Black box photovoltaic unit fault ride-through control identification method and system based on BPA

Abstract

The invention discloses a BPA-based black box photovoltaic unit fault ride-through control identification method. The method comprises the steps of constructing a power flow and stability program of a photovoltaic unit BPA model, and filling in basic parameters of a power flow and stability program data card; constructing a full-parameter identification library of the model fault ride-through control card under the optimization of an identification algorithm; measuring fault condition data of the black box photovoltaic unit; and training the power flow and stability program by using data to obtain a BPA control model for black box photovoltaic unit fault ride-through control identification. The invention further designs a system which comprises a photovoltaic actual measurement training set data interval segmentation module, a hybrid algorithm multi-control mode parameter identification module of the photovoltaic BPA model, a photovoltaic actual measurement test set data interval segmentation module, a photovoltaic BPA model three-section test module and a photovoltaic BPA model ride-through performance evaluation module. According to the black box photovoltaic fault control mode and the parameter identification and detection method, the accuracy of safety and stability simulation analysis of a complex power system under photovoltaic large-scale grid connection is improved.

Description

Technical field:

[0001] The present invention relates to the technical field of photovoltaic unit fault ride-through control identification technology, and in particular to a black box photovoltaic unit fault ride-through control identification method and system based on BPA. Background technology:

[0002] With the vigorous development of the new energy industry, photovoltaics have been developed on a large scale, connected to the grid at a high ratio, and a large number of power electronic equipment have been used. However, photovoltaics have randomness and volatility in the power generation process. Therefore, in order to reliably and correctly evaluate the impact of photovoltaic units on the operation of the power system after access, it is necessary to establish a simulation model to characterize the performance of the actual photovoltaic unit. In the analysis of the power system, the simulation model of the photovoltaic unit usually adopts a general model and typical parameters. For example, the invention patent with the publication number of CN118920577A discloses a method and system for electromechanical transient modeling of photovoltaic power generation system. The structure and parameters of this type of model are simple, and the practicality is not strong in the simulation of large power grids, and the simulation accuracy of fault ride-through characteristics is not high. At the same time, as an important component of the photovoltaic unit, the inverter, its control performance will directly affect whether the entire new energy grid-connected system can operate safely and stably. However, the control mode and parameters of the inverter that has been put into use are often unknown due to factors such as commercial confidentiality. Therefore, it is urgent to design a method for identifying the key control parameters and methods of black box photovoltaic units. Summary of the invention:

[0003] In view of the above problems, the present invention provides a black-box photovoltaic unit fault ride-through control identification method and system based on BPA to solve the difficulties in establishing a simulation model that accurately reflects the fault ride-through performance of black-box photovoltaic units, thereby improving the accuracy of simulation analysis of complex power systems under large-scale photovoltaic grid connection.

[0004] A black-box photovoltaic unit fault ride-through control identification method based on BPA, the method specifically comprising:

[0005] Collect basic data of the photovoltaic unit, and build the power flow and stability program of the BPA model of the photovoltaic unit in the PSDEdit simulation software based on the basic data, wherein the basic data includes the unit topology and parameters of each component; based on the parameters of each component, calculate and fill in the basic parameters of the power flow and stability program data card of the BPA model of the photovoltaic unit;

[0006] Based on the high and low fault ride-through control card mechanism of the photovoltaic unit BPA model, an identification model and an objective function are established, and a hybrid identification optimization algorithm is introduced; a full parameter identification library of multiple control modes of the photovoltaic unit BPA model fault ride-through control card under the optimization of the hybrid identification algorithm is constructed; wherein the hybrid identification optimization algorithm is an improved stochastic gradient descent algorithm, and the specific method is:

[0007] Given n groups of observed sample data points, a model with known structure but unknown parameters is used to fit these data points, and the model parameters are determined to minimize the sum of the squares of the deviations between the model prediction value and the observed sample value. The specific process is as follows:

[0008] Assume that the model to be identified is y i =f(x i ,k), where y i 、x i is the input measured data, k is the unknown parameter of the model, and the goal of identification is to find the optimal parameter k so that the model predicts the value As close as possible to the measured data y i , and for each measured data point (x i ,y i ) must have a corresponding prediction point The deviation vector between the two can be expressed as:

[0009] In order to evaluate the average difference between the measured data and the predicted data of the entire data segment, the objective function is defined as the arithmetic mean of the difference between the measured data and the predicted data. The objective function is expressed as follows:

[0010]

[0011] n is the number of sample data points;

[0012] The original objective function is squared and then minimized, so the minimum value formula of the objective function can be updated as follows:

[0013]

[0014] The optimization goal is to minimize the objective function J(k) by adjusting the model parameter k. In order to optimize the objective function J(k) more quickly and accurately, we choose to combine the stochastic gradient descent and Adam algorithms to form an improved stochastic gradient descent algorithm. The optimization process of the algorithm is as follows:

[0015] Initialization parameters: Randomly initialize the control parameter k0, select the learning rate η, the momentum term m0=0, the variance term v0=0, the momentum decay factors β1, β2, and the small constant ∈ to prevent zero division.

[0016] Optimization stage 1: Pre-training with SGD: In the early stages of training, the SGD algorithm is used for pre-training to help parameters converge quickly. Iterate through the following steps:

[0017] Computing Gradients In each iteration, a mini-batch (or a single sample) is randomly selected from the sample dataset to calculate the gradient, and the gradient of the objective function J(k) with respect to the parameter k is calculated as follows:

[0018]

[0019] Update momentum and variance:

[0020]

[0021] Where m t is the momentum after t iterations, v t is the variance after t iterations;

[0022] Bias correction: To correct the initial momentum and variance bias, the following corrections are made:

[0023]

[0024] Update parameters: Update parameters according to the corrected momentum and variance:

[0025]

[0026] Repeat the above steps and check the convergence condition after each iteration. If the convergence condition is met, stop the Adam iteration. When the objective function converges to less than 0.1 or reaches the maximum number of iterations of 5000, the optimization algorithm terminates and the final optimization control parameter k is obtained.

