent function, with superscript . k For the number of iterations,p For the above dimensionless parameters, v For speed, m For sample batch size, , These represent the optimized batch sample size and speed, respectively. β 1 , β 2 and η ∈ (0, 1), ξ It is the minimum value. g For gradient.

[0038] The present invention also provides: a method for setting parameters of a power distribution device based on online learning and digital twins, for implementing the aforementioned method, comprising the following steps:

[0039] Step 1: Construct a digital twin architecture for the test system, including the actual test loop and the corresponding digital twin model. Build a virtual mapping of the test system through multiphysics modeling to achieve real-time data interaction between the test system and the virtual model.

[0040] Step 2: By installing sensors at key nodes in the actual test circuit, collect voltage, current, temperature and electromagnetic interference data in real time during the test process, and use the Internet of Things cloud platform to receive the real-time collected data and transmit it to the digital twin model.

[0041] Step 3: The digital twin model receives real-time data, performs simulation calculations, and simulates the corresponding TRV waveform; the real-time acquired data is compared and analyzed with the model output features, and the particle swarm optimization algorithm is used to dynamically correct the model parameters to ensure that the digital twin model is consistent with the experimental system.

[0042] Step 4: Based on the corrected digital twin model, the actual test loop parameters are dimensionless, and a mapping relationship between the dimensionless parameters and the TRV characteristic parameters is established.

[0043] Step 5: Using dimensionless parameters as optimization variables and the differential equations output by the digital twin model as constraints, construct an optimization function with the objective of minimizing the differences between the TRV amplitude coefficient, time delay parameter and standard value.

[0044] Step 6: Run the optimization algorithm in the digital twin model, correct the data characteristics, obtain the optimal dimensionless parameter combination through multiple iterations, and convert it into actual inductance, capacitance and resistance values;

[0045] Step 7: Feed back the optimized actual inductance, capacitance and resistance values ​​to the actual test system, perform simulation verification through a digital twin model, compare and analyze the degree of conformity between the simulation results and the standard requirements, and complete the parameter self-tuning process.

[0046] The present invention also provides: an electronic device, comprising:

[0047] One or more processors;

[0048] A storage device on which one or more programs are stored;

[0049] When the one or more programs are executed by the one or more processors, the one or more processors implement the aforementioned power distribution device parameter setting test method.

[0050] The present invention also provides a computer-readable storage medium storing a computer program thereon, which, when executed by a processor, implements the steps in the aforementioned power distribution device parameter setting test method.

[0051] Compared with the prior art, the present invention, employing the above technical solution, has the following technical effects:

[0052] 1. Real-time optimization: This invention enables dynamic correction of model parameters through real-time data interaction between the digital twin model and the actual experimental system, achieving online optimization during the experimental process and significantly reducing the number of repeated debugging attempts and time required by traditional methods.

[0053] 2. Precise Adaptation: The digital twin model based on multi-physics coupling modeling can comprehensively simulate the electrical, thermodynamic and electromagnetic characteristics under different test conditions, adapt to various test methods such as T100, T60, T30, and T10, and ensure that the parameter tuning results accurately meet the standard requirements.

[0054] 3. Platform-based expansion: The digital twin model constructed in this invention has good scalability. It is not only suitable for optimizing and tuning the parameters of the current test loop, but also can be adapted to complex test scenarios such as composite tests through model reconstruction and algorithm expansion. At the same time, it supports the integration and application of a variety of intelligent optimization algorithms. Attached Figure Description

[0055] Figure 1 This is a structural diagram of a three-phase direct test circuit according to an embodiment of the present invention;

[0056] Figure 2 This is a single-phase equivalent circuit diagram of the first-phase circuit breaker in an embodiment of the present invention;

[0057] Figure 3 This is the overall architecture of the digital twin model optimization system of the present invention;

[0058] Figure 4 A complete workflow diagram for optimizing parameters of a digital twin model;

[0059] Figure 5 Optimize the neural network structure diagram for the BP in a digital twin model;

[0060] Figure 6 The training loss curves for a BP neural network using historical and real-time data;

