Relay protection simulation system

Abstract

The invention relates to the technical field of relay protection, in particular to a relay protection simulation system, which comprises a simulation unit and a parameter optimization unit, and is characterized in that the simulation unit comprises a power system model used for creating a simulated power system environment and simulating the operation condition of a power system; the protection device simulation module is configured to simulate the working mode and response of an actual relay protection device, namely test action parameters of the protection device in various fault scenes and system states; and the fault simulation module creates various fault conditions and is used for testing the working condition of the protection device when the fault occurs. According to the relay protection simulation system provided by the invention, real-time optimization of a simulation result is realized by combining the relay protection simulation system and a machine learning algorithm, and action parameters of a protection device are dynamically adjusted based on an optimization result, so that the response capability and accuracy of the protection system are improved, and the risks of false action and missing action are effectively reduced.

Description

Technical Field

[0001] The present invention relates to the technical field of relay protection, and particularly to a relay protection simulation system. Background Art

[0002] A relay protection simulation system is a computerized tool used in the power system to simulate, test, and optimize relay protection devices and schemes. Its main function is to verify and evaluate the response performance, accuracy, response speed, reliability, etc. of relay protection devices by simulating various fault scenarios in the power grid, so as to ensure that when a fault occurs in the real power grid, the relay protection system can cut off the faulty part in a timely and accurate manner, protecting the power grid and equipment from further damage.

[0003] The Chinese patent application with the publication number CN109596925A discloses a relay protection device simulation test system. It includes: a host computer for building a power system model based on RT-LAB; an RT-LAB simulator for running the power system model in real time; a power amplifier for amplifying the voltage and current signals in the power system model output by the simulator to the voltage and current signal ranges of the first type of relay protection device to be tested; a signal conversion device for conditioning the signals transmitted between the first type of relay protection device to be tested and the simulator; and a switch for converting the signals transmitted between the simulator and the second type of relay protection device to be tested. This invention can realize large-scale power system modeling and simulation, simultaneously meet the tests of multiple traditional and intelligent relay protection devices, and has a simple structure and is easy to implement.

[0004] As in the above application, most of the existing relay protection simulation systems are tested based on preset fault scenarios and static parameter configurations, and cannot adjust the parameters of the protection device in real time according to the dynamic changes of the power grid. Even if the simulation results show that the parameters of the protection device are inappropriate, manual intervention and multiple adjustments are often required, resulting in low efficiency. Secondly, the identification and diagnosis of fault types in traditional simulation systems rely on pre-set rules and lack intelligent adaptive capabilities. Summary of the Invention

[0005] To solve the above problems, the present invention provides a relay protection simulation system.

[0006] The present invention adopts the following technical solutions. A relay protection simulation system includes a simulation unit and a parameter optimization unit. The simulation unit includes:

[0007] A power system model for creating a simulated power system environment and simulating the operation of the power system;

[0008] A protection device simulation module, which is configured to simulate the working mode and response of an actual relay protection device, that is, to test the action parameters of the protection device under various fault scenarios and system states;

[0009] A measurement module, which is configured to simulate various measurement devices in a power system, collect data on the real-time operating state of the power system, record it as power system operating state data, and transmit these data as input signals to the protection device simulation module for analysis;

[0010] A simulation result evaluation module, which is configured to evaluate whether the simulation results meet the standards, mark the simulation results that do not meet the evaluation results, and mark them as non-compliant simulation results;

[0011] The parameter optimization unit includes:

[0012] A data collection module, which is used to collect the power system operating state data corresponding to the marked simulation results;

[0013] A first data analysis module, which inputs the collected power system operating state data into a pre-constructed first machine learning model and outputs the fault type;

[0014] A second data analysis module, which inputs the operating state data and the fault type into a second machine learning model, outputs the protection device adjustment parameters, and updates the protection device simulation module based on the protection device adjustment parameters.

[0015] As a further description of the above technical solution: The simulation unit further includes:

[0016] A fault simulation module, which creates various fault conditions for testing the working conditions of the protection device when a fault occurs;

[0017] A user interface module, which is used to provide an operation interface for the user to configure system parameters, start the simulation, observe the results, and view the simulation data information during the simulation process. The simulation data information includes a graphical display of the simulation process, the status of the protection device, and the system waveform display function, which is convenient for the user to operate and analyze.

