Simulation test system for test switch of spare power automatic switching device

Abstract

The invention discloses a spare power automatic switching device test switch simulation test system, and belongs to the technical field of power grid simulation test. Probability distribution generated through kernel density estimation covers a fault scene of a 99% confidence interval, and the characterization capability of an extreme scene is enhanced; the time delay precision of a switch action sequence is controlled timely and reliably by combining the global search capability of quantum annealing and the local fine tuning of the particle swarm algorithm, and equipment damage caused by time sequence dislocation is avoided by combining the UWB technology and the phase compensation algorithm; through an objective function weight coefficient and a closed-loop feedback mechanism, an optimization objective can be dynamically adjusted according to a fault type, through real-time comparison of a physical device and digital twinning and through logic criterion goodness of fit analysis, control logic consistency can be guaranteed, and through loop response time difference analysis and two-dimensional index quantification, verification precision can be guaranteed. And the strict requirements of the power system on correct logic and quick action of the spare power automatic switching device are met.

Description

Technical Field

[0001] This invention relates to the field of power grid simulation testing technology, and specifically to a simulation testing system for an automatic transfer switch. Background Technology

[0002] With the continuous development of the power grid, the load carried by the power grid is increasing day by day. During centralized maintenance, in order to ensure the reliability of power supply, it is no longer possible to conduct equipment verification by completely shutting down the equipment. Instead, the equipment can only be shut down in rotation. However, the verification of the automatic transfer switch (ATS) has always been a difficult point during the rotational shutdown verification process. This is because the ATS involves interlocking circuits. If appropriate safety measures are not taken, it is very easy to cause the operating line or switch to trip, thereby causing the whole station to lose power.

[0003] According to the requirements of relay protection technical supervision, newly commissioned automatic transfer switches (ATS) and their periodic inspections must pass a switch-driven test to ensure the reliability of the device and the correctness of the external circuit. This ensures that the device can operate correctly and guarantee power supply after the working power supply or equipment is disconnected due to faults. Furthermore, according to relay protection testing practices, a simulation test must be conducted before the switch-driven test to ensure the reliability of the device itself, improve testing efficiency, and reduce the number of switch operations, thus benefiting the safe operation of primary equipment. However, the contradiction is even more pronounced in the phased construction of substations with incomplete primary wiring and the periodic inspection under normal switch operation after commissioning, making a drive test impossible. In actual field conditions, there are no suitable simulation test methods and equipment; therefore, the reliability and correctness of the logic and circuit of the ATS cannot be guaranteed.

[0004] The existing solutions, when implemented, cannot safely and accurately simulate the operation of the automatic transfer switch during rotational shutdowns, thus failing to effectively verify the logic and output circuit of the automatic transfer switch, resulting in poor reliability of the automatic transfer switch in power grid operation. Summary of the Invention

[0005] The purpose of this invention is to provide a test system for automatic transfer switch (ATS) simulation, which solves the technical problem that existing solutions cannot safely and accurately simulate the operation of ATS during rotational shutdowns, thus failing to effectively verify the logic and output circuit of ATS and resulting in poor reliability of ATS in power grid operation.

[0006] Technical solution: This invention discloses a simulation test system for an automatic transfer switch, comprising:

[0007] Fault scenario probability distribution analysis and processing module: Collects and preprocesses distributed power grid data during rotational outages through edge nodes, constructs a power grid state prediction model based on federated learning, trains the power grid state prediction model using the preprocessed distributed power grid data, and generates fault scenario probability distribution.

[0008] The timing calibration module for optimizing the opening and closing delay parameters: Based on the probability distribution of the generated fault scenarios, a quantum annealing-optimized switch action sequence generator is designed. The opening and closing delay parameters are dynamically adjusted through the particle swarm optimization algorithm, and sub-nanosecond timing calibration of multiple switch actions is achieved based on ultra-wideband synchronization technology.

[0009] Virtual-Real Fusion Multidimensional Verification and Analysis Module: A virtual-real fusion verification platform is built, and the optimized switching action sequence is simultaneously injected into the digital twin model and the physical standby automatic transfer device. The output relay action waveforms of the two are compared in real time through photonic crystal fiber transmission, and a two-dimensional verification report including the logical criterion consistency and the loop response time difference is output.

