An Adaptive Method for Inverter Control Parameters Based on Dynamic Identification of Grid Impedance

Abstract

This invention discloses an adaptive method for inverter control parameters based on dynamic identification of grid impedance. The method includes: S1, acquiring operating data of the inverter under different disturbance currents Δi in different operating conditions; S2, building a hybrid neural network model based on the MATLAB neural network toolbox; S3, training the hybrid neural network model until it converges; S4, calculating the real-time inductance and resistance values ​​of the grid-connected terminals of each phase of the inverter online; and S5, adjusting the inverter's control parameters in real-time based on the real-time inductance and resistance values ​​of the grid-connected terminals of each phase of the inverter, so that the inverter's phase margin is maintained within a preset range. This invention identifies the inductance and resistance of each grid-connected terminal of the inverter through a neural network, allowing for real-time adjustment of the inverter's control parameters, thereby maintaining the inverter's phase margin within a preset range, significantly improving the inverter's stability, and making it suitable for large-scale application.

Description

Technical Field

[0001] This invention relates to the field of inverter control technology, and in particular to an adaptive method for inverter control parameters based on dynamic identification of grid impedance. Background Technology

[0002] Energy storage systems have become a driving force for energy revolution and transformation. In traditional energy storage systems, when the energy storage inverter is in grid-connected power supply mode, the controller parameters of the entire machine are fixed values. Generally, after laboratory testing, the control parameters are fixed.

[0003] However, under different grid environments, especially in weak grids, inverters in energy storage systems are prone to oscillation and resonance problems, which can lead to grid-connected system failure in severe cases, seriously affecting the stability of the energy storage system and limiting its large-scale application. Summary of the Invention

[0004] The purpose of this invention is to provide an adaptive method for inverter control parameters based on dynamic identification of grid impedance. This method uses a neural network to identify the inductance and resistance of each grid-connected terminal of the inverter, so as to change the inverter control parameters in real time, thereby maintaining the phase margin of the inverter within a preset range, greatly improving the stability of the inverter, and making it suitable for large-scale application.

[0005] To achieve the above objectives, this invention discloses an adaptive method for inverter control parameters based on dynamic identification of grid impedance. The inverter has three phases of grid-connected terminals, which are respectively connected to the power grid. Each phase of the inverter has an inductance and a resistance between itself and the power grid. The adaptive method for inverter control parameters based on dynamic identification of grid impedance includes the following steps:

[0006] S1. Obtain the operating data of the inverter under different disturbance currents Δi in different operating conditions. The operating data under different disturbance currents Δi in different operating conditions are randomly divided into training data and test data. The different operating conditions are the operating conditions under which the grid-connected terminal of each phase of the inverter has different inductance values ​​and different resistance values ​​between the grid and the grid. The operating data includes the pre-injection voltage U0, pre-injection current I0, post-injection voltage U1, and post-injection current I1 of the grid-connected terminal of each phase of the inverter. The pre-injection voltage U0 and pre-injection current I0 are the voltage and current values ​​of the grid-connected terminal of each phase of the inverter before the corresponding disturbance current Δi is injected. The post-injection voltage U1 and post-injection current I1 are the voltage and current values ​​of the grid-connected terminal of each phase of the inverter after the corresponding disturbance current Δi is injected.

[0007] S2. Build a hybrid neural network model based on the MATLAB Neural Network Toolbox. The hybrid neural network model includes an input layer, a Dropout layer, a neural network layer group, a fully connected layer, a weight layer, and an output layer. The neural network layer group includes at least two different types of neural network layers.

[0008] S3. Using the inductance and resistance values ​​of the grid-connected terminal of each phase of the inverter under different operating conditions in the training data, and the operating data under different disturbance currents Δi under different operating conditions as data input, and the inductance and resistance values ​​of the grid-connected terminal of each phase of the inverter under real-time operating conditions as data output, train the hybrid neural network model until the hybrid neural network model converges.

[0009] S4. Using the real-time voltage U and real-time current I of the grid-connected terminal of each phase of the inverter as the data input of the converged hybrid neural network model, calculate the real-time inductance and real-time resistance of the grid-connected terminal of each phase of the inverter online.

[0010] S5. Based on the real-time inductance and resistance values ​​of the grid-connected terminals of each phase of the inverter, adjust the control parameters of the inverter in real time to maintain the phase margin of the inverter within a preset range.

