A new energy unit full working condition impedance identification method and device

Abstract

The application discloses a new energy unit full working condition impedance identification method and device, relates to the field of power system impedance identification, and the scheme comprises the following steps: acquiring electric parameter information and a concerned frequency, inputting the electric parameter information and the concerned frequency into a neural network to obtain positive sequence impedance; the neural network training step comprises the following steps: constructing an electromagnetic simulation model to determine a basic training set and a test set according to the electromagnetic simulation model; processing the corresponding sample of the target concerned frequency band based on a preset sample expansion strategy to obtain an enhanced sample and determine an enhanced training set; and optimizing the structure parameters of the neural network by using an improved grey wolf algorithm to determine optimal structure parameters, which is beneficial to effectively relieving overfitting and underfitting of the impedance identification result, corresponding to constructing a neural network and training based on the enhanced training set to guide the neural network to mainly learn the characteristic information of the target concerned frequency band until a preset training termination condition is reached. The application can reliably and accurately realize impedance identification under any working condition and any frequency point by relying on the trained neural network.

Description

Technical Field

[0001] This invention relates to the field of power system impedance identification technology, and in particular to a method and device for impedance identification of new energy generating units under all operating conditions. Background Technology

[0002] With the widespread integration of new energy generating units and the increasing power electronics level of power systems, broadband oscillations are easily induced, affecting the safe and stable operation of power systems. Therefore, oscillation research is necessary. Specifically, traditional stability analysis methods are based on mechanistic models, requiring detailed structural and parameter information of the power system and often relying on certain physical assumptions and simplifications. However, the control of new energy generating units is a black-box or gray-box model, making traditional analysis methods highly limited in practical applications. Impedance methods measure external characteristics at the ports, and due to their convenience, they are currently commonly used for oscillation research. However, frequency sweep impedance measurement is not only time-consuming and costly, but it can only obtain impedance at specific operating conditions, making it difficult to guarantee model completeness. To meet engineering needs, data-driven modeling methods based on artificial intelligence are gradually gaining attention, currently mainly including the following two approaches: The first approach involves constructing a white-box model of the new energy unit, deriving the transfer function using typical control theory, and then fitting key control parameters based on measured impedance characteristics. However, this approach involves significant human intervention and has limited accuracy. The second approach uses neural networks to identify impedance under all operating conditions, based on electromagnetic simulation results constructed from the mechanism. This approach is somewhat simplified and shows relatively good consistency under different operating conditions. However, in practical applications, it has been found that the control of new energy units in the 40-60Hz frequency band is complex, with the amplitude and phase of impedance changing drastically with operating conditions. The impedance identification accuracy of existing solutions in this frequency band needs improvement. Furthermore, existing solutions typically set the structural parameters of the neural network based on experience, which carries the risk of overfitting and underfitting.

[0003] Therefore, how to provide a more effective impedance identification scheme for new energy units under all operating conditions is an urgent problem to be solved. Summary of the Invention

[0004] In view of this, the present invention provides a method and device for impedance identification of new energy units under all operating conditions, which can realize impedance identification under any operating condition and at any frequency point by relying on a trained neural network.

[0005] To address the aforementioned technical problems, this application provides a method for impedance identification of new energy generating units under all operating conditions, including: Obtain electrical parameter information and frequency of attention for new energy units under the operating conditions to be identified; The electrical parameter information and the frequency of interest are used as inputs to a pre-trained neural network to obtain the positive sequence impedance corresponding to the frequency of interest under the operating condition to be identified. The pre-training step of the neural network includes: An electromagnetic simulation model corresponding to the new energy unit is constructed so as to determine the basic dataset based on the electromagnetic simulation model, and then the basic dataset is divided into a basic training set and a test set. Based on a preset sample augmentation strategy, data augmentation processing is performed on the samples corresponding to the target frequency band under the basic training set to obtain augmented samples, thereby determining the augmented training set, which includes the basic training set and the augmented samples. The improved Grey Wolf algorithm is used to optimize the structural parameters of the neural network to determine the optimal structural parameters of the neural network, so as to construct a neural network corresponding to the optimal structural parameters, and train it based on the enhanced training set until a preset training termination condition is reached to obtain a trained neural network.

[0006] Furthermore, the underlying dataset includes multiple sets of input-output samples; After determining the basic dataset based on the electromagnetic simulation model, the following is also included: For the input data under each group of samples, the physical reference data corresponding to the input data is used to perform per-unit processing to obtain the input data mapped to the first interval; For the output data under each group of samples, the output data is sequentially subjected to sign-preserving logarithmic transformation and linear mapping to obtain output data mapped to the second interval.

[0007] Furthermore, based on a preset sample augmentation strategy, data augmentation processing is performed on the samples corresponding to the target frequency band in the basic training set to obtain augmented samples, including: For target samples whose frequencies are within the target interest frequency band under the basic training set, the frequencies in the output data and input data of the target samples are kept unchanged, and the electrical parameter information in the input data is subjected to at least one noise enhancement process to obtain at least one enhanced sample.

[0008] Furthermore, the improved gray wolf algorithm is used to optimize the structural parameters of the neural network to determine the optimal structural parameters of the neural network, including: S11: Configure the initial parameters of the improved Grey Wolf algorithm; S12: Initialize the population based on the preset initialization strategy and the initial parameters to generate an initial population; S13: Calculate the fitness value of each gray wolf individual in the initial population, and determine Alpha wolf, Beta wolf and Delta wolf based on the fitness value; S14: Based on the preset position update strategy, the position information of the Alpha wolf, the Beta wolf and the Delta wolf are used as guiding information to update the position information of each gray wolf individual in the current iteration number, so as to obtain the corresponding candidate gray wolf individuals. S15: Calculate the fitness value of each of the candidate gray wolves, so as to determine the updated Alpha wolf, Beta wolf and Delta wolf based on the calculated fitness values, the fitness values ​​of the Alpha wolf, Beta wolf and Delta wolf before the update and the preset alpha wolf position update strategy. S16: The updated Alpha, Beta, and Delta wolves are selected as the first three gray wolf individuals in the new generation population. After removing the candidate gray wolf individuals that have been selected as the updated alpha wolf from all candidate gray wolf individuals, Ng-3 candidate gray wolf individuals are selected as fillers based on the fitness values ​​of the remaining candidate gray wolf individuals to obtain the new generation population; Ng is the population size set in the initial parameters. S17: Determine whether the preset iteration stop condition has been met; if yes, determine the position information corresponding to the current Alpha wolf as the optimal structural parameters of the neural network; if no, set the current iteration number = current iteration number + 1, take the new generation population as the current iteration population and return to step S14.