[0010] Furthermore, the collected distributed power grid data includes voltage sag waveforms, load mutation rates, and topology switching timing. The preprocessing performed includes voltage sag waveform denoising and load mutation rate calculation.

[0011] Furthermore, the power grid state prediction model is constructed using an improved FedAvg federated learning framework. The local model is a CNN-LSTM hybrid network. During federated model training and encrypted aggregation, each edge node trains the model based on local data and uses homomorphic encryption to encrypt the gradient ∇L of the local model. The central server decrypts the data and then performs parameter aggregation. When generating the probability distribution of fault scenarios, kernel density estimation is used to construct the probability density function of the fault scenarios. Based on the probability density function, the 99% confidence interval is obtained through calculation.

[0012] Furthermore, based on the generated fault scenario probability distribution, the optimization objective function J and constraints for the switching action sequence are defined. The optimization objective function J aims to minimize the total switching action delay and energy consumption. The constraints include opening and closing delay accuracy constraints, timing logic constraints, and equipment safety constraints.

[0013]

[0014] in, To optimize the weighting coefficients, 0 < <1. The total delay of the switch action is dynamically adjusted according to the urgency of the fault. The energy consumption is obtained by summing the action delays of all switches. It is obtained by summing the operating power of all switches;

[0015] The time delay accuracy constraint for opening and closing the circuit is: , For different switch indices, For the first The action delay of each switch; This represents the theoretically optimal delay.

[0016] The timing logic constraint is: for series switch groups, the following must be satisfied. ; Minimum action interval;

[0017] Equipment safety constraints are: operating voltage ; These are the minimum operating voltage and the maximum operating voltage, respectively. , , This is the rated voltage.

[0018] Furthermore, when generating the initial switch action sequence based on the quantum annealing algorithm, the action timing of m switches is encoded into N qubits;

[0019] With the objective function J as the energy function and the constraints as penalty terms, the expression for calculating the total Hamiltonian H is: ;in, To constrain penalty items; This is the penalty coefficient;

[0020] When performing quantum annealing scheduling, the quantum state is updated by calculation based on the initial temperature, annealing coefficient, and number of iterations;

[0021] When the temperature meets the requirements, it collapses to the ground state and outputs the initial action sequence;

[0022] Using the initial action sequence generated by quantum annealing as the initial population, the opening and closing delay parameters are dynamically optimized through the particle swarm optimization algorithm.

[0023] Furthermore, during sub-nanosecond calibration, the local clock correction value for the slave node is calculated; calibration is performed every 10ms, and the slave node adjusts its local crystal oscillator according to the local clock correction value. The specific calculation is as follows:

[0024] ;

[0025] in, Correction value for the local clock of the slave node; Master node clock; For the slave node's local clock; The phase difference of the UWB signal; This is the carrier angular frequency.

[0026] Furthermore, the optimized switching action sequence and calibrated timing parameters are input into the digital twin system to simulate the action response under fault scenarios, and closed-loop feedback adjustment is achieved through error analysis.

[0027] Furthermore, the time delay accuracy error is calculated. and synchronization error The expression involved is:

[0028] ;in, These represent the simulated action delay and the optimized target delay, respectively; D is the action sequence index.

[0029] ;in, These are the local clock correction values ​​for switch node f and switch node g, respectively, where f and g are any two switch nodes;

[0030] like >0.1ms or If the value is greater than 0.1 ns, restart particle swarm optimization and perform UWB calibration enhancement until the performance requirements are met.

[0031] Furthermore, when calculating the logical criterion consistency, features are extracted from the collected waveforms, and the relay action logic states of the digital twin model and the physical device are compared to obtain the logical criterion consistency.

[0032] When calculating the loop response time difference, the moment when 50% of the amplitude of the rising edge of the waveform is extracted as the response trigger time. The time difference between the digital twin and the physical device is calculated to obtain the loop response time difference.

[0033] Furthermore, when generating a two-dimensional verification report for logical criterion consistency and loop response time difference, the report includes the logical criterion consistency value and trend chart, and the statistical distribution of loop response time difference. When making a conclusion, if both the logical criterion consistency and the maximum loop response time difference meet the corresponding verification requirements, the verification is deemed successful; otherwise, the sequence number of the non-compliant action and the reason for the deviation are marked.

