Online stability analysis method for new energy grid-connected system based on neural network

Abstract

The application discloses a neural network-based online stability analysis method for a new energy grid-connected system. A preset new energy grid-connected system structure under offline data is used to deduce a mathematical model for judging the stability of the grid-connected system, to obtain a function relationship between the inverter side admittance, the grid system impedance and the return rate matrix and the preset new energy grid-connected system steady-state value and frequency, and to reconstruct a neural network framework accordingly. A neural network offline simulation model is established by using the preset new energy grid-connected system, a new energy grid-connected system offline stability analysis model is obtained by inputting the reconstructed neural network with a data set obtained based on simulation, an online stability analysis model for the new energy grid-connected system is obtained by using transfer learning for training, and the online stability analysis model for the new energy grid-connected system is used to analyze the stability of the grid-connected system under a steady-state working point based on a small amount of online measurement data. The application realizes online stability analysis of the new energy grid-connected system by using a small amount of online measurement data.

Description

Technical Field

[0001] This invention relates to the field of new energy power generation technology, and in particular to an online stability analysis method, terminal, and computer-readable storage medium for a new energy grid-connected system based on neural networks. Background Technology

[0002] With the rapid development of renewable energy, the integration of numerous power electronic devices has reduced system inertia and damping, posing a severe challenge to the stable operation of power systems. In stability analysis studies of renewable energy grid-connected systems, impedance analysis divides the grid-connected system into two subsystems by establishing Norton models for the grid-connected inverter and Thevenin models for the grid system. Further stability analysis is then performed using the generalized Nyquist criterion, reducing the difficulty of stability analysis and facilitating the analysis of grid-connected system instability mechanisms. Therefore, it has become one of the mainstream methods for stability analysis of renewable energy grid-connected systems. However, actual renewable energy grid-connected systems are often "black box" or "grey box" models, making it difficult to obtain complete control structures and parameters, which greatly hinders stability analysis. Furthermore, even with all system parameters obtained, the derivation process of the impedance model is complex, often requiring offline data for stability analysis results, resulting in significant time lags in guiding system parameter design. Therefore, it is essential to establish a method for online analysis of the stability of black-box grid-connected systems.

[0003] Impedance analysis, derived mathematically, is commonly used to perform stability analysis on grid-connected systems. Different linearization methods can be employed to obtain the stability of the grid-connected inverter and the power grid system. dq Impedance and admittance models in the black-box domain and order domain are used, and stability is analyzed using the generalized Nyquist criterion. This process requires a complete control structure and parameters of the system, is complex in derivation, and is computationally slow, allowing only offline analysis using existing data. A neural network-based stability analysis method simply requires inputting measured data into the neural network, continuously optimizing the weights through training, and finally obtaining the stability analysis results for the black-box system. However, the accuracy of this method depends on the amount of data used in model training; a limited amount of measurement data can lead to overfitting, thus affecting the accuracy of the stability analysis.

[0004] Patent application CN114611676A, entitled "Impedance Model Identification Method and System for New Energy Power Generation Systems Based on Neural Networks," includes: obtaining and normalizing training and test datasets of a new energy power generation system at different steady-state operating points using a frequency sweep method; training a neural network using the training dataset to obtain a neural network with impedance characteristics of the new energy power generation system; inputting the input data of the test dataset into the neural network to obtain impedance identification results; calculating the mean square error by combining the output data of the test dataset, adjusting the number of hidden layers and the number of neurons in each hidden layer of the neural network so that the mean square error is less than a set threshold; and inputting input data at any steady-state operating point into the obtained neural network to obtain the impedance output at the corresponding operating point. This impedance identification method normalizes and preprocesses the dataset to the same value range, which speeds up the calculation to some extent, but still has the following drawbacks: This impedance identification method still relies on a large dataset. Insufficient datasets can lead to the mean square error not meeting the conditions, increasing the computational load and hindering online impedance identification. Furthermore, this method only performs impedance identification at the renewable energy sending end, i.e., the inverter side, without analyzing the impedance at the receiving end of the grid, which is detrimental to the stability analysis of the grid-connected system. Summary of the Invention

[0005] The purpose of this invention is to address the problems of existing technologies by providing an online stability analysis method for new energy grid-connected systems based on neural networks. This method utilizes a small amount of online measurement data to train an online stability analysis model for the new energy grid-connected system, thereby performing online stability analysis and solving the problem of lag in offline stability analysis caused by real-time changes in operating points in actual engineering projects.