[0011] Preferably, step S1 specifically includes:

[0012] S11. Record the inductance and resistance values ​​between the grid-connected terminal of the inverter and the power grid at the current moment, and the inlet voltage U0 and inlet current I0 of the grid-connected terminal of the inverter at the current moment.

[0013] S12. At preset intervals, inject disturbance currents Δi of different frequencies into the power grid, and record the post-injection voltage U1 and post-injection current I1 of the grid-connected terminal of each phase of the inverter under the disturbance currents Δi of different frequencies.

[0014] S13. Adjust the inductance and resistance values ​​between the grid-connected terminal of each phase of the inverter and the power grid, and repeat the above steps to obtain the operating data of the inverter under different disturbance currents Δi in different operating conditions.

[0015] Preferably, step S1 further includes:

[0016] S101. Perform Fourier transform on the operating data of the inverter under different disturbance currents Δi in different operating conditions to obtain the harmonic content and fundamental frequency content corresponding to the pre-injection voltage U0, pre-injection current I0, post-injection voltage U1, and post-injection current I1 of each phase of the inverter in different operating conditions.

[0017] Preferably, step S101 further includes:

[0018] S1011. Normalize the inductance, resistance, pre-injection voltage U0, pre-injection current I0, post-injection voltage U1, and post-injection current I1 of the grid-connected terminal of each phase of the inverter, as well as the harmonic content and fundamental frequency content corresponding to the pre-injection voltage U0, pre-injection current I0, post-injection voltage U1, and post-injection current I1 of the grid-connected terminal of each phase of the inverter, to obtain the normalized inductance, resistance, pre-injection voltage U0, pre-injection current I0, post-injection voltage U1, and post-injection current I1 of the grid-connected terminal of each phase of the inverter, as well as the harmonic content and fundamental frequency content corresponding to the normalized pre-injection voltage U0, pre-injection current I0, post-injection voltage U1, and post-injection current I1 of the grid-connected terminal of each phase of the inverter.

[0019] Preferably, the neural network layer group includes GRU layers and LSTM layers.

[0020] Preferably, step S1011 specifically includes:

[0021] S10111. Map the inductance, resistance, pre-injection voltage U0, pre-injection current I0, post-injection voltage U1, and post-injection current I1 of each phase of the inverter, as well as the harmonic content and fundamental frequency content corresponding to the pre-injection voltage U0, pre-injection current I0, post-injection voltage U1, and post-injection current I1 of each phase of the inverter, onto the [0, 1] interval to obtain the normalized inductance, resistance, pre-injection voltage U0, pre-injection current I0, post-injection voltage U1, and post-injection current I1 of each phase of the inverter, as well as the harmonic content and fundamental frequency content corresponding to the pre-injection voltage U0, pre-injection current I0, post-injection voltage U1, and post-injection current I1 of each phase of the inverter.

[0022] Preferably, step S3 specifically includes:

[0023] S31. The normalized inductance, resistance, pre-injection voltage U0, pre-injection current I0, post-injection voltage U1, and post-injection current I1 of each phase of the inverter in the training data, as well as the harmonic content and fundamental frequency content corresponding to the normalized pre-injection voltage U0, pre-injection current I0, post-injection voltage U1, and post-injection current I1 of each phase of the inverter in the real-time operating condition, are used as data inputs and the normalized inductance and resistance of each phase of the inverter in the real-time operating condition are used as data outputs and imported into the hybrid neural network model.

[0024] S32. Use the backpropagation algorithm function in the MATLAB Neural Network Toolbox to train the hybrid neural network model until the hybrid neural network model converges.

[0025] Preferably, step S32 specifically includes:

[0026] S321. Train a hybrid neural network model using the backpropagation algorithm function in the MATLAB Neural Network Toolbox;

[0027] S322. Set the initial learning rate to 0.0001, randomly select a small batch of samples from the training data, and train the hybrid neural network model for a preset number of rounds using the Adam stochastic gradient descent algorithm to continuously update the weight matrix parameters and bias values ​​of the GRU layer and LSTM layer until the hybrid neural network model converges.

