High-power new energy converter actual measurement modeling system based on convolutional neural network

Abstract

The invention discloses a high-power new energy converter actual measurement modeling system based on a convolutional neural network, and belongs to the technical field of new energy converter parameter identification. The system comprises a semi-physical simulation system, a virtual simulation system, a parameter identification system based on a convolutional neural network, a deviation measurement and calculation system, a semi-physical simulation experiment working condition configuration module (8), a virtual simulation experiment working condition configuration module (9), a semi-physical simulation experiment data acquisition and processing module (10) and a virtual simulation experiment data acquisition and processing module (11). A model verification module (21), a model verification result report module (22), and a new energy converter actual measurement model output module (23). The high-power new energy converter actual measurement modeling system based on the convolutional neural network can maintain the accuracy of parameter identification under the background that the proportion of new energy to be accessed to a power grid is continuously increased and new energy converter control model parameters are gradually complex, and meanwhile, the calculation time of the identification process is shortened; conditions are provided for accurately cognizing the operation mechanism of the new energy converter with complex characteristics and checking the safety and stability of a power system.

Description

Technical Field

[0001] This invention belongs to the field of parameter identification technology for new energy converters, and in particular relates to a test modeling system for high-power new energy converters based on convolutional neural networks. Background Technology

[0002] As the proportion of new energy sources connected to the grid continues to increase, the large-scale connection of new energy power generation equipment has led to the introduction of new equipment such as power electronics, resulting in a series of new, wide-ranging, and high-frequency electromagnetic transient problems, such as overvoltage, broadband oscillation, bidirectional power flow, and harmonic pollution. These problems can cause drastic changes in system state parameters, leading to inaccurate identification of new energy grid connection models.

[0003] Current simulation work for renewable energy power plants mainly uses typical models, which cannot accurately describe the transient processes and actual operational characteristics of the plants. Many power outages both domestically and internationally have been related to inaccurate renewable energy simulations. Similarly, the situation regarding domestic renewable energy experimental modeling is also not optimistic, severely restricting accurate simulation and understanding of system characteristics. Against this backdrop, the National Dispatch Center (NDC) requires the implementation of simulation verification of renewable energy power generation model parameters, replacing typical models with real-time renewable energy models for experimental modeling. However, conventional experimental modeling relies on real-time renewable energy simulation systems for fault ride-through characteristic testing, which presents challenges such as simulation complexity and uncontrollable time constraints. Directly using physical renewable energy converter controllers for testing significantly improves the efficiency of renewable energy experimental modeling. Furthermore, current experimental modeling faces problems such as high testing thresholds, strong technological barriers, difficulty in model parameter identification, and high rates of simulation non-convergence. In addition, the experimental modeling process is lengthy and complex, involving professional knowledge from multiple fields such as power systems, control theory, and equivalent calculations. It suffers from low automation, high manpower and material costs, and a lack of dedicated model verification and deviation calculation programs, resulting in a long modeling cycle. To address the aforementioned issues, this invention constructs a high-power new energy converter experimental modeling system based on convolutional neural networks. This system is used for standardized parameter identification, model verification, and deviation calculation, shortening the modeling cycle and providing support for conducting experimental modeling work, accurately understanding the complex operating mechanism of new energy converters, and verifying the safety and stability of power systems. Summary of the Invention

[0004] This invention constructs a high-power renewable energy converter experimental modeling system based on convolutional neural networks. This system is used for standardized parameter identification, model verification, and deviation calculation, shortening the modeling cycle and providing support for conducting experimental modeling work, accurately understanding the complex operating mechanisms of renewable energy converters, and verifying the safety and stability of power systems. The overall composition of this invention is as follows: Figure 1As shown, the system specifically includes: a hardware-in-the-loop (HIL) simulation system, a virtual simulation system, a parameter identification system based on a convolutional neural network, a deviation calculation system, a HIL experiment condition configuration module (8), a virtual simulation experiment condition configuration module (9), a HIL experiment data acquisition and processing module (10), a virtual simulation experiment data acquisition and processing module (11), a model verification module (21), a model verification result reporting module (22), and a new energy converter measured model output module (23). The functions and working principles of each system and module are described in detail below.