[0027] The RTDS platform is used to measure the fault operating condition data of the black box photovoltaic unit, and the operating condition data is divided into four types of fault data sets, including: low-throughput symmetrical fault set, low-throughput asymmetrical fault set, high-throughput symmetrical fault set and high-throughput asymmetrical fault set; each type of fault data set is divided into a training set and a test set according to the proportion, and each type of fault crossing process is divided into three stages for separate control, that is, the data in each training set and test set are segmented according to the fault crossing period, the fault crossing recovery starting point, and the fault crossing recovery process, and the power, fault type, fault start and end time and fault impedance of each test set working condition are calculated;

[0028] Among them, the rolling standard deviation coefficient is used to divide each type of fault crossing process into three stages for separate control. The specific calculation method is as follows:

[0029] First, the overall fault ride-through interval of the photovoltaic unit is determined based on the high and low voltage threshold values. Then, the sliding window technology is used to continuously calculate the rolling mean and rolling standard deviation of the voltage, active power, and reactive power data within this fault ride-through time series interval. The calculation method is as follows:

[0030]

[0031] Among them, Rolling Mean t Rolling Std t Rolling standard deviation, t is the end position of the current window data, w is the size of the data size under the window;

[0032] Then, the rolling standard deviation coefficient of each variable is calculated based on the rolling mean and standard deviation, and the data interval segment where the rolling standard deviation coefficient of each variable is lower than the set threshold is identified. By intersecting these variable data interval segments, the data interval segment during the fault crossing period of the fault condition can be obtained. The calculation formula of the rolling standard deviation coefficient is as follows:

[0033]

[0034] Where, Rolling CV t is the rolling standard deviation coefficient;

[0035] Finally, continue to process the data of the remaining fault ride-through intervals, find out the data point position where each variable data first recovers to its normal operating steady-state value within the ±5% deviation range in the remaining interval, and take the maximum value to obtain the end point position of the fault recovery process; for the fault ride-through recovery starting point stage, the default starting point is the end point of the fault ride-through period, and the end point is located at a position where the time difference between the end point of the fault ride-through period and the end point of the fault recovery process is 10%;

[0036] Input each type of fault training set into the full parameter identification library of multiple control modes of the fault ride-through control card of the photovoltaic unit BPA model in turn, and obtain the parameter fitting values ​​and fitting errors of the mathematical models of multiple control modes in each section of the fault ride-through control card for each type of fault type; select a group of control mode combinations with the smallest fitting error in each section of the control mode as the possible optimal control mode for each section of the control mode, fill the corresponding control parameters into the corresponding position of the photovoltaic unit BPA model program card, cyclically import the test set working power and fault settings of the same fault type into the corresponding position of the photovoltaic unit BPA model program card, and call the power flow and stability program of the photovoltaic unit BPA model for simulation, wherein the fault setting includes Fault type, fault start and end time and fault impedance; calculate the deviation between the simulation data and the measured data of the PV unit. If the deviation is less than the national standard, the control mode combination and identification parameters are the BPA optimal control for this type of fault. Otherwise, modify or reselect the initial value of the identification algorithm until the deviation of all fault type test set operating conditions meets the national standard requirements, and obtain the PV unit BPA control model of the black box PV unit fault ride-through control identification. After obtaining the optimal PV BPA unit control model, the model can also be used for multiple groups of simulations of low-through and high-through faults, and the reactive support coefficient and active recovery coefficient of each simulation condition are calculated and compared with the national standard values, and finally the fault ride-through performance of the BPA model is evaluated.

[0037] Preferably, the parameters of each component include: photovoltaic power generation unit parameters, inverter normal steady-state control parameters, output line impedance and transformer parameters; the flow and stability program data cards of the photovoltaic unit BPA model include: PSL card, WES card, WEV card, WLP card, WLQ card, WEU card, WEQ card, WAF card and WAI card.

[0038] Preferably, the step length of the black-box photovoltaic unit fault condition data measured by the RTDS platform is converted into a simulation setting step length of the photovoltaic unit BPA model, and the original data is classified according to the fault type and purpose.

[0039] Preferably, the start point and the end point of the data fault crossing period and the fault recovery process are determined by the standard deviation coefficient, and the fault data set is accurately segmented.

[0040] Preferably, the specific method for calculating the deviation between the simulation data and the actual measured data of the photovoltaic unit is:

[0041] Calculate the average deviation of steady-state and transient intervals in each period:

[0042]

[0043] Among them, X S is the standard value of the BPA model simulation data of the electrical quantity to be tested; XM K is the standard value of the measured data of the electrical quantity to be tested; SSart and K SEend K is the first and last serial number of the BPA model simulation data within the calculation deviation interval; MSart and K MEend To calculate the first and last serial numbers of the measured data within the deviation interval;

[0044] Maximum deviation F3 in the steady-state interval:

[0045]

[0046] The weighted average total deviation F of the whole process G :

[0047] F G =10% F 1_故障前正常稳态 +60% F 1,2_故障穿越期间 +30% F 1,2_故障恢复

[0048] Among them, F 1_故障前正常稳态 is the average deviation of the normal steady state before the fault; F 1,2_故障穿越期间 is the average deviation of the transient steady state during the fault ride-through period; F 1,2_故障恢复 is the average deviation of the fault recovery transient steady state.