[0009] Further, step S12 includes: Based on the Tent mapping, the location information of the j-th gray wolf individual in the initial population is determined as follows:

[0010] This represents the location information of the j-th individual gray wolf. The lower limit for the search of structural parameters set under the initial parameters. This is the upper limit for the search of the structural parameters. This represents the j-th chaotic parameter in the chaotic sequence. Based on the first preset relation, the first preset relation is:

[0011] in, Let j be the (j-1)th chaotic parameter in the chaotic sequence. u This indicates the preset boundary parameter.

[0012] Further, step S14 includes: Based on the second preset relation, the nonlinear convergence factor for the current iteration number is determined. The second preset relation is:

[0013] in, This represents the nonlinear convergence factor. t Indicates the current iteration number. T This indicates the maximum number of iterations set under the initial parameters; For each individual gray wolf at the current iteration number, the position information of the individual gray wolf is updated based on the nonlinear convergence factor, the position information of the Alpha wolf, the position information of the Beta wolf, and the position information of the Delta wolf, so as to obtain the corresponding candidate gray wolf individuals; The location information of the candidate gray wolf individuals is sequentially discretized and rounded, and then subjected to boundary protection processing to obtain updated candidate gray wolf individuals and proceed to step S15.

[0014] Further, step S15 includes: S151: Calculate the updated fitness value of each of the candidate gray wolf individuals; S152: Incorporate all candidate gray wolf individuals into the same comparison pool as the Alpha wolf, Beta wolf, and Delta wolf mentioned before the update; S153: Based on the fitness values ​​of all gray wolf individuals in the comparison pool, perform a unified sorting, and select the gray wolf individual with the best fitness as the new Alpha wolf, the second best gray wolf individual as the new Beta wolf, and the third best gray wolf individual as the new Delta wolf based on the sorting results.

[0015] Furthermore, following step S153, the following steps are also included: Generate a random number within a third interval, and determine whether the random number is less than a preset threshold; If the value is less than the preset threshold, the new Alpha wolf undergoes Levy flight mutation processing to obtain the mutated Alpha wolf. Calculate the fitness value of the mutated Alpha Wolf, and if it is determined that the fitness value of the mutated Alpha Wolf is better than that of the new Alpha Wolf, replace the new Alpha Wolf with the mutated Alpha Wolf.

[0016] Furthermore, the new Alpha wolf undergoes Levy flight mutation processing to obtain the mutated Alpha wolf, including: Based on the third preset relation and the location information of the new Alpha Wolf, the location information of the mutated Alpha Wolf is determined; the third preset relation is:

[0017] in, The lower limit for the search of structural parameters set under the initial parameters. This is the upper limit for the search of the structural parameters. This indicates element-wise multiplication. The location information of the new Alpha wolf. The location information of the mutated Alpha wolf; step Denotes the random step size that follows a Levy distribution and step Based on the fourth preset relation, the fourth preset relation is:

[0018] in, w The first random variable and w Follows a mean of 0 and a standard deviation of Normal distribution, standard deviation Determined based on the sixth pre-defined relation; v The second random variable and v It follows a standard normal distribution; the sixth preset relation is:

[0019] in, The default shape parameter of the Levy distribution is Γ, where Γ is the gamma function. right Discretization and rounding processes, as well as out-of-bounds protection processes, are performed sequentially to obtain the position information of the mutated Alpha wolf.

[0020] To address the aforementioned technical problems, this invention also provides a full-condition impedance identification device for new energy generating units, comprising: Memory, used to store computer programs; A processor is used to execute the computer program to implement the steps of the impedance identification method for all operating conditions of new energy units as described above.

[0021] The beneficial effects of this invention are as follows: This application provides a method and apparatus for impedance identification of new energy power generation units under all operating conditions. The scheme includes acquiring electrical parameter information and the frequency of interest of the new energy power generation unit under the operating condition to be identified; inputting the electrical parameter information and the frequency of interest as input terms into a pre-trained neural network to obtain the positive-sequence impedance corresponding to the frequency of interest under the operating condition to be identified; wherein the pre-training step of the neural network includes: constructing an electromagnetic simulation model corresponding to the new energy power generation unit to determine a basic dataset based on the electromagnetic simulation model, and then dividing the basic dataset into a basic training set and a test set; performing data augmentation processing on samples corresponding to the target frequency band in the basic training set based on a preset sample augmentation strategy to obtain augmented samples, and then determining an augmented training set, which includes the basic training set and augmented samples; using an improved gray wolf algorithm to optimize the structural parameters of the neural network to determine the optimal structural parameters of the neural network, so as to construct a neural network corresponding to the optimal structural parameters, and training it based on the augmented training set until a preset training termination condition is reached to obtain a trained neural network. As can be seen, this application relies on a trained neural network to achieve impedance identification under any operating condition and at any frequency point, and can reliably and accurately determine the positive sequence impedance. Furthermore, the determination of the enhanced training set can guide the neural network to focus on learning the feature information of the target frequency band, which is conducive to improving the impedance identification accuracy of the trained neural network in the target frequency band. At the same time, the improved gray wolf algorithm is used to adaptively optimize the structural parameters of the neural network to determine the optimal structural parameters, which helps to effectively alleviate the overfitting and underfitting of the impedance identification results, and ensures that the trained neural network can accurately achieve impedance identification under all operating conditions of new energy units.