[0034] Compared to existing solutions, the beneficial effects achieved by this invention are:

[0035] This invention achieves encrypted transmission and decryption aggregation of model parameters through homomorphic encryption, which can avoid cross-domain leakage of original power grid data; the probability distribution generated by kernel density estimation covers fault scenarios with a 99% confidence interval, enhancing the ability to represent extreme scenarios; and the parameter aggregation through federated learning can effectively improve the prediction accuracy of the model on multi-regional power grid data, providing reliable scenario input for subsequent switching action simulation.

[0036] This invention combines the global search capability of quantum annealing with the local fine-tuning of particle swarm optimization to achieve timely and reliable control of the time delay accuracy of switching action sequences, meeting the requirements for rapid fault response. By combining UWB technology with a phase compensation algorithm, it avoids equipment damage caused by timing misalignment. Through objective function weighting coefficients and a closed-loop feedback mechanism, the optimization objective can be dynamically adjusted according to the fault type to adapt to complex power grid scenarios. By incorporating equipment safety thresholds and timing logic constraints into the constraints, combined with digital twin verification, it can ensure that the action sequence meets the power system safety regulations.

[0037] This invention avoids the limitations of single simulation or physical experiment by comparing physical devices with digital twins in real time, thereby effectively improving the accuracy and reliability of simulation testing. Through logic criterion consistency analysis, it can ensure the consistency of control logic. Through loop response time difference analysis, it can ensure dynamic characteristic matching. Dual-dimensional index quantification can ensure verification accuracy and meet the stringent requirements of power systems for the correct logic and rapid action of automatic transfer switch devices. Attached Figure Description

[0038] The invention will now be further described with reference to the accompanying drawings.

[0039] Figure 1 This is a flowchart illustrating the operation of a test switch simulation system for an automatic transfer switch according to the present invention. Detailed Implementation

[0040] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0041] like Figure 1 As shown, the present invention is a simulation test system for an automatic transfer switch, comprising:

[0042] 1. Fault Scenario Probability Distribution Analysis and Processing Module: This module collects and preprocesses distributed power grid data during off-peak periods via edge nodes, constructs a power grid state prediction model based on federated learning, and trains the model using the preprocessed distributed power grid data to generate fault scenario probability distributions. Specific steps include:

[0043] The collected distributed power grid data includes voltage sag waveforms, load mutation rates, and topology switching timing. The preprocessing performed includes voltage sag waveform denoising and load mutation rate calculation.

[0044] Specifically, when denoising the voltage sag waveform, a wavelet threshold denoising algorithm is used, the expression of which is:

[0045]

[0046] in, These are the original wavelet coefficients; These are the coefficients after thresholding. For adaptive threshold, , The noise standard deviation is estimated from the high-frequency coefficients of wavelet decomposition; N is the signal length; j is the level of wavelet decomposition, j=1,2,…,J, where J is the total number of decomposition levels; k represents the temporal and spatial sampling point index at the j-th scale, k=1,2,… ; The length of the coefficients in the j-th layer;

[0047] When calculating the load mutation rate, the mutation rate index is calculated based on the collected load data, and the expression is:

[0048]

[0049] in, The mutation rate of the load; The sampling interval is 1 second by default. This represents the load value at the current moment. t represents the load value at the previous moment; t is the time variable.

[0050] It should be explained that by quantifying the denoising and outlier handling processes through mathematical models, standardized input features can be provided for federated learning, avoiding the bias caused by interference in the original data during model training.

[0051] When constructing a power grid state prediction model based on federated learning, the federated learning model initialization and parameter configuration include the model architecture and federated training parameters.

[0052] The model architecture employs an improved FedAvg federated learning framework, with the local model being a CNN-LSTM hybrid network whose output layer predicts feature vectors for fault scenarios. d represents the dimension, with a value of 64, and the loss function is defined as:

[0053]

[0054] Where L is the total loss function value; The loss weight coefficient has a default value of 0.7, but it can be customized according to the application requirements of the actual application scenario. MSE() is the mean square error of feature vector prediction. To predict the feature vector; y represents the feature vector of the real scene; CE() is the cross-entropy loss for fault type classification; y is the real fault label, such as voltage sag or topology switching. To predict the probability distribution of labels.

[0055] When configuring federated training parameters, set the local training epochs E=10, the learning rate η=0.001, and the global aggregation formula as follows:

[0056]

[0057] in, These are the aggregated global model parameters; Let be the sample size of the i-th node; Here are the local model parameters for the i-th node; i is the node index, i = 1, 2, 3, ... ; The number of edge nodes participating in federated training.