[0018] As a further description of the above technical solution: The power system operating state data includes phase current data, line current data, load data, phase voltage data, line voltage data, and grid frequency data, and the power system operating state data is collected through the measurement module.

[0019] As a further description of the above technical solution: The construction method of the first machine learning model includes:

[0020] Initialize the structure of the first machine learning model. The first machine learning model structure adopts a multi-layer forward network structure of the MLP type, with four input layers, two hidden layers, and one output layer. The input layers include the first input layer, the second input layer, the third input layer, and the fourth input layer. The number of nodes in the first input layer is 2, corresponding to phase current data and line current data respectively. The number of nodes in the second input layer is 1, corresponding to load data. The number of nodes in the third input layer is 2, corresponding to phase voltage data and line voltage data respectively. The number of nodes in the fourth input layer is 1, corresponding to power grid frequency data. The hidden layers include the first hidden layer and the second hidden layer. The number of nodes in the first hidden layer is 128, with the ReLU function as the activation function. The number of nodes in the second hidden layer is 64, with the ReLU function as the activation function. The output layer is the first output layer, and the number of nodes in the first output layer is [1, n], predicting the fault type, where n is the number of fault types.

[0021] After initializing the structure of the first machine learning model, use the power system operating state data and the corresponding fault types of the power system operating state data to train the first machine learning model. The power system operating state data and the corresponding fault types of the power system operating state data are obtained from the database. The database records the power system operating state data corresponding to each fault type of each power system model for optimizing the model. The optimizer is Adam, the loss function is the MSE loss function, the batch size is set to 32, and the number of iterations is 200 rounds. When the loss function converges, stop training, indicating that the training is completed.

[0022] As a further description of the above technical solution: The training method of the first machine learning model includes:

[0023] Convert the power system operating state data into a corresponding set of feature vectors.

[0024] Use the power system operating state data as the input of the machine learning model. The machine learning model takes the fault type corresponding to each set of power system operating state data as the output, takes the actual corresponding fault type of each set of power system operating state data as the prediction target, and takes minimizing the loss function value of the machine learning model as the training target. Stop training when the loss function value of the machine learning model is less than or equal to the preset target loss value.

[0025] As a further description of the above technical solution: The second machine learning model includes 3 input layers, 2 convolutional layers, 2 pooling layers, 1 RNN layer and 2 fully connected layers; the number of neurons in the input layer is 5; the 2 convolutional layers include a first convolutional layer and a second convolutional layer, the convolutional kernel sizes of the first convolutional layer and the second convolutional layer are both 3×3, the activation functions are both ReLU, the strides are both 1, the margins are both 0, the number of convolutional kernels in the first convolutional layer is 32, and the number of convolutional kernels in the second convolutional layer is 64; the 2 pooling layers include a first pooling layer and a second pooling layer, the structures of the 2 pooling layers are the same, the window sizes are both 2×2, and the pooling method is the maximum pooling method, that is, within the pooling window, only the maximum value is taken as the eigenvalue at the corresponding position of the window after pooling; the RNN layer includes 64 LSTM units, 64 input channels and 64 output channels; the LSTM unit includes 5 components, and the 5 components include an input gate, a forget gate, a cell state, an output gate and a hidden state; the number of time steps of the input channels of the RNN layer is 1; the RNN is provided with a random weakening mechanism, and the random weakening probability is 0.2; the 2 fully connected layers include a first fully connected layer and a second fully connected layer; the first fully connected layer includes 128 neurons, and the activation function is ReLU; the second fully connected layer includes 1 neuron, which is used as the final output to output the adjustment data of the protection device.

[0026] As a further description of the above technical solution: The training method of the second machine learning model includes:

[0027] The training method of the second machine learning model includes:

[0028] Divide the adjustment data of V groups of historical protection devices into a training set and a validation set. There is adjustment data of A groups of historical protection devices in the training set. The adjustment data of the historical protection device includes operation state data, fault types and the adjustment parameters of the actual protection device. Initialize the weight parameter set Q of all layers and the bias parameter set B of all layers, and repeatedly calculate and update the parameters until the loss function loos converges, and the model is completed training.