[0049] A black-box photovoltaic unit fault ride-through control identification system based on BPA includes a photovoltaic measured training set data interval segmentation module, a photovoltaic BPA model hybrid algorithm multi-control mode parameter identification module, a photovoltaic measured test set data interval segmentation module, a photovoltaic BPA model three-stage test module, and a photovoltaic BPA model ride-through performance evaluation module.

[0050] Preferably, the photovoltaic measured training data interval segmentation module is used for step size conversion of measured photovoltaic single fault type training set data and segmentation of three-stage intervals of fault crossing process;

[0051] The hybrid algorithm multi-control mode parameter identification module of the photovoltaic BPA model is used to identify the control parameters of all control modes in the fault ride-through WLP and WLQ control cards of the photovoltaic BPA unit model, and determine the most likely control mode combination of the BPA three-stage control based on the size of the identification error;

[0052] The photovoltaic actual measurement test set data interval segmentation module is used for the step size conversion, interval segmentation and acquisition of the power size of each working condition of the actual measurement photovoltaic single fault type test set data, and the calculation of the fault setting parameters;

[0053] The photovoltaic BPA model three-stage inspection module is used to import and solidify the most likely control mode combination and control parameters into the SWI program card, and cyclically modify the photovoltaic BPA model DAT card output power size and SWI card fault settings according to the inspection set working condition sequence, and cyclically call the BPA software power flow calculation PFNT.EXE program and the stability calculation SWNT.EXE program to complete the inspection simulation of all inspection set working conditions under the single fault type, and finally use the simulation SWX result data and the measured data to perform deviation calculation;

[0054] The photovoltaic BPA model evaluation module is used to evaluate the high and low fault ride-through performance of the photovoltaic BPA model under the optimal control mode.

[0055] The present invention designs a black-box photovoltaic unit fault ride-through control identification method and system based on BPA. According to the fault ride-through characteristics of the photovoltaic unit, the fault ride-through process is divided into three stages for accurate simulation of three-stage control. At the same time, after studying the simulation operation principle of the photovoltaic BPA mode fault ride-through control card, the present invention proposes a black-box photovoltaic unit identification method that determines the most likely control mode combination and control parameters of the BPA three-stage according to the size of the identification error under the hybrid algorithm multi-model fitting. It not only avoids the complicated combination of the control modes of each section of the BPA, greatly reduces the workload of the inspection, but also improves the comprehensiveness of the inspection. In addition, the black-box photovoltaic fault control mode and parameter identification and inspection method proposed by the present invention improves the accuracy of the simulation analysis of the safety and stability of complex power systems under large-scale photovoltaic grid connection. Description of the drawings:

[0056] Attached Figure 1 This is a block diagram of the steps of the black box photovoltaic unit fault ride-through control identification method based on BPA of the present invention. Attached Figure 2 This is the overall control block diagram of the photovoltaic stand-alone BPA model of the present invention.

[0057] Attached Figure 3 This is a schematic diagram of measuring actual data of the photovoltaic unit of the present invention.

[0058] Attached Figure 4 This is a comparison diagram of a two-phase 10% voltage drop fault in an embodiment of the present invention.

[0059] Attached Figure 5 This is a comparison diagram of a two-phase 20% voltage drop fault in an embodiment of the present invention.

[0060] Attached Figure 6 This is a comparison diagram of a two-phase 50% voltage drop fault in an embodiment of the present invention.

[0061] Attached Figure 7 This is a comparison diagram of a two-phase 125% voltage rise fault in an embodiment of the present invention. Specific implementation method:

[0062] In order to make the technical solution of the present invention easier to understand, a black-box photovoltaic unit fault ride-through control identification method and system based on BPA disclosed in the present invention is now clearly and completely described in combination with embodiments and drawings.

[0063] like Figure 1 As shown, a method for fault ride-through control identification of a black-box photovoltaic unit based on BPA control, the method specifically includes:

[0064] Collect the basic data of the photovoltaic unit, and build the power flow and stability program of the BPA model of the photovoltaic unit in the PSDEdit simulation software based on the basic data, wherein the basic data includes the unit topology and the parameters of each component, wherein the parameters of each component include the parameters of the photovoltaic power generation unit, the normal steady-state control parameters of the inverter, the output line impedance and the transformer parameters; based on the parameters of each component, calculate and fill in the basic parameters of the power flow and stability program data card of the BPA model of the photovoltaic unit, and the control cards that must be filled in with parameters are: PSL card (photovoltaic power generation model); WES card (converter current control and limiting model); WEV card (fault ride-through state judgment model); WLP card (active power control model under fault ride-through state); WLQ card (reactive power control model under fault ride-through state); WEU card (active power control model under normal operation state); WEQ card (reactive power control model under normal operation state); the control cards that are not required to be filled in are: WAF card (frequency control model); WAI card (inertia control model), and the overall control block diagram of the BPA control model of the photovoltaic unit is as follows: Figure 2 ;

[0065] Based on the high and low fault ride-through control card mechanism of the BPA model of the photovoltaic unit, the identification model and objective function are established, and a hybrid identification optimization algorithm is introduced; a full parameter identification library of multiple control modes of the BPA model fault ride-through control card of the photovoltaic unit under the optimization of the hybrid identification algorithm is constructed; wherein, the hybrid identification optimization algorithm is an improved stochastic gradient descent algorithm. It can be seen that there are a large number of linear and nonlinear models in the identification target model, and the measured sample set data of each model is large and complex, which has caused certain difficulties in the parameter identification of the model. Therefore, this paper adopts the improved stochastic gradient descent algorithm to optimize the identification of the control model.