[0005] 1. Hardware-in-the-loop simulation system. This system includes a new energy converter main circuit simulation module (1), a new energy DC power supply simulation module (2), a power grid / load simulation module (3), a new energy power supply controller (24), and a simulator (25). The new energy converter main circuit is built in the simulation environment. The new energy DC power supply simulation model and the power grid / load simulation model are connected and uploaded to the simulator. Digital signal commands are input to the simulator through the physical new energy power supply controller. The simulator performs encoding calculations and outputs analog voltage and current signals back to the new energy power supply controller, realizing hardware-in-the-loop hardware-in-the-loop simulation. A schematic diagram of the hardware-in-the-loop simulation is shown below. Figure 2 As shown.

[0006] 2. Virtual Simulation System. This system includes a new energy converter simulation module (4), a DC power supply simulation module (5), a virtual power grid / load simulation module (6), and a new energy power controller simulation module (7). The main circuit of the new energy converter simulation is built in the simulation environment, and the DC power supply simulation model and the power grid / load simulation model are connected. Signal control is provided through the new energy power controller simulation model.

[0007] 3. A parameter identification system based on a convolutional neural network. The system includes a training group parameter initialization module (12), a target output vector setting module (13), an actual output vector calculation module (14), an error calculation module (15), a parameter adjustment module (16), and a parameter output module (17). The system uses the measured data collected by the hardware-in-the-loop simulation experimental data acquisition and processing module (10) to identify the control parameters of the new energy converter through a convolutional neural network.

[0008] A typical structure of a convolutional neural network includes convolutional layers, pooling layers, and fully connected layers. Its core feature is the convolution operation, which reduces the number of weights, lowers computational cost, and preserves the spatial structure information of the object. In a convolutional neural network, the convolution formula is expressed as:

[0009]

[0010] Pooling layers can reduce computation and improve model robustness without changing the dimension of the feature maps. Following the pooling layers is a fully connected layer, which transforms all the feature matrices from the pooling layers into one-dimensional feature vectors. Fully connected layers are typically placed at the end of a convolutional neural network structure to classify the data and then output the results.

[0011] Convolutional neural networks (CNNs) perform supervised training, so their sample set consists of vector pairs of the form (input vector, ideal output vector). All these vector pairs should originate from the actual operating structure of the network, i.e., the simulated system; they are collected from the actual operating system. Before training begins, all weights should be initialized with different random numbers. The training process consists of two phases. The first phase is the forward propagation phase, where a sample is taken from the sample set, input into the network, and the corresponding actual output is calculated. During this phase, information is transmitted from the input layer to the output layer through successive transformations; this process is also the normal execution process after the network completes training. The second phase is the backpropagation phase, where the difference between the actual output and the corresponding ideal output is calculated, and the weight matrix is ​​adjusted according to the method of minimizing the error. The specific training process of the network is as follows: Figure 3 As shown, the process is executed by 6 modules. The training group parameter initialization module (12) randomly selects N samples as the training group, sets each weight and threshold to a small random value close to 0, and initializes the precision control parameters and learning rate. The target output vector setting module (13) gives the target output vector. The actual output vector calculation module (14) calculates the intermediate layer output vector and the final actual output vector of the network. The error calculation module (15) compares the elements in the actual output vector with the elements in the target output vector and calculates the output error. The parameter adjustment module (16) calculates the adjustment amount of each weight and threshold in turn, adjusts the weight and threshold, and judges whether the precision requirements are met. If not, the target output vector is modified and retraining is performed. If the requirements are met, the process enters the parameter output module (17). The parameter output module (17) ends the training and outputs the weights and thresholds.

[0012] 4. Deviation Calculation System. This system includes a transient interval average deviation calculation module (18), a steady-state interval average deviation and maximum deviation calculation module (19), and a weighted average total deviation calculation module (20). The deviation calculation system uses the measured data and simulation data collected by the hardware-in-the-loop simulation experiment data acquisition and processing module (10) and the virtual simulation experiment data acquisition and processing module (11) to calculate the transient interval average deviation, steady-state interval average deviation, and maximum deviation of voltage, active current, reactive current, active power, and reactive power at each time period, and calculates the weighted average total deviation.

[0013] Methods for calculating average deviation and maximum deviation:

[0014] a) The formula for calculating the average deviation F1 in the steady-state interval is:

[0015]

[0016] In the formula, X S X represents the per-unit value of the model simulation data for the electrical quantity to be evaluated. M K represents the per-unit value of the test data for the electrical quantity to be tested. S_Start K S_End To calculate the first and last indexes of the model simulation data within the error interval, K M_Start K M_End These are the first and last serial numbers of the test data within the error interval.

