New energy unit broadband impedance rapid measurement method and system based on transfer learning

Abstract

The invention discloses a new energy unit broadband impedance rapid measurement method and system based on transfer learning, and the method comprises the steps: an offline training stage: building a standard simulation model based on the electrical characteristics and topological structure of an actual new energy unit; performing all-condition broadband impedance scanning on the standard simulation model by adopting a single sinusoidal signal injection method to obtain an offline impedance data set; and training the pre-training model. An online measurement stage: constructing a to-be-trained model; measuring the impedance data of the actual new energy unit at the key frequency point on line by adopting a single sinusoidal signal injection method to obtain an online impedance data set; and finely adjusting the to-be-trained model by using a transfer learning technology to realize the rapid reconstruction of the broadband impedance characteristic. The method is combined with transfer learning, online measurement frequency points are reduced by 80%-90%, and meanwhile the high signal-to-noise ratio of a single sinusoidal signal is kept. The method solves the contradiction between precision and efficiency in broadband impedance measurement, and is especially suitable for the frequent fluctuation of the grid-connected working condition of a new energy unit.

Description

Technical Field

[0001] The present invention relates to the technical field of new energy power generation, and specifically, to a method and system for rapid measurement of broadband impedance of new energy generating units based on transfer learning. Background Art

[0002] Under the background of energy structure transformation and sustainable development, new energy power generation technologies represented by wind energy and solar energy have received unprecedented attention and development. The power system is developing towards the direction of high proportion of renewable energy access and high proportion of power electronic device applications, which has triggered a series of stability problems, seriously threatening the safe and stable operation of the power system and the efficient consumption of new energy. Therefore, realizing the stability analysis and online evaluation of new energy grid-connected systems has become an urgent demand in current engineering applications.

[0003] As one of the mainstream methods for stability analysis of new energy grid-connected systems, the core of the impedance analysis method lies in accurately obtaining the broadband impedance characteristics of new energy power generation equipment and the power grid, and then using impedance stability criteria to analyze the stability of the grid-connected system.

[0004] Impedance measurement is a method of obtaining the impedance characteristics of a measured system by injecting small-signal disturbances into the system. Measurement accuracy and efficiency are two important indicators for evaluating the performance of impedance measurement. Currently, impedance measurement methods are mainly divided into two categories: passive and active measurements. Passive measurement methods use the inherent harmonic or noise signals at the ports of the measured system to estimate the impedance characteristics without the need for an external disturbance source, with low cost but limited measurement accuracy. In contrast, active measurement methods inject self-designed disturbance signals into the measured system through external devices to obtain more accurate impedance characteristics, with higher measurement accuracy and a wider application range, so they have been widely studied and applied.

[0005] According to the type of disturbance signal, active measurement methods are mainly divided into single-sine signal injection method and broadband signal injection method.

[0006] The single-sine signal injection method injects disturbance signals of a single frequency at each frequency point, with the advantages of high signal-to-noise ratio and high measurement accuracy, but the full-frequency band measurement efficiency is low, making it difficult to meet the rapid measurement requirements under dynamic conditions. The broadband signal injection method injects disturbance signals containing multiple frequency components at one time, which can significantly improve the measurement efficiency. However, due to the high signal complexity and limited disturbance amplitude, the measurement accuracy decreases. Therefore, existing methods are difficult to achieve an effective balance between measurement accuracy and efficiency, and are difficult to meet the rapid measurement requirements of broadband impedance of new energy generating units under dynamic conditions. Summary of the Invention

[0007] Aiming at the defects in the prior art, the purpose of the present invention is to provide a method and system for rapid measurement of broadband impedance of new energy generating units based on transfer learning.

[0008] According to one aspect of the present invention, there is provided a method for rapidly measuring the broadband impedance of a new energy unit based on transfer learning, including:

[0009] Offline training stage:

[0010] Based on the electrical characteristics and topological structure of the actual new energy unit, construct a standard simulation model of the new energy unit;

[0011] Adopt the single sine signal injection method to perform full-condition broadband impedance scanning on the standard simulation model to obtain an offline impedance data set;

[0012] Based on the offline impedance data set, train a pre-trained model;

[0013] Online measurement stage:

[0014] Based on the pre-trained model, construct a model to be trained;

[0015] Adopt the single sine signal injection method to online measure the impedance data of the actual new energy unit at key frequency points to obtain an online impedance data set;

[0016] Based on the online impedance data set, use transfer learning technology to fine-tune the parameters of the model to be trained;

[0017] Use the fine-tuned model to be trained to measure the broadband impedance.