[0066] The hybrid identification optimization algorithm is specifically:

[0067] Given n groups of observed sample data points, a model with known structure but unknown parameters is used to fit these data points, and the model parameters are determined to minimize the sum of the squares of the deviations between the model prediction value and the observed sample value. The specific process is as follows:

[0068] Assume that the model to be identified is y i =f(x i ,k), where y i 、x i is the input measured data, k is the unknown parameter of the model, and the goal of identification is to find the optimal parameter k so that the model predicts the value As close as possible to the measured data y i , and for each measured data point (x i ,y i ) must have a corresponding prediction point The deviation vector between the two can be expressed as:

[0069] In order to evaluate the average difference between the measured data and the predicted data of the entire data segment, the objective function is defined as the arithmetic mean of the difference between the measured data and the predicted data. The objective function is expressed as follows:

[0070]

[0071] n is the number of sample data points;

[0072] The original objective function is squared and then minimized, so the minimum value formula of the objective function can be updated as follows:

[0073]

[0074] The optimization goal is to minimize the objective function J(k) by adjusting the model parameter k. In order to optimize the objective function J(k) more quickly and accurately, we choose to combine the stochastic gradient descent and Adam algorithms to form an improved stochastic gradient descent algorithm. The optimization process of the algorithm is as follows:

[0075] Initialization parameters: Randomly initialize the control parameter k0, select the learning rate η, the momentum term m0=0, the variance term v0=0, the momentum decay factors β1, β2, and the small constant ∈ to prevent zero division.

[0076] Optimization phase: Use SGD for pre-training: In the early stages of training, use the SGD algorithm for pre-training to help parameters converge quickly. Iterate through the following steps:

[0077] Computing Gradients In each iteration, a mini-batch (or a single sample) is randomly selected from the sample dataset to calculate the gradient, and the gradient of the objective function J(k) with respect to the parameter k is calculated as follows:

[0078]

[0079] Update momentum and variance:

[0080]

[0081] Where m t is the momentum after t iterations, v t is the variance after t iterations;

[0082] Bias correction: To correct the initial momentum and variance bias, the following corrections are made:

[0083]

[0084] Update parameters: Update parameters according to the corrected momentum and variance:

[0085]

[0086] Repeat the above steps and check the convergence condition after each iteration. If the convergence condition is met, stop the Adam iteration. When the objective function converges to less than 0.1 or reaches the maximum number of iterations of 5000, the optimization algorithm terminates and the final optimization control parameter k is obtained.

[0087] Construct a full parameter identification library of multiple control modes of the fault ride-through control card of the photovoltaic unit BPA model under the optimization of the identification algorithm, and conduct measured data of single fault type ( Figure 3 ) During parameter identification, it is necessary to identify all control modes and parameters of a corresponding set of fault ride-through cards, and then select the most likely control mode combination for verification. The control mode models that need to be identified in a set of fault ride-through control cards are as follows:

[0088] The identification model of the active power control mode specified during the WLP card fault ride-through period is: ref =K p *P0+P SET , where P ref is the specified active power, K p is the active power coefficient, P0 is the initial active power, P SET To set the active power;

[0089] The identification model of the active current control mode specified during the WLP card fault ride-through period is: Ip ref =K VP *V t +K IP *IP0+IP SET , where Ip ref is the specified active current, K VP , K IP is the active current coefficient, V t is the terminal voltage amplitude, IP0 is the initial active current, IP SET To set the active current;

[0090] The identification model of low voltage limit active current control mode during WLP card fault ride-through is:

[0091] Among them, LvPI is the low voltage limit active current, LvPI1 is the current upper threshold, Brkpt is the voltage upper threshold, Zerox is the voltage lower threshold, and V is the terminal voltage standard value.

[0092] The identification model of the specified active current control mode under the WLP card fault ride-through recovery starting point control is:

[0093] Ip ref_start =min(K IP *IP0+IP SET ,IP0);Ip ref_start Initially specify the active power for the active power recovery process.

[0094] The identification model of the specified slope control mode under the WLP card fault ride-through recovery process control is:

[0095] P ref_pro =K*t BPA +P BPA ; Among them, P ref_pro is the specified active power of the active power recovery process; K is the specified slope; t BPA is the simulation time based on the photovoltaic BPA unit model; P BPA is the active intercept of the photovoltaic BPA model recovery process.

[0096] The identification model of the inertia curve recovery control mode under the WLP card fault ride-through recovery process control is:

[0097] In the formula, The active power is initially specified for the active power recovery process, T P is the active power recovery time constant of the inertia curve, and t3 is the initial time of the active power recovery process under BPA simulation.

[0098] The identification model of the reactive power control mode specified during the WLQ card fault ride-through period is: Q ref =K Q *Q0+Q SET ; Among them, Q ref is the specified reactive power, K Q is the reactive power coefficient, Q0 is the initial reactive power, Q SET To set the reactive power.

[0099] The identification model of the specified reactive current control mode during the WLQ card fault ride-through period is:

[0100] IQ ref =K VQ *(V set -V t )+K IQ *IQ0+IQ SET Among them, IQ ref is the specified reactive current, K VQ , K IQ is the reactive current coefficient, V set With V t are the fault ride-through set voltage and terminal voltage amplitude, IQ0 is the initial reactive current, IQ SET To set the reactive current.

[0101] The identification model of the specified reactive current control mode under the WLQ card fault ride-through recovery starting point control is:

[0102] IQ ref_start =K IQ *IQ0+IQ SET Among them, IQ ref_start Specifies the reactive current at which reactive power recovery starts.