[0028] Preferably, after step S322, the method further includes:

[0029] S323. Using the post-injection voltage U1 and post-injection current I1 of the grid-connected terminal of each phase of the inverter in the test data as data input, calculate the predicted inductance and predicted resistance between the grid-connected terminal of each phase of the inverter and the power grid, verify the converged hybrid neural network model, and confirm that the current hybrid neural network model has converged when the maximum error MAX of the hybrid neural network model is less than or equal to 5% and the average error MAE is less than or equal to 3%.

[0030] Specifically, let R1 be the actual resistance between the grid-connected terminal of each phase of the inverter and the power grid, R0 be the predicted resistance between the grid-connected terminal of each phase of the inverter and the power grid, L1 be the actual inductance between the grid-connected terminal of each phase of the inverter and the power grid, and L0 be the predicted inductance between the grid-connected terminal of each phase of the inverter and the power grid. The maximum error resistance (MAX) of the hybrid neural network model is calculated using the following formula:

[0031]

[0032] The mean error (MAE) resistance of the hybrid neural network model is calculated using the following formula:

[0033]

[0034] The maximum error (MAX) inductance of the hybrid neural network model is calculated using the following formula:

[0035]

[0036] The mean error (MAE) of the inductance in the hybrid neural network model is calculated using the following formula:

[0037]

[0038] Compared with existing technologies, this invention trains a converged hybrid neural network model using operating data of the inverter under different disturbance currents Δi in various operating conditions. The real-time voltage U and real-time current I of the grid-connected terminal of each phase of the inverter are used as data inputs to the converged hybrid neural network model. The model calculates the real-time inductance and resistance values ​​of the grid-connected terminal of each phase of the inverter online. Based on these values, the control parameters of the inverter are adjusted in real time to maintain the phase margin within a preset range. The neural network identifies the inductance and resistance of each grid-connected terminal of the inverter, allowing for real-time changes to the inverter's control parameters, thereby maintaining the phase margin within the preset range and significantly improving the inverter's stability, making it suitable for large-scale application. Attached Figure Description

[0039] Figure 1 This is a flowchart of the inverter control parameter adaptive method based on dynamic identification of grid impedance according to the present invention.

[0040] Figure 2 This is a schematic diagram of harmonic injection at the grid-connected end of each phase of the inverter of the present invention;

[0041] Figure 3 This is a schematic diagram of voltage and current acquisition at the grid-connected terminal of each phase of the inverter of the present invention;

[0042] Figure 4 This is a schematic diagram of the hybrid neural network model of the present invention;

[0043] Figure 5 This is a schematic diagram illustrating the error convergence of the loss function during the training process of the hybrid neural network model of this invention. Detailed Implementation

[0044] To illustrate the technical content, structural features, objectives, and effects of the present invention in detail, the following description is provided in conjunction with the embodiments and accompanying drawings.

[0045] Please see Figures 1-5 As shown in this embodiment, the inverter control parameter adaptive method based on dynamic identification of grid impedance has three phase grid-connected terminals, which are respectively connected to the power grid. Each phase grid-connected terminal of the inverter has an inductance and a resistance between itself and the power grid. The inverter control parameter adaptive method based on dynamic identification of grid impedance includes the following steps:

[0046] S1. Obtain the operating data of the inverter under different disturbance currents Δi under different operating conditions. The operating data under different disturbance currents Δi under different operating conditions are randomly divided into training data and test data. The different operating conditions refer to the conditions where the grid-connected terminals of each phase of the inverter have different inductance values ​​and different resistance values ​​between them. The operating data includes the pre-injection voltage U0, pre-injection current I0, post-injection voltage U1, and post-injection current I1 of the grid-connected terminals of each phase of the inverter. The pre-injection voltage U0 and pre-injection current I0 are the voltage and current values ​​of the grid-connected terminals of each phase of the inverter before the corresponding disturbance current Δi is injected. The post-injection voltage U1 and post-injection current I1 are the voltage and current values ​​of the grid-connected terminals of each phase of the inverter after the corresponding disturbance current Δi is injected. It can be understood that the inductance and resistance values ​​of the grid-connected terminals of each phase of the inverter are the grid impedance.

[0047] S2. Build a hybrid neural network model based on the MATLAB Neural Network Toolbox. The hybrid neural network model includes an input layer, a Dropout layer, a neural network layer group, a fully connected layer, a weight layer, and an output layer. The neural network layer group includes at least two different types of neural network layers.

