Method and system for improving stability of LCL type grid-connected inverter
By constructing a stability determination model and a prediction model, and using an SVM training dataset to optimize control parameters and gains, the stability problem of LCL grid-connected inverters in weak grid environments was solved, achieving high-precision stability determination and dynamic stability improvement.
Patent Information
- Application Number
- CN202511290745.0
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-09-10
- Publication Date
- 2025-12-12
AI Technical Summary
Existing stability analysis methods for LCL-type grid-connected inverters are difficult to handle complex nonlinear systems in weak grid environments. Hardware parameter deviations and grid impedance changes affect inverter stability, and traditional methods cannot meet dynamic stability requirements.
By constructing a stability determination model, a control parameter prediction model, and a control gain prediction model, and using a support vector machine (SVM) model to train the dataset, the system predicts and optimizes the control parameters and gains to adapt to changes in hardware parameters, thereby achieving system stability determination.
In a weak grid environment, high-precision stability determination and control parameter optimization of LCL grid-connected inverters were achieved, improving the dynamic stability of the system, with strong adaptability and suitability for complex nonlinear relationships.
Smart Images
Figure CN121124599A_ABST
Abstract
Description
TECHNICAL FIELD
[0001] The present application relates to the field of power electronics, more particularly to a method and system for improving the stability of an LCL grid-connected inverter. BACKGROUND
[0002] Under the guidance of the "double carbon" goal, renewable energy technologies have rapidly developed. LCL three-level grid-connected inverters have become the mainstream choice in photovoltaic power generation systems due to their excellent high harmonic filter performance and high power density. As a third-order circuit, the LCL filter itself has a resonance effect, and changes in hardware parameter deviations and grid impedance will affect the stability of the inverter.
[0003] Existing stability analysis methods usually rely on accurate system models and parameters, making it difficult to handle complex nonlinear systems.
[0004] However, in a weak grid environment, the dynamic changes in grid impedance and the uncertainty of LCL filter parameters make it impossible for traditional stability analysis methods to meet the dynamic stability requirements in practical applications. SUMMARY
[0005] To address the problems in the above field, the present application proposes a method and system for improving the stability of an LCL grid-connected inverter. By changing the input and output, different training set data are constructed to establish a stability determination model, a control parameter prediction model, and a control gain prediction model. The prediction results of the control parameter prediction model and the control gain prediction model are used as inputs to the stability determination model, and the system stability label output by the stability determination model is used to determine the system stability. An optimization method is provided to re-predict control parameters and control gains when the system is unstable. This method can accurately determine the system stability under changes in hardware parameters and is suitable for LCL grid-connected inverters in a weak grid environment.
[0006] To solve the above technical problems, the present application discloses a method for improving the stability of an LCL grid-connected inverter, comprising the following steps: Based on the control scheme for controlling the stability of an LCL grid-connected inverter, determine the control parameters representing the state of the LCL filter side inductance and the control gain representing the state of the grid side inductance under different operating conditions; Using the control parameters, control gains, LCL filter side inductance, grid side inductance, capacitance, and grid impedance as inputs, and the system stability label as output, construct a system stability training data set and train it to obtain a stability determination model; Using the LCL filter side inductance, capacitance, and system stability label as inputs, and the control parameters as outputs, construct a control parameter training data set and train it to obtain a control parameter prediction model; Using grid-side inductance, capacitance, grid impedance, and system stability labels as inputs and control gain as output, a control gain training dataset is constructed and trained to obtain a control gain prediction model. The LCL filter-side inductance, grid-side inductance, capacitance, and grid impedance of the LCL grid-connected inverter under the current operating conditions are collected. The control parameter prediction model and control gain prediction model are input for prediction, and the control parameters and control gain are output. These are then input into the stability judgment model. When the output system stability label represents instability, the control parameters and control gain are re-predicted by changing the LCL filter-side inductance or grid impedance. The system stability label output by the stability determination model represents stability. The control parameters and control gain output at this point are then used as the optimal parameter combination for the stable operation of the LCL grid-connected inverter.