[0058] It should be explained that by using a loss function to weight and fuse regression and classification tasks, the model can be ensured to learn both the feature representation and type discrimination capabilities of the fault scenario simultaneously; the federated aggregation formula balances the differences in data scale between different nodes by weighting the sample size.

[0059] When performing federated model training and encrypted aggregation, each edge node trains the model based on local data, and the parameter update formula is:

[0060]

[0061] in, These are the updated model parameters; These are the current model parameters; The learning rate has a range of values. ; Let be the local dataset of the i-th node, and ∇L be the gradient of the loss function; For iteration rounds;

[0062] Homomorphic encryption is used to encrypt the gradient ∇L of the local model. The encryption formula is as follows:

[0063]

[0064] in, This is the encrypted gradient ciphertext; For public encryption keys, For the product of large prime numbers, , It is a 1024-bit prime number; For model Calculation.

[0065] The central server decrypts and then performs parameter aggregation.

[0066] It should be explained that by optimizing the local model through gradient descent, combined with homomorphic encryption and weighted aggregation, the local knowledge of distributed nodes can be integrated while protecting data privacy, which can effectively improve the model's generalization ability.

[0067] When generating the probability distribution of fault scenarios, kernel density estimation is used to construct the probability density function of the fault scenarios. The relevant expression is as follows:

[0068]

[0069] in, Let z be the estimated probability density function of the fault scenario; z is the feature vector of the target fault scenario. For the first A feature vector of historical fault scenarios; n is the number of historical fault samples; h is the bandwidth; d is the dimension of the feature vector; K() is the kernel function; This is the normalization factor.

[0070] Based on this probability density function, the 99% confidence interval is obtained through calculation, involving the following expression:

[0071]

[0072] in, The symbol for the entire interval; The lower confidence limit; The infimum; The set of feature vectors representing the target fault scenarios that satisfy the conditions, where the part after "|" represents the condition; This indicates that the cumulative probability that the feature vector z of the target fault scenario is less than or equal to a preset value is greater than or equal to 0.005. The upper confidence limit; For supremum; condition This indicates that the cumulative probability that the feature vector z of the target fault scenario is less than or equal to a preset value is less than or equal to 0.995. It is the cumulative distribution function; Let be the estimated probability density function of the failure scenario. These are characteristic variables.

[0073] The resulting probability distribution covers key parameters such as voltage drop depth and duration. The voltage drop depth ranges from 0 to 30%Un, and the duration ranges from 10 to 500ms. The coverage rate for extreme scenarios reaches 99.2%, such as when all three phases of voltage drop by 25% simultaneously.

[0074] It should be explained that mapping high-dimensional feature vectors to continuous probability distributions through kernel density estimation, and combining this with confidence interval calculation to achieve a quantitative characterization of extreme fault scenarios, can provide probabilistic inputs for subsequent switching action simulations.

[0075] In this embodiment of the invention, homomorphic encryption is used to achieve encrypted transmission and decryption aggregation of model parameters, which can avoid cross-domain leakage of original power grid data; the probability distribution generated by kernel density estimation covers fault scenarios with a 99% confidence interval, enhancing the ability to represent extreme scenarios; parameter aggregation through federated learning can effectively improve the prediction accuracy of the model on multi-regional power grid data, providing reliable scenario input for subsequent switching action simulation; through the coordinated implementation of the above steps, a full-process quantitative modeling from distributed data acquisition to probabilistic scenario output is realized, providing a theoretically rigorous and engineering-feasible technical solution for the experimental switching simulation test of automatic transfer switch devices.

[0076] 2. Opening and Closing Delay Parameter Optimization and Timing Calibration Module: Based on the generated fault scenario probability distribution, a quantum annealing optimized switch action sequence generator is designed. The opening and closing delay parameters are dynamically adjusted using a particle swarm optimization algorithm, and sub-nanosecond timing calibration of multiple switch actions is achieved based on ultra-wideband synchronization technology. Specific steps include:

[0077] Based on the generated probability distribution of fault scenarios, the optimization objective function and constraints of the switching action sequence are defined;

[0078] The objective function J is optimized to minimize the total delay of the switching action. and motion energy consumption For the target, the relevant expression is:

[0079]

[0080] in, To optimize the weighting coefficients, 0 < <1, dynamically adjust according to the urgency of the fault, for example, take emergency faults. =0.8 to prioritize reducing latency; total delay of switching action The energy consumption is obtained by summing the action delays of all switches. It is obtained by summing the operating power of all switches;

[0081] The constraints include opening and closing delay accuracy constraints, timing logic constraints, and equipment safety constraints;

[0082] The timing delay accuracy constraint for opening and closing the circuit breaker is as follows: , For different switch indices, For the first The action delay of each switch; This represents the theoretically optimal delay.