[0103] The identification model of the inertia curve recovery control mode under the WLQ card fault ride-through recovery process control is:

[0104]

[0105] In the formula, is the initial specified active power of the reactive power recovery process, T Q is the reactive recovery time constant of the inertia curve.

[0106] Referring to the national standard GB / T 32892-2016, the RTDS platform is used to measure the fault condition data of the black box photovoltaic unit. The step size of the measured data set is converted into the simulation setting step size of the photovoltaic BPA model (the minimum simulation step size of BPA is 0.01 cycle). In addition, it is necessary to refine the voltage drop step size near the high and low voltage threshold values ​​of photovoltaics specified in the national standard (the high and low voltage threshold values ​​of the photovoltaic national standard are 1.1 and 0.85, and the voltage drop step size near the threshold value is generally set to 0.01pu). When it is impossible to obtain the high and low voltage threshold values ​​of the photovoltaic unit through collection, the change rules of the active and reactive waveforms of the fault condition near the voltage threshold value specified in the national standard can be observed to determine the exact high and low voltage threshold values. At the same time, the maximum value of the active and reactive current under extreme fault conditions can also be used to obtain the maximum current limit value of the system. The actual unit data measurement schematic diagram is shown in the figure below. Figure 2 .

[0107] The working condition data is divided into four types of fault data sets, and the fault types include: low-throughput symmetrical fault set, low-throughput asymmetrical fault set, high-throughput symmetrical fault set and high-throughput asymmetrical fault set; each type of fault data set is divided into a training set and a test set according to a ratio, and each type of fault crossing process is divided into three stages for separate control, that is, the data in each training set and the test set are segmented according to the fault crossing period, the fault crossing recovery starting point, and the fault crossing recovery process, and the power, fault type, fault start and end time and fault impedance of each test set working condition are calculated; the key to segmenting the working condition data is to determine the starting point and end point of the fault crossing period and the fault recovery process, and the standard deviation coefficient is a mathematical tool that reflects the degree of data fluctuation. Whether it is periodic data or trend data, by selecting a suitable window, the standard deviation coefficient can be used to effectively identify the steady-state interval of the data. Therefore, the present invention detects the interval range of the fault crossing period in the data by rolling the standard deviation coefficient, and the process is as follows:

[0108] First, the overall fault ride-through interval of the unit is determined based on the high and low voltage threshold values. Then, the sliding window technology is used to continuously calculate the rolling mean and rolling standard deviation of the voltage, active power, and reactive power data within this fault ride-through time series interval. The calculation method is as follows:

[0109]

[0110] Among them, Rolling Mean t Rolling Std t Rolling standard deviation, t is the end position of the current window data, and w is the size of the data in the window.

[0111] Then, the rolling standard deviation coefficient of each variable is calculated based on the rolling mean and standard deviation, and the data interval segment where the rolling standard deviation coefficient of each variable is lower than the set threshold is identified. By intersecting these variable data interval segments, the data interval segment during the fault crossing period of the fault condition can be obtained. The calculation formula of the rolling standard deviation coefficient is as follows:

[0112]

[0113] Where, Rolling CV t is the rolling standard deviation coefficient.

[0114] Finally, continue to process the data of the remaining fault-through intervals, find the data point position where each variable data first recovers to its normal operating steady-state value within the ±5% deviation range in the remaining interval, and take the maximum value to obtain the end point position of the fault recovery process. For the fault-through recovery starting point stage, the starting point is generally defaulted to the end point of the fault-through period, and the end point is located at a position where the time difference between the end point of the fault-through period and the end point of the fault recovery process is 10%. In this way, the three-stage intervals of the fault-through period, fault-through recovery starting point, and fault-through of the fault recovery process can be determined for the single fault type training set and test set data.

[0115] Input each type of fault training set into the full parameter identification library of the photovoltaic unit BPA model fault ride-through control card multiple control modes in turn, and obtain the parameter fitting values ​​and fitting errors of the mathematical models of the multiple control modes of each section of the fault ride-through control card for each type of fault type;

[0116] A group of control mode combinations with the smallest fitting error in each section of the control mode is selected as the possible optimal control mode of each section of the control mode. After the corresponding control parameters are filled into the corresponding position of the BPA model program card of the photovoltaic unit, the test set working power and fault settings (including fault type, fault start and end time and fault impedance) of the same fault type are cyclically imported into the corresponding position of the BPA model program card of the photovoltaic unit, and the power flow and stability program of the BPA model of the photovoltaic unit is called for simulation. The deviation between the simulation data and the measured data of the photovoltaic unit is calculated. If the deviation is less than the national standard, Then the control mode combination and identification parameters are the BPA optimal control for this type of fault. Otherwise, the initial value of the identification algorithm is modified or reselected until the operating condition deviations of all fault type test sets meet the requirements of the national standard, and the BPA control model of the photovoltaic unit for black-box photovoltaic unit fault ride-through control identification is obtained; after obtaining the optimal photovoltaic BPA unit control model, multiple groups of simulations of low-through and high-through faults can be performed on the model, and the reactive support coefficient and active recovery coefficient of each simulation condition can be calculated and compared with the values ​​specified in the national standard, and finally the fault ride-through performance of the BPA model is evaluated.