[0006] A method for online stability analysis of a renewable energy grid-connected system based on neural networks, comprising the following steps: S1. Using the preset new energy grid-connected system under offline data, derive the mathematical model for determining the stability of the grid-connected system, and obtain the functional relationship between the inverter side admittance, grid system impedance and hysteresis matrix and the preset new energy grid-connected system steady-state value and frequency; S2. Reconstruct the neural network based on the functional relationship between the inverter side admittance, grid system impedance and hysteresis matrix and the preset steady-state value and frequency of the new energy grid-connected system; S3. Establish an offline simulation model of the target new energy grid-connected system based on the preset new energy grid-connected system, and obtain a dataset for offline stability analysis of the target new energy grid-connected system through offline simulation; S4. Input the offline stability analysis dataset of the target new energy grid-connected system into the reconstructed neural network for training to obtain the offline stability analysis model of the target new energy grid-connected system; S5. Using transfer learning, an online stability analysis model for the target new energy grid-connected system is trained based on the offline stability analysis model of the target new energy grid-connected system; S6. Input the dataset of the online stability analysis of the new energy grid-connected system to be analyzed obtained by measurement into the online stability analysis model of the target new energy grid-connected system to obtain the online grid-connected inverter admittance model and online return matrix of the new energy grid-connected system to be analyzed; S7. Perform online stability analysis on the online cyclicity matrix according to the generalized Nyquist criterion.

[0007] The online stability analysis of the online retracement matrix based on the generalized Nyquist criterion includes: Calculate and output the phase margin of the online renewable energy grid-connected system to be analyzed at the current time; if the phase margin is less than the threshold, output the online inverter admittance model, perform frequency domain analysis on the online inverter admittance model, correct the inverter control framework or control parameters, and return to step S6 to re-perform the online stability analysis of the renewable energy grid-connected system to be analyzed; otherwise, output the phase margin of the online renewable energy grid-connected system to be analyzed and end.

[0008] The present invention provides an online stability analysis method for new energy grid-connected systems based on neural networks. It utilizes the structure of new energy grid-connected systems under offline data to derive a mathematical model for determining the stability of the grid-connected system. Based on the mathematical model of the new energy grid-connected system, the neural network framework is reconstructed. Online stability analysis of the new energy grid-connected system is achieved using a small amount of online measurement data. This solves the problems of inaccurate model output caused by the small amount of dataset in actual engineering or the increased computation time required to obtain sufficient dataset.

[0009] The present invention provides an online stability analysis method for new energy grid-connected systems based on neural networks. It establishes a neural network model using a general new energy grid-connected system, establishes an offline stability analysis model for the new energy grid-connected system based on offline simulation, and obtains an online stability analysis model for the new energy grid-connected system by training it using transfer learning. The method performs stability analysis on the grid-connected system under the steady-state operating point, solving the problem of difficulty or inaccuracy in stability analysis of black-box systems under changing operating points in actual engineering. Attached Figure Description

[0010] Figure 1 This is a flowchart of an online stability analysis method for a new energy grid-connected system based on a neural network, according to an embodiment of the present invention.

[0011] Figure 2 This is a schematic diagram of the structure of a preset new energy grid connection system provided in an embodiment of the present invention.

[0012] Figure 3The equivalent block diagram provided for embodiments of the present invention does not consider phase-locked loops.

[0013] Figure 4 This is a block diagram of an inverter control system provided in an embodiment of the present invention.

[0014] Figure 5 A diagram of the reconstructed neural network framework provided in an embodiment of the present invention. Detailed Implementation

[0015] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below in conjunction with the embodiments of the present invention. Obviously, the described embodiments are some embodiments of the present invention, but not all embodiments.

[0016] Please see Figures 1 to 5 As shown in the figure, an online stability analysis method for a new energy grid-connected system based on a neural network provided by an embodiment of the present invention includes the following steps: S1. Using the preset new energy grid-connected system structure under offline data, a mathematical model for determining the stability of the grid-connected system is derived, and the functional relationships between the inverter-side admittance, grid system impedance, and hysteresis matrix and the preset steady-state values ​​and frequency of the grid-connected system are obtained: In this embodiment of the invention, a current-dual closed-loop three-phase LCL grid-connected inverter is used as the basis for the preset new energy grid-connected system. Figure 2 This diagram illustrates the overall control structure of the pre-designed renewable energy grid-connected system. The LCL filter is composed of… , C and composition, , , Inductance that forms a three-phase power grid DC side voltage, For the common coupling point voltage, This refers to the inductor current on the inverter bridge side. For inverter output current, For capacitor current, This is the active damping feedback coefficient. q Output phase angle for PLL , For current loop dq The axis is given a reference value.