[0017] b) The formula for calculating the average deviation F2 of the transient interval is:

[0018]

[0019] c) The formula for calculating the maximum deviation F3 in the steady-state interval is:

[0020]

[0021] The average deviations of each time period are calculated by weighted average. The weights of each time period in the fault condition disturbance test are: A (before disturbance) 10%, B (during disturbance) 60%, C (after disturbance) 30%; the weights of each time period in the normal condition test are: A (before command step) 30%, B (after command step) 70%.

[0022] 5. Hardware-in-the-loop simulation test condition configuration module (8). This module configures the test conditions for the hardware-in-the-loop simulation system and performs hardware-in-the-loop simulation experiments.

[0023] 6. Virtual simulation experiment condition configuration module (9). This module configures the experimental conditions for the virtual simulation system and conducts virtual simulation experiments. The parameters of the new energy power controller simulation model in the virtual simulation experiment are obtained by the parameter identification system based on the convolutional neural network.

[0024] The experimental conditions configured by the hardware-in-the-loop simulation test condition configuration module (8) and the virtual simulation test condition configuration module (9) include normal operating conditions and fault operating conditions. The normal operating conditions include high power, medium power and low power experiments, and the fault operating conditions include symmetrical fault, asymmetrical fault, low penetration four operating conditions (0%, 20%, 50%, 80%) and high penetration two operating conditions (120%, 130%) disturbance experiments. The control parameters of the new energy converter under normal and fault operating conditions are obtained through parameter identification. During actual operation, the corresponding mode parameters are switched according to the working environment of the new energy converter.

[0025] According to GB / T 19964—2024, the State Grid Corporation of China has clearly stipulated the requirements for the ability of large-scale photovoltaic power plants to operate without disconnecting from the grid when encountering low-voltage faults. The operational scope that such power plants should meet is as follows: Figure 4 As shown. Photovoltaic power plants connected to the grid via high-voltage levels require dynamic reactive power support during grid faults to assist in voltage recovery within the affected area. Currently, a typical low-voltage ride-through control strategy prioritizes ensuring the required reactive current output while controlling the renewable energy converter to balance active current output within total current limits. Under grid voltage-oriented vector control, the reference value I for reactive current during faults, considering various factors, is... q_ref The calculation formula is:

[0026]

[0027] In the formula, K qU U is the reactive current support factor; U is the per-unit value of the voltage amplitude at the grid connection point; I q0 K represents the per-unit value of reactive current before the fault. qI This is the reactive current proportionality coefficient; This is the baseline value for low-voltage ride-through reactive current; I qmax This represents the maximum value of the reactive current.

[0028] The calculation of active current reference values ​​during a fault can be broadly categorized into the following three types depending on the control scheme:

[0029]

[0030] In the formula, P0 is the per-unit value of the active power output of the new energy converter before the fault; I max This represents the maximum current of the new energy converter; I d0 K represents the per-unit value of the active current before the fault. dI K is the active current proportionality coefficient. dU This is the active current support factor; This is the baseline value for low-voltage ride-through active current.

[0031] Furthermore, to mitigate overvoltage issues on the DC side of renewable energy converters during grid faults, low-voltage control strategies often include a DC voltage regulation stage. Since overvoltage is essentially a manifestation of power imbalance, the explanation of this stage requires starting with power analysis. According to instantaneous power theory, the power input from the renewable energy converter to the grid is:

[0032]

[0033] At the instant a low-voltage fault occurs, the active current I output by the new energy converter decreases rapidly due to the rapid drop in the grid connection point voltage U. dBefore any changes occur, the active power P fed into the grid will suddenly decrease, but the DC-side power will remain unchanged due to the temporary instability of the DC voltage. Before the AC and DC active power of the renewable energy converter returns to equilibrium, the differential power will continue to charge the DC bus capacitor, causing the DC voltage to rise. During this process, the operating point of the photovoltaic array will gradually shift from the maximum power point to... Figure 5 Point A on the PV curve is shown. To ensure DC-side voltage safety, the maximum power point tracking algorithm of the new energy converter needs to be locked first, and instead, the active current reference value I during the low-voltage ride-through period should be used. d_ref Calculate the allowable active power P injected into the grid based on the grid connection voltage U. LV This refers to the power that the photovoltaic array needs to generate at this time. This power is used to adjust the voltage at the photovoltaic array terminals to keep the active power balanced and achieve DC voltage stability.