[0117] The calculation of transient and steady-state deviations in the three stages of fault ride-through control can be found in the national standard GB / T 32892-2016. Its main purpose is to verify the accuracy of the model by calculating the deviation between the simulation data and the measured data of the photovoltaic BPA model. After completing the segmentation of the BPA simulation data and the measured data of the test conditions, it is necessary to calculate the deviations of various electrical quantities in the normal steady state before the fault occurs, the transient steady state during the fault ride-through, and the transient steady state interval of fault recovery. The electrical quantities that need to calculate the deviations include output voltage, active current, reactive current, active power, and reactive power. The calculated deviation types include the average deviation of the transient interval of each time period, the average deviation and maximum deviation of the steady state interval, and the weighted average total deviation of the whole process. The calculation formulas for the average deviation, maximum deviation, and weighted average total deviation are as follows:

[0118] Calculate the average deviation of steady-state and transient intervals in each period:

[0119]

[0120] Among them, X S is the standard value of the BPA model simulation data of the electrical quantity to be tested; X M K is the standard value of the measured data of the electrical quantity to be tested; SSart and K SEend K is the first and last serial number of the BPA model simulation data within the calculation deviation interval; MSart and K MEend To calculate the first and last serial numbers of the measured data within the deviation interval;

[0121] Maximum deviation F3 in the steady-state interval:

[0122]

[0123] The weighted average total deviation F of the whole process G :

[0124] F G =10% F 1_故障前正常稳态 +60% F 1,2_故障穿越期间 +30% F 1,2_故障恢复

[0125] Among them, F 1_故障前正常稳态 is the average deviation of the normal steady state before the fault; F 1,2_故障穿越期间 is the average deviation of the transient steady state during the fault ride-through period; F 1,2_故障恢复 is the average deviation of the fault recovery transient steady state.

[0126] The deviation calculation results between the photovoltaic BPA model simulation data and the measured data of the test conditions should meet the following conditions: 1) The voltage deviations of the high-voltage side of the power generation unit of the photovoltaic BPA model of all test conditions should not be greater than the maximum allowable voltage deviation in Table 1; 2) The average deviation of the current, reactive current, active power and reactive power in the steady-state and transient intervals of all test conditions, the maximum deviation in the steady-state interval and the weighted average total deviation should not be greater than the maximum allowable deviation in Table 1; 3) For the model simulation test under the two-phase asymmetric fault condition, the maximum allowable deviation value of the fundamental positive sequence component is 1.5 times the value in Table 1.

[0127] Table 1 shows the maximum allowable deviation values ​​of electrical quantities tested as specified in the national standard.

[0128]

[0129] This embodiment also provides a method for evaluating the fault ride-through performance of the BPA control model of the photovoltaic unit: the model performs multiple simulations of two types of faults, low-through and high-through, and calculates the reactive support coefficient and active recovery coefficient of each simulated working condition, and compares them with the values ​​specified in the national standard, and finally evaluates the fault ride-through performance of the BPA model. The calculation formulas for evaluating the reactive support coefficient and active recovery coefficient for the fault ride-through performance of the photovoltaic BPA model are as follows:

[0130]

[0131] In the formula, K LVRT_P K is the low-power dynamic reactive current support coefficient of the photovoltaic power generation system, and its value range is generally 1.5 to 3; HVRT_P ΔI is the high-voltage dynamic reactive current support coefficient of the photovoltaic power generation system, and its value range is generally greater than 1.5; t The dynamic reactive current increment injected into the photovoltaic power generation system, in ampere (A); U t I is the voltage standard value of the photovoltaic power generation system grid connection point; N is the rated current of the photovoltaic power generation system, in ampere (A);

[0132]

[0133] Among them, K Q is the low-breakdown dynamic active power recovery coefficient of the photovoltaic power generation system, and its value range is generally not less than 0.3; S N Installed capacity of photovoltaic power generation system.

[0134] A black box photovoltaic unit fault ride-through control identification system based on BPA includes a photovoltaic measured training set data interval segmentation module, a photovoltaic BPA model hybrid algorithm multi-control mode parameter identification module, a photovoltaic measured test set data interval segmentation module, a photovoltaic BPA model three-stage test module, and a photovoltaic BPA model ride-through performance evaluation module. The photovoltaic measured training data interval segmentation module is used for the step size conversion of the measured photovoltaic single fault type training set data and the segmentation of the three-stage intervals of the fault ride-through process;

[0135] The hybrid algorithm multi-control mode parameter identification module of the photovoltaic BPA model is used to identify the control parameters of all control modes in the fault ride-through WLP and WLQ control cards of the photovoltaic BPA unit model, and determine the most likely control mode combination of the BPA three-stage control based on the size of the identification error;

[0136] The photovoltaic actual measurement test set data interval segmentation module is used for the step size conversion, interval segmentation and acquisition of the power size of each working condition of the actual measurement photovoltaic single fault type test set data, and the calculation of the fault setting parameters;

[0137] The photovoltaic BPA model three-stage inspection module is used to import and solidify the most likely control mode combination and control parameters into the SWI program card, and cyclically modify the photovoltaic BPA model DAT card output power size and SWI card fault settings according to the inspection set working condition sequence, and cyclically call the BPA software power flow calculation PFNT.EXE program and the stability calculation SWNT.EXE program to complete the inspection simulation of all inspection set working conditions under the single fault type, and finally use the simulation SWX result data and the measured data to perform deviation calculation;

[0138] The photovoltaic BPA model evaluation module is used to evaluate the high and low fault ride-through performance of the photovoltaic BPA model under the optimal control mode.

[0139] In order to verify the effectiveness of the method of the present invention, a black box photovoltaic unit in an actual photovoltaic power station was selected to test and verify the scheme of the present invention. The basic situation of the test system is as follows:

[0140] The actual rated power of the photovoltaic unit is 1.25MW, the maximum output power is 1.5MW, the DC side bus rated voltage is 1.250kV, the grid side rated voltage is 690V, the rated grid frequency is 50Hz, the maximum output current of the grid side inverter is 1.1pu, and the high and low voltage thresholds of the inverter are 0.9 and 1.1.