[0018] Preferably, the constructing a standard simulation model of the new energy unit based on the electrical characteristics and topological structure of the actual new energy unit includes:

[0019] Construct a standard simulation model based on the Simulink or PSCAD simulation platform;

[0020] Set the main circuit topology, electrical parameters and control strategy to be consistent with the actual new energy unit;

[0021] Set control parameters to ensure that the standard simulation model can operate stably within the full-condition range.

[0022] Preferably, the adopting the single sine signal injection method to perform full-condition broadband impedance scanning on the standard simulation model to obtain an offline impedance data set includes:

[0023] Determine the condition range, where the condition range includes the active power output and reactive power output ranges. The active power output range is 0 to 1 p.u., and the reactive power output range is from the maximum capacitive reactive power required by the grid for the unit to output to the maximum inductive reactive power;

[0024] Determine the frequency scanning range, where the frequency scanning range covers at least the frequency band below half of the switching frequency;

[0025] Set the active power resolution, reactive power resolution, and frequency sweep resolution to ensure that impedance data can be collected.

[0026] Under the conditions of the working condition range, frequency sweep range, and resolution, use the single sine signal injection method to collect the impedance data of the standard simulation model point by point to obtain an offline impedance data set.

[0027] Preferably, the pre-trained model is a BP neural network model.

[0028] Preferably, constructing a model to be trained based on the pre-trained model includes:

[0029] Construct the structure of the model to be trained to be consistent with the structure of the pre-trained model, including an input layer, a hidden layer, and an output layer;

[0030] Inherit the hidden layer parameters of the pre-trained model;

[0031] Randomly initialize the output layer parameters.

[0032] Preferably, the process of selecting the key frequency points includes:

[0033] Select boundary frequency points, including the lower limit frequency point and the upper limit frequency point of the offline frequency sweep range;

[0034] Within the frequency sweep range formed by the boundary frequency points, evenly select basic frequency points at logarithmic intervals;

[0035] According to the smoothness of the impedance characteristic change, increase or decrease the density of the already selected basic frequency points.

[0036] Preferably, based on the online impedance data set, using transfer learning technology to fine-tune the parameters of the model to be trained includes:

[0037] According to the data volume and impedance characteristic complexity of the online impedance data set, set the learning rate and the number of training rounds. The learning rate range is 1e-5 to 1e-3, and the number of training rounds range is 50 to 200 times;

[0038] Use the learning rate and the number of training rounds to optimize the model to be trained.

[0039] Preferably, the effectiveness of the pre-trained model and the fine-tuned model to be trained is verified by the coefficient of determination.

[0040] Preferably, the new energy unit includes a direct-drive wind turbine, a doubly-fed wind turbine, a photovoltaic power generation unit, and an energy storage unit and the new energy power generation system composed of them; the broadband impedance includes traditional positive and negative sequence impedances, improved positive and negative sequence impedances, and dq-domain impedances.

[0041] According to a second aspect of the present invention, there is provided a fast broadband impedance measurement system for new energy units based on transfer learning, including an offline training module and an online measurement module;

[0042] The offline training module includes:

[0043] A simulation unit: Based on the electrical characteristics and topological structure of the actual new energy unit, a standard simulation model of the new energy unit is constructed;

[0044] An offline data unit: The standard simulation model is scanned for broadband impedance under all operating conditions by using the single sine signal injection method to obtain an offline impedance data set;

[0045] A pre-training unit: Based on the offline impedance data set, a pre-trained model is trained;

[0046] The online measurement module includes:

[0047] A unit to be trained: Based on the pre-trained model, a model to be trained is constructed;

[0048] An online data unit: The impedance data of the actual new energy unit at key frequency points is measured online by using the single sine signal injection method to obtain an online impedance data set;

[0049] A fine-tuning unit: Based on the online impedance data set, the parameters of the model to be trained are fine-tuned by using transfer learning technology;

[0050] A measurement unit: The broadband impedance is measured by using the fine-tuned model to be trained.