[0007] Preferably, the step of determining the control parameters representing the inductor state on the LCL filter side and the control gain representing the inductor state on the grid side under different operating conditions specifically includes: Data was collected from the LCL-type grid-connected inverter under different operating conditions, including DC bus voltage. U dc LCL filter side inductor L xi , grid-side inductor L xg ,capacitance C x and grid impedance L xwg Grid voltage U PCCx ; The control scheme includes controlling the LCL-type grid-connected inverter using the DD-Σ control algorithm, with the control law being: ; ; ; ; in, T s For the switching cycle, K p1 For control parameters, K p2 To control the gain, i xi ( n ) is the first n Inverter-side inductor current at each sampling time, i xg ( n ) is the first na grid-connected current at a sampling moment, I xgref representing a grid-connected current reference value.
[0008] Preferably, the obtaining the stability determination model specifically comprises: taking the control parameter, the control gain, the LCL filter side inductance, the grid side inductance, the capacitance and the grid impedance as inputs, labeling the system stability, establishing a feature space taking the system stability label as an output, and constructing an A model; the feature space taking the system stability label as an output is: ; wherein 1 represents stability and 0 represents instability; the SVM model is trained through the system stability training set data, and the stability determination model is obtained.
[0009] Preferably, the obtaining the control parameter prediction model comprises: taking the LCL filter side inductance, the capacitance and the system stability label as inputs, establishing a space feature taking the control parameter as an output, and constructing an SVM1 model; the feature space taking the control parameter as an output is: ; the SVM1 model is trained through the control parameter training set data, and the control parameter prediction model is obtained.
[0010] Preferably, the obtaining the control gain prediction model comprises: taking the grid side inductance, the capacitance, the grid impedance and the system stability label as inputs, establishing a space feature taking the control gain as an output, and constructing an SVM2 model; the feature space taking the control gain as an output is: ; the SVM2 model is trained through the control gain training set data, and the control gain prediction model is obtained.
[0011] Preferably, the re-predicting the control parameter and the control gain comprises: when the output of the stability determination model is 0, that is, the system is unstable, the input control parameter and the control gain are re-selected through a cross-validation manner until the selected control parameter and control gain input the output corresponding to the stability determination model is 1.
[0012] Preferably, the re-selecting the input control parameter and control gain through the cross-validation manner specifically comprises: by kCross-validation divides the system stability training dataset into two parts. k Each subset consists of a set of parameters, each containing a combination of control parameters and control gain; for each candidate parameter combination, the following steps are performed: k Independent rounds of SVM model training and validation set evaluation; Each time, select one parameter combination as the validation set, and the rest... k -1 parameters are combined into a training set; the SVM model is trained using the training set, and its performance is evaluated on the validation set; For each parameter combination, the classification accuracy, recall, and stability of the stability determination model are calculated by comparing the system stability label output by the stability determination model with the actual system stability. F 1. Score indicator; Will k The classification accuracy, recall, and F1 score were averaged to obtain the result. k Average performance score of the second validation; The parameter combination with the highest average performance is selected as the optimal parameter combination for stable operation of the LCL-type grid-connected inverter.
[0013] Preferably, the method further includes cleaning and normalizing the data in the constructed system stability training dataset, control parameter training dataset, and control gain training dataset, scaling the data to the [0,1] interval.
[0014] Preferably, it also includes an LCL-type grid-connected inverter stability improvement system, comprising: The control parameter and control gain determination module is used to determine the control parameters and control gain representing the inductor state on the LCL filter side and the grid-side inductor state under different operating conditions based on the control scheme for the stable operation of the LCL grid-connected inverter. The stability determination model construction module is used to construct and train a system stability training dataset by taking control parameters, control gain, LCL filter-side inductance, grid-side inductance, capacitance, and grid impedance as inputs and system stability labels as outputs, in order to obtain a stability determination model. The control parameter prediction model building module is used to construct a control parameter training dataset and train it to obtain a control parameter prediction model by taking the inductance, capacitance and system stability labels of the LCL filter side as inputs and the control parameters as outputs. The control gain prediction model building module is used to construct a control gain training dataset and train it to obtain a control gain prediction model by taking grid-side inductance, capacitance, grid impedance and system stability labels as inputs and control gain as output. The stability determination module collects the LCL filter-side inductance, grid-side inductance, capacitance, and grid impedance of the LCL grid-connected inverter under the current operating conditions. It inputs these parameters into the control parameter prediction model and the control gain prediction model for prediction, outputting control parameters and control gain. These are then input into the stability determination model. When the output system stability label indicates instability, the control parameters and control gain are re-predicted by changing the LCL filter-side inductance or the grid impedance. This process continues until the system stability label output by the stability determination model indicates stability. The output control parameters and control gain at this point are then used as the optimal parameter combination for stable operation of the LCL grid-connected inverter.