[0061] Figure 7 For inductance parameter L s The trend of the value changing with temperature and the predicted inductance value output by the digital twin model as the actual system operating state changes;

[0062] Figure 8 The changing trend of the capacitance C0 parameter value with temperature and the predicted capacitance value output by the digital twin model as the actual system operating state changes;

[0063] Figure 9 The changing trend of the capacitor C1 parameter value with temperature and the predicted capacitance value output by the digital twin model as the actual system operating state changes;

[0064] Figure 10 For resistor R s The changing trend of parameter values ​​with temperature and the predicted capacitance value output by the digital twin model as the actual system operating state changes;

[0065] Figure 11 For resistor R L The changing trend of parameter values ​​with temperature and the predicted capacitance value output by the digital twin model as the actual system operating state changes;

[0066] Figure 12 A comparison chart of the multiphysics parameter values ​​predicted by the BP neural network and the actual values;

[0067] Figure 13 These are the prediction error values ​​for temperature, magnetic field, electric field, voltage, and current in a multiphysics environment.

[0068] Figure 14 Error convergence plot before and after online update optimization of digital twin model. Detailed Implementation

[0069] To make the objectives, technical solutions, and advantages of this invention clearer, the technical solutions of the application will be further described in detail below with reference to the accompanying drawings. The described embodiments are only a part of the embodiments involved in this invention. All non-innovative embodiments based on these embodiments by other researchers in the art are within the protection scope of this invention. Furthermore, the step numbers in the embodiments of this invention are only set for ease of explanation and do not limit the order of the steps. The execution order of each step in the embodiments can be adaptively adjusted according to the understanding of those skilled in the art.

[0070] Example 1:

[0071] A method for setting parameters of power distribution devices based on online learning and digital twins is provided, as follows:

[0072] Step 1: Construct the digital twin architecture of the test system, including the actual test loop and the corresponding digital twin model.

[0073] In this embodiment, the three-phase direct test circuit, such as... Figure 1 As shown, AC represents a three-phase AC source, and MB represents a protective circuit breaker, which is used to interrupt the short-circuit current in time when the test fails. This represents a reactor, T represents a three-phase winding transformer, and resistance... and capacitors , To form a frequency modulation device, resistors Used for filtering, TB represents the circuit breaker under test.

[0074] Will Figure 1 The three-phase direct test circuit in the model is equivalent to the circuit breaker of the first phase by performing a symmetry test. The phase in which the circuit breaker is first opened is taken as the equivalent single phase. At the same time, the transformer and reactor in the circuit are equivalent to inductors, and the power grid or impulse generator is equivalent to a voltage source. The equivalent model is then established.

[0075] In this embodiment, the single-phase equivalent test circuit is as follows: Figure 2 As shown, where, Represents an equivalent voltage source. Represents the equivalent inductance. , It is the transient recovery voltage across the circuit breaker.

[0076] Based on KCL, KVL, and the characteristics of inductor and capacitor components. Figure 2 The equations and initial conditions of the equivalent circuit are expressed as follows:

[0077] ;

[0078] In the formula, and respectively flowing through the inductor and The current is indicated by an asterisk (*), which signifies a dimensional parameter.

[0079] Step 2: Based on the circuit equivalent circuit diagram, establish a real-time interactive relationship architecture between the actual test circuit and the digital twin model.

[0080] In this embodiment, the digital twin model optimization system is as follows: Figure 3 As shown:

[0081] The circuit breaker entity is the circuit breaker in the breaking test circuit, including: AC source, reactor, frequency modulation device and the circuit breaker under test entity.

[0082] A sensor array is precisely deployed at the key nodes of the test circuit, including: voltage sensors at both ends of the circuit breaker body, loop current sensors, reactor temperature sensors, and system frequency sensors.

[0083] Configure a data acquisition unit to collect actual data, and use a fiber optic transmission protocol to ensure real-time data transmission and anti-interference capabilities.

[0084] Complete system clock synchronization to ensure time consistency of data from all sensors, and utilize the IoT cloud platform to receive real-time collected data.