[0029] As a further description of the above technical solution: The method for evaluating whether the simulation result meets the standard includes:

[0030] Response time evaluation, reliability evaluation and accuracy evaluation;

[0031] Among them, the response time evaluation method includes:

[0032] According to the design requirements of the power grid, the equipment protection requirements and the real-time operation requirements of the system, set the response time standard, denoted as the standard response time;

[0033] During the simulation process, record the time from the occurrence of the fault to the action of the protection device after each fault occurs, as the real-time response time;

[0034] When the real-time response time is greater than the standard response time, the simulation results are marked.

[0035] As a further description of the above technical solution: The reliability evaluation method includes:

[0036] By simulating different fault conditions multiple times, check whether the protection device simulation module can stably execute protection actions under different scenarios. When the protection device simulation module fails to execute protection actions, mark the simulation results.

[0037] As a further description of the above technical solution: The accuracy evaluation method includes:

[0038] Through the fault simulation module, simulate different types of faults, and check whether the protection device simulation module can accurately identify and make correct responses. When misoperation or omission occurs, mark the simulation results.

[0039] Beneficial effects:

[0040] A relay protection simulation system provided by the present invention marks the simulation results that do not meet the evaluation standards by evaluating whether the simulation results meet the standards, then collects the power system operation state data corresponding to the marked simulation results, inputs them into a pre-constructed first machine learning model to output the fault type, and then inputs the operation state data and the fault type into a second machine learning model to output the adjustment parameters of the protection device. Based on the adjustment parameters of the protection device, the protection device simulation module is updated. That is, by combining the relay protection simulation system with machine learning algorithms, the real-time optimization of the simulation results is achieved, and the action parameters of the protection device are dynamically adjusted based on the optimization results. Through real-time analysis and automatic adjustment, not only the response ability and accuracy of the protection system are improved, but also the risks of misoperation and omission are effectively reduced. This solution significantly shortens the design and debugging cycle of the relay protection simulation system, enhances the intelligence and adaptability of the power system, and provides a more reliable guarantee for the safe operation of the power grid. Description of the Drawings

[0041] The following further explains the present invention in conjunction with the drawings and embodiments:

[0042] Figure 1 It is the module connection diagram of the simulation unit provided by the embodiment of the present invention;

[0043] Figure 2 It is the module connection diagram of the optimization unit provided by the embodiment of the present invention;

[0044] Figure 3 It is the flowchart of the response time evaluation method provided by the embodiment of the present invention. Specific Embodiments

[0045] In order to make the technical means, creative features, achieved objectives and effects realized by the present invention easy to understand, the present invention will be further described below in conjunction with specific illustrations. It should be noted that, without conflict, the embodiments in the present application and the features in the embodiments can be combined with each other.

[0046] Embodiment 1

[0047] Please refer to Figures 1 - 3 , an embodiment of the present invention provides a technical solution: a relay protection simulation system, including: a simulation unit and an optimization unit, wherein, the simulation unit includes:

[0048] A power system model, which creates a simulated power system environment and simulates the operation of the power system. Among them, the operation of the power system includes the topological structure of the power grid, parameters of equipment (transformers, circuit breakers, transmission lines, etc.), load and power generation;

[0049] It should be noted that the power system model is used to provide input signals and an operating background for relay protection.

[0050] A protection device simulation module, which is configured to simulate the working mode and response of an actual relay protection device, that is, to test the action parameters of the protection device under various fault scenarios and system states.

[0051] A measurement module, which is configured to simulate various measurement devices in the power system, collect data on the real-time operating state of the power system, and transmit these data as input signals to the protection device simulation module for analysis;

[0052] A fault simulation module, which creates various fault conditions for testing the working conditions of the protection device when a fault occurs;

[0053] A simulation result evaluation module, which is configured to evaluate whether the simulation result meets the standard, mark the simulation results that do not meet the evaluation result as non-compliant simulation results;

[0054] A user interface module, which is used to provide an operation interface for users to configure system parameters, start simulations, observe results, and view simulation data information during the simulation process. The simulation data information includes graphical display of the simulation process, protection device status, and system waveform display functions, facilitating user operation and analysis;

[0055] A parameter optimization unit, which is used to analyze according to the simulation data and optimize the protection setting parameters.

[0056] Among them, the parameter optimization unit includes:

[0057] A data collection module, which collects data on the operating state of the power system obtained by the measurement module;

[0058] The operation state data of the power system includes phase current data, line current data, load data, phase voltage data, line voltage data, and grid frequency data.

[0059] The first data analysis module inputs the collected operation state data of the power system into a pre-constructed first machine learning model to output the fault type.