[0083] The timing logic constraint is: for series switch groups, the following must be satisfied. ; The minimum action interval is set to 5ms by default.

[0084] Equipment safety constraints are: operating voltage ; These are the minimum operating voltage and the maximum operating voltage, respectively. , , Rated voltage;

[0085] It should be explained that by clearly defining the optimization direction and boundary conditions, a mathematical model can be provided for subsequent quantum annealing optimization, ensuring that the switching action sequence achieves global optimality while meeting safety and accuracy requirements.

[0086] When generating the initial switch action sequence based on the quantum annealing algorithm, the action timing of m switches is... Encoded as N qubits, ;in, For the maximum allowable delay, For time resolution.

[0087] With the objective function J as the energy function and the constraints as penalty terms, the expression for calculating the total Hamiltonian H is: ;in, To constrain penalties, ; This is the penalty coefficient, with a default value of 100.

[0088] When performing quantum annealing scheduling, the initial temperature =1000, annealing coefficient =0.95, iteration number L=1000, the quantum state is updated using the following formula:

[0089]

[0090] in, Let be the quantum state at time t; exp() is the exponential function; Here, j is the time evolution operator; j is the imaginary unit. To reduce Planck's constant; This is the initial quantum state.

[0091] Finally, it collapses to the ground state when the temperature W→0, outputting the initial action sequence. ;

[0092] It should be explained that by utilizing the global optimization capability of quantum annealing, an initial action sequence that satisfies the objective function can be generated quickly, avoiding the problem of traditional optimization algorithms getting stuck in local optima, and providing a high-quality initial solution for subsequent parameter fine-tuning of the particle swarm optimization algorithm.

[0093] Initial action sequence generated by quantum annealing As the initial population, the opening and closing delay parameters are dynamically optimized using the particle swarm optimization algorithm. The particle velocity and position update formulas are as follows:

[0094]

[0095] in, For the first The particle in the first Wei, Di The speed of each iteration; For particle indexing; For dimension indexing; This is the number of the next iteration. This is the inertia weight, with a value ranging from 0.4 to 0.9; For the first The particle in the first Wei, Di The speed of each iteration; This is a cognitive learning factor, with a value range of 1.4 to 2.0; The random number is a cognitive random number that follows a uniform distribution in the interval [0,1]. For the first The particle in the first The optimal position of an individual in dimension; For the first The particle in the first Wei, Di The position of the next iteration; The social learning factor has a value range of 1.4 to 2.0. For social random numbers, and Independent, uniformly distributed random numbers in the [0,1] range; For the entire particle swarm in the th The global optimal position of the dimension; For the first The particle in the first Wei, Di The position of the next iteration; For the first The particle in the first Wei, Di The speed of each iteration.

[0096] Furthermore, when the particle swarm optimization algorithm dynamically optimizes the opening and closing delay parameters, the fitness function expression is F=1 / (1+J); F is the fitness score of the candidate solution, and J is the objective function mentioned above.

[0097] It should be noted that by using the swarm intelligence optimization of the particle swarm optimization algorithm to further fine-tune the delay parameters based on the initial solution of quantum annealing, the accuracy can be controlled within a controllable range, while dynamically adapting to changes in fault scenarios, such as adjusting delay requirements caused by load fluctuations.

[0098] When performing sub-nanosecond timing calibration of multiple switching actions based on ultra-wideband synchronization technology, one UWB master node and M UWB slave nodes are deployed. The UWB master node is installed in the control center, and bidirectional timestamp transmission is performed between the master and slave nodes through UWB pulse signals.

[0099] Get the sending time of the master node's synchronization frame. The time of reception from the node is recorded as ,Delay The master node then sends a response frame, and the time of reception is recorded as follows: Then the one-way propagation delay of the UWB signal is .