[0141] The two-phase asymmetric low-through fault data set with large and small power voltage drops of 10%, 20%, and 50% is set as the training set, and the two-phase asymmetric low-through fault data set with medium power voltage drops of 10%, 20%, and 50% is selected as the test set. The comparison between the simulation results of the optimal control photovoltaic BPA model and the measured results is shown in Figure 2. Figure 4 , 5 , 6. The calculation results of identification error are shown in Table 2, Table 3, and Table 4.

[0142] The two-phase asymmetric high-through fault data set with high and low power voltage rise of 120% and 130% is set as the training set, and the two-phase asymmetric high-through fault data set with medium power voltage rise of 125% is selected as the test set. The comparison between the simulation results of the optimal control photovoltaic BPA model and the measured results is shown in Figure 2. Figure 7 The calculation results of the identification error are shown in Table 5.

[0143] Table 2 Calculation results of two-phase 10% low-through fault identification error

[0144]

[0145] Table 3 Calculation results of two-phase 20% low-through fault identification error

[0146]

[0147] Table 4 Calculation results of two-phase 50% low-through fault identification error

[0148]

[0149] Table 5 Calculation results of two-phase 125% high-voltage fault identification error

[0150]

[0151] It should be pointed out that for ordinary technicians in this technical field, several improvements, substitutions, modifications and embellishments can be made without departing from the principles and purpose of the present invention. These improvements, substitutions, modifications and embellishments should also be regarded as the scope of protection of the present invention.

Claims

1. A black-box photovoltaic unit fault ride-through control identification method based on BPA, characterized in that: The method specifically comprises: Collect basic data of the photovoltaic unit, and build the power flow and stability program of the BPA model of the photovoltaic unit in the PSDEdit simulation software based on the basic data, wherein the basic data includes the unit topology and parameters of each component; based on the parameters of each component, calculate and fill in the basic parameters of the power flow and stability program data card of the BPA model of the photovoltaic unit; Based on the high and low fault ride-through control card mechanism of the photovoltaic unit BPA model, the identification model and objective function are established, and a hybrid identification optimization algorithm is introduced; a full parameter identification library of multiple control modes of the photovoltaic unit BPA model fault ride-through control card under the optimization of the hybrid identification algorithm is constructed; wherein the hybrid identification optimization algorithm is an improved stochastic gradient descent algorithm, and the specific method is: If the model to be identified is y i =f(x i ,k), where y i 、x i is the input measured data, k is the unknown parameter of the model, and the goal of identification is to find the optimal parameter k so that the model predicts the value As close as possible to the measured data y i , and for each measured data point (x i ,y i ) must have a corresponding prediction point The deviation vector between the two can be expressed as: The objective function is defined as the arithmetic mean of the difference between the measured data and the predicted data. The objective function is expressed as follows: n is the number of sample data points; After squaring the original objective function and then performing minimization optimization, the formula for solving the minimum value of the objective function can be updated as follows: Combining the stochastic gradient descent and Adam algorithms, an improved stochastic gradient descent algorithm is constructed. The optimization process of the algorithm is as follows: Initialization parameters: Randomly initialize the control parameter k0, select the learning rate η, the momentum term m0=0, the variance term v0=0, the momentum decay factors β1, β2, and the small constant ∈ to prevent zero division; Optimization phase: Use the gradient descent algorithm for pre-training, specifically: Computing Gradients In each iteration, a mini-batch (or a single sample) is randomly selected from the sample dataset to calculate the gradient, and the gradient of the objective function J(k) with respect to the parameter k is calculated as follows: Update momentum and variance: Where m t is the momentum after t iterations, v t is the variance after t iterations; Bias correction: To correct the initial momentum and variance bias, the following corrections are made: Update parameters: Update parameters according to the corrected momentum and variance: Repeat the above steps and check the convergence condition after each iteration; if the convergence condition is met, stop the Adam iteration; when the objective function converges to less than 0.1 or reaches the maximum number of iterations of 5000, the optimization algorithm terminates and the final optimization control parameter k is obtained; The RTDS platform is used to measure the fault operating data of the black-box photovoltaic unit, and the operating data is divided into four types of fault data sets, including: low-throughput symmetrical fault set, low-throughput asymmetrical fault set, high-throughput symmetrical fault set and high-throughput asymmetrical fault set; each type of fault data set is divided into a training set and a test set according to the proportion, and each type of fault crossing process is divided into three stages for separate control, that is, the data in each training set and test set are segmented according to the fault crossing period, the fault crossing recovery starting point, and the fault crossing recovery process, and the power, fault type, fault start and end time and fault impedance of each test set working condition are calculated; Among them, the rolling standard deviation coefficient is used to divide each type of fault crossing process into three stages for separate control. The specific calculation method is as follows: First, the overall fault ride-through interval of the photovoltaic unit is determined based on the high and low voltage threshold values. Then, the sliding window technology is used to continuously calculate the rolling mean and rolling standard deviation of the voltage, active power, and reactive power data within this fault ride-through time series interval. The calculation method is as follows: Among them, Rolling Mean t Rolling Std t Rolling standard deviation, t is the end position of the current window data, w is the size of the data size under the window; Then, the rolling standard deviation coefficient of each variable is calculated based on the rolling mean and standard deviation, and the data interval segment where the rolling standard deviation coefficient of each variable is lower than the set threshold is identified. By intersecting these variable data interval segments, the data interval segment during the fault crossing period of the fault condition can be obtained. The calculation formula of the rolling standard deviation coefficient is as follows: Where, Rolling CV t is the rolling standard deviation coefficient; Finally, continue to process the data of the remaining fault ride-through intervals, find out the data point position where each variable data first recovers to its normal operating steady-state value within the ±5% deviation range in the remaining interval, and take the maximum value to obtain the end point position of the fault recovery process; for the fault ride-through recovery starting point stage, the default starting point is the end point of the fault ride-through period, and the end point is located at a position where the time difference between the end point of the fault ride-through period and the end point of the fault recovery process is 10%; Each type of fault training set is input into the full parameter identification library of multiple control modes of the fault ride-through control card of the photovoltaic unit BPA model in turn, and the parameter fitting values ​​and fitting errors of the mathematical models of multiple control modes in each section of the fault ride-through control card of each type of fault type are obtained; a group of control mode combinations with the smallest fitting error in each section of the control mode are selected as the possible optimal control mode of each section of the control mode, and the corresponding control parameters are filled into the corresponding position of the photovoltaic unit BPA model program card, and the test set working condition power and fault setting of the same fault type are cyclically imported into the corresponding position of the photovoltaic unit BPA model program card, and the power flow and stability program of the photovoltaic unit BPA model is called for simulation, wherein the fault setting includes the fault type, the fault start and end time and the fault impedance; the deviation between the simulation data and the measured data of the photovoltaic unit is calculated, and if the deviation is less than the national standard, the control mode combination and the identification parameter are the BPA optimal control of this type of fault, otherwise the initial value of the identification algorithm is modified or reselected until the working condition deviation of all fault type test sets meets the national standard requirements, and the photovoltaic unit BPA control model of the black box photovoltaic unit fault ride-through control identification is obtained.