[0051] Compared with the prior art, the present invention has at least one of the following beneficial effects:

[0052] The fast broadband impedance measurement method for new energy units based on transfer learning according to the embodiment of the present invention, by fusing the single sine signal injection and transfer learning technologies in stages, while retaining the high signal-to-noise ratio advantage of the single sine signal, significantly reduces the online measurement frequency points, effectively solves the contradiction between accuracy and efficiency in broadband impedance measurement, realizes the fast and accurate reconstruction of broadband impedance characteristics, provides reliable technical support for the stability analysis and online evaluation of the new energy grid-connected system, and thus effectively guarantees the safe and stable operation of the new energy grid-connected system. It is particularly suitable for the fast broadband impedance measurement in scenarios where the new energy grid-connected conditions such as photovoltaic power generation units and wind turbines fluctuate frequently. Description of the Drawings

[0053] By reading the detailed description of the non-limiting embodiments with reference to the following drawings, other features, objects, and advantages of the present invention will become more apparent:

[0054] Figure 1It is the basic flowchart of the fast measurement method for broadband impedance of new energy units based on transfer learning in an embodiment;

[0055] Figure 2 It is the electrical structure and control structure of the direct-drive permanent magnet synchronous wind turbine DDPMSG-WTG in an embodiment;

[0056] Figure 3 It is the measurement result of the broadband admittance characteristics of the actual direct-drive permanent magnet synchronous wind turbine DDPMSG-WTG under rated conditions in an embodiment;

[0057] Figure 4 It is the structural schematic diagram of the fast measurement system for broadband impedance of new energy units based on transfer learning in an embodiment. Detailed implementation manners

[0058] The present invention will be described in detail below in conjunction with specific embodiments. The following embodiments will help those skilled in the art to further understand the present invention, but do not limit the present invention in any form. It should be noted that for those of ordinary skill in the art, without departing from the concept of the present invention, several deformations and improvements can still be made. These all belong to the protection scope of the present invention.

[0059] The traditional single sine signal injection method has the advantages of high signal-to-noise ratio and high measurement accuracy, but the full-band measurement efficiency is low, and it is difficult to meet the fast measurement requirements under dynamic conditions; while the broadband signal injection method can significantly improve the measurement efficiency by injecting a disturbance signal containing multiple frequency components at one time, but due to the high signal complexity and limited disturbance amplitude, etc., the measurement accuracy decreases. In summary, the existing methods are difficult to achieve an effective balance between measurement accuracy and efficiency. The embodiment of the present invention proposes a fast measurement method for broadband impedance of new energy units based on transfer learning, which through offline training and online measurement in stages, while ensuring the measurement accuracy, greatly improves the measurement efficiency and effectively solves the contradiction between measurement accuracy and efficiency.

[0060] Refer to Figure 1 As shown, it is the basic flowchart of the fast measurement method for broadband impedance of new energy units based on transfer learning in an embodiment. The measurement method adopts the following steps:

[0061] The first stage: Offline training stage:

[0062] Offline training stage:

[0063] S1. Based on the electrical characteristics and topological structure of the actual new energy unit, construct a standard simulation model of the new energy unit;

[0064] S2. Use the single sine signal injection method to perform full-condition broadband impedance scanning on the standard simulation model to obtain an offline impedance data set;

[0065] S3. Train a pre-trained model based on the offline impedance dataset;

[0066] Online measurement stage:

[0067] S4. Construct a model to be trained based on the pre-trained model;

[0068] S5. Adopt the single sine signal injection method to online measure the impedance data of the actual new energy unit at key frequency points and obtain the online impedance dataset;

[0069] S6. Based on the online impedance dataset, use transfer learning technology to fine-tune the parameters of the model to be trained;

[0070] S7. Use the fine-tuned model to be trained to measure the broadband impedance.

[0071] Specifically, the input and output data format of the fine-tuned model to be trained is {Input: U d 、U q 、I d 、I q 、f; Output: Z RS}. Among them, U d 、U q 、I d 、I q 、f are the port dq-axis voltage, current and disturbance frequency of the actual new energy unit under the real-time operating conditions respectively; Z RS is the port impedance of the actual new energy unit under the real-time operating conditions. According to the actual application requirements, the impedance type can be selected as positive and negative sequence impedance or dq-domain impedance, and the impedance form can also be selected as real part and imaginary part, or amplitude and phase angle.

[0072] It should be noted that the fine-tuned model to be trained does not directly measure the impedance data at each frequency point through the single sine signal injection method. Its essence is the broadband impedance characteristic of the actual new energy unit under the real-time operating conditions. By inputting U d 、U q 、I d 、I q under the real-time operating conditions, and the frequency f, it predicts the impedance characteristics of the full frequency band, saving the time-consuming process of actual measurement at each frequency point.

[0073] In the above embodiments, by fusing the single sine signal injection and transfer learning technologies in stages, while retaining the high signal-to-noise ratio advantage of the single sine signal, the online measurement frequency points are significantly reduced, effectively solving the contradiction between accuracy and efficiency in broadband impedance measurement.

[0074] Refer to Figure 2As shown in the figure, the electrical structure and control structure diagram of the actual new energy unit in an embodiment of the present invention is a direct-drive permanent magnet synchronous wind turbine generator (DDPMSG-WTG). Based on this structure, in this embodiment, step S1 described above is implemented to construct a standard simulation model of the direct-drive permanent magnet synchronous wind turbine generator.