[0100] When performing sub-nanosecond calibration, the local clock correction value of the slave node is calculated, involving the following expression: ;in, Correction value for the local clock of the slave node; Master node clock; For the slave node's local clock; The phase difference of the UWB signal; This is the carrier angular frequency.

[0101] A calibration is performed every 10ms, with the slave node correcting the value based on its local clock. Adjust the local crystal oscillator with an adjustment step size of 0.01ns to ensure that the master-slave node clock deviation is ≤0.1ns.

[0102] It should be explained that by using the high-precision time transfer and phase compensation of UWB, the timing synchronization error of multiple switching actions can be reduced from the microsecond level to the sub-nanosecond level, which can effectively avoid switching malfunctions caused by timing misalignment, such as closing arcs and load impacts.

[0103] The optimized switching action sequence and calibrated timing parameters are input into the digital twin system to simulate the action response under fault scenarios, and closed-loop feedback adjustment is achieved through error analysis.

[0104] Among them, the time delay accuracy error is calculated. and synchronization error The expression involved is:

[0105]

[0106] in, These represent the simulated action delay and the optimized target delay, respectively; D is the action sequence index.

[0107]

[0108] in, These are the local clock correction values ​​for switch node f and switch node g, respectively; f and g are any two switch nodes.

[0109] like >0.1ms or If the value is greater than 0.1 ns, restart particle swarm optimization, increase the number of iterations to 1500, and perform UWB calibration enhancement, such as increasing the phase sampling rate to 10 GSa / s, until the performance requirements are met.

[0110] In this embodiment of the invention, the global search capability of quantum annealing and the local fine-tuning of particle swarm optimization are combined to achieve timely and reliable control of the time delay accuracy of the switching action sequence, meeting the requirements for rapid fault response. By combining UWB technology with a phase compensation algorithm, the synchronization error of multiple switches can be reduced to below 0.1 ns, avoiding equipment damage caused by timing misalignment, such as reducing the probability of switch arc reignition by 99%. Through the objective function weight coefficient and closed-loop feedback mechanism, the optimization objective can be dynamically adjusted according to the fault type to adapt to complex power grid scenarios. By incorporating equipment safety thresholds and timing logic constraints into the constraints, combined with digital twin verification, it can be ensured that the action sequence meets the power system safety regulations.

[0111] 3. Virtual-Real Fusion Multidimensional Verification and Analysis Module: A virtual-real fusion verification platform is built, simultaneously injecting the optimized switching action sequence into the digital twin model and the physical automatic transfer switch (ATS). The output relay action waveforms of both are compared in real time via photonic crystal fiber transmission, outputting a two-dimensional verification report including logical criterion consistency and loop response time difference. Specific steps include:

[0112] Deploy physical automatic transfer switches, including output relays and current / voltage sampling modules, and connect them to actual power simulation systems, such as RTDS real-time digital simulators.

[0113] Setting up a digital twin model operating environment: An industrial-grade server is used to deploy a backup automatic switching digital twin model based on MATLAB / Simulink, including an electromagnetic transient simulation module and a relay logic module.

[0114] A two-way data transmission channel is constructed by connecting physical devices and digital twin servers through photonic crystal optical fibers.

[0115] A high-speed current sensor is connected in parallel across the output relay of the physical backup automatic transfer device. A virtual oscilloscope module is connected to the relay output port of the digital twin model to synchronously acquire the current waveform at the moment of action, including characteristic quantities such as rise time, peak current, and duration.

[0116] The collected waveform data is transmitted to the data processing server via photonic crystal fiber.

[0117] When calculating the logical criterion consistency, features are extracted from the collected waveforms, and the relay action logic states of the digital twin model and the physical device are compared. The relay action logic states include either action or no action, and the relevant expressions are as follows:

[0118]

[0119] Where C represents the logical criterion consistency; The number of actions that are consistent between the two logical states, such as both taking action or neither taking action. This represents the total number of verification actions.

[0120] When calculating the loop response time difference, the moment when 50% of the amplitude of the rising edge of the waveform is extracted as the response trigger time. The time difference between the digital twin and the physical device is then calculated, involving the following expression:

[0121]

[0122] in, This refers to the loop response time difference; The response trigger time of the digital twin model's output relay is based on the digital twin system clock; The response trigger time of the output relay of the physical backup automatic transfer device is calibrated using the PTP protocol based on the physical system clock.