[0085] The IoT cloud platform receives data from the sensor network deployed at key measurement points, including real-time collected physical parameters such as voltage, current, and temperature, as well as static parameters of the circuit breaker entities such as stationary contacts, nozzles, and moving contacts. It then combines these data with multi-physics coupling to obtain a digital model of the circuit breaker entity, thus forming a digital twin model of the circuit breaker breaking test circuit.

[0086] The IoT cloud platform continuously receives operational status data from the test system, including transient recovery voltage waveforms, loop current characteristics, component temperature changes, and system oscillation frequency.

[0087] The system preprocesses the raw signals obtained from the data acquisition unit, including signal amplification, noise filtering, AD conversion, and data packaging, and then transmits the processed digital signals to the digital twin model through a high-speed communication interface.

[0088] On the control end, through the autonomous learning and predictive deduction of the BP neural network model, the output parameters are visualized and analyzed and the state is predicted, and the diagnostic results are output to update and iterate the historical database.

[0089] Step 3: The digital twin model receives real-time data and performs simulation calculations.

[0090] After receiving real-time data, the digital twin model uses a BP neural network to perform high-precision simulation calculations based on the current circuit physical parameters (R, L, C) and static parameters to simulate the corresponding TRV waveform.

[0091] In particular, the digital twin model employs an improved numerical algorithm to ensure that the simulation results not only reflect the physical characteristics of the system but also meet real-time requirements.

[0092] Then, by comparing the differences between the actual measurement data and the simulation results through the optimization algorithm module, the parameter optimization program is started and autonomous learning and prediction are performed.

[0093] In this embodiment, a BP neural network is used for prediction, and the changed parameters are output and sent to the experimental system through a feedback control channel to realize the dynamic adjustment of the digital twin model.

[0094] Step 4: Optimize the workflow based on digital twin parameters, and conduct data collection and optimization verification. The workflow is as follows: Figure 4 As shown.

[0095] Step 4.1: The system synchronously collects multi-dimensional sensor data such as voltage, current, and temperature, and uses data fusion technology to eliminate measurement errors from a single sensor.

[0096] Establish a timestamp alignment mechanism to ensure accurate matching of data from different sampling rates across the time dimension. Identify and remove outlier data to guarantee the quality and reliability of the input data.

[0097] Step 4.2: Automatically extract key feature parameters from the acquired waveform data, including peak voltage, rise time, peak time, oscillation frequency, and decay characteristics.

[0098] The sliding window algorithm is used to calculate feature values ​​in real time, and a feature change trend model is established by combining historical data to provide a quantitative basis for subsequent optimization.

[0099] Step 4.3: Based on the current circuit parameters, the digital twin model performs multi-scenario simulation calculations, including normal interruption, reignition, and different load conditions.

[0100] The simulation results not only include TRV waveform prediction, but also provide sensitivity analysis and parameter impact assessment, providing theoretical guidance for optimization.

[0101] Step 4.4: Establish a comprehensive deviation evaluation index system and conduct multi-dimensional comparative analysis of actual measurement characteristics and simulation prediction characteristics.

[0102] Deviation assessment considers not only absolute differences but also the consistency of trends.

[0103] Step 4.5: For deviations or changes in system characteristics, utilize the online learning and update mechanism of the BP neural network model. Based on historical optimization data and system response characteristics, adjust the structural parameters and learning rate of the digital twin model to enhance its adaptability to system changes;

[0104] Step 4.6: Conduct a comprehensive effect verification of the optimized system, including parameter rationality check, simulation accuracy evaluation and stability analysis.

[0105] Simultaneously establish optimization history archives to record the input conditions, optimization process and output results of each optimization, providing experience accumulation for subsequent optimizations;

[0106] Step 5: Use historical and real-time data to pre-train and update the parameters of the BP neural network model.

[0107] The BP neural network model first obtains the error of forward propagation of the solution signal, and then modifies the weights of the coefficients of each layer based on the back propagation of the error, continuously approaching the true value, thus achieving the purpose of training neurons.

[0108] In this embodiment, the BP neural network structure is as follows: Figure 5 As shown, it includes: input layer, hidden layer, and output layer. Figure 5 In the diagram, P1, P2, and P3 are input layer nodes, b1 is a hidden layer node, b2 is an output layer node, w1 represents the connection weight between the input layer and the hidden layer, and w2 represents the connection weight between the hidden layer and the output layer. The sample y with the actual value is transmitted to the loss function for processing.