[0015] Compared with the prior art, the present invention has the following beneficial effects: This invention proposes a stability improvement method for LCL-type grid-connected inverters. Based on a control scheme for stable operation of the LCL-type grid-connected inverter, it determines the control parameters representing the inductor state on the LCL filter side and the control gain representing the inductor state on the grid side under different operating conditions, which are then used as parameters to be optimized. By collecting data on the LCL filter-side inductance, grid-side inductance, capacitance, and grid impedance during the current operating condition of the LCL-type grid-connected inverter, the constructed control parameter prediction model and control gain prediction model are input for prediction, respectively. The output control parameters and control gain are then input into a stability judgment model. When the output system stability label indicates instability, the control parameters and control gain are re-predicted by changing the LCL filter-side inductor or grid impedance, further optimizing the control parameters and control gain to improve the dynamic stability of the system. This method achieves high-precision determination of the stability of the LCL-type grid-connected inverter under different hardware parameters and grid conditions, as well as the selection of control parameters. This method does not require the establishment of an accurate stability analysis model for the LCL-type grid-connected inverter and weak grid, making it suitable for LCL-type grid-connected inverters in weak grid environments. It can effectively handle complex nonlinear relationships and only requires re-collection of data for training, demonstrating strong adaptability. Attached Figure Description
[0016] Figure 1 This is a flowchart of the LCL-type grid-connected inverter stability improvement method proposed in this invention; Figure 2 This is a topology diagram of an LCL-type T-type three-level grid-connected inverter under a weak power grid, provided in an embodiment of the present invention. Figure 3 The embodiments of the present invention provide methods for fixing hardware parameters and changing control parameters before data cleaning. K p1 and control gain K p2 Stability data at that time; Figure 4 The data cleaning process provided in this embodiment of the invention involves fixing hardware parameters and changing control parameters.K p1 and control gain K p2 Stability data at that time; Figure 5 The present invention provides a method for selecting control parameters for an LCL grid-connected inverter based on a multi-SVM model. K p1 and control gain K p2 Optimized flowchart; Figure 6 The three-phase grid-connected current waveforms for system stability using the DD-Σ algorithm provided in this embodiment of the invention; Figure 7 Waveform diagrams for verifying the stability criterion based on the SVM model when the side inductance changes, as provided in the embodiments of the present invention; Figure 8 Waveform diagrams for verifying the stability criterion based on the SVM model when the grid-side inductance changes, as provided in the embodiments of the present invention; Figure 9 The waveform diagram is used to verify the parameter combination optimization method based on the SVM model when the side inductance changes, as provided in the embodiment of the present invention. Figure 10 The waveform diagram is used to verify the parameter combination optimization method based on the SVM model when the grid-side inductance changes, as provided in the embodiments of the present invention. Detailed Implementation
[0017] The following will refer to the appendices in the embodiments of the present invention. Figures 1-10 The technical solutions in the embodiments of the present invention will be clearly and completely described. It should be understood that the terminology used in the present invention is only for describing particular implementation methods and is not intended to limit the present invention.
[0018] Example like Figure 1 As shown, this invention proposes a method for improving the stability of an LCL-type grid-connected inverter, comprising the following steps: S1: Based on the control scheme for controlling the stable operation of LCL grid-connected inverters, determine the control parameters representing the inductor state on the LCL filter side and the control gain representing the inductor state on the grid side under different operating conditions; S2: Using control parameters, control gain, LCL filter-side inductance, grid-side inductance, capacitance, and grid impedance as inputs, and system stability labels as outputs, construct a system stability training dataset and train it to obtain a stability determination model; S3: Using the inductance, capacitance, and system stability labels of the LCL filter side as inputs and the control parameters as outputs, construct a control parameter training dataset and train it to obtain a control parameter prediction model; S4: Using grid-side inductance, capacitance, grid impedance, and system stability labels as inputs and control gain as output, construct a control gain training dataset and train it to obtain a control gain prediction model. S5: Collect the LCL filter-side inductance, grid-side inductance, capacitor, and grid impedance of the LCL grid-connected inverter under the current operating conditions. Input the control parameter prediction model and control gain prediction model to make predictions respectively, output the control parameters and control gain, and input them into the stability judgment model. When the output system stability label represents instability, the control parameters and control gain are re-predicted by changing the LCL filter-side inductance or grid impedance. S6: Until the system stability label output by the stability determination model represents stability, the output control parameters and control gain at this time are taken as the optimal parameter combination when the LCL grid-connected inverter is working stably.