[0048] S3. Using the inductance and resistance values ​​of the grid-connected terminal of each phase of the inverter under different operating conditions in the training data, and the operating data under different disturbance currents Δi under different operating conditions as data input, and the inductance and resistance values ​​of the grid-connected terminal of each phase of the inverter under real-time operating conditions as data output, train the hybrid neural network model until the hybrid neural network model converges.

[0049] S4. Using the real-time voltage U and real-time current I of the grid-connected terminal of each phase of the inverter as the data input of the converged hybrid neural network model, calculate the real-time inductance and real-time resistance of the grid-connected terminal of each phase of the inverter online.

[0050] S5. Based on the real-time inductance and resistance values ​​of the grid-connected terminals of each phase of the inverter, adjust the control parameters of the inverter in real time to maintain the phase margin of the inverter within a preset range.

[0051] Preferably, step S1 specifically includes:

[0052] S11. Record the inductance and resistance values ​​between the grid-connected terminal of the inverter and the power grid at the current moment, and the inlet voltage U0 and inlet current I0 of the grid-connected terminal of the inverter at the current moment.

[0053] S12. At preset intervals, inject disturbance currents Δi of different frequencies into the power grid, and record the post-injection voltage U1 and post-injection current I1 of the grid-connected terminal of each phase of the inverter under the disturbance currents Δi of different frequencies.

[0054] S13. Adjust the inductance and resistance values ​​between the grid-connected terminal of each phase of the inverter and the power grid, and repeat the above steps to obtain the operating data of the inverter under different disturbance currents Δi in different operating conditions.

[0055] Preferably, step S1 further includes:

[0056] S101. Perform Fourier transform on the operating data of the inverter under different disturbance currents Δi in different operating conditions to obtain the harmonic content and fundamental frequency content corresponding to the pre-injection voltage U0, pre-injection current I0, post-injection voltage U1, and post-injection current I1 of each phase of the inverter in different operating conditions.

[0057] Preferably, step S101 further includes:

[0058] S1011. Normalize the inductance, resistance, pre-injection voltage U0, pre-injection current I0, post-injection voltage U1, and post-injection current I1 of the grid-connected terminal of each phase of the inverter, as well as the harmonic content and fundamental frequency content corresponding to the pre-injection voltage U0, pre-injection current I0, post-injection voltage U1, and post-injection current I1 of the grid-connected terminal of each phase of the inverter, to obtain the normalized inductance, resistance, pre-injection voltage U0, pre-injection current I0, post-injection voltage U1, and post-injection current I1 of the grid-connected terminal of each phase of the inverter, as well as the harmonic content and fundamental frequency content corresponding to the normalized pre-injection voltage U0, pre-injection current I0, post-injection voltage U1, and post-injection current I1 of the grid-connected terminal of each phase of the inverter.

[0059] Preferably, the neural network layer group includes GRU layers and LSTM layers.

[0060] Preferably, step S1011 specifically includes:

[0061] S10111. Map the inductance, resistance, pre-injection voltage U0, pre-injection current I0, post-injection voltage U1, and post-injection current I1 of each phase of the inverter, as well as the harmonic content and fundamental frequency content corresponding to the pre-injection voltage U0, pre-injection current I0, post-injection voltage U1, and post-injection current I1 of each phase of the inverter, onto the [0, 1] interval to obtain the normalized inductance, resistance, pre-injection voltage U0, pre-injection current I0, post-injection voltage U1, and post-injection current I1 of each phase of the inverter, as well as the harmonic content and fundamental frequency content corresponding to the pre-injection voltage U0, pre-injection current I0, post-injection voltage U1, and post-injection current I1 of each phase of the inverter.

[0062] Preferably, step S3 specifically includes:

[0063] S31. The normalized inductance, resistance, pre-injection voltage U0, pre-injection current I0, post-injection voltage U1, and post-injection current I1 of each phase of the inverter in the training data, as well as the harmonic content and fundamental frequency content corresponding to the normalized pre-injection voltage U0, pre-injection current I0, post-injection voltage U1, and post-injection current I1 of each phase of the inverter in the real-time operating condition, are used as data inputs and the normalized inductance and resistance of each phase of the inverter in the real-time operating condition are used as data outputs and imported into the hybrid neural network model.

[0064] S32. Use the backpropagation algorithm function in the MATLAB Neural Network Toolbox to train the hybrid neural network model until the hybrid neural network model converges.