[0034] Meanwhile, when a power system fault causes the voltage at the grid connection point of the photovoltaic power station to rise, the photovoltaic power station should have the following characteristics: Figure 6 The high voltage ride-through capability is shown. When the grid connection voltage of the photovoltaic power station is between 110% and 130% of the nominal voltage, the reactive current injected into the grid by the photovoltaic power station should be the reactive current output value I during normal operation before the voltage rise. q0 With dynamic reactive current increment ΔI t The difference in dynamic reactive current increment of the photovoltaic power station should respond to the voltage change at the grid connection point and should satisfy:

[0035] ΔI t =K qI ×(U-1.1)×I N (1.1≤U≤1.3) (7)

[0036] In the formula, ΔI t The dynamic reactive current increment injected into a photovoltaic power plant, measured in amperes (A); I N This is the rated current of the photovoltaic power station, expressed in amperes (A).

[0037] During periods of voltage rise at the grid connection point, photovoltaic power plants should have active power control capabilities to maintain the active power unchanged before the fault, provided that dynamic reactive current support is met.

[0038] 7. Hardware-in-the-loop simulation experiment data acquisition and processing module (10). This module acquires hardware-in-the-loop simulation experiment data and records voltage, active current, reactive current, active power, and reactive power data.

[0039] 8. Virtual simulation experiment data acquisition and processing module (11). This module acquires virtual simulation experiment data and records voltage, active current, reactive current, active power, and reactive power data accordingly.

[0040] 9. Model Validation Module (21). This module determines whether the deviation meets the standard based on the deviation calculation results of the deviation calculation system. If the deviation meets the standard, the model validation result is input into the model validation result report module (22). If the deviation does not meet the standard, the identification parameters are adjusted or the parameter identification system based on convolutional neural networks is used to re-identify the parameters. The maximum allowable deviation is shown in Table 1.

[0041] Table 1

[0042]

[0043] 10. Model Validation Result Report Module (22). This module generates a result report based on the deviation calculation and model validation results.

[0044] 11. Module for Outputting Measured Model of New Energy Converter (23). This module provides the measured model of the new energy converter. Attached Figure Description

[0045] Figure 1 Overall block diagram of the experimental modeling system for high-power new energy converters based on convolutional neural networks

[0046] Figure 2 This is a schematic diagram of a hardware-in-the-loop simulation.

[0047] Figure 3 This describes the specific training process of a convolutional neural network.

[0048] Figure 4 This is to meet the low-voltage ride-through capability requirements for photovoltaic power plants.

[0049] Figure 5 This is a schematic diagram showing the changes in the operating point of the photovoltaic array during low voltage periods.

[0050] Figure 6 This is to meet the high-voltage ride-through capability requirements for photovoltaic power plants. Detailed Implementation

[0051] The technical solution of the present invention will be further described in detail below with reference to the embodiments of the present invention. Obviously, the embodiments described herein are only a part of the embodiments of the present invention, and not all of them. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort should fall within the scope of protection of the present invention.

[0052] This invention constructs a high-power renewable energy converter experimental modeling system based on convolutional neural networks. This system standardizes parameter identification, model verification, and deviation calculation, shortening the modeling cycle and providing support for conducting experimental modeling work, accurately understanding the complex operating mechanisms of renewable energy converters, and verifying the safety and stability of power systems. Specific implementation methods are as follows:

[0053] 1. The hardware-in-the-loop simulation system performs hardware-in-the-loop simulation experiments after configuring the experimental conditions using the hardware-in-the-loop simulation experimental condition configuration module (8). The hardware-in-the-loop simulation experimental condition configuration module (8) configures disturbance tests for four operating conditions (0%, 20%, 50%, 80%): high power, medium power, low power, symmetrical fault, asymmetrical fault, and low penetration, and two operating conditions (120%, 130%): high penetration.

[0054] 2. Semi-physical simulation experiment data acquisition and processing module (10) acquires semi-physical simulation experiment data, including voltage, active current, reactive current, active power, and reactive power, and inputs them into the parameter identification system based on convolutional neural network.