[0019] Specifically, such as Figure 2 The diagram shown is a topology diagram of an LCL-type T-type three-level grid-connected inverter under a weak power grid, with DC bus voltage... U dc The voltages of the upper and lower supporting capacitors are respectively u C1 and u C2 LCL filter side inductor L xi ,capacitance C x , grid-side inductor L xg and grid impedance by L xwg , U PCCx This represents the grid voltage; each phase arm contains four power switching transistors. S x1 - S x4 ,in, x = a , b , c .
[0020] In step S1, the LCL-type grid-connected inverter is controlled using the DD-Σ control algorithm, and the control law is as follows: ; ; ; ; in, T s For the switching cycle, Kp1 For control parameters, K p2 To control the gain, i xi ( n ) is the first n Inverter-side inductor current at each sampling time, i xg ( n ) is the first n The grid-connected current at each sampling time, I xgref This indicates the reference value for grid-connected current.
[0021] The raw data of LCL filter-side inductance, grid-side inductance, capacitance and grid impedance collected under the current operating conditions of the LCL grid-connected inverter are cleaned and normalized. Through cleaning, obvious noise data and outliers are removed, and missing data is filled in.
[0022] Normalization is performed using a standard normalization process to scale the data to the [0,1] interval, thereby improving the stability and convergence speed of training SVM, SVM1, and SVM2 models. ; in, X ={ x 1, x 2,..., x n} represents the training data sequence to be normalized. i =1,..., n , x j This represents the mean of the training data.
[0023] like Figure 3 and 4 As shown, these are the hardware parameters. Lxi =3mH、 Lxg When =1mH, respectively K p1 Increased from 0.06 to 0.6 K p2 The stability change graph from 0.02 to 0.2 shows that... Figure 3 Abnormal data exists; it is supplemented by running the simulation system to determine its stability. Figure 4 The correct data is obtained; other cases are handled similarly.
[0024] In step S2, based on the normalized data, the system stability is labeled using the control parameters, control gain, LCL filter-side inductance, grid-side inductance, capacitance, and grid impedance as inputs. A model is constructed by establishing a feature space with the system stability labels as outputs, where A model is an SVM model.
[0025] The feature space with system stability labels as output is established as follows: ; Where 1 represents stable and 0 represents unstable; The SVM model is trained using the system stability training set data to obtain a stability determination model.
[0026] Based on the stability determination model, a stability criterion for LCL-type grid-connected inverters is formed.
[0027] In step S3, the SVM1 model is constructed by taking the inductance, capacitance and system stability labels of the LCL filter side as inputs and establishing spatial features with control parameters as outputs. The feature space with control parameters as output is established as follows: ; The SVM1 model is trained using the control parameter training set data to obtain the control parameter prediction model.
[0028] In step S4, the SVM2 model is constructed by establishing spatial features with control gain as the output, using grid-side inductance, capacitance, grid impedance, and system stability labels as inputs. The feature space with control gain as the output is established as follows: ; The SVM2 model is trained using the control gain training set data to obtain the control gain prediction model.
[0029] In the simulation system, changing the control parameters K p1 With other parameters remaining unchanged, the resonant frequency of the LCL filter remains unchanged because the hardware parameters are the same. The stability of the system can be judged based on two indicators: the total harmonic distortion of the grid-connected current and the harmonic amplitude at the resonant frequency. This method can quickly obtain training data and save time by changing parameters and rerunning the simulation system.
[0030] The system stability criterion requirements are as follows: ; In the formula, I fr It is the harmonic amplitude at the resonant frequency. Ibase It is the fundamental frequency amplitude.