[0065] Preferably, step S32 specifically includes:

[0066] S321. Train a hybrid neural network model using the backpropagation algorithm function in the MATLAB Neural Network Toolbox;

[0067] S322. Set the initial learning rate to 0.0001, randomly select a small batch of samples from the training data, and train the hybrid neural network model for a preset number of rounds using the Adam stochastic gradient descent algorithm to continuously update the weight matrix parameters and bias values ​​of the GRU layer and LSTM layer until the hybrid neural network model converges.

[0068] Preferably, after step S322, the method further includes:

[0069] S323. Using the post-injection voltage U1 and post-injection current I1 of the grid-connected terminal of each phase of the inverter in the test data as data input, calculate the predicted inductance and predicted resistance between the grid-connected terminal of each phase of the inverter and the power grid, verify the converged hybrid neural network model, and confirm that the current hybrid neural network model has converged when the maximum error MAX of the hybrid neural network model is less than or equal to 5% and the average error MAE is less than or equal to 3%.

[0070] Specifically, let R1 be the actual resistance between the grid-connected terminal of each phase of the inverter and the power grid, R0 be the predicted resistance between the grid-connected terminal of each phase of the inverter and the power grid, L1 be the actual inductance between the grid-connected terminal of each phase of the inverter and the power grid, and L0 be the predicted inductance between the grid-connected terminal of each phase of the inverter and the power grid. The maximum error resistance (MAX) of the hybrid neural network model is calculated using the following formula:

[0071]

[0072] The mean error (MAE) resistance of the hybrid neural network model is calculated using the following formula:

[0073]

[0074] The maximum error (MAX) inductance of the hybrid neural network model is calculated using the following formula:

[0075]

[0076] The mean error (MAE) of the inductance in the hybrid neural network model is calculated using the following formula:

[0077]

[0078] It is worth noting that since the inverter has three phases, all the data described in this embodiment are collected for each phase. The maximum error (MAX) resistance, average error (MAE) resistance, maximum error (MAX) inductance, and average error (MAE) inductance all need to be collected for each phase. Of course, for inverters with other numbers of phases, the data is collected for the corresponding number of phases, which will not be elaborated here.

[0079] Combination Figures 1-5This invention trains a converged hybrid neural network model using operating data of the inverter under different disturbance currents Δi in various operating conditions. The real-time voltage U and real-time current I at the grid-connected terminal of each phase of the inverter are used as data inputs to the converged hybrid neural network model. The model calculates the real-time inductance and resistance values ​​at the grid-connected terminal of each phase of the inverter online. Based on these values, the control parameters of the inverter are adjusted in real-time to maintain the phase margin within a preset range. The neural network identifies the inductance and resistance at each grid-connected terminal of the inverter, allowing for real-time adjustments to the inverter's control parameters, thereby maintaining the phase margin within the preset range and significantly improving the inverter's stability, making it suitable for large-scale application.

[0080] The above-disclosed embodiments are merely preferred embodiments of the present invention and should not be construed as limiting the scope of the present invention. Therefore, any equivalent variations made in accordance with the claims of the present invention are still within the scope of the present invention.

Claims

1. An adaptive method for inverter control parameters based on dynamic identification of grid impedance, wherein the inverter has three phase grid-connected terminals, the three phase grid-connected terminals of the inverter are respectively connected to the power grid, and each phase grid-connected terminal of the inverter has an inductance and a resistance between itself and the power grid, characterized in that, The inverter control parameter adaptive method based on dynamic identification of grid impedance includes the following steps: The operation data of the inverter under different disturbance currents Δi under different operating conditions are obtained. The operation data under different disturbance currents Δi under different operating conditions are randomly divided into training data and test data. The different operating conditions are the operating conditions under which the grid-connected terminal of each phase of the inverter has different inductance values ​​and different resistance values ​​between the grid and the grid. The operation data includes the pre-injection voltage U0, pre-injection current I0, post-injection voltage U1, and post-injection current I1 of the grid-connected terminal of each phase of the inverter. The pre-injection voltage U0 and pre-injection current I0 are the voltage and current values ​​of the grid-connected terminal of each phase of the inverter before the corresponding disturbance current Δi is injected. The post-injection voltage U1 and post-injection current I1 are the voltage and current values ​​of the grid-connected terminal of each phase of the inverter after the corresponding disturbance current Δi is injected. A hybrid neural network model is built based on the MATLAB Neural Network Toolbox. The hybrid neural network model includes an input layer, a Dropout layer, a neural network layer group, a fully connected layer, a weight layer, and an output layer. The neural network layer group includes at least two different types of neural network layers. The hybrid neural network model is trained by taking the inductance and resistance values ​​of the grid-connected terminal of each phase of the inverter under different operating conditions in the training data, and the operating data under different disturbance currents Δi in different operating conditions as data inputs, and the inductance and resistance values ​​of the grid-connected terminal of each phase of the inverter under real-time operating conditions as data outputs, until the hybrid neural network model converges. Using the real-time voltage U and real-time current I at the grid-connected terminal of each phase of the inverter as the data input of the converged hybrid neural network model, the real-time inductance and real-time resistance at the grid-connected terminal of each phase of the inverter are calculated online. Based on the real-time inductance and resistance values ​​at the grid-connected terminal of each phase of the inverter, the control parameters of the inverter are adjusted in real time to maintain the phase margin of the inverter within a preset range.