[0075] Due to restrictions such as trade secrets and other factors, the control system of the direct-drive permanent magnet synchronous wind turbine generator usually exhibits "black / gray box" characteristics, and the detailed internal control structure and parameters cannot be obtained. In this embodiment, the system operator can know the main circuit topology, electrical parameters, and the basic control structure adopted by the system (such as V dc Q control, PQ control, etc.). Based on the above structure and parameter information, a standard simulation model is constructed on the Simulink and PSCAD simulation platforms. Specifically, the standard simulation model is consistent with the actual new energy unit in terms of the main circuit topology, electrical parameters, and the basic control structure adopted; for the control parameters that cannot be obtained, common control parameter design methods can be used for design to ensure that the standard simulation model can operate stably within the full operating range.

[0076] As an example, the control parameter design method can refer to the following literature, specifically as follows (References: [1] Dai Jinshui, Lü Jing. Design of Double Closed-loop PI Regulators for Grid-connected Inverters in Wind Power Generation [J]. Electrical Engineering & Electric Appliance, 2011, (09): 5-9.)

[0077] In an embodiment of the present invention, S2 is implemented. The single sine signal injection method is used to perform a full operating range wide-frequency impedance scan on the standard simulation model to obtain an offline impedance data set. Specifically, the following steps can be adopted:

[0078] S21, determine the operating range, including the active power output and reactive power output ranges.

[0079] To ensure that the subsequent pre-trained model is applicable to the entire steady-state operating range and meets the reactive power requirements of the power grid, the active power output range is set to 0 to 1 p.u., and the reactive power output range is set to the maximum capacitive reactive power to the maximum inductive reactive power required by the power grid for the unit to output, and has sufficient resolution.

[0080] It should be noted that: the numerical range referred to by "sufficient resolution" needs to be determined according to specific scenario requirements. The core principle is to avoid missing key features of the impedance characteristics due to insufficient sweep resolution.

[0081] The resolutions of different numerical ranges can adapt to the impedance characteristic differences of different new energy units (such as wind power, photovoltaic), and retain the generality and flexibility of the technical solutions.

[0082] Exemplary: In the frequency band where the impedance characteristics change significantly, the sweep resolution needs to be high enough (e.g., ≤5 Hz, more precisely, ≤1 Hz); in the frequency band where the impedance characteristics change gently, the sweep resolution can be appropriately reduced (e.g., ≥20 Hz).

[0083] Of course, it is also necessary to consider the acquisition cost of impedance data, the annotation cost, and the training cost of the model in practical applications, and the sweep resolution cannot be too detailed.

[0084] Exemplary: The active power output range is set to 0 - 1 p.u., the reactive power output range is set to -0.3 - 0.3 p.u., the resolution is set to 0.1 p.u., and a total of 70 operating conditions are obtained.

[0085] S22. Determine the sweep range, that is, the frequency range of the disturbance signal injected into the new energy unit port.

[0086] The sweep range here covers at least the frequency band below half of the switching frequency. Similarly, sufficient resolution is also required.

[0087] Exemplary: The sweep range is set to 1 - 1001 Hz, where the resolution of the low - frequency band (1 - 200 Hz) is set to 1 Hz, and the resolution of the high - frequency band (201 - 1001 Hz) is set to 25 Hz, with a total of 233 frequency points.

[0088] S23. Measure the impedance data of the standard simulation model within the above - mentioned operating condition range, sweep range, and resolution settings, and obtain the offline impedance dataset.

[0089] It should be noted that: Considering that the offline measurement has low requirements for efficiency, the single - sine - wave signal injection method with the highest measurement accuracy can be preferred to measure point - by - point. The format of the obtained offline impedance dataset is {input: U d 、U q 、I d 、I q 、f; output: Z SM}. Among them, U d 、U q 、I d 、I q 、f are the dq - axis voltages, currents, and disturbance frequencies at the ports of the standard simulation model respectively, with a total of five inputs; Z SM is the traditional positive - and - negative - sequence impedance at the port of the standard simulation model, expressed in the form of real part and imaginary part, with a total of four outputs.

[0090] In the above embodiments, S21 - S23 constructed a high - quality and high - coverage offline impedance dataset by finely designing the working conditions and sweep parameters and combining high - precision data acquisition methods. This provided a reliable pre - training basis for the subsequent steps, enabling the online measurement stage to quickly reconstruct the broadband impedance characteristics with only a small number of key frequency points, improving the online measurement efficiency by more than 80% while ensuring accuracy, and effectively solving the contradiction between impedance measurement accuracy and efficiency under dynamic working conditions of new energy units.