[0109] Factors affecting the propagation performance of a backpropagation neural network include the number of hidden layers, initial values, activation function type, and learning rate. If the learning rate is too large, the iteration step may be too large and the optimal solution may be missed, while if the learning rate is too small, the iteration speed will be slow.

[0110] The specific parameters of the BP neural network in this embodiment are shown in Table 1 below.

[0111] Table 1

[0112]

[0113] In this embodiment, the BP neural network processes the data as follows:

[0114] (1) In the data forward propagation stage, the input signal is first transmitted from the input layer to the hidden layer, and at this time the output result H of the hidden layer is obtained. j It can be represented as:

[0115] ;

[0116] in, j It is a positive integer; n Indicates the number of neurons in the input layer; Represents the connection weights between the input layer and the hidden layer; X i This represents the number of neurons in the hidden layer.

[0117] (2) The data stream is then transmitted from the hidden layer to the output layer, and the calculation results of the output layer are... O k It can be represented as:

[0118] ;

[0119] in, k It is a positive integer;l Indicates the number of neurons in the hidden layer; This represents the connection weights between the hidden layer and the output layer. H j This is the output value.

[0120] (3) After the neural network completes its first forward propagation, due to the model output value O k Compared with the true value y k There is a deviation between them, and the resulting error E can be expressed as:

[0121] ;

[0122] in, m This represents the number of nodes in the output layer. y k This is the actual value.

[0123] (4) The error propagates along the reverse path. The BP neural network uses gradient descent to optimize and adjust the weight parameters. The weight update formula is shown below:

[0124] ;

[0125] in, This represents the learning rate parameter; e k Indicates the first k The error of each sample.

[0126] The learning rate of the BP neural network was set to 0.01. Historical data was input to train the BP neural network. The pre-trained BP neural network was then applied to the actual circuit. Real-time data was input into the pre-trained BP neural network model to obtain the predicted TRV voltage. The TRV voltage was compared with the actual TRV voltage to obtain the error. If the error exceeded a threshold, the gradient descent algorithm was used to optimize the physical parameters, focusing on the key parameter, inductance L. s The twin model is updated by predicting the capacitance C0 and changes continuously as the system's operating state changes.

[0127] Step 6: Based on the corrected digital twin model, the actual test loop parameters are made dimensionless, and a mapping relationship between the dimensionless parameters and the TRV characteristic parameters is established.

[0128] Among them, dimensionless parameters include:

[0129] ;

[0130] in, p norm L represents the value of dimensionless parameters such as inductance, capacitance, and resistance.s The equivalent inductance is C0, C1 are frequency modulation capacitors, and R is R. s R is a frequency modulation resistor. L For the filter resistor, the subscripts max and min represent the maximum and minimum values, respectively.

[0131] The mapping relationship between dimensionless parameters and TRV characteristic parameters is established as follows:

[0132] ;

[0133] in, ω This is the weight matrix. σ It is the ReLU activation function. b For bias, F TRV This represents the approximate relationship between the dimensionless parameter and the TRV parameter.

[0134] Then, using dimensionless parameters as optimization variables and the differential equations output by the digital twin model as constraints, an optimization function is constructed with the objective of minimizing the differences between the TRV amplitude coefficient, time delay parameter, and standard value.

[0135] The differential equation output by the digital twin model is expressed as:

[0136] ;

[0137] in, A dimensionless state vector, a matrix The structure is as follows:

[0138] ;

[0139] Furthermore, an optimization function is constructed with the objective of minimizing the differences between the TRV amplitude coefficient, time delay parameter, and standard values, expressed as:

[0140] ;

[0141] in, p This represents the parameter vector to be optimized. Indicates the peak voltage of TRV. This is the standard value for TRV peak voltage. Peak time, This is the standard value for the delay parameter. α 1 and α 2 These are the weighting coefficients. λ ∈ (0, 1), J It is the minimum optimization function.