[0022] The above description is only an overview of the technical solution of this application. In order to better understand the technical means of this application, it can be implemented according to the contents of the specification. In order to make the above and other objects, features and advantages of this application more obvious and understandable, the following are specific embodiments of this application. Attached Figure Description

[0023] The accompanying drawings, which are included to provide a further understanding of this application and form part of this application, illustrate exemplary embodiments and are used to explain this application, but do not constitute an undue limitation of this application. In the drawings: Figure 1 A flowchart of a method for impedance identification of new energy generating units under all operating conditions provided by the present invention; Figure 2 This is a schematic diagram of the structure of a new energy unit impedance identification device under all operating conditions provided by the present invention. Detailed Implementation

[0024] The core of this invention is to provide a method and device for impedance identification of new energy generating units under all operating conditions. Based on a trained neural network, impedance identification can be achieved under any operating condition and at any frequency point.

[0025] The technical solutions of the embodiments of this application will be clearly described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of this application. All other embodiments obtained by those skilled in the art based on the embodiments of this application are within the scope of protection of this application.

[0026] The terms "first," "second," etc., used in this application are used to distinguish similar objects and not to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that embodiments of this application can be implemented in orders other than those illustrated or described herein, and the objects distinguished by "first," "second," etc., are generally of the same class, without limiting the number of objects; for example, a first object can be one or more. Furthermore, in this application, "and / or" indicates at least one of the connected objects, and the character " / " generally indicates that the preceding and following objects have an "or" relationship.

[0027] Please refer to Figure 1 , Figure 1 The flowchart is a method for impedance identification of new energy units under all operating conditions provided by the present invention.

[0028] The impedance identification method for this new energy unit under all operating conditions includes: S21: Obtain electrical parameter information and frequency of attention for new energy units under the operating conditions to be identified; S22: Input the electrical parameter information and the frequency of interest as input terms into the pre-trained neural network to obtain the positive sequence impedance corresponding to the frequency of interest under the operating condition to be identified. The pre-training steps for the neural network include: An electromagnetic simulation model corresponding to the new energy unit is constructed so that the basic dataset can be determined based on the electromagnetic simulation model, and then the basic dataset can be divided into a basic training set and a test set. Based on a pre-defined sample augmentation strategy, data augmentation processing is performed on samples in the basic training set that correspond to the target frequency band of interest to obtain augmented samples, thereby determining the augmented training set, which includes the basic training set and the augmented samples. An improved gray wolf algorithm is used to optimize the structural parameters of a neural network to determine the optimal structural parameters. This allows for the construction of a neural network corresponding to the optimal structural parameters, which is then trained on an enhanced training set until a preset training termination condition is met, resulting in a fully trained neural network.

[0029] Specifically, the electrical parameter information under the operating condition to be identified can include voltage information and current information. The voltage information can specifically be the d-axis voltage and q-axis voltage of the AC port of the new energy unit, that is, the voltage obtained by transforming the three-phase AC voltage in the three-phase coordinate system to the two-phase rotating dq coordinate system. The current information can specifically be the d-axis current and q-axis current of the AC port of the new energy unit, that is, the current obtained by transforming the three-phase AC current in the three-phase coordinate system to the two-phase rotating dq coordinate system. The focus frequency can be understood as any frequency point to be identified within the focus frequency band corresponding to the current electrical parameter information.

[0030] Based on the pre-trained neural network, the positive sequence impedance corresponding to the frequency of interest under the operating condition to be identified can be determined in step S22. It can be understood that the positive sequence impedance is in complex form, containing a real part and an imaginary part; where the real part corresponds to the equivalent resistance and the imaginary part corresponds to the equivalent reactance.

[0031] The neural network here can specifically be a BP neural network (Back Propagation Neural Network). The input of the neural network is set to 5 dimensions, including the d-axis voltage, q-axis voltage, d-axis current, q-axis current and frequency of the new energy unit. The output of the neural network is set to 2 dimensions, including the real part and imaginary part of the positive sequence impedance of the new energy unit at the frequency corresponding to the input, that is, the equivalent resistance and equivalent reactance.

[0032] The specific steps for constructing an electromagnetic simulation model and determining the basic dataset accordingly can include: conducting hardware-in-the-loop simulation tests based on the controller provided by the new energy manufacturer, selecting an active power output level covering from minimum output to full power, so that the constructed electromagnetic simulation model can be applied to the entire steady-state operating range of the new energy unit; measuring and recording the d-axis voltage, q-axis voltage, d-axis current, and q-axis current at the AC port of the new energy unit at different steady-state operating points, and measuring the positive sequence impedance of the new energy unit at different frequency values ​​within the frequency band of interest at that operating point; that is, the four-dimensional parameters of d-axis voltage, q-axis voltage, d-axis current, and q-axis current define a steady-state operating point, and tests are conducted at multiple frequency values ​​within the frequency band of interest corresponding to that operating point to obtain the positive sequence impedance corresponding to each frequency. The positive sequence impedance corresponding to different frequencies at the same steady-state operating point is different. Based on the above test results, a basic dataset was determined. This dataset includes multiple sets of input-output samples. Each set includes 5-dimensional input data and 2-dimensional output data. The 5-dimensional input data includes d-axis voltage, q-axis voltage, d-axis current, q-axis current, and frequency. The 2-dimensional output data includes the real and imaginary parts of the positive-sequence impedance of the new energy unit at its steady-state operating point and frequency, i.e., the equivalent resistance and equivalent reactance. The specific implementation steps for constructing the electromagnetic simulation model have been detailed in relevant technical documents and will not be repeated here.

[0033] The basic dataset can be divided into a basic training set and a test set. Specific division methods include, but are not limited to, proportional division, that is, 80% of the samples in the basic dataset are used as the basic training set and 20% of the samples are used as the test set. No special limitation is made here.

[0034] Furthermore, the target frequency band refers to the frequency band that needs to be focused on, which in this application can be the 40-60Hz band. Of course, depending on the actual application requirements, the target frequency band can also be set to other frequency bands that need to be focused on, and no special limitation is made here. Based on the preset sample augmentation strategy, data augmentation processing can be performed on the samples corresponding to the target frequency band to obtain augmented samples. These augmented samples can be added to the basic training set to form an augmented training set.