[0091] In a preferred embodiment of the present invention, the pre - training model used is a BP neural network model. This BP neural network model has several significant advantages.

[0092] In terms of the balance between efficiency and cost, due to its lightweight structure and efficient training characteristics, the BP neural network can achieve a coefficient of determination of more than 99.99% (this coefficient reflects the prediction accuracy of the model) with a relatively low training cost in the task of identifying the broadband impedance model of new energy units. In contrast, if other network models (such as CNN, RNN, etc.) are used, not only is the improvement in model prediction accuracy extremely limited, but also several times the training cost needs to be invested, resulting in a marginal benefit far lower than the increase in cost, thus significantly reducing the engineering cost - effectiveness.

[0093] In terms of the adaptability between task characteristics and the model, both the input (including port dq voltage, current, and disturbance frequency) and output (port impedance) of the broadband impedance identification task of new energy units are independent physical quantities without spatio - temporal correlation, and essentially belong to a multi - input multi - output static regression problem. The BP neural network has a high degree of adaptability to these task characteristics and is sufficient to meet the high - precision requirements of this task. Other networks (such as CNN, RNN), which have advantages in dealing with data with spatio - temporal continuity or graphical dependence in their designs, cannot play these advantages in this task, but instead lead to an increased risk of overfitting and higher deployment complexity due to the sharp increase in the number of parameters.

[0094] In terms of the friendliness of transfer learning, the shallow structure of the BP neural network is more convenient for inheriting and fine - tuning parameters in transfer learning, making the transfer process more stable and efficient.

[0095] Further, in a preferred embodiment, step S3 is implemented. Based on the offline impedance dataset, the pre - training model of the BP neural network is trained. Specifically, the following steps can be adopted:

[0096] S31, normalize the offline impedance dataset and divide it into a training set and a test set.

[0097] Exemplarily, the Z - score normalization method is used for normalization, and the impedance data of 70 working conditions are evenly divided into a training set and a test set in an interleaved order.

[0098] S32. Construct a BP neural network, where the number of neurons in the input and output layers is the same as the dimensions of the input and output features respectively, while the number of hidden layers, the number of neurons in each hidden layer, the training algorithm, and the hyperparameters need to be adjusted and optimized according to the performance of the neural network. After completing the network construction and parameter setting, train the constructed BP neural network based on the training set to obtain a pre-trained model.

[0099] Exemplarily, the number of neurons in the input and output layers of the BP neural network is 5 and 4 respectively; the number of hidden layers is 5, and the number of neurons in each hidden layer is 8, 10, 12, 14, and 12 respectively; the training algorithm is set to the Levenberg - Marquardt algorithm, the learning rate is set to 0.003, and the number of training epochs is set to 800.

[0100] S33. Use the coefficient of determination to verify and evaluate the effectiveness of the pre-trained model to guide the structural optimization and parameter adjustment of the pre-trained model, ensuring the reliability and applicability of the pre-trained model. The calculation formula of the coefficient of determination is as follows:

[0101]

[0102] where n is the sample size; y i is the impedance measurement value obtained by sweep frequency, is the impedance prediction value output by the model, is the average value of the sample impedance measurement values. The range of the coefficient of determination is 0 to 1. The closer the coefficient of determination is to 1, the better the effectiveness of the pre-trained model. The above effectiveness includes two aspects: accuracy and generalization ability, which are characterized by the coefficient of determination calculated based on the training set and the test set respectively. In this embodiment, after calculation, the accuracy and generalization ability of the pre-trained model are both as high as 99.99996%, indicating that the model has extremely high fitting accuracy and excellent generalization ability, providing a highly reliable pre-training basis for the subsequent application of transfer learning technology.

[0103] After the pre-trained model is completed, enter the online measurement stage. In a preferred embodiment, implement step S4 to construct a model to be trained with the pre-trained initialization of the structure and hidden layer parameters and the random initialization of the output layer parameters. Specifically:

[0104] Construct a new BP neural network with the same architecture as the pre-trained model, covering the input layer, hidden layer, and output layer. The input layer serves as a channel for information transmission, and the number of its nodes matches the dimension of the input features, and no parameter initialization is required. The initialization process only involves the weights and biases of the hidden layer and the output layer. Specifically, the parameters of the hidden layer are inherited from the pre-trained model, while the parameters of the output layer are randomly initialized to form the model to be trained.