2. A black-box photovoltaic unit fault ride-through control identification method based on BPA as claimed in claim 1, characterized in that: The parameters of each component include: photovoltaic power generation unit parameters, inverter normal steady-state control parameters, output line impedance and transformer parameters; the power flow and stability program data cards of the photovoltaic unit BPA model include: PSL card, WES card, WEV card, WLP card, WLQ card, WEU card, WEQ card, WAF card and WAI card.

3. A black-box photovoltaic unit fault ride-through control identification method based on BPA as claimed in claim 1, characterized in that: The step length of the black-box photovoltaic unit fault condition data measured by the RTDS platform is converted into the simulation setting step length of the photovoltaic unit BPA model, and the original data is classified according to the fault type and purpose.

4. A black-box photovoltaic unit fault ride-through control identification method based on BPA as claimed in claim 1, characterized in that: The standard deviation coefficient is used to determine the start and end points of the data fault crossing period and the fault recovery process, and the fault data set is accurately segmented.

5. A black-box photovoltaic unit fault ride-through control identification method based on BPA as claimed in claim 1, characterized in that: The specific method for calculating the deviation between the simulation data and the actual measured data of the photovoltaic unit is: calculating the average deviation of the steady-state and transient intervals in each time period: Among them, X S is the standard value of the BPA model simulation data of the electrical quantity to be tested; X M K is the standard value of the measured data of the electrical quantity to be tested; SSart and K SEend K is the first and last serial number of the BPA model simulation data within the calculation deviation interval; MSart and K MEend To calculate the first and last serial numbers of the measured data within the deviation interval; Maximum deviation F3 in the steady-state interval: The weighted average total deviation F of the whole process G : F G =10%F 1_故障前正常稳态 +60%F 1,2_故障穿越期间 +30%F 1,2_故障恢复 Among them, F 1_故障前正常稳态 is the average deviation of the normal steady state before the fault; F 1,2_故障穿越期间 is the average deviation of the transient steady state during the fault ride-through period; F 1,2_故障恢复 is the average deviation of the fault recovery transient steady state.

6. A black box photovoltaic unit fault ride-through control identification system based on BPA, characterized in that: The system includes a photovoltaic measured training set data interval segmentation module, a photovoltaic BPA model hybrid algorithm multi-control mode parameter identification module, a photovoltaic measured test set data interval segmentation module, a photovoltaic BPA model three-stage test module, and a photovoltaic BPA model ride-through performance evaluation module.

7. A black-box photovoltaic unit fault ride-through control identification system based on BPA as claimed in claim 6, characterized in that: The photovoltaic measured training data interval segmentation module is used for the step size conversion of the photovoltaic single fault type training set data and the segmentation of the three-stage intervals of the fault crossing process; The hybrid algorithm multi-control mode parameter identification module of the photovoltaic BPA model is used to identify the control parameters of all control modes in the fault ride-through WLP and WLQ control cards of the photovoltaic BPA unit model, and determine the most likely control mode combination of the BPA three-stage control based on the size of the identification error; The photovoltaic actual measurement test set data interval segmentation module is used for the step size conversion, interval segmentation and acquisition of the power size of each working condition of the actual measurement photovoltaic single fault type test set data, and the calculation of the fault setting parameters; The photovoltaic BPA model three-stage inspection module is used to import and solidify the most likely control mode combination and control parameters into the SWI program card, and cyclically modify the photovoltaic BPA model DAT card output power size and SWI card fault settings according to the inspection set working condition sequence, and cyclically call the BPA software power flow calculation PFNT.EXE program and the stability calculation SWNT.EXE program to complete the inspection simulation of all inspection set working conditions under the single fault type, and finally use the simulation SWX result data and the measured data to perform deviation calculation; The photovoltaic BPA model evaluation module is used to evaluate the high and low fault ride-through performance of the photovoltaic BPA model under the optimal control mode.

Images