[0035] An improved Grey Wolf algorithm is used to adaptively optimize the structural parameters of a neural network to determine its optimal parameters. For example, this corresponds to a backpropagation (BP) neural network. The structural parameters include the number of layers and the number of nodes (i.e., the number of neurons) in each layer. The optimal structural parameters refer to the best number of layers and the best number of nodes in each layer. After determining the optimal structural parameters based on the improved Grey Wolf algorithm, a neural network with these optimal parameters is constructed. Specific steps may include configuring the transfer function and initializing the weights and biases of the neural network. Subsequently, the constructed neural network is trained based on the augmented training set. The neural network learns the mapping features of the augmented dataset and adjusts the weights and biases of each neuron until a preset training termination condition is met, resulting in a trained neural network. It is understood that the preset training termination condition can be flexibly set according to actual application needs and is not specifically limited here.

[0036] In summary, this application provides a method for impedance identification of new energy power units under all operating conditions. This method relies on a trained neural network to achieve impedance identification under any operating condition and at any frequency point, and can reliably and accurately determine the positive sequence impedance. Furthermore, the determination of the enhanced training set can guide the neural network to focus on learning the feature information of the target frequency band, which is conducive to improving the impedance identification accuracy of the trained neural network in the target frequency band. At the same time, the improved gray wolf algorithm is used to adaptively optimize the structural parameters of the neural network to determine the optimal structural parameters, which helps to effectively alleviate the overfitting and underfitting of the impedance identification results, and ensures that the trained neural network can accurately achieve impedance identification under all operating conditions of new energy power units.

[0037] Based on the above embodiments: In some embodiments, the underlying dataset includes multiple sets of input-output samples; After determining the basic dataset based on the electromagnetic simulation model, the following is also included: For the input data under each group of samples, the physical reference data corresponding to the input data is used to perform per-unit processing to obtain the input data mapped to the first interval; For the output data of each group of samples, the output data is subjected to sign-preserving logarithmic transformation and linear mapping in sequence to obtain the output data mapped to the second interval.

[0038] Specifically, the basic dataset includes multiple sets of input-output samples. The input data for each set of samples includes electrical parameter information (i.e., d-axis voltage, q-axis voltage, d-axis current, and q-axis current) and frequency. The output data includes the equivalent resistance and equivalent reactance corresponding to the input data. In order to improve the training efficiency of the neural network, the data in the basic dataset can be preprocessed in the manner described above, so as to perform subsequent steps based on the preprocessed basic dataset.

[0039] To elaborate, for the input data, it is normalized using its corresponding physical reference data to obtain the input data mapped to the first interval, which refers to the (0,1) interval. For example, for the d-axis voltage and q-axis voltage, the rated voltage of the new energy unit can be used as the physical reference data to complete the above normalization process. For the d-axis current and q-axis current, the rated current of the new energy unit can be used as the physical reference data to complete the above normalization process. For the frequency, a reference frequency can be selected as the physical reference data to complete the above normalization process.

[0040] For the output data, a sign-preserving logarithmic transformation and a linear mapping are performed sequentially to obtain output data mapped to the second interval, where the second interval refers to the (-1, 1) interval. More specifically, the output data for the i-th sample group includes equivalent resistance and equivalent reactance, where 1 ≤ i ≤ the total number of samples and i is an integer. The steps of performing the sign-preserving logarithmic transformation and linear mapping on the output data sequentially include: Using the seventh preset relation, determine the equivalent resistance for the i-th sample group. The first intermediate value after sign-preserving logarithmic transformation Equivalent reactance The second intermediate value after sign-preserving logarithmic transformation The seventh preset relation is:

[0041] Using the eighth preset relation, determine the first intermediate value respectively. The equivalent resistance mapped to the (-1,1) interval after linear mapping and the second intermediate value The equivalent reactance after linear mapping to the (-1,1) interval The eighth preset relation is:

[0042] in, Indicates all The minimum value in, Indicates all The maximum value in; Indicates all The minimum value in, Indicates all The maximum value in.

[0043] As can be seen, in order to address the problem of large impedance amplitude ranges in the output data, this application adopts the above processing, which can improve the sensitivity of the neural network to magnitude differences through linear mapping.

[0044] In some embodiments, based on a preset sample augmentation strategy, data augmentation processing is performed on samples in the basic training set corresponding to the target frequency band of interest to obtain augmented samples, including: For target samples whose frequencies are within the target interest frequency band under the basic training set, keep the frequencies in the output data and input data of the target samples unchanged, and perform at least one noise enhancement process on the electrical parameter information in the input data to obtain at least one enhanced sample.

[0045] Specifically, the target samples are those whose frequencies fall within the target interest frequency band (the target interest frequency band can be the 40~60Hz band), and the number of target samples is greater than 1. You can select all the samples in the basic training set whose frequencies fall within the target interest frequency band, or you can select a portion of the samples. There are no special restrictions here.

[0046] Keeping the output data of the target sample unchanged and the frequency term in the input data unchanged, perform at least one noise enhancement process on the electrical parameter information in the input data to obtain at least one enhanced sample. Specifically, the electrical parameter information includes d-axis voltage, q-axis voltage, d-axis current, and q-axis current. Gaussian white noise is injected into these four dimensions to simulate various random measurement disturbances and small fluctuations in operating conditions that actually exist in the industrial field, thereby obtaining enhanced samples. This noise injection operation is performed at least once to obtain at least one enhanced sample, so that these enhanced samples can be added to the basic training set to obtain an enhanced training set.

[0047] It is evident that, considering the drastic fluctuations in impedance amplitude and phase data of new energy power units in the 40-60Hz frequency band, the above settings can expand the target focus frequency band, that is, the proportion of samples in the 40-60Hz frequency band in the entire enhanced training set. During subsequent training, the error in this range will contribute more to the total loss, achieving the effect of focusing on learning the key target focus frequency band (i.e., the 40-60Hz frequency band), which is beneficial to enhancing the robustness of the neural network.