[0031] Steps S3 and S4 show that the control parameters are obtained through the SVM1 model. K p1 Training is performed for the output, from the control parameters K p1 As can be seen from the design rules, K p1 Only with the LCL filter side inductor L xi and system stability S Related.
[0032] Control parameters were analyzed using the SVM2 model. K p2 Train for output. K p2 Only the grid-side inductance of the LCL filter L xg Grid impedance L xwg and system stability S Related.
[0033] The purpose of constructing the stability judgment model is to further verify the correctness of the control parameters and control gains selected by the SVM1 and SVM2 models, and to use it as a basis for judging the correctness of the selection of control parameters and control gains.
[0034] The SVM1 and SVM2 models were trained respectively to obtain the control parameter prediction model and the control gain prediction model.
[0035] In step S5, the LCL filter-side inductance, grid-side inductance, capacitance and grid impedance of the LCL grid-connected inverter under the current operating conditions are collected. The control parameter prediction model and control gain prediction model are input for prediction, and the control parameters and control gain are output.
[0036] The output control parameters and control gain are input into the stability determination model. When the system stability label output by the stability determination model represents stability, that is, when the output of the SVM model is 1, it is used as the basis for judging the correctness of the input control parameters and control gain.
[0037] When the output of the SVM model is 1, the control parameters and control gain of the corresponding stability determination model input are taken as the optimal parameter combination for the stable operation of the LCL grid-connected inverter.
[0038] Since the performance of SVM models is highly dependent on the configuration of key parameters, especially the penalty coefficient C and kernel function parameters, this invention employs cross-validation to select parameters for the SVM model.
[0039] When the output of the stability determination model is 0, the input control parameters are tested using cross-validation. K p1* and control gain K p2* Repeat the selection process until the selected control parameter is reached. K p1* and control gain K p2* The output corresponding to the input stability determination model is 1.
[0040] like Figure 5 As shown, this is a method for cross-validating the input control parameters. K p1* and control gain K p2* The procedure for reselecting is as follows: Dataset partitioning The constructed system stability training dataset is divided into training data and test data. During the stability assessment model validation process, the test data is used for final performance evaluation, while the training data is used for training the stability assessment model and optimizing parameters.
[0041] Define parameter range Choose the range of parameters to be optimized. For the SVM model, the core hyperparameters to be optimized include the penalty parameter C and the kernel function parameters.
[0042] Cross-validation process use k Cross-validation divides the system stability training dataset into two parts. k Each subset is a "fold," which is a parameter combination consisting of control parameters and control gain. For each candidate parameter combination, execution is performed. k Each round of independent SVM model training and validation set evaluation; each time, one "fold" is selected as the validation set, and the rest... k -1 folds are used as the training set. By dividing the data into training and validation sets, it is ensured that there is no duplicate data between the two. The SVM model is trained using the training set, and its performance is evaluated on the validation set. For each parameter combination, the classification accuracy, recall, and stability determination model are calculated by comparing the system stability label output by the stability determination model with the actual system stability. F 1. Score indicator; k The classification accuracy, recall, and F1 score were averaged to obtain the result. k The average performance score of the second verification is used to select the parameter combination with the highest average performance as the optimal parameter combination for stable operation of the LCL grid-connected inverter.
[0043] Classification accuracy is defined as the percentage of correctly predicted samples out of the total sample count. Recall is the percentage of correctly predicted stable (or unstable) samples out of the actual stable (or unstable) samples. The F1 score is a metric that considers both classification accuracy and recall, and its calculation formula is:
[0044] ; This invention also proposes a stability improvement system for LCL-type grid-connected inverters, comprising: The control parameter and control gain determination module is used to determine the control parameters and control gain representing the inductor state on the LCL filter side and the grid-side inductor state under different operating conditions based on the control scheme for the stable operation of the LCL grid-connected inverter. The stability determination model construction module is used to construct and train a system stability training dataset by taking control parameters, control gain, LCL filter-side inductance, grid-side inductance, capacitance, and grid impedance as inputs and system stability labels as outputs, in order to obtain a stability determination model. The control parameter prediction model building module is used to construct a control parameter training dataset and train it to obtain a control parameter prediction model by taking the inductance, capacitance and system stability labels of the LCL filter side as inputs and the control parameters as outputs. The control gain prediction model building module is used to construct a control gain training dataset and train it to obtain a control gain prediction model by taking grid-side inductance, capacitance, grid impedance and system stability labels as inputs and control gain as output. The stability determination module collects the LCL filter-side inductance, grid-side inductance, capacitance, and grid impedance of the LCL grid-connected inverter under the current operating conditions. It inputs these parameters into the control parameter prediction model and the control gain prediction model for prediction, outputting control parameters and control gain. These are then input into the stability determination model. When the output system stability label indicates instability, the control parameters and control gain are re-predicted by changing the LCL filter-side inductance or the grid impedance. This process continues until the system stability label output by the stability determination model indicates stability. The output control parameters and control gain at this point are then used as the optimal parameter combination for stable operation of the LCL grid-connected inverter.