2. The inverter control parameter adaptive method based on dynamic identification of grid impedance as described in claim 1, characterized in that, The acquisition of operating data of the inverter under different disturbance currents Δi in different operating conditions specifically includes: Record the inductance and resistance values ​​between the grid-connected terminal of the inverter and the power grid at each phase at the current moment, as well as the inlet voltage U0 and inlet current I0 of the grid-connected terminal of the inverter at each phase at the current moment. At preset time intervals, a disturbance current Δi of different frequencies is injected into the power grid, and the voltage U1 and current I1 of the grid-connected terminal of each phase of the inverter under the disturbance current Δi of different frequencies are recorded. Adjust the inductance and resistance values ​​between the grid-connected terminal of each phase of the inverter and the power grid, and repeat the above steps to obtain the operating data of the inverter under different disturbance currents Δi in different operating conditions.

3. The inverter control parameter adaptive method based on dynamic identification of grid impedance as described in claim 2, characterized in that, The step of acquiring the operating data of the inverter under different disturbance currents Δi in different operating conditions further includes: Fourier transform is performed on the operating data of the inverter under different disturbance currents Δi in different operating conditions to obtain the harmonic content and fundamental frequency content corresponding to the pre-injection voltage U0, pre-injection current I0, post-injection voltage U1, and post-injection current I1 of each phase of the inverter in different operating conditions.

4. The inverter control parameter adaptive method based on dynamic identification of grid impedance as described in claim 3, characterized in that, The step of performing a Fourier transform on the operating data of the inverter under different disturbance currents Δi in different operating conditions to obtain the harmonic content and fundamental frequency content corresponding to the pre-injection voltage U0, pre-injection current I0, post-injection voltage U1, and post-injection current I1 of each phase of the inverter under different operating conditions further includes: The inductance, resistance, pre-injection voltage U0, pre-injection current I0, post-injection voltage U1, and post-injection current I1 of each phase of the inverter, as well as the harmonic content and fundamental frequency content corresponding to the pre-injection voltage U0, pre-injection current I0, post-injection voltage U1, and post-injection current I1 of each phase of the inverter, are normalized to obtain the normalized harmonic content and fundamental frequency content corresponding to the inductance, resistance, pre-injection voltage U0, pre-injection current I0, post-injection voltage U1, and post-injection current I1 of each phase of the inverter.

5. The inverter control parameter adaptive method based on dynamic identification of grid impedance as described in claim 4, characterized in that, The neural network layer group includes GRU layers and LSTM layers.