[0017] The process of deriving a mathematical model for determining the stability of the grid-connected system based on the preset new energy grid-connected system structure, and deriving the functional relationships between the grid-connected inverter-side admittance, grid system impedance, hysteresis matrix, and preset steady-state values ​​and frequencies for determining the stability of the grid-connected system, is as follows: First, the admittance of the grid-connected inverter is derived, considering only the LCL filter, current loop control, active damping, and pulse width modulation process. After block diagram simplification, an equivalent control block diagram without considering the phase-locked loop is obtained. The steps for establishing the equivalent control block diagram without considering the phase-locked loop are as follows: For the current loop proportional-integral element, K C For active damping compensation coefficient, in dq In the coordinate system, its transfer function can be represented by a 2×2 matrix, and the specific expression is as follows: ; ; Because the LCL filter stage is located in the inverter abc A three-phase stationary coordinate system, through the Park transformation dq LCL filtering stage in coordinate system , C and They can be used respectively with transfer functions , , The specific expression is: ; ; ; The time delay of the grid-connected inverter control system is dq Available in coordinate system The expression is: ; in, , Ts The sampling period.

[0018] After simplification, the equivalent control block diagram of the grid-connected inverter, without considering the phase-locked loop, is as follows: Figure 3 As shown.

[0019] by For input, For the output, the admittance of the grid-connected inverter, neglecting the influence of the phase-locked loop, can be obtained as follows: ; in, ; ; .

[0020] The inverter admittance is further modified to consider the influence of the phase-locked loop (PLL), resulting in an equivalent small-signal model considering the PLL. The steps for establishing the small-signal model considering the PLL are as follows: When a small-signal disturbance is introduced into the grid voltage, due to the presence of a phase-locked loop (PLL), two sets of coordinate systems with different meanings exist in the grid-connected system. dq Coordinate System and Control dq There are small perturbation phase angles between the coordinate systems. θ, U d 、U q For the voltage at point PCC d shaft and q Axial components ，I 2d 、I 2q For the inverter output current d shaft and q Axial components ， Including superscript " s The quantity "" represents the actual power grid parameters in the system coordinate system, including the superscript "". c The quantity represented by “” indicates the parameter acting on the control system, including “ The quantity of “” is a small signal component and does not contain “”. The quantity is the steady-state component. The voltage at point PCC and the grid-connected current are in two... dq The relationships in the coordinate system are as follows: ; ; The small disturbance phase angle of the phase-locked loop output can be expressed as: ; In the formula: ,in , These are the PI parameters for a phase-locked loop proportional-integral controller.

[0021] Substitute the above equation back to eliminate the control. dq The grid voltage in the coordinate system, neglecting the secondary small disturbance component, can be obtained as follows: θ With system dq The expression for the relationship between grid voltages in the coordinate system is as follows: ; make ; Substituting the above equation back, we can obtain the system that takes into account the dynamic effects of the phase-locked loop. dq Coordinate System and Control dq Modulation voltage in coordinate system Relationship and grid current The relationship is: ; ; Based on the above derivation, the equivalent small-signal block diagram of the inverter considering the influence of the phase-locked loop is as follows: Figure 4 As shown. For input, For the output, the simplified block diagram yields the grid-connected inverter admittance considering the phase-locked loop effect as follows: ; in: ; ; ; The admittance of the grid-connected inverter is transformed into the following form: ; in ; ; ; ; ; in For matrix The four corresponding elements.

[0022] The transformed admittance of the grid-connected inverter can be expressed as a function of the steady-state value, i.e., defined as follows: ; in, R 1(s), R 2(s), R 3(s), R 4(s), R 5(s) represents the frequency domain component in the simplified admittance of the grid-connected inverter that is independent of the steady-state value.

[0023] Furthermore, the impedance of the power grid is deduced. dq The mathematical model of a three-phase power grid system in the coordinate system is expressed as follows: ; in, These are the mains voltages. d, q Axial components, for dq Power grid impedance matrix in coordinate system The four corresponding elements.

[0024] Under a three-phase balanced power grid dq The power grid impedance in the coordinate system presents a symmetric matrix, that is: ; Expanding and simplifying, we get: ; ; The impedance of the power grid system can be expressed as a function of its steady-state value, i.e., defined as follows: ; Define the return matrix L dq for: ; The return matrix can be expressed as a function of the steady-state value, that is: .