[0045] In summary, the stability improvement method for LCL grid-connected inverters proposed in this invention does not require the establishment of an accurate stability analysis model for LCL grid-connected inverters and weak power grids, and can effectively handle complex nonlinear relationships, demonstrating strong adaptability.
[0046] The constructed stability determination model, control parameter prediction model, and control gain prediction model, through optimization of control parameters and control gain, can accurately determine system stability under changes in hardware parameters, and are suitable for LCL-type grid-connected inverters in weak grid environments.
[0047] The proposed stability improvement method can be adapted to other LCL grid-connected inverter control schemes besides the DD-Σ control algorithm. It does not require rebuilding the stability analysis model according to different control schemes; it only requires re-collecting data for training.
[0048] Simulation verification The proposed scheme was verified in MATLAB / Simulink software using the simulation parameters shown in Table 1.
[0049] Table 1 Simulation Parameters Table 2 shows the training set for partial grid impedance changes. Since changes in the filter capacitor are not considered, the filter capacitor is set to 20. μ F, only considers the changes of other variables.
[0050] Table 2 Training set for partial grid impedance changes In Table 2, the step size of each change in grid impedance is 0.1 mH. Using the controlled variable method, whenever the grid impedance increases by 0.1 mH, the control gain is gradually modified according to the parameter design rules of the DD-Σ algorithm. K p2 Then, the system stability is verified through simulation waveforms to form a training set. However, when forming the verification set, the simulation parameters and hardware parameters used must be avoided to prevent the training set from overlapping with the verification set.
[0051] The training set in Table 2 was used to train the SVM model, resulting in a stability criterion model. The validation set data was then used to validate the SVM model. Considering the experimental conditions, Table 3 presents the validation results of the SVM model's stability criterion under varying inductance on the LCL filter side and grid impedance.
[0052] Table 3. Verification results of the stability criterion based on SVM like Figure 6 As shown, the LCL filter-side inductance, grid-side inductance, grid impedance, and control parameters are obtained using the first row of Table 3. K p1 and control gain K p2Below is the stable three-phase grid-connected current waveform of the system running through the DD-Σ algorithm. The horizontal axis t / s represents the inverter's operating time, and the vertical axis I / A represents the current. This three-phase grid-connected current waveform is consistent with the output of the SVM model, indicating that the system is stable.
[0053] exist Figure 6 Based on the parameters, Figure 7 When the inductance on the LCL filter side increases by 1mH, the control parameters before the change are used. K p1 and control gain K p2 The three-phase grid-connected current waveform is shown. The horizontal axis t / s represents the inverter's operating time, and the vertical axis I / A represents the current. These correspond to the parameters in the second and third rows of Table 3, respectively. When the inductance on the LCL filter side increases at 0.04s, harmonics appear in the current. When the control parameters in the third row are modified at 0.07s... K p1 and control gain K p2 The system stabilized again.
[0054] exist Figure 6 Based on the parameters, Figure 8 When the grid-side inductance increases by 1mH (similar to changes in grid impedance), the control parameters before the change are used. K p1 and control gain K p2 The three-phase grid-connected current waveform is shown. The horizontal axis t / s represents the inverter's operating time, and the vertical axis I / A represents the current. These correspond to the parameters in the fourth and fifth rows of Table 3, respectively. When the inverter-side inductance increases at 0.04s, harmonics appear in the current. When the inductance increases at 0.07s, the control parameters in the fifth row are modified. K p1 and control gain K p2 The system stabilized again, and the simulation waveform was consistent with the stability criterion results.