[0048] In some embodiments, the improved gray wolf algorithm is used to optimize the structural parameters of the neural network to determine the optimal structural parameters of the neural network, including: S11: Configure the initial parameters of the improved Grey Wolf algorithm; S12: Initialize the population based on the preset initialization strategy and initial parameters to generate the initial population; S13: Calculate the fitness value of each gray wolf individual in the initial population, and determine the Alpha wolf, Beta wolf and Delta wolf based on the fitness value; S14: Based on the preset position update strategy, the position information of Alpha wolf, Beta wolf and Delta wolf are used as guiding information to update the position information of each gray wolf individual in the current iteration number, so as to obtain the corresponding candidate gray wolf individuals. S15: Calculate the fitness value of each candidate gray wolf individual, so as to determine the updated Alpha wolf, Beta wolf and Delta wolf based on the calculated fitness values, the fitness values ​​of Alpha wolf, Beta wolf and Delta wolf before the update and the preset alpha wolf position update strategy. S16: Select the updated Alpha, Beta, and Delta wolves as the first three gray wolf individuals in the new generation population. After removing the candidate gray wolf individuals that have been selected as the new alpha wolf from all candidate gray wolf individuals, select Ng-3 candidate gray wolf individuals based on the fitness values ​​of the remaining candidate gray wolf individuals to fill the gap and obtain the new generation population; Ng is the population size set in the initial parameters. S17: Determine whether the preset iteration stop condition has been met; if yes, determine the position information corresponding to the current Alpha wolf as the optimal structural parameters of the neural network; if no, set the current iteration number = current iteration number + 1, take the new generation population as the current iteration population and return to step S14.

[0049] Specifically, the initial parameters configured in step S11 may include the population size (i.e., the total number of gray wolves included in the population), the maximum number of iterations, and the boundary information of the structural parameters of the neural network. For example, for a BP neural network, the boundary information may include the upper and lower limits of the number of layers that can be set in the neural network and the upper and lower limits of the number of nodes in each layer.

[0050] Based on step S12, population initialization can generate an initial population, which includes Ng gray wolf individuals, where Ng is the population size set in the initial parameters and is an integer greater than 3. The gray wolf individuals are used to represent the setting of the structural parameters of the neural network, that is, the position information of a single gray wolf individual corresponds to a setting scheme of the structural parameters of the neural network.

[0051] Specifically, the optimization objective of this application is to minimize the impedance prediction error under the test set. Therefore, the specific method for calculating the fitness value of a gray wolf individual in this application can be as follows: based on the location information of the gray wolf individual, using this location information as the structural parameters of the neural network, a corresponding neural network is constructed. The input data under the test set is input into the neural network, and the impedance prediction error under the test set is calculated based on the network's predicted output data and the actual output data recorded in the test set, so that this error is used as the fitness value of the gray wolf individual. It can be seen that, corresponding to this optimization objective, the smaller the fitness value, the better the structural parameters of the neural network defined for the corresponding gray wolf individual.

[0052] In step S13, the fitness values ​​of each individual gray wolf in the initial population are determined according to the above-described fitness value calculation method. The gray wolves are sorted in ascending order of fitness value, with the first gray wolf being the Alpha wolf, the second gray wolf being the Beta wolf, and the third gray wolf being the Delta wolf. That is, the fitness value of the Alpha wolf (the best) is less than the fitness value of the Beta wolf (the second best) and the fitness value of the Delta wolf (the third best) and the fitness value of any remaining gray wolf.

[0053] Based on step S14, the positions of individual gray wolves are updated. That is, the Ng gray wolf individuals before the update in the current iteration can be updated to obtain Ng candidate gray wolf individuals through step S14. Based on step S15, the Alpha wolf, Beta wolf, and Delta wolf are updated. It should be noted that the updated Alpha wolf, Beta wolf, and Delta wolf may have the following possible scenarios: they may all be different from the Alpha wolf, Beta wolf, and Delta wolf before the update; they may also be partially consistent with the Alpha wolf, Beta wolf, and Delta wolf before the update; or, since no better solution has been found for the candidate gray wolf individuals in this round, the three alpha wolves will not be replaced, that is, they will remain consistent with the ones before the update. The specific result is determined by the comparison of individual fitness during the iteration process.

[0054] In step S16, the updated alpha wolf refers to the Alpha wolf, Beta wolf, and Delta wolf. For example, assuming Ng=10, the updated Alpha wolf is still the same as the original Alpha wolf, and the updated Beta and Delta wolves are all taken from the candidate gray wolf individuals in this round, then Ng-3 candidate gray wolf individuals are selected from the remaining Ng-2 candidate gray wolf individuals as fillers. Specifically, Ng is selected based on the order of fitness values. Three candidate gray wolf individuals, together with the three newly appointed alpha wolves, constitute the new generation of the population, which remains at a size of Ng.

[0055] When the preset iteration stopping condition is met, the position information corresponding to the current Alpha wolf is the optimal structural parameters of the neural network, and the fitness value corresponding to the current Alpha wolf is the minimum impedance prediction error under the test set. If not, the current iteration number is set to the current iteration number + 1, the new generation population is used as the current iteration population, and the process returns to step S14 to continue iterating for adaptive optimization. It can be understood that determining whether the preset iteration stopping condition has been met can be equivalent to determining whether the current iteration number has reached the maximum iteration number.

[0056] As can be seen, this application can achieve adaptive optimization of neural network structure parameters through the above settings, which helps to effectively alleviate overfitting and underfitting of impedance identification results.

[0057] In some embodiments, step S12 includes: Based on the Tent mapping, the location information of the j-th gray wolf individual in the initial population is determined as follows:

[0058] This represents the location information of the j-th individual gray wolf. The lower bound for the search of the structure parameters set under the initial parameters. This represents the upper limit of the search for structural parameters. This represents the j-th chaotic parameter in the chaotic sequence. Based on the first preset relation, the first preset relation is:

[0059] in, Let j be the (j-1)th chaotic parameter in the chaotic sequence. u This indicates the preset boundary parameter.

[0060] Specifically, a chaotic sequence is generated based on the Tent mapping. It should be noted that the first term of the chaotic sequence, i.e., the first chaotic parameter x1, can be randomly selected within the interval (0,1). Starting with the second chaotic parameter, it can be determined using a first preset relational formula, thus obtaining the chaotic sequence. The j-th chaotic parameter in the chaotic sequence is assigned to the j-th gray wolf individual for position calculation, j=1,…,Ng, where j is an integer, so that… Upper limit mapped to the physical search space and lower limit Within.

[0061] Furthermore, by way of example, u It can be set to 0.75.