[0060] The second data analysis module inputs the operation state data and the fault type into a second machine learning model to output the protection device adjustment parameters, and updates the protection device simulation module based on the protection device adjustment parameters.

[0061] In this embodiment, by evaluating whether the simulation results meet the standards, the simulation results that do not meet the evaluation results are marked, and then the operation state data of the power system corresponding to the marked simulation results is collected and input into the pre-constructed first machine learning model to output the fault type. Then, the operation state data and the fault type are input into the second machine learning model to output the protection device adjustment parameters, and the protection device simulation module is updated based on the protection device adjustment parameters. That is, by combining the relay protection simulation system with the machine learning algorithm, the real-time optimization of the simulation results is realized, and the action parameters of the protection device are dynamically adjusted based on the optimization results. Through real-time analysis and automatic adjustment, not only the response ability and accuracy of the protection system are improved, but also the risks of misoperation and missed operation are effectively reduced. This solution greatly shortens the design and debugging cycle of the relay protection simulation system, enhances the intelligence and adaptability of the power system, and provides a more reliable guarantee for the safe operation of the power grid.

[0062] Embodiment 2

[0063] On the basis of the above embodiment, this embodiment further discloses the construction and training method of the first machine learning model:

[0064] The construction method of the first machine learning model includes:

[0065] Initialize the first machine learning model structure. The first machine learning model structure adopts a multi-layer forward network structure of the MLP type, with four input layers, two hidden layers, and one output layer. The input layers include the first input layer, the second input layer, the third input layer, and the fourth input layer. The number of nodes in the first input layer is 2, corresponding to phase current data and line current data respectively. The number of nodes in the second input layer is 1, corresponding to load data. The number of nodes in the third input layer is 2, corresponding to phase voltage data and line voltage data respectively. The number of nodes in the fourth input layer is 1, corresponding to grid frequency data. The hidden layers include the first hidden layer and the second hidden layer. The number of nodes in the first hidden layer is 128, with the ReLU function as the activation function. The number of nodes in the second hidden layer is 64, with the ReLU function as the activation function. The output layer is the first output layer, and the number of nodes in the first output layer is [1, n], predicting the fault type, where n is the number of fault types.

[0066] After initializing the first machine learning model structure, use the power system operating state data and the corresponding fault types to train the first machine learning model. The power system operating state data and the corresponding fault types are obtained from the database. The database records the power system operating state data corresponding to each fault type of the power system model each time, which is used to optimize the model. The optimizer is Adam, the loss function is the MSE loss function, the batch size is set to 32, and the number of iterations is 200 rounds. When the loss function converges, the training ends, indicating that the training is completed.

[0067] It should be noted specifically that: The Adam optimizer is one of the five main optimizers commonly used in machine learning, and its full name is Adaptive Moment Estimation.

[0068] It should be noted that the fault types include short circuit faults, grounding faults, overload faults, open circuit faults, and unbalance faults;

[0069] Among them, short circuit faults include single-phase ground short circuit, two-phase short circuit, two-phase ground short circuit, three-phase short circuit, and three-phase ground short circuit; grounding faults include single-phase grounding fault, two-phase grounding fault, and three-phase grounding fault; overload faults include continuous overload faults and instantaneous overload faults; open circuit faults include single-phase open circuit fault, two-phase open circuit fault, and three-phase open circuit fault; unbalance faults include negative sequence current faults and zero sequence current faults.

[0070] The training method of the first machine learning model includes:

[0071] Convert the power system operating state data into a corresponding set of feature vectors;

[0072] Use the power system operation status data as the input of the machine learning model. The machine learning model takes the fault type corresponding to each group of power system operation status data as the output, takes the actual fault type corresponding to each group of power system operation status data as the prediction target, and takes minimizing the loss function value of the machine learning model as the training target. Stop training when the loss function value of the machine learning model is less than or equal to the preset target loss value.

[0073] The loss function value of the first machine learning model is the mean square error.

[0074] The mean square error is one of the commonly used loss functions. By minimizing the loss function formula as the goal to train the model, the machine learning model can better fit the data, thereby improving the performance and accuracy of the model.

[0075] In the loss function, MSE is the loss function value of the first machine learning model, x is the feature vector group number; e is the number of feature vector groups; y x is a group of fault types output by the first machine learning model corresponding to the xth group of feature vectors, is the actual fault type corresponding to the xth group of feature vectors.