[0105] In an embodiment of the present invention, in step S5, the single sine signal injection method is used to online measure the impedance data of the actual new energy unit at key frequency points, and an online impedance data set is obtained. Specifically, the following steps can be adopted:

[0106] S51, Select key frequency points.

[0107] First, determine the boundary frequency points: covering the starting frequency point and the ending frequency point of the above-mentioned frequency sweep range;

[0108] Subsequently, perform uniform sampling at logarithmic intervals: within the determined boundary of the frequency sweep range, uniformly select basic frequency points at logarithmic intervals;

[0109] Finally, implement dynamic adjustment of the frequency band density: for the selected basic frequency points, reduce the density of frequency points in the low-frequency band and high-frequency band where the impedance characteristics change relatively gently; while in the power frequency band where the impedance characteristics change significantly, increase the density of frequency points.

[0110] Exemplarily, a total of 24 key frequency points are selected, which are 1, 6, 11, 21, 31, 41, 52, 62, 72, 82, 92, 102, 122, 142, 162, 182, 201, 326, 451, 576, 701, 826, 951, 1001Hz respectively.

[0111] S52, Inject single sine perturbation signals of each key frequency point into the ports of the actual new energy unit in sequence, and measure the online impedance data set. It should be ensured that the data format of the online impedance data set is exactly the same as that of the offline impedance data set.

[0112] Among them, the characteristics of the single sine perturbation signal are jointly determined by three parameters: its amplitude, frequency, and initial phase.

[0113] Regarding the amplitude, the amplitude of the perturbation must be controlled within a sufficiently small range to ensure that it will not affect the stable operation of the unit. At the same time, in order to meet the measurement requirements, a sufficiently large signal-to-noise ratio must be ensured. Generally, the perturbation amplitude is set between 1% and 5% of the voltage or current amplitude at the injection point. In terms of frequency, in step S52, the frequencies are sequentially set to the selected key frequency points, such as the 24 key frequency points selected in the above embodiment. As for the initial phase, generally, it can be set to 0 degrees.

[0114] It should be noted that the generation method of the single sine perturbation signal in step S23 is the same as above, the difference being that in S23, it is a full-frequency band sweep (such as 233 points in the embodiment).

[0115] Of course, in some other embodiments, when the measurement accuracy requirement is low, the multi-sine signal injection method can be adopted in S52 to improve the measurement efficiency.

[0116] In the above embodiment, by injecting single - sine disturbance signals at each key frequency point into the ports of the actual new - energy generating units, compared with the traditional single / multi - sine disturbance injection method, the number of online - measured frequency points is reduced by about 89.70%, improving the measurement efficiency.

[0117] In an embodiment of the present invention, in step S6, based on the online impedance data set, transfer learning technology is used to fine - tune the parameters of the model to be trained. Specifically, the following steps can be adopted:

[0118] S61, set the learning rate and the number of training epochs.

[0119] Since there are only differences in control parameters and structural details between the actual new - energy generating units and the standard simulation model, as shown in Table 1, the two belong to a weakly heterogeneous or even homogeneous system. Therefore, during the parameter fine - tuning process, a relatively small learning rate and number of training epochs must be used to optimize the model to be trained to avoid a decline in model performance caused by overfitting.

[0120] Exemplarily: The initial values of the learning rate and the number of training epochs can be preset according to the online data volume, impedance characteristic complexity, etc. Learning rate: 1e - 5 to 1e - 3; Number of training epochs: 50 to 200.

[0121] It should be noted that the main purpose of setting the "relatively low learning rate and number of training epochs" here is to prevent overfitting and maintain the knowledge in the pre - training stage. Therefore, the above - mentioned numerical range is not fixed and should be dynamically adjusted accordingly based on the evaluation of the model training effect. For example:

[0122] If the model shows high accuracy but insufficient generalization ability, this may mean that overfitting has occurred, and at this time, the learning rate should be reduced;

[0123] By observing the trend of the loss function changing with the training cycle, if the loss function has tended to be stable before reaching the preset maximum training cycle, the training process can be considered to be terminated early.

[0124] S62, based on the online impedance data, fine - tune the parameters of the model to be trained to obtain the broadband impedance characteristic model of the actual new - energy generating units under the current operating conditions.

[0125] Furthermore, in a specific implementation, the learning rate is set to 0.00001 and the number of training epochs is set to 80. Compared with the traditional data - driven broadband impedance identification method, the number of training epochs is reduced by 90%, significantly improving the efficiency of model training, and thus further improving the overall efficiency of impedance measurement of the present invention.