[0123] Furthermore, the waveforms of both were overlaid using MATLAB, and the rise time was compared. Peak current The deviation was analyzed to determine whether the requirements were met. , ; These are the rise time of the output relay of the digital twin model and the rise time of the output relay of the physical backup automatic transfer device, respectively. These are the peak currents of the output relays of the digital twin model and the output relays of the physical backup automatic transfer device, respectively.

[0124] When generating a two-dimensional verification report of logical criterion consistency and loop response time difference, it includes the logical criterion consistency C value and a trend chart, with the trend chart drawn according to the action sequence time axis;

[0125] Loop response time difference The statistical distribution, including the maximum, minimum, and average values;

[0126] Waveform characteristic parameter deviation table, including rise time, peak current, etc.;

[0127] When making a conclusion, if C ≥ 99.5% and If the time is ≤5ms, the verification is considered successful. This represents the maximum loop response time difference.

[0128] Otherwise, mark the sequence number of the mismatched action and the reason for the deviation; for example, action 156: physical device relay sticking causes a mismatch in logic state.

[0129] In this embodiment of the invention, by comparing the physical device with the digital twin in real time, the limitations of single simulation or physical experiment can be avoided. For example, the digital twin can reproduce extreme working conditions, and the physical device verifies the actual hardware characteristics, thereby effectively improving the accuracy and reliability of simulation testing. Through logic criterion consistency analysis, the consistency of control logic can be ensured. Through loop response time difference analysis, dynamic characteristic matching can be ensured. Dual-dimensional index quantification can ensure verification accuracy and meet the stringent requirements of the power system for the correct logic and fast action of the automatic transfer switch. Compared with traditional optical fiber, the low loss and high bandwidth characteristics of photonic crystal fiber can avoid the interference of transmission delay on response time difference calculation.

[0130] It is worth noting that the formulas and expressions involved in the embodiments of the present invention all perform dimensional processing on the calculated data items and implement targeted standardization processing according to the characteristics of the calculated data items. For example, the numerical values ​​of the data items are uniformly mapped to the [0,1] interval. These are all existing conventional technical solutions, and the specific implementation steps will not be elaborated here.

[0131] In the several embodiments provided by this invention, it should be understood that the disclosed system can be implemented in other ways. For example, the embodiments of the invention described above are merely illustrative; for example, the division of modules is only a logical functional division, and there may be other division methods in actual implementation.

[0132] The modules described as separate components may or may not be physically separate. The components shown as modules may or may not be physical modules; they may be located in one place or distributed across multiple network modules. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs.

[0133] Furthermore, the functional modules in the various embodiments of the present invention can be integrated into one processing module, or each module can exist physically separately, or two or more modules can be integrated into one module. The integrated module can be implemented in hardware or in the form of hardware plus software functional modules.

[0134] It will be apparent to those skilled in the art that the present invention is not limited to the details of the exemplary embodiments described above, and that the present invention can be implemented in other specific forms without departing from the essential characteristics of the present invention.

[0135] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention.

Claims

1. A simulation test system for an automatic transfer switch, characterized in that, include: Fault scenario probability distribution analysis and processing module: Collects and preprocesses distributed power grid data during rotational outages through edge nodes, constructs a power grid state prediction model based on federated learning, trains the power grid state prediction model using the preprocessed distributed power grid data, and generates fault scenario probability distribution. The timing calibration module for optimizing the opening and closing delay parameters: Based on the generated probability distribution of fault scenarios, a quantum annealing-optimized switch action sequence generator is designed. The opening and closing delay parameters are dynamically adjusted through the particle swarm optimization algorithm, and sub-nanosecond timing calibration of multiple switch actions is achieved based on ultra-wideband synchronization technology. Virtual-Real Fusion Multidimensional Verification and Analysis Module: A virtual-real fusion verification platform is built, and the optimized switching action sequence is simultaneously injected into the digital twin model and the physical standby automatic transfer device. The output relay action waveforms of the two are compared in real time through photonic crystal fiber transmission, and a two-dimensional verification report including the logical criterion consistency and the loop response time difference is output.

2. The automatic transfer switch simulation test system according to claim 1, characterized in that, The collected distributed power grid data includes voltage sag waveforms, load mutation rates, and topology switching timing. The preprocessing includes voltage sag waveform denoising and load mutation rate calculation.