[0055] 3. The parameter identification system based on convolutional neural network identifies the control parameters of the new energy converter. The training process of the convolutional neural network is executed by 6 modules. The training group parameter initialization module (12) randomly seeks N samples as the training group, sets each weight and threshold to a small random value close to 0, and initializes the precision control parameters and learning rate. The target output vector setting module (13) gives the target output vector. The actual output vector calculation module (14) calculates the intermediate layer output vector and the final actual output vector of the network. The error calculation module (15) compares the elements in the actual output vector with the elements in the target output vector and calculates the output error. The parameter adjustment module (16) calculates the adjustment amount of each weight and threshold in turn, adjusts the weight and threshold, and judges whether the precision requirements are met. If not, the target output vector is modified and retrained. If it is met, the parameter output module (17) is entered. The parameter output module (17) ends the training and outputs the weight and threshold.

[0056] 4. The virtual simulation experiment configuration module (9) configures the experimental conditions for the virtual simulation system and conducts virtual simulation experiments. The experimental conditions are consistent with those of the semi-physical simulation experiments. The parameters of the new energy power controller simulation model in the virtual simulation experiment are obtained by the parameter identification system based on the convolutional neural network.

[0057] 5. Virtual simulation experiment data acquisition and processing module (11) Acquires and processes virtual simulation experiment data, including voltage, active current, reactive current, active power and reactive power.

[0058] 6. The deviation calculation system uses the measured data and simulation data collected by the hardware-in-the-loop simulation experiment data acquisition and processing module (10) and the virtual simulation experiment data acquisition and processing module (11) to calculate the average deviation of voltage, active current, reactive current, active power, and reactive power in the transient interval, the average deviation of steady-state interval, and the maximum deviation in each time period, and calculates the weighted average total deviation.

[0059] 7. Model verification module (21) determines whether the deviation meets the standard based on the deviation calculation result of the deviation calculation system. If the deviation meets the standard, the model verification result is input into the model verification result report module (22). If the deviation does not meet the standard, the identification parameters are adjusted or the parameter identification system based on convolutional neural network is used to re-identify the parameters.

[0060] 8. The model verification result report module (22) generates a result report based on the deviation calculation and model verification results, and the new energy converter measured model output module (23) provides the new energy converter measured model.

Claims

1. A high-power new energy converter measured modeling system based on a convolutional neural network, characterized in that, The system mainly comprises a semi-physical simulation system, a virtual simulation system, a parameter identification system based on a convolutional neural network, a deviation measurement system, a semi-physical simulation experiment working condition configuration module (8), a virtual simulation experiment working condition configuration module (9), a semi-physical simulation experiment data acquisition and processing module (10), a virtual simulation experiment data acquisition and processing module (11), a model verification module (21), a model verification result reporting module (22), a new energy converter actual measurement model output module (23); The semi-physical simulation system comprises a new energy converter main circuit simulation module (1), a new energy direct current power supply simulation module (2), a power grid / load simulation module (3), a new energy power supply controller (24), and a simulation machine (25), a new energy converter main circuit is built in a simulation environment, a new energy direct current power supply simulation model and a power grid / load simulation model are connected and uploaded to the simulation machine, and a signal control is provided through the new energy power supply controller; The virtual simulation system comprises a new energy converter simulation module (4), a direct current power supply simulation module (5), a virtual power grid / load simulation module (6), and a new energy power supply controller simulation module (7), a new energy converter simulation main circuit is built in a simulation environment, a direct current power supply simulation model and a power grid / load simulation model are connected, and a signal control is provided through the new energy power supply controller simulation model; The parameter identification system based on the convolutional neural network comprises a training group parameter initialization module (12), a target output vector setting module (13), an actual output vector calculation module (14), an error calculation module (15), a parameter adjustment module (16), and a parameter output module (17); The deviation measurement system comprises a transient interval average deviation calculation module (18), a steady-state interval average deviation and maximum deviation calculation module (19), and a weighted average total deviation calculation module (20).

2. The system according to claim 1, wherein, The semi-physical simulation system is configured with experimental conditions by the semi-physical simulation experiment condition configuration module (8), and then performs a semi-physical simulation experiment, the semi-physical simulation experiment data acquisition and processing module (10) acquires semi-physical simulation experiment data and inputs the semi-physical simulation experiment data to the parameter identification system based on a convolutional neural network, the parameter identification system based on a convolutional neural network identifies new energy converter control parameters, the virtual simulation experiment condition configuration module (9) configures experimental conditions for a virtual simulation system, and performs a virtual simulation experiment, in the virtual simulation experiment, the new energy converter control parameters are obtained by the parameter identification system based on a convolutional neural network, the virtual simulation experiment data acquisition and processing module (11) acquires and processes virtual simulation experiment data, the deviation measurement system measures the deviation by using the measured data and simulation data acquired by the semi-physical simulation experiment data acquisition and processing module (10) and the virtual simulation experiment data acquisition and processing module (11), inputs the deviation measurement result to the model verification module (21), the model verification module (21) judges whether the deviation meets the standard according to the deviation measurement result, if the deviation meets the standard, inputs the model verification result to the model verification result report module (22), if the deviation does not meet the standard, adjusts the identified parameters or re-performs parameter identification by the parameter identification system based on a convolutional neural network, and the model verification result report module (22) generates a result report according to the deviation measurement and the model verification result, and outputs a new energy converter measured model by the new energy converter measured model output module (23).