[0142] Then, an optimization algorithm is run in the digital twin model to obtain the optimal combination of dimensionless parameters through multiple iterations, and then converts it into actual inductance, capacitance and resistance values.

[0143] In this embodiment, the optimization algorithm is the adaptive momentum gradient descent algorithm, and the iterative process is as follows:

[0144]

[0145] Among them, superscript k For the number of iterations, J For the above minimum optimization function, p For the above dimensionless parameters, v For speed, m For sample batch size, β 1 , β 2 and η ∈ (0, 1), ξ It is the minimum value. g For gradient.

[0146] Step 7: Feed back the optimized actual inductance, capacitance and resistance values ​​to the actual test system, and at the same time perform simulation verification through a digital twin model. Compare and analyze the degree of conformity between the simulation results and the standard requirements to complete the parameter self-tuning process.

[0147] In this embodiment, the training loss curves of the BP neural network using historical data and real-time data are as follows: Figure 6 As shown. In actual operation, the inductance L s Parameter values, capacitor C0 parameter value, capacitor C1 parameter value, resistor R s Parameter values, resistance R L The changing trends of parameter values ​​with temperature and the predicted inductance values ​​output by the digital twin model as the actual system operating state changes are respectively as follows: Figure 7 , Figure 8 , Figure 9 , Figure 10 , Figure 11 As shown.

[0148] Furthermore, in this embodiment, the comparison between the multiphysics parameter values ​​predicted by the BP neural network and the actual values ​​is as follows: Figure 12 As shown. The prediction errors for temperature, magnetic field, electric field, voltage, and current in a multiphysics environment are as follows: Figure 13 As shown. The error convergence of the digital twin model before and after online update optimization is as follows. Figure 14 As shown.

[0149] Example 2

[0150] A power distribution device parameter setting system based on online learning and digital twin is provided. The system performs short-circuit breaking tests on high-voltage AC circuit breakers using the method described in Embodiment 1. The system includes: an actual test circuit, a data acquisition module, an Internet of Things cloud module, a digital twin module, and a control terminal.

[0151] In this embodiment, the actual test circuit includes: a voltage source, an equivalent inductance, a TRV frequency modulation device, and a high-voltage AC circuit breaker; the digital twin system constructs a virtual mapping of the test system through multiphysics modeling, including an electrical parameter model, a thermodynamic model, and an electromagnetic compatibility model.

[0152] The system works as follows: First, a digital twin model of the test system is constructed, and operating data such as voltage, current, and temperature are collected in real time through sensors during the test. The real-time data is compared with the output of the twin model, and the model parameters are dynamically corrected using an adaptive algorithm. Then, the dimensionless test loop parameters are used as optimization variables, and the differential equations output by the digital twin model are used as constraints to construct an optimization function with the goal of minimizing the difference between the TRV characteristic parameters and the standard values. Finally, the parameters are optimized and verified through simulation on the digital twin platform to achieve accurate tuning of the test parameters.

[0153] In this embodiment of the invention, an electronic device is also provided, comprising: one or more processors; a storage device having one or more programs stored thereon; when the one or more programs are executed by the one or more processors, the one or more processors implement the power distribution device parameter setting method described in any of the above embodiments.

[0154] In this embodiment of the invention, a computer-readable storage medium is also provided, on which a computer program is stored. When the program is executed by a processor, it implements the steps in any of the power distribution device parameter setting methods described in the above embodiments.

[0155] The above description is only a preferred embodiment of the present invention. It should be noted that for those skilled in the art, several improvements and modifications can be made without departing from the principle of the present invention, and these improvements and modifications should also be considered within the scope of protection of the present invention.