[0055] Using the data in Table 3, and taking system stability as input, control parameters... K p1 and control gain K p2 Convert to output, for control parameters under different selected hardware parameters. K p1 and control gain K p2Verification was conducted. As shown in Table 4, the results of the selection of control parameters and control gain for grid-connected inverters under two different hardware parameters are presented. Since the grid impedance change has the same effect as the grid-side inductance change, only the LCL filter inductance value and grid impedance value are changed. The training set and validation set are also completely avoided, and there are no cases of identical data.
[0056] Table 4. Validation results of the SVM-based method for selecting control parameters and control gains. For the control parameters selected in Table 4 K p1 and control gain K p2 The simulation was performed again, and the simulation results are as follows: Figure 9 and Figure 10 The graphs shown correspond to the waveforms used to verify the parameter combination optimization method based on the SVM model when the inductance on the LCL filter side changes and when the inductance on the grid side changes. The horizontal axis t / s represents the inverter's operating time, and the vertical axis I / A represents the current. When the inductance on the LCL filter side or the grid impedance increases by 1mH, if the current continues to be used... L xi =3.8mH L xg =1.1mH and L xwg =0mH, from Figure 9 and Figure 10 As can be seen, the system exhibits resonant instability, with harmonics appearing in the grid-connected current. However, by using the control parameters and control gain selected through the SVM model in Table 4, the system can be stabilized again, demonstrating the effectiveness of the dynamic stability improvement method for the LCL-type grid-connected inverter proposed in this invention.
[0057] The above description is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.
[0058] Furthermore, unless otherwise stated, all technical and scientific terms used in this invention have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains. All references to this specification are incorporated by way of citation to disclose and describe methods relating to those references. In the event of any conflict with any incorporated reference, the content of this specification shall prevail.
Claims
1. A method for improving the stability of an LCL-type grid-connected inverter, characterized in that, Includes the following steps: Based on the control scheme for controlling the stable operation of LCL grid-connected inverters, the control parameters representing the inductor state on the LCL filter side and the control gain representing the inductor state on the grid side under different operating conditions are determined. Using control parameters, control gain, LCL filter-side inductance, grid-side inductance, capacitance, and grid impedance as inputs, and system stability labels as outputs, a system stability training dataset is constructed and trained to obtain a stability determination model. Using the inductance, capacitance, and system stability labels of the LCL filter side as inputs and the control parameters as outputs, a control parameter training dataset is constructed and trained to obtain a control parameter prediction model. Using grid-side inductance, capacitance, grid impedance, and system stability labels as inputs and control gain as output, a control gain training dataset is constructed and trained to obtain a control gain prediction model. The LCL filter-side inductance, grid-side inductance, capacitance, and grid impedance of the LCL grid-connected inverter under the current operating conditions are collected. The control parameter prediction model and control gain prediction model are input for prediction, and the control parameters and control gain are output. These are then input into the stability judgment model. When the output system stability label represents instability, the control parameters and control gain are re-predicted by changing the LCL filter-side inductance or grid impedance. The system stability label output by the stability determination model represents stability. The control parameters and control gain output at this point are then used as the optimal parameter combination for the stable operation of the LCL grid-connected inverter.
2. The method for improving the stability of an LCL-type grid-connected inverter according to claim 1, characterized in that, The determination of the control parameters representing the inductor state on the LCL filter side and the control gain representing the inductor state on the grid side under different operating conditions specifically includes: Data was collected from the LCL-type grid-connected inverter under different operating conditions, including DC bus voltage. U dc LCL filter side inductor L xi , grid-side inductor L xg ,capacitance C x and grid impedance L xwg Grid voltage U PCCx ; The control scheme includes controlling the LCL-type grid-connected inverter using the DD-Σ control algorithm, with the control law being: ； ； ； ； in, T s For the switching cycle, K p1 For control parameters, K p2 To control the gain, i xi ( n ) is the first n Inverter-side inductor current at each sampling time, i xg ( n ) is the first n The grid-connected current at each sampling time, I xgref This indicates the reference value for grid-connected current.