3. The system of claim 1, wherein the system is configured to: The semi-physical simulation system connects the new energy direct current power supply simulation model and the power grid / load simulation model of the main circuit of the new energy converter built in the simulation environment, uploads the main circuit to the RT_LAB simulation machine, inputs a digital signal instruction to the simulation machine through the new energy power supply controller, performs coding calculation by the simulation machine, feeds back the voltage and current output analog signals to the new energy power supply controller by the simulation machine, and realizes the semi-physical simulation of hardware in the loop.

4. The system according to claim 1, wherein, The semi-physical simulation experiment condition configuration module (8) and the virtual simulation experiment condition configuration module (9) respectively configure experimental conditions for the semi-physical simulation system and the virtual simulation system, the experimental conditions include normal conditions and fault conditions, the normal conditions are high-power, medium-power and low-power experiments, and the fault conditions are symmetric fault, asymmetric fault, low penetration four conditions (0%, 20%, 50%, 80%) and high penetration two conditions (120%, 130%) disturbance experiments, the new energy converter control parameters under the normal and fault conditions are obtained by parameter identification, and corresponding mode parameters are switched according to the working environment of the new energy converter during actual operation.

5. The system according to claim 1, wherein, The semi-physical simulation experiment data acquisition and processing module (10) and the virtual simulation experiment data acquisition and processing module (11) respectively acquire semi-physical simulation experiment data and virtual simulation experiment data, record voltage, active current, reactive current, active power and reactive power data, and the simulation model parameters of the new energy power supply controller in the virtual simulation experiment are obtained by the parameter identification system based on a convolutional neural network.

6. The system according to claim 1, wherein, The parameter identification system based on the convolutional neural network uses the data collected by the semi-physical simulation experiment data acquisition and processing module (10) to realize the new energy converter control parameter identification through the convolutional neural network, and the expression of the convolutional neural network is: The training process of the convolutional neural network is performed by six modules. The training group parameter initialization module (12) randomly seeks N samples as a training group, sets each weight and threshold to a small random value close to 0, initializes the precision control parameter and the learning rate, and the target output vector setting module (13) gives the target output vector. The actual output vector calculation module (14) calculates the intermediate layer output vector and the final actual output vector of the network. The error calculation module (15) compares the elements in the actual output vector with the elements in the target output vector to calculate the output error. The parameter adjustment module (16) sequentially calculates the adjustment amount of each weight and threshold, adjusts the weight and threshold, and judges whether the precision requirement is met. If not, the target output vector is modified and retrained. If yes, the parameter output module (17) is entered, which ends the training and outputs the weight and threshold.

7. The system according to claim 1, wherein, The deviation calculation system calculates the voltage, active current, reactive current, active power, reactive power, transient interval average deviation, steady interval average deviation and maximum deviation by using the measured data and simulation data collected by the semi-physical simulation experiment data acquisition and processing module (10) and the virtual simulation experiment data acquisition and processing module (11); The average deviation and maximum deviation calculation method is: a) The steady interval average deviation F1 calculation formula is: In the formula, X S is a model simulation data standard value of an electrical quantity to be evaluated, X M is an experimental data standard value of the electrical quantity to be evaluated, K S_Start , K S_End are the first and last serial numbers of model simulation data in a calculation error interval, K M_Start , K M_End are the first and last serial numbers of experimental data in the calculation error interval b) The transient interval average deviation F2 calculation formula is: c) The steady interval maximum deviation F3 calculation formula is: The average deviations of each period are weighted and averaged. The fault condition disturbance experiment period weights are A (before disturbance) 10%, B (during disturbance) 60%, and C (after disturbance) 30%. The normal condition experiment period weights are A (before command step) 30%, B (after command step) 70%.

Images