[0126] Table 1 Parameters of the standard simulation model and the actual new - energy generating units

[0127]

[0128]

[0129] Similarly, in one embodiment, the coefficient of determination is used to verify and evaluate the effectiveness of the obtained broadband impedance characteristic model, so as to guide the adjustment of the learning rate and the number of training epochs during the fine-tuning process, and ensure that the measurement method proposed by the present invention has a sufficiently high measurement accuracy.

[0130] In the offline training stage of the above embodiment, based on the standard simulation model of a certain type of new energy unit, the common laws of its broadband impedance characteristics are obtained through pre-training, and a general knowledge base of its impedance characteristics is constructed; in the online stage, for the characteristic offset caused by differences in control structure, parameters, etc. of the actual new energy unit, the pre-trained knowledge is reused through fine-tuning to solve the engineering problem of "one unit, one model".

[0131] In practical applications, new energy units cannot operate safely and stably under some extreme working conditions. Not only can the broadband impedance characteristics of the units under extreme working conditions not be measured, but it may also pose a threat to the safety of equipment and personnel. The staged strategy of the above embodiment completes the impedance scanning of the full working conditions and full frequency bands in the simulation environment through the offline measurement stage, effectively avoiding the measurement risks of extreme working conditions of actual new energy units, and at the same time ensuring the comprehensiveness and reliability of the measurement.

[0132] In addition, in a specific embodiment, a model trained through fine-tuning is used for measurement, and the measurement results of the broadband admittance characteristics of the actual DDPMSG-WTG under rated working conditions are obtained as Figure 3 shown. After calculation, the effectiveness of the obtained broadband admittance characteristic model is as high as 99.91%, and the entire measurement process takes no more than 5 seconds, indicating that the impedance measurement method proposed by the embodiment of the present invention effectively solves the contradiction between the measurement accuracy and efficiency of the broadband impedance of new energy units.

[0133] In some specific embodiments, the impedance can be represented in the form of imaginary part and real part or amplitude and phase angle to meet different application requirements.

[0134] In some specific embodiments, the single sine signal injection method is used to measure impedance data frequency by frequency; when the measurement accuracy requirement is low, the multi-sine signal injection method can be used to improve the measurement efficiency.

[0135] The applicable scope of the present invention includes, but is not limited to, the rapid measurement of the broadband impedance of direct-drive wind turbine units, doubly-fed wind turbine units, photovoltaic power generation units, energy storage units, and new energy power generation systems composed of them. The broadband impedance includes traditional positive and negative sequence impedances, improved positive and negative sequence impedances, and dq-domain impedances, and the admittance is also applicable.

[0136] In other embodiments of the present invention based on the same inventive concept, a fast broadband impedance measurement system for new energy units based on transfer learning is provided, as Figure 4 shown, which includes an offline training module and an online measurement module;

[0137] The offline training module includes:

[0138] A simulation unit: Based on the electrical characteristics and topological structure of the actual new energy unit, a standard simulation model of the new energy unit is constructed;

[0139] An offline data unit: The single sine signal injection method is used to perform a full-condition broadband impedance scan on the standard simulation model to obtain an offline impedance data set;

[0140] A pre-training unit: Based on the offline impedance data set, a pre-training model is trained;

[0141] The online measurement module includes:

[0142] A unit to be trained: Based on the pre-training model, a model to be trained is constructed;

[0143] An online data unit: The single sine signal injection method is used to online measure the impedance data of the actual new energy unit at key frequency points to obtain an online impedance data set;

[0144] A fine-tuning unit: Based on the online impedance data set, transfer learning technology is used to fine-tune the parameters of the model to be trained;

[0145] A measurement unit: Using the fine-tuned model to be trained, the broadband impedance is measured.

[0146] In the above examples of the present invention, the specific implementation of each module / unit can specifically refer to the implementation technology of the corresponding steps in the method for fast broadband impedance measurement of new energy units based on transfer learning in the above embodiments, which will not be elaborated here.

[0147] The specific embodiments of the present invention have been described above. It should be understood that the present invention is not limited to the above specific implementation manners. Those skilled in the art can make various deformations or modifications within the scope of the claims, which do not affect the essence of the present invention. The above preferred features can be combined arbitrarily without conflict.

Claims

1. A fast broadband impedance measurement method for new energy generating units based on transfer learning, characterized in that, Including: Offline training stage: Based on the electrical characteristics and topological structure of the actual new energy unit, construct a standard simulation model of the new energy unit; Use the single sine signal injection method to perform full-condition wide-frequency impedance scanning on the standard simulation model to obtain an offline impedance dataset; Based on the offline impedance dataset, train a pre-trained model; Online measurement stage: Based on the pre-trained model, construct a model to be trained; Use the single sine signal injection method to online measure the impedance data of the actual new energy unit at key frequency points to obtain an online impedance dataset; Based on the online impedance dataset, use transfer learning technology to fine-tune the parameters of the model to be trained; Use the fine-tuned model to be trained to measure wide-frequency impedance.