6. The inverter control parameter adaptive method based on dynamic identification of grid impedance as described in claim 5, characterized in that, The normalization process is performed on the inductance, resistance, pre-injection voltage U0, pre-injection current I0, post-injection voltage U1, and post-injection current I1 of the grid-connected terminal of each phase of the inverter, as well as the harmonic content and fundamental frequency content corresponding to the pre-injection voltage U0, pre-injection current I0, post-injection voltage U1, and post-injection current I1 of the grid-connected terminal of each phase of the inverter, to obtain the normalized inductance, resistance, pre-injection voltage U0, pre-injection current I0, post-injection voltage U1, and post-injection current I1 of the grid-connected terminal of each phase of the inverter, as well as the normalized harmonic content and fundamental frequency content corresponding to the pre-injection voltage U0, pre-injection current I0, post-injection voltage U1, and post-injection current I1 of the grid-connected terminal of each phase of the inverter, specifically including: The inductance, resistance, pre-injection voltage U0, pre-injection current I0, post-injection voltage U1, and post-injection current I1 of each phase of the inverter, as well as the harmonic content and fundamental frequency content corresponding to the pre-injection voltage U0, pre-injection current I0, post-injection voltage U1, and post-injection current I1 of each phase of the inverter, are mapped onto the [0, 1] interval to obtain the normalized inductance, resistance, pre-injection voltage U0, pre-injection current I0, post-injection voltage U1, and post-injection current I1 of each phase of the inverter, as well as the harmonic content and fundamental frequency content corresponding to the pre-injection voltage U0, pre-injection current I0, post-injection voltage U1, and post-injection current I1 of each phase of the inverter.

7. The inverter control parameter adaptive method based on dynamic identification of grid impedance as described in claim 6, characterized in that, The process involves using the inductance and resistance values ​​of the grid-connected terminal of each phase of the inverter under different operating conditions in the training data, and the operating data under different disturbance currents Δi under different operating conditions, as data inputs, and using the inductance and resistance values ​​of the grid-connected terminal of each phase of the inverter under real-time operating conditions as data outputs, to train the hybrid neural network model until the hybrid neural network model converges. Specifically, this includes: The normalized inductance, resistance, pre-injection voltage U0, pre-injection current I0, post-injection voltage U1, and post-injection current I1 of each phase of the inverter in the training data, as well as the harmonic content and fundamental frequency content corresponding to the normalized pre-injection voltage U0, pre-injection current I0, post-injection voltage U1, and post-injection current I1 of each phase of the inverter in the grid-connected terminal, are used as data inputs. The normalized inductance and resistance of each phase of the inverter in the real-time operating condition are used as data outputs and imported into the hybrid neural network model. Train the hybrid neural network model using the backpropagation algorithm function in the MATLAB Neural Network Toolbox until the hybrid neural network model converges.

8. The inverter control parameter adaptive method based on dynamic identification of grid impedance as described in claim 7, characterized in that, The process of training a hybrid neural network model using the backpropagation algorithm function in the MATLAB Neural Network Toolbox until the hybrid neural network model converges specifically includes: Train a hybrid neural network model using the backpropagation algorithm function in the MATLAB Neural Network Toolbox; The initial learning rate is set to 0.0001. A small batch of samples is randomly selected from the training data. The hybrid neural network model is trained for a preset number of rounds using the Adam stochastic gradient descent algorithm to continuously update the weight matrix parameters and bias values ​​of the GRU and LSTM layers until the hybrid neural network model converges.

9. The inverter control parameter adaptive method based on dynamic identification of grid impedance as described in claim 8, characterized in that, The initial learning rate is set to 0.0001. A mini-batch of samples is randomly selected from the training data. The hybrid neural network model is trained for a preset number of rounds using the Adam stochastic gradient descent algorithm to continuously update the weight matrix parameters and bias values ​​of the GRU and LSTM layers until the hybrid neural network model converges. This process then includes: Using the post-injection voltage U1 and post-injection current I1 of each phase of the inverter in the test data as data input, the predicted inductance and predicted resistance between the grid-connected terminal of each phase of the inverter and the power grid are calculated. The converged hybrid neural network model is verified, and the current hybrid neural network model is confirmed to have converged when the maximum error MAX of the hybrid neural network model is less than or equal to 5% and the average error MAE is less than or equal to 3%.

10. The inverter control parameter adaptive method based on dynamic identification of grid impedance as described in claim 9, characterized in that, Let R1 be the actual resistance between the grid-connected terminal of each phase of the inverter and the power grid, R0 be the predicted resistance between the grid-connected terminal of each phase of the inverter and the power grid, L1 be the actual inductance between the grid-connected terminal of each phase of the inverter and the power grid, and L0 be the predicted inductance between the grid-connected terminal of each phase of the inverter and the power grid. Calculate the maximum error resistance (MAX) of the hybrid neural network model using the following formula: The mean error (MAE) resistance of the hybrid neural network model is calculated using the following formula: The maximum error (MAX) inductance of the hybrid neural network model is calculated using the following formula: The mean error (MAE) of the inductance in the hybrid neural network model is calculated using the following formula:

Images