[0062] In some embodiments, step S14 includes: Based on the second preset relation, the nonlinear convergence factor for the current iteration number is determined. The second preset relation is:

[0063] in, Represents the nonlinear convergence factor. t Indicates the current iteration number. T This indicates the maximum number of iterations set under the initial parameters; For each individual gray wolf at the current iteration number, the position information of the individual gray wolf is updated based on the nonlinear convergence factor, the position information of Alpha wolf, the position information of Beta wolf and Delta wolf, so as to obtain the corresponding candidate gray wolf individuals. The location information of the candidate gray wolf individuals is sequentially discretized and rounded, and then subjected to boundary protection processing to obtain the updated candidate gray wolf individuals and proceed to step S15.

[0064] Specifically, the nonlinear convergence factor based on sine function decay is determined through the second preset relation. a Therefore, it is evident that in this application, regarding a The update method enables a Taking larger values ​​in the early stages of iteration is beneficial for the algorithm's global search in the early stages; taking smaller values ​​in the later stages of iteration is beneficial for the algorithm to explore local areas, thus improving the algorithm's convergence speed.

[0065] More specifically, the steps for updating the location information of individual gray wolves based on the nonlinear convergence factor, the location information of Alpha wolves, the location information of Beta wolves, and the location information of Delta wolves to obtain candidate gray wolf individuals include: For each individual gray wolf in the current iteration, the distance between the gray wolf and the Alpha wolf, Beta wolf, and Delta wolf is determined based on the ninth preset relation; the ninth preset relation is:

[0066] in, This indicates the position information of the individual gray wolf at the current iteration number. This refers to the location information of the Alpha wolf. This refers to the location information of the Beta wolf. This refers to the location information of the Delta wolf. express and The distance between them express and The distance between them express and The distance between them; C 1, C 2, C 3] is the first collaborative coefficient vector and is determined based on the tenth preset relation, which is:

[0067] in, r 21 , r 22 , r 23 All are random numbers within the interval (0,1); The gray wolf individuals are updated based on the eleventh preset relation to obtain candidate gray wolf individuals; the eleventh preset relation is:

[0068] in, Indicating targeting The obtained candidate gray wolf individuals, [ X 1, X 2, X 3] can be understood as an intermediate vector, [ A 1, A 2, A 3] is the second cooperative coefficient vector and is determined based on the twelfth preset relation, which is:

[0069] in, r 11 ,r 12 , r 13 All are random numbers within the interval (0,1).

[0070] Furthermore, considering that the position information of candidate gray wolves updated in the above manner is usually a continuous floating-point number, while the structural parameters of the neural network are integers, it is also necessary to further refine the information. Discretization and rounding, followed by boundary protection, are performed sequentially to obtain updated candidate gray wolf individuals, and then the process proceeds to step S15. Discretization and rounding yields integer structural parameters; boundary protection prevents integer structural parameters from exceeding the search upper and lower limits, eliminates invalid solutions, and ensures the stability and feasibility of the optimization process.

[0071] In some embodiments, step S15 includes: S151: Calculate the fitness value of each candidate gray wolf individual after the update; S152: Include all candidate gray wolf individuals in the same comparison pool as the Alpha, Beta, and Delta wolves before the update; S153: Based on the fitness values ​​of all gray wolf individuals in the comparison pool, perform a unified sorting, and select the gray wolf individual with the best fitness as the new Alpha wolf, the second best gray wolf individual as the new Beta wolf, and the third best gray wolf individual as the new Delta wolf based on the sorting results.

[0072] In this embodiment, the Alpha Wolf, Beta Wolf, and Delta Wolf are updated through the above-mentioned sequential inheritance mechanism, which helps to prevent the loss of historical optimal solutions while not discarding any potential high-quality solutions found in the current iteration.

[0073] Specifically, corresponding to the optimization objective in this application, in step S153, the fitness values ​​of all gray wolf individuals in the comparison pool can be uniformly sorted in ascending order. The gray wolf individual ranked first is taken as the new Alpha wolf, the gray wolf individual ranked second is taken as the new Beta wolf, and the gray wolf individual ranked third is taken as the new Delta wolf.

[0074] In some embodiments, after step S153, the method further includes: Generate a random number within the third interval and determine whether the random number is less than a preset threshold; If the value is less than a preset threshold, the new Alpha wolf will undergo Levy flight mutation processing to obtain the mutated Alpha wolf. Calculate the fitness value of the mutated Alpha Wolf, and if the fitness value of the mutated Alpha Wolf is determined to be better than that of the new Alpha Wolf, replace the new Alpha Wolf with the mutated Alpha Wolf.

[0075] In this embodiment, in order to prevent the alpha wolf (i.e., the alpha wolf) from getting stuck in a local optimum, the above steps are also set; specifically, the preset threshold is also located in the third interval. For example, the third interval can be the interval (0,1), and the preset threshold can be 0.2.

[0076] If the random number is not less than a preset threshold, there is no need to perform Levy flight mutation on the new Alpha wolf; if the random number is less than the preset threshold, Levy flight mutation is required to obtain a mutated Alpha wolf. Based on the position information of the mutated Alpha wolf, its corresponding fitness value is calculated. If it is determined that the fitness value of the mutated Alpha wolf is better than the fitness value of the new Alpha wolf, the mutated Alpha wolf replaces the new Alpha wolf; if it is determined that the fitness value of the mutated Alpha wolf is not better than the fitness value of the new Alpha wolf, there is no need to replace it. For example, corresponding to the optimization objective in this application, which is to minimize the impedance prediction error under the test set, if the fitness value of the mutated Alpha wolf is less than the fitness value of the new Alpha wolf, the mutated Alpha wolf replaces the new Alpha wolf; otherwise, there is no need to replace it.