[0076] Embodiment 2

[0077] Based on the above embodiment, this embodiment further discloses the construction and training method of the second machine learning model:

[0078] The second machine learning model includes 3 input layers, 2 convolutional layers, 2 pooling layers, 1 RNN layer and 2 fully connected layers; the number of neurons in the input layer is 5; the 2 convolutional layers include the first convolutional layer and the second convolutional layer, and the kernel sizes of the first convolutional layer and the second convolutional layer are both 3×3, the activation functions are both ReLU, the strides are both 1, and the paddings are both 0. The number of kernels in the first convolutional layer is 32, and the number of kernels in the second convolutional layer is 64; the 2 pooling layers include the first pooling layer and the second pooling layer, and the structures of the 2 pooling layers are the same. The window sizes are both 2×2, and the pooling method uses the max pooling method, that is, within the pooling window, only the maximum value is taken as the feature value at the corresponding position of the pooled window; the RNN layer includes 64 LSTM units, 64 input channels and 64 output channels; the LSTM unit includes 5 components, and the 5 components include an input gate, a forget gate, a cell state, an output gate and a hidden state; the number of time steps of the input channels of the RNN layer is 1; the RNN is provided with a random weakening mechanism, and the random weakening probability is 0.2; the 2 fully connected layers include the first fully connected layer and the second fully connected layer; the first fully connected layer includes 128 neurons, and the activation function is ReLU; the second fully connected layer includes 1 neuron, which is used as the final output to output the adjustment data of the protection device. The regression calculation formula for the adjustment data Y of the protection device is:

[0079] Y = E×t + C;

[0080] In the formula, E is the weight of the second fully connected layer, C is the bias value of the second fully connected layer; t is a three-dimensional tensor.

[0081] The training method of the second machine learning model includes:

[0082] Divide the adjustment data of V groups of historical protection devices into a training set and a validation set. There are adjustment data of A groups of historical protection devices in the training set. The adjustment data of the historical protection device includes operation state data, fault types and the actual adjustment parameters of the protection device. Initialize the weight parameter set Q of all layers and the bias parameter set B of all layers. Calculate the predicted value y through forward propagation. The actual adjustment parameter of the protection device is set as yt. The formula for calculating the loss function loos is:

[0083]

[0084] In the formula yt r represents the actual adjustment parameter of the protection device in the adjustment data of the r-th group of historical protection devices, and y r represents the predicted adjustment parameter of the protection device in the adjustment data of the r-th group of historical protection devices. r represents the group number of the adjustment data of the historical protection device, and r ≤ A; calculate the gradient through backpropagation. The formulas for updating the parameters using the gradient descent method include:

[0085]

[0086] In the formula, γ represents the learning rate, which is a preset value; represents the gradient of Q, represents the gradient of B; Q′ represents the updated set of weight parameters Q; B′ represents the updated set of bias parameters B;

[0087] Repeatedly calculate and update the parameters until loos converges, that is, the model training is completed.

[0088] It should be noted that the adjustment parameters include action time, action current, overcurrent multiple, delay setting, and differential protection sensitivity;

[0089] Among them, the action time refers to the time from when the protection device detects a fault or abnormal situation to when it actually cuts off the circuit. When the simulation results show that the response time is insufficient or too long, the action time needs to be adjusted. According to the operation requirements of the power grid and the type of fault, the action time is reasonably set;

[0090] The action current refers to the trigger current value determined for the protection device when an overcurrent or short - circuit fault occurs. When the simulation results show that the protection device fails to cut off the current in time, the action current needs to be appropriately adjusted. By increasing or decreasing the set value of the action current for adjustment, where increasing the set current can reduce misoperation, and decreasing the set current can improve sensitivity and quickly respond to faults;

[0091] The overcurrent multiple refers to the multiple relationship between the action current of the protection device and the rated current of the equipment. A reasonable overcurrent multiple can ensure that the protection device can quickly respond when the fault current exceeds the normal operating current;

[0092] The delay setting refers to setting a certain delay after a fault occurs for the protection device to avoid false responses to instantaneous fluctuations or short - term disturbances. According to the simulation results, the delay setting value is adjusted so that the protection device can avoid misoperation when the power grid load fluctuates or the type of fault is not clear, and at the same time ensure timely response when a fault occurs;

[0093] The differential protection sensitivity is used to detect internal faults of the equipment. Too low sensitivity may lead to missed detection of internal faults, and too high sensitivity may cause misoperation. Adjust the sensitivity setting of the differential protection according to the simulation results. If the system detects too frequent misoperations or missed operations, the sensitivity setting needs to be optimized to ensure that faults can be quickly and accurately identified.