2. The broadband impedance rapid measurement method for new energy units based on transfer learning according to claim 1, wherein The construction of the standard simulation model of the new energy unit based on the electrical characteristics and topological structure of the actual new energy unit includes: Construct a standard simulation model based on the Simulink or PSCAD simulation platform; Set the main circuit topology, electrical parameters, and control strategy to be consistent with the actual new energy unit; Set control parameters to ensure that the standard simulation model can operate stably within the full-condition range.

3. A method for rapid measurement of broadband impedance of a new energy unit based on transfer learning according to claim 1, characterized in that, The use of the single sine signal injection method to perform full-condition wide-frequency impedance scanning on the standard simulation model to obtain an offline impedance dataset includes: Determine the operating condition range, which includes the active power output and reactive power output ranges. The active power output range is 0 to 1 p.u., and the reactive power output range is from the maximum capacitive reactive power required by the grid for the unit to the maximum inductive reactive power; Determine the frequency scanning range, which at least covers the frequency band below half of the switching frequency; Set the active power resolution, reactive power resolution, and frequency scanning resolution to ensure that impedance data can be collected; Under the conditions of the operating condition range, frequency scanning range, and resolution, use the single sine signal injection method to collect the impedance data of the standard simulation model at each frequency point to obtain an offline impedance dataset.

4. A broadband impedance rapid measurement method for a new energy unit based on transfer learning according to claim 1, characterized in that, The pre-trained model is a BP neural network model.

5. A method for rapid measurement of broadband impedance of a new energy unit based on transfer learning according to claim 1, characterized in that, The construction of the model to be trained based on the pre-trained model includes: Construct the structure of the model to be trained to be consistent with the structure of the pre-trained model, including an input layer, a hidden layer, and an output layer; Inherit the hidden layer parameters of the pre-trained model; Randomly initialize the output layer parameters.

6. The method for rapidly measuring the broadband impedance of a new energy unit based on transfer learning according to claim 3, wherein The selection process of the key frequency points includes: Select boundary frequency points, including the lower limit frequency point and the upper limit frequency point of the offline frequency scanning range; Evenly select basic frequency points at logarithmic intervals within the frequency scanning range formed by the boundary frequency points; According to the smoothness of the impedance characteristic change, increase or decrease the density of the already selected basic frequency points.

7. A broadband impedance rapid measurement method for new energy generating units based on transfer learning according to claim 1, characterized in that The use of transfer learning technology to fine-tune the parameters of the model to be trained based on the online impedance dataset includes: According to the data volume and impedance characteristic complexity of the online impedance dataset, set the learning rate and the number of training epochs. The learning rate range is 1e-5 to 1e-3, and the number of training epochs range is 50 to 200 times; Use the learning rate and the number of training epochs to optimize the model to be trained.

8. A broadband impedance rapid measurement method for new energy generating units based on transfer learning according to claim 1, characterized in that The effectiveness of both the pre-trained model and the fine-tuned model to be trained is verified by the coefficient of determination.

9. A method for fast measurement of broadband impedance of a new energy generating unit based on transfer learning according to claim 1, characterized in that, The new energy unit includes a direct-drive wind turbine unit, a doubly-fed wind turbine unit, a photovoltaic power generation unit, a energy storage unit, and a new energy power generation system composed thereof; the broadband impedance includes traditional positive and negative sequence impedances, improved positive and negative sequence impedances, and dq-domain impedances.

10. A broadband impedance rapid measurement system for new energy generating units based on transfer learning, characterized in that, It includes an offline training module and an online measurement module; The offline training module includes: Simulation unit: Based on the electrical characteristics and topological structure of the actual new energy unit, a standard simulation model of the new energy unit is constructed; Offline data unit: The single sine signal injection method is used to perform a full-condition broadband impedance scan on the standard simulation model to obtain an offline impedance data set; Pre-training unit: Based on the offline impedance data set, a pre-training model is trained; The online measurement module includes: Unit to be trained: Based on the pre-training model, a model to be trained is constructed; Online data unit: The single sine signal injection method is used to online measure the impedance data of the actual new energy unit at key frequency points to obtain an online impedance data set; Fine-tuning unit: Based on the online impedance data set, the parameters of the model to be trained are fine-tuned using transfer learning technology; Measurement unit: The broadband impedance is measured using the fine-tuned model to be trained.

Images