[0077] More specifically, in some embodiments, the new Alpha wolf undergoes Levy flight mutation to obtain the mutated Alpha wolf, including: Based on the third preset relation and the new Alpha Wolf's location information, the location information of the mutated Alpha Wolf is determined; the third preset relation is:

[0078] in, The lower bound for the search of the structure parameters set under the initial parameters. This represents the upper limit of the search for structural parameters. This indicates element-wise multiplication. Location information for the new Alpha wolf. The location information of the mutated Alpha wolf; step Denotes the random step size that follows a Levy distribution and step Based on the fourth preset relation, the fourth preset relation is:

[0079] in,w The first random variable and w Follows a mean of 0 and a standard deviation of Normal distribution, standard deviation Determined based on the sixth pre-defined relation; v The second random variable and v It follows a standard normal distribution; the sixth presupposed relation is:

[0080] in, The default shape parameter of the Levy distribution is Γ, where Γ is the gamma function. right Discretization and rounding processes, as well as out-of-bounds protection processes, are performed sequentially to obtain the position information of the mutated Alpha wolf.

[0081] Specifically, considering the relationship determined by the third pre-set relation... Typically, parameters are consecutive floating-point numbers, while the structural parameters of a neural network are integers, therefore, it is necessary to... Discretization and rounding processes are performed sequentially, followed by out-of-bounds protection.

[0082] w Follows a mean of 0 and a standard deviation of The normal distribution (i.e. w ~N(0, 2 )), v Follows the standard normal distribution (i.e.) v ~N(0,1)), The value range is [0, 2], preferably, =1.5.

[0083] To verify the effectiveness of the neural network trained in this application, and to examine its training accuracy and generalization ability, tests were conducted using samples from the test set, and the mean absolute error was used for evaluation, as shown in the following formula:

[0084] in, This indicates that the test set is used for the first... l The predicted equivalent resistance of a single sample and the output of the neural network. This indicates that the test set is used for the first... l The predicted equivalent reactance of each sample and the output of the neural network. Indicates the first test set l The true equivalent resistance in each sample. Indicates the first test set l The true equivalent reactance in each sample; L The total number of samples in the test set. This represents the mean absolute error corresponding to the equivalent resistance. This represents the mean absolute error corresponding to the equivalent reactance.

[0085] In polar coordinates, the magnitude-phase error is evaluated as follows:

[0086] in, This indicates that the test set is used for the first... l Each sample, and the impedance magnitude calculated based on the output of the neural network. Indicates based on the test set, the first... l The impedance amplitude calculated from the output data of each sample; This indicates that the test set is used for the first... l Each sample, and the phase angle calculated based on the output of the neural network. Indicates based on the test set, the first... l The phase angle is calculated from the output data of each sample; it should also be noted that, for the amplitude, the impedance amplitude is first calculated in reverse using the equivalent resistance R of the real part and the equivalent reactance X of the imaginary part. Then, the amplitude error is obtained by using the average relative error assessment. For the phase, first calculate the phase angle using the four-quadrant arctangent function. A complex plane exponential operation mechanism is introduced and converted into angle values ​​to achieve smooth constraint of phase angle difference.

[0087] The new energy generator unit here can be a doubly fed wind turbine, with an impedance frequency that can vary from 1 to 350 Hz. The enhanced training set is {P1.0, P0.9, P0.8, P0.6, P0.5, P0.3, P0.2, P0.1}, and the test set is {P0.7, P0.4}. The control group uses a BP neural network with a fixed network structure [25,25,25,25,25], i.e., 5 hidden layers, with 25 nodes in each layer. After optimization using the improved Grey Wolf algorithm, the network structure parameters of the neural network are [9,5,6,15], i.e., 4 hidden layers, with fewer nodes in each layer than the control group, thus simplifying the overall network structure. The errors are shown in Table 1. The IGWO group refers to the group corresponding to the improved Grey Wolf algorithm. It can be seen that the accuracy and generalization ability of the IGWO group are significantly better than those of the control group, which proves that the proposed solution can significantly improve the impedance identification accuracy of new energy generator units under all operating conditions.

[0088] Table 1

[0089] Please refer to Figure 2 , Figure 2 This is a schematic diagram of the structure of a new energy unit impedance identification device under all operating conditions provided by the present invention.

[0090] The impedance identification device for the new energy unit under all operating conditions includes: Memory 31 is used to store computer programs; The processor 32 is used to execute computer programs to implement the steps of the impedance identification method for all operating conditions of new energy units as described above.

[0091] For a description of the impedance identification device for all operating conditions of new energy generating units provided in this application, please refer to the embodiments of the above-mentioned impedance identification method for all operating conditions of new energy generating units, which will not be repeated here.

[0092] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on its differences from other embodiments. Similar or identical parts between embodiments can be referred to interchangeably. For the apparatus disclosed in the embodiments, since it corresponds to the method disclosed in the embodiments, the description is relatively simple; relevant parts can be referred to in the method section. Relational terms such as "first" and "second" are used merely to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising one" does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0093] The above description of the disclosed embodiments enables those skilled in the art to make or use the invention. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the invention. Therefore, the invention is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A method for impedance identification of new energy generating units under all operating conditions, characterized in that, include: Obtain electrical parameter information and frequency of attention for new energy units under the operating conditions to be identified; The electrical parameter information and the frequency of interest are used as inputs to a pre-trained neural network to obtain the positive sequence impedance corresponding to the frequency of interest under the operating condition to be identified. The pre-training step of the neural network includes: An electromagnetic simulation model corresponding to the new energy unit is constructed so as to determine the basic dataset based on the electromagnetic simulation model, and then the basic dataset is divided into a basic training set and a test set. Based on a preset sample augmentation strategy, data augmentation processing is performed on the samples corresponding to the target frequency band under the basic training set to obtain augmented samples, thereby determining the augmented training set, which includes the basic training set and the augmented samples. The improved Grey Wolf algorithm is used to optimize the structural parameters of the neural network to determine the optimal structural parameters of the neural network, so as to construct a neural network corresponding to the optimal structural parameters, and train it based on the enhanced training set until a preset training termination condition is reached to obtain a trained neural network.

2. The impedance identification method for new energy generating units under all operating conditions as described in claim 1, characterized in that, The underlying dataset includes multiple sets of input-output samples; After determining the basic dataset based on the electromagnetic simulation model, the following is also included: For the input data under each group of samples, the physical reference data corresponding to the input data is used to perform per-unit processing to obtain the input data mapped to the first interval; For the output data under each group of samples, the output data is sequentially subjected to sign-preserving logarithmic transformation and linear mapping to obtain output data mapped to the second interval.