[0094] In this embodiment, a deep - learning algorithm is used to establish the first machine - learning model and the second machine - learning model, which can accurately evaluate the fault type and the adjustment parameters of the protection device, realize the automatic identification of the fault type and ensure the automatic adjustment of the device parameters, reduce manual intervention, and improve the operation efficiency.

[0095] Embodiment 3

[0096] The method for evaluating whether the simulation result meets the standard includes:

[0097] Response time evaluation, reliability evaluation, and accuracy evaluation;

[0098] Among them, the response time evaluation method includes:

[0099] According to the design requirements of the power grid, the equipment protection requirements, and the real-time operation requirements of the system, set the response time standard, denoted as the standard response time;

[0100] During the simulation process, record the time from the occurrence of a fault to the action of the protection device after each fault occurs, as the real-time response time;

[0101] When the real-time response time is greater than the standard response time, mark the simulation result;

[0102] The reliability evaluation method includes:

[0103] By simulating different fault conditions multiple times, check whether the protection device simulation module can stably execute the protection action under different scenarios. When the protection device simulation module fails to execute the protection action, mark the simulation result.

[0104] The accuracy evaluation method includes:

[0105] Through the fault simulation module, simulate different types of faults, and check whether the protection device simulation module can accurately identify and make a correct response. When there is misoperation or omission, mark the simulation result.

[0106] The above shows and describes the basic principles, main features, and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited by the above embodiments. The descriptions in the above embodiments and the specification only illustrate the principles of the present invention. Without departing from the spirit and scope of the present invention, the present invention will have various changes and improvements, and these changes and improvements all fall within the scope of the present invention claimed. The scope of protection claimed by the present invention is defined by the appended claims and their equivalents.

Claims

1. A relay protection simulation system, characterized in that: It includes a simulation unit and a parameter optimization unit, and the simulation unit includes: Power system model, used to create a simulated power system environment and simulate the operation of the power system; A protection device simulation module, which is constructed to simulate the working mode and response of an actual relay protection device, i.e., to test the action parameters of the protection device under various fault scenarios and system states; The measurement module is constructed to simulate various measurement devices in the power system, collect data on the real-time operation status of the power system, record them as power system operation status data, and pass these data as input signals to the protection device simulation module for analysis; A simulation result evaluation module is configured to evaluate whether the simulation result meets the standard, and mark the simulation result that does not meet the standard as a simulation result that does not meet the standard; The parameter optimization unit comprises: A data collection module, used for collecting power system operation status data corresponding to the marked simulation results; a first data analysis module, inputting the collected power system operation status data into a pre-built first machine learning model, and outputting a fault type; The second data analysis module inputs the operating status data and the fault type into the second machine learning model, outputs the protection device adjustment parameters, and updates the protection device simulation module based on the protection device adjustment parameters.

2. A relay protection simulation system according to claim 1, characterized in that: The simulation unit also includes: Fault simulation module, which creates various fault conditions to test the operation of protection devices in the event of a fault; The user interface module is used to provide users with an operation interface for configuring system parameters, starting simulation, observing results, and viewing simulation data information during the simulation process. The simulation data information includes a graphical display of the simulation process, protection device status, and system waveform display function.

3. A relay protection simulation system according to claim 2, characterized in that: The power system operation status data includes phase current data, line current data, load data, phase voltage data, line voltage data and grid frequency data, and the power system operation status data is collected and acquired by a measurement module.