3. The method for improving the stability of an LCL-type grid-connected inverter according to claim 2, characterized in that, The stability determination model specifically includes: Using control parameters, control gain, LCL filter-side inductance, grid-side inductance, capacitance, and grid impedance as inputs, the system stability is labeled, and an A-model is constructed by establishing a feature space with the system stability labels as outputs. The feature space with system stability labels as output is established as follows: ； Where 1 represents stable and 0 represents unstable; The SVM model is trained using the system stability training set data to obtain a stability determination model.
4. The method for improving the stability of an LCL-type grid-connected inverter according to claim 1, characterized in that, The obtained control parameter prediction model includes: Using the inductance, capacitance, and system stability labels of the LCL filter side as inputs, an SVM1 model is constructed by establishing spatial features with control parameters as outputs. The feature space with control parameters as output is established as follows: ； The SVM1 model is trained using the control parameter training set data to obtain the control parameter prediction model.
5. The method for improving the stability of an LCL-type grid-connected inverter according to claim 1, characterized in that, The obtained control gain prediction model includes: Using grid-side inductance, capacitance, grid impedance, and system stability labels as inputs, an SVM2 model is constructed by establishing spatial features with control gain as the output. The feature space with control gain as the output is established as follows: ； The SVM2 model is trained using the control gain training set data to obtain the control gain prediction model.
6. The method for improving the stability of an LCL-type grid-connected inverter according to claim 1, characterized in that, The re-prediction of control parameters and control gain includes: When the output of the stability determination model is 0, i.e. the system is unstable, the input control parameters and control gain are reselected through cross-validation until the output of the stability determination model corresponding to the selected control parameters and control gain is 1.
7. The method for improving the stability of an LCL-type grid-connected inverter according to claim 6, characterized in that, The method of reselecting the input control parameters and control gain through cross-validation specifically includes: pass k Cross-validation divides the system stability training dataset into two parts. k Each subset consists of a set of parameters, each containing a combination of control parameters and control gain; for each candidate parameter combination, the following steps are performed: k Independent rounds of SVM model training and validation set evaluation; Each time, select one parameter combination as the validation set, and the rest... k -1 parameters are combined into a training set; the SVM model is trained using the training set, and its performance is evaluated on the validation set; For each parameter combination, the classification accuracy, recall, and stability of the stability determination model are calculated by comparing the system stability label output by the stability determination model with the actual system stability. F 1. Score indicator; Will k The classification accuracy, recall, and F1 score were averaged to obtain the result. k Average performance score of the second validation; The parameter combination with the highest average performance is selected as the optimal parameter combination for stable operation of the LCL-type grid-connected inverter.
8. The method for improving the stability of an LCL-type grid-connected inverter according to claim 1, characterized in that, It also includes cleaning and normalizing the data in the constructed system stability training dataset, control parameter training dataset, and control gain training dataset, scaling the data to the [0,1] interval.
9. A stability improvement system for an LCL-type grid-connected inverter, characterized in that, include: The control parameter and control gain determination module is used to determine the control parameters and control gain representing the inductor state on the LCL filter side and the grid-side inductor state under different operating conditions based on the control scheme for the stable operation of the LCL grid-connected inverter. The stability determination model construction module is used to construct and train a system stability training dataset by taking control parameters, control gain, LCL filter-side inductance, grid-side inductance, capacitance, and grid impedance as inputs and system stability labels as outputs, in order to obtain a stability determination model. The control parameter prediction model building module is used to construct a control parameter training dataset and train it to obtain a control parameter prediction model by taking the inductance, capacitance and system stability labels of the LCL filter side as inputs and the control parameters as outputs. The control gain prediction model building module is used to construct a control gain training dataset and train it to obtain a control gain prediction model by taking grid-side inductance, capacitance, grid impedance and system stability labels as inputs and control gain as output. The stability determination module collects the LCL filter-side inductance, grid-side inductance, capacitance, and grid impedance of the LCL grid-connected inverter under the current operating conditions. It inputs these parameters into the control parameter prediction model and the control gain prediction model for prediction, outputting control parameters and control gain. These are then input into the stability determination model. When the output system stability label indicates instability, the control parameters and control gain are re-predicted by changing the LCL filter-side inductance or the grid impedance. This process continues until the system stability label output by the stability determination model indicates stability. The output control parameters and control gain at this point are then used as the optimal parameter combination for stable operation of the LCL grid-connected inverter.