3. The impedance identification method for new energy generating units under all operating conditions as described in claim 2, characterized in that, Based on a preset sample augmentation strategy, data augmentation processing is performed on the samples in the basic training set corresponding to the target frequency band of interest to obtain augmented samples, including: For target samples whose frequencies are within the target interest frequency band under the basic training set, the frequencies in the output data and input data of the target samples are kept unchanged, and the electrical parameter information in the input data is subjected to at least one noise enhancement process to obtain at least one enhanced sample.

4. The method for impedance identification of new energy generating units under all operating conditions as described in any one of claims 1 to 3, characterized in that, The improved gray wolf algorithm is used to optimize the structural parameters of the neural network to determine the optimal structural parameters of the neural network, including: S11: Configure the initial parameters of the improved Grey Wolf algorithm; S12: Initialize the population based on the preset initialization strategy and the initial parameters to generate an initial population; S13: Calculate the fitness value of each gray wolf individual in the initial population, and determine Alpha wolf, Beta wolf and Delta wolf based on the fitness value; S14: Based on the preset position update strategy, the position information of the Alpha wolf, the Beta wolf and the Delta wolf are used as guiding information to update the position information of each gray wolf individual in the current iteration number, so as to obtain the corresponding candidate gray wolf individuals. S15: Calculate the fitness value of each of the candidate gray wolves, so as to determine the updated Alpha wolf, Beta wolf and Delta wolf based on the calculated fitness values, the fitness values ​​of the Alpha wolf, Beta wolf and Delta wolf before the update and the preset alpha wolf position update strategy. S16: The updated Alpha, Beta, and Delta wolves are selected as the first three gray wolf individuals in the new generation population. After removing the candidate gray wolf individuals that have been selected as the updated alpha wolf from all candidate gray wolf individuals, Ng-3 candidate gray wolf individuals are selected as fillers based on the fitness values ​​of the remaining candidate gray wolf individuals to obtain the new generation population; Ng is the population size set in the initial parameters. S17: Determine whether the preset iteration stop condition has been met; if yes, determine the position information corresponding to the current Alpha wolf as the optimal structural parameters of the neural network; if no, set the current iteration number = current iteration number + 1, take the new generation population as the current iteration population and return to step S14.

5. The impedance identification method for new energy generating units under all operating conditions as described in claim 4, characterized in that, Step S12 includes: Based on the Tent mapping, the location information of the j-th gray wolf individual in the initial population is determined as follows: This represents the location information of the j-th individual gray wolf. The lower limit for the search of structural parameters set under the initial parameters. This is the upper limit for the search of the structural parameters. This represents the j-th chaotic parameter in the chaotic sequence. Based on the first preset relation, the first preset relation is: in, Let j be the (j-1)th chaotic parameter in the chaotic sequence. u This indicates the preset boundary parameter.

6. The impedance identification method for new energy generating units under all operating conditions as described in claim 4, characterized in that, Step S14 includes: Based on the second preset relation, the nonlinear convergence factor for the current iteration number is determined. The second preset relation is: in, This represents the nonlinear convergence factor. t Indicates the current iteration number. T This indicates the maximum number of iterations set under the initial parameters; For each individual gray wolf at the current iteration number, the position information of the individual gray wolf is updated based on the nonlinear convergence factor, the position information of the Alpha wolf, the position information of the Beta wolf, and the position information of the Delta wolf, so as to obtain the corresponding candidate gray wolf individuals; The location information of the candidate gray wolf individuals is sequentially discretized and rounded, and then subjected to boundary protection processing to obtain updated candidate gray wolf individuals and proceed to step S15.

7. The impedance identification method for new energy generating units under all operating conditions as described in claim 6, characterized in that, Step S15 includes: S151: Calculate the updated fitness value of each of the candidate gray wolf individuals; S152: Incorporate all candidate gray wolf individuals into the same comparison pool as the Alpha wolf, Beta wolf, and Delta wolf mentioned before the update; S153: Based on the fitness values ​​of all gray wolf individuals in the comparison pool, perform a unified sorting, and select the gray wolf individual with the best fitness as the new Alpha wolf, the second best gray wolf individual as the new Beta wolf, and the third best gray wolf individual as the new Delta wolf based on the sorting results.

8. The impedance identification method for new energy generating units under all operating conditions as described in claim 7, characterized in that, Following step S153, the method further includes: Generate a random number within a third interval, and determine whether the random number is less than a preset threshold; If the value is less than the preset threshold, the new Alpha wolf undergoes Levy flight mutation processing to obtain the mutated Alpha wolf. Calculate the fitness value of the mutated Alpha Wolf, and if it is determined that the fitness value of the mutated Alpha Wolf is better than that of the new Alpha Wolf, replace the new Alpha Wolf with the mutated Alpha Wolf.

9. The impedance identification method for new energy generating units under all operating conditions as described in claim 8, characterized in that, The new Alpha wolf is subjected to Levy flight mutation to obtain the mutated Alpha wolf, including: Based on the third preset relation and the location information of the new Alpha Wolf, the location information of the mutated Alpha Wolf is determined; the third preset relation is: in, The lower limit for the search of structural parameters set under the initial parameters. This is the upper limit for the search of the structural parameters. This indicates element-wise multiplication. The location information of the new Alpha wolf. The location information of the mutated Alpha wolf; step Denotes the random step size that follows a Levy distribution and step Based on the fourth preset relation, the fourth preset relation is: in, w The first random variable and w Follows a mean of 0 and a standard deviation of Normal distribution, standard deviation Determined based on the sixth pre-defined relation; v The second random variable and v It follows a standard normal distribution; the sixth preset relation is: in, The default shape parameter of the Levy distribution is Γ, where Γ is the gamma function. right Discretization and rounding processes, as well as out-of-bounds protection processes, are performed sequentially to obtain the position information of the mutated Alpha wolf.

10. A full-condition impedance identification device for new energy generating units, characterized in that, include: Memory, used to store computer programs; A processor, configured to execute the computer program to implement the steps of the impedance identification method for all operating conditions of a new energy unit as described in any one of claims 1 to 9.

Images