4. A relay protection simulation system according to claim 3, characterized in that: The method for constructing the first machine learning model includes: Initialize the first machine learning model structure. The first machine learning model structure adopts an MLP type multi-layer forward network structure with four input layers, two hidden layers, and one output layer. The input layer includes the first input layer, the second input layer, the third input layer, and the fourth input layer. The number of nodes in the first input layer is 2, corresponding to the phase current data and the line current data respectively. The number of nodes in the second input layer is 1, corresponding to the load data. The number of nodes in the third input layer is 2, corresponding to the phase voltage data and the line voltage data respectively. The number of nodes in the fourth input layer is 1, corresponding to the power grid frequency data. The hidden layer includes the first hidden layer and the second hidden layer. The number of nodes in the first hidden layer is 128, and the ReLU function is used as the activation function. The number of nodes in the second hidden layer is 64, and the ReLU function is used as the activation function. The output layer is the first output layer, and the number of nodes in the first output layer is [1, n], predicting the fault type, where n is the number of fault types. After initializing the first machine learning model structure, the power system operating status data and the fault types corresponding to the power system operating status data are used to train the first machine learning model. The power system operating status data and the fault types corresponding to the power system operating status data are obtained from a database. The database records the power system operating status data corresponding to each fault type of each power system model for each time, which is used to optimize the model. The optimizer is Adam, the loss function is the MSE loss function, the batch size is set to 32, the number of iterations is 200 rounds, and the training is ended when the loss function converges, indicating that the training is completed.

5. A relay protection simulation system according to claim 4, characterized in that: The first machine learning model training method comprises: Convert the power system operation status data into a corresponding set of feature vectors; The power system operating status data is used as the input of the machine learning model, the machine learning model takes the fault type corresponding to each group of power system operating status data as the output, takes the fault type actually corresponding to each group of power system operating status data as the prediction target, and takes minimizing the loss function value of the machine learning model as the training target; training is stopped when the loss function value of the machine learning model is less than or equal to the preset target loss value.

6. A relay protection simulation system according to claim 5, characterized in that: The second machine learning model includes 3 input layers, 2 convolutional layers, 2 pooling layers, 1 RNN layer and 2 fully connected layers; the number of neurons in the input layer is 5; the 2 convolutional layers include the first convolutional layer and the second convolutional layer, the convolution kernel sizes of the first convolutional layer and the second convolutional layer are both 3×3, the activation functions are both ReLU, the step sizes are both 1, the margins are both 0, the number of convolution kernels of the first convolutional layer is 32, and the number of convolution kernels of the second convolutional layer is 64; the 2 pooling layers include the first pooling layer and the second pooling layer, the two pooling layers have the same structure, the window size is both 2×2, and the pooling method adopts the maximum pooling method, that is, within the pooling window, only take The maximum value is used as the feature value of the corresponding position of the window after pooling; the RNN layer includes 64 LSTM units, 64 input channels and 64 output channels; the LSTM unit includes 5 components, and the 5 components include input gate, forget gate, cell state, output gate and hidden state; the time step number of the input channel of the RNN layer is 1; RNN is equipped with a random attenuation mechanism, and the probability of random attenuation is 0.2; the two fully connected layers include the first fully connected layer and the second fully connected layer; the first fully connected layer includes 128 neurons, and the activation function is ReLU; the second fully connected layer includes 1 neuron, which outputs the adjustment data of the protection device as the final output.

7. A relay protection simulation system according to claim 6, characterized in that: The training method of the second machine learning model includes: The adjustment data of historical protection devices in group V are divided into a training set and a validation set. The training set contains the adjustment data of historical protection devices in group A. The adjustment data of historical protection devices include operating status data, fault types and actual adjustment parameters of protection devices. The weight parameter set Q of all layers and the bias parameter set B of all layers are initialized. The parameters are repeatedly calculated and updated until the loss function loos converges and the model completes training.

8. A relay protection simulation system according to claim 1, characterized in that: The method for evaluating whether the simulation result meets the standard includes: response time assessment, reliability assessment, and accuracy assessment; The response time evaluation methods include: According to the design requirements of the power grid, the protection requirements of the equipment and the real-time operation requirements of the system, the response time standard is set and recorded as the standard response time; During the simulation, the time from the occurrence of each fault to the action of the protection device is recorded as the real-time response time; When the real-time response time is greater than the standard response time, the simulation result is marked.

9. A relay protection simulation system according to claim 8, characterized in that: The reliability evaluation method comprises: By simulating different fault conditions multiple times, check whether the protection device simulation module can stably perform protection actions in different scenarios. When the protection device simulation module fails to perform protection actions, the simulation results will be marked.

10. A relay protection simulation system according to claim 8, characterized in that: The accuracy assessment method includes: The fault simulation module is used to simulate different types of faults to check whether the protection device simulation module can accurately identify and respond correctly. If a false operation or missed operation occurs, the simulation result will be marked.

Images