Claims

1. A power distribution device parameter setting system based on online learning and digital twins, characterized in that, include: Actual test circuit, data acquisition module, IoT cloud module, digital twin module and control terminal; The actual test circuit includes: a voltage source, an equivalent inductor, a TRV frequency modulation device, and a high-voltage AC circuit breaker; The data acquisition module collects voltage, current, temperature and electromagnetic interference data in real time during the test process through sensors installed at key nodes of the actual test circuit, and transmits the data to the Internet of Things cloud module. The IoT cloud module receives real-time collected data and preprocesses the raw signal, automatically extracts key feature parameters from the collected waveform data, calculates feature values, and transmits the processed digital signal to the digital twin module. The digital twin module receives digital signals from the IoT cloud module, constructs a digital twin model, performs simulation calculations, simulates the corresponding TRV waveform, compares and analyzes the real-time collected data with the model output features, dynamically corrects the model parameters using a particle swarm optimization algorithm, uses a BP neural network model online learning and update mechanism for prediction, and verifies the effect of the optimized parameters. The control terminal receives the prediction results from the digital twin module, performs diagnosis and displays the results visually, establishes an optimization history archive, and records the input conditions, optimization process and output results for each optimization. The digital twin module processes data as follows: Receive real-time data and perform multi-scenario simulation calculations, including: normal start-up, reignition, and different load conditions, simulate the corresponding TRV waveforms, and perform sensitivity analysis and parameter impact assessment. Establish a comprehensive deviation evaluation index system and compare the characteristics of actual measurement data with simulation prediction results; For deviations or changes in system characteristics, the BP neural network model is used for online learning and updating to make predictions. Based on historical optimization data and system response characteristics, the data features are corrected, the structural parameters and learning rate of the digital twin model are adjusted, and the changed parameters are output and sent to the actual system through the feedback control channel to realize the dynamic adjustment of the digital twin model. Based on the adjusted digital twin model, the actual test loop parameters are dimensionless, and a mapping relationship between the dimensionless parameters and the TRV characteristic parameters is established. Using the dimensionless parameters as optimization variables and the differential equations output by the digital twin model as constraints, an optimization function is constructed with the goal of minimizing the differences between the TRV amplitude coefficient, time delay parameters and standard values. The optimization algorithm is run in the digital twin model to correct the data characteristics. The optimal combination of dimensionless parameters is obtained through multiple iterations and converted into actual inductance, capacitance and resistance values. The optimized system is validated, including: parameter rationality check, simulation accuracy assessment and stability analysis, and optimization history archive is established to record the input conditions, optimization process and output results of each optimization. The mapping relationship between dimensionless parameters and TRV characteristic parameters is established as follows: ； in, p norm The value of the dimensionless parameter. ω (1) , ω (2) This is the weight matrix. σ It is the ReLU activation function. b (1) , b (2) For bias, F TRV This represents an approximate relationship between the dimensionless parameter and the TRV parameter; The differential equation output by the digital twin model is expressed as: ； in, A dimensionless state vector, a matrix : ； in, L s For equivalent inductance, C 0 , C 1 For frequency modulation capacitors, R s For frequency modulation resistor, R L For filtering resistors; An optimization function is constructed with the objective of minimizing the differences between the TRV amplitude coefficient, time delay parameter, and standard values, and is expressed as: ； in, p This represents the dimensionless parameter vector to be optimized. p 0 This represents the initial dimensionless parameter vector. Indicates the peak voltage of TRV. This is the standard value for TRV peak voltage. Peak time, This is the standard value for the delay parameter. α 1 and α 2 λ represents the weighting coefficient, where λ ∈ (0, 1). J It is the minimum optimization function.

2. The power distribution device parameter setting system based on online learning and digital twin as described in claim 1, characterized in that, The data acquisition module deploys a sensor array at key nodes of the actual test circuit, including: voltage sensors at both ends of the circuit breaker, loop current sensors, reactor temperature sensors, and system frequency sensors. The system clock synchronization ensures the time consistency of the data from each sensor.

3. The power distribution device parameter setting system based on online learning and digital twins according to claim 1, characterized in that, The IoT cloud module processes the following: The system receives real-time acquired physical and static parameters, including: transient recovery voltage waveform, loop current characteristics, component temperature changes, and system oscillation frequency; and the static parameters include: parameters of stationary contact, moving contact, nozzle, and shield. The acquired raw signals are preprocessed, including signal amplification, noise filtering, AD conversion, and data packaging. Key feature parameters are automatically extracted from the acquired waveform data, including: peak voltage, rise time, peak time, oscillation frequency, and decay characteristics. The sliding window algorithm is used to calculate feature values ​​in real time, and a feature change trend model is established by combining historical data. The processed digital signal is transmitted to the digital twin module via a high-speed communication interface.