3. The automatic transfer switch simulation test system according to claim 2, characterized in that, The power grid state prediction model is constructed using an improved FedAvg federated learning framework. The local model is a CNN-LSTM hybrid network. During federated model training and encrypted aggregation, each edge node trains the model based on local data and uses homomorphic encryption to encrypt the gradient ∇L of the local model. The central server decrypts the data and then performs parameter aggregation. When generating the probability distribution of fault scenarios, kernel density estimation is used to construct the probability density function of the fault scenarios. Based on the probability density function, the 99% confidence interval is obtained through calculation.

4. The automatic transfer switch simulation test system according to claim 1, characterized in that, Based on the generated fault scenario probability distribution, the optimization objective function J and constraints of the switching action sequence are defined. The optimization objective function J aims to minimize the total switching action delay and energy consumption. The constraints include opening and closing delay accuracy constraints, timing logic constraints, and equipment safety constraints. ； in, To optimize the weighting coefficients, 0 < <1. The total delay of the switch action is dynamically adjusted according to the urgency of the fault. The energy consumption is obtained by summing the action delays of all switches. It is obtained by summing the operating power of all switches; The time delay accuracy constraint for opening and closing the circuit is: , For different switch indices, For the first The action delay of each switch; This represents the theoretically optimal delay. The timing logic constraint is: for series switch groups, the following must be satisfied. ; Minimum action interval; The equipment safety constraint is: operating voltage ; These are the minimum operating voltage and the maximum operating voltage, respectively. , , This is the rated voltage.

5. The automatic transfer switch simulation test system according to claim 4, characterized in that, When generating the initial switch action sequence based on the quantum annealing algorithm, the action timing of m switches is encoded into N qubits; With the objective function J as the energy function and the constraints as penalty terms, the expression for calculating the total Hamiltonian H is: ;in, To constrain penalty items; This is the penalty coefficient; When performing quantum annealing scheduling, the quantum state is updated by calculation based on the initial temperature, annealing coefficient, and number of iterations; When the temperature meets the requirements, it collapses to the ground state and outputs the initial action sequence; Using the initial action sequence generated by quantum annealing as the initial population, the opening and closing delay parameters are dynamically optimized through the particle swarm optimization algorithm.

6. The automatic transfer switch simulation test system according to claim 1, characterized in that, During sub-nanosecond calibration, the local clock correction value for the slave node is calculated. Calibration is performed every 10ms, and the slave node adjusts its local crystal oscillator based on the local clock correction value. The specific calculation is as follows: ； in, Correction value for the local clock of the slave node; Master node clock; For the slave node's local clock; The phase difference of the UWB signal; This is the carrier angular frequency.

7. The automatic transfer switch simulation test system according to claim 6, characterized in that, The optimized switching action sequence and calibrated timing parameters are input into the digital twin system to simulate the action response under fault scenarios, and closed-loop feedback adjustment is achieved through error analysis.

8. The automatic transfer switch simulation test system according to claim 7, characterized in that, Calculate and obtain the time delay accuracy error and synchronization error The expression involved is: ;in, These represent the simulated action delay and the optimized target delay, respectively; D is the action sequence index. ;in, These are the local clock correction values ​​for switch node f and switch node g, respectively, where f and g are any two switch nodes; like >0.1ms or If the value is greater than 0.1 ns, restart particle swarm optimization and perform UWB calibration enhancement until the performance requirements are met.

9. The automatic transfer switch simulation test system according to claim 1, characterized in that, When calculating the logical criterion consistency, feature extraction is performed on the collected waveforms, and the relay action logic states of the digital twin model and the physical device are compared to obtain the logical criterion consistency. When calculating the loop response time difference, the moment when 50% of the amplitude of the rising edge of the waveform is extracted as the response trigger time. The time difference between the digital twin and the physical device is calculated to obtain the loop response time difference.

10. The automatic transfer switch simulation test system according to claim 9, characterized in that, When generating a two-dimensional verification report for logical criterion consistency and loop response time difference, the report includes the logical criterion consistency value and trend chart, and the statistical distribution of loop response time difference. When making a conclusion, if both the logical criterion consistency and the maximum loop response time difference meet the corresponding verification requirements, the verification is deemed successful; otherwise, the sequence number of the non-compliant action and the reason for the deviation are marked.

Images