4. The power distribution device parameter setting system based on online learning and digital twins according to claim 1, characterized in that, The digital twin model constructs a virtual mapping of the actual system through multiphysics modeling, realizing real-time data interaction between the actual system and the virtual model. It includes electrical parameter models, thermodynamic models, and electromagnetic compatibility models, and is used to establish multiphysics coupling that includes electrical characteristics, thermodynamic effects, and electromagnetic interference.

5. The power distribution device parameter setting system based on online learning and digital twins according to claim 1, characterized in that, The propagation process of the BP neural network is as follows: During the data forward propagation phase, the input signal is transmitted from the input layer to the hidden layer, and the output of the hidden layer... Represented as: ； in, j It is a positive integer; n Indicates the number of neurons in the input layer; Represents the connection weights between the input layer and the hidden layer; This represents the number of neurons in the hidden layer. f Forward propagation; Data flows from the hidden layer to the output layer, and the computation results of the output layer... O k Represented as: ； in, k It is a positive integer; l Indicates the number of neurons in the hidden layer; This represents the connection weights between the hidden layer and the output layer. H j This is the output value; g For gradient; After the neural network completes its first forward propagation, due to the model output value O k Compared with the true value y k There is a deviation between them, and the resulting error E is expressed as: ； in, m This represents the number of nodes in the output layer. Error propagates along the reverse path. The BP neural network uses gradient descent to optimize and adjust the weight parameters. The weight update formula is as follows: ； in, This represents the learning rate parameter; e k Indicates the first k The error of each sample.

6. The power distribution device parameter setting system based on online learning and digital twins according to claim 5, characterized in that, The optimization algorithm is an adaptive momentum gradient descent algorithm, and the iterative process is as follows: ； in, Let k be the gradient descent function, and k be the number of iterations. v For speed, , These represent the optimized batch sample size and speed, respectively. β 1 , β 2 and η ∈ (0, 1), ξ It is the minimum value. g For gradient.

7. A method for setting parameters of a power distribution device based on online learning and digital twins, applied to the system described in any one of claims 1 to 6, characterized in that, Includes the following steps: Step 1: Construct a digital twin architecture for the actual test system, including the actual test loop and the corresponding digital twin model. Build a virtual mapping of the test system through multiphysics modeling to achieve real-time data interaction between the test system and the virtual model. Step 2: By installing sensors at key nodes in the actual test circuit, collect voltage, current, temperature and electromagnetic interference data in real time during the test process, and use the Internet of Things cloud platform to receive the real-time collected data and transmit it to the digital twin model. Step 3: The digital twin model receives real-time data, performs simulation calculations, and simulates the corresponding TRV waveform; the real-time acquired data is compared and analyzed with the model output features, and the particle swarm optimization algorithm is used to dynamically correct the model parameters to ensure that the digital twin model is consistent with the experimental system; Step 4: Based on the corrected digital twin model, the actual test loop parameters are dimensionless, and a mapping relationship between the dimensionless parameters and the TRV characteristic parameters is established. Step 5: Using dimensionless parameters as optimization variables and the differential equations output by the digital twin model as constraints, construct an optimization function with the objective of minimizing the differences between the TRV amplitude coefficient, time delay parameter and standard value. Step 6: Run the optimization algorithm in the digital twin model, correct the data characteristics, obtain the optimal dimensionless parameter combination through multiple iterations, and convert it into actual inductance, capacitance and resistance values; Step 7: Feed back the optimized actual inductance, capacitance and resistance values ​​to the test system, perform simulation verification through a digital twin model, compare and analyze the degree of conformity between the simulation results and the standard requirements, and complete the parameter self-tuning process.

8. An electronic device, characterized in that, include: One or more processors; A storage device on which one or more programs are stored; When the one or more programs are executed by the one or more processors, the one or more processors implement the power distribution device parameter setting method as described in claim 7.

9. A computer-readable storage medium, characterized in that, It stores a computer program, which, when executed by a processor, implements the steps in the power distribution device parameter setting method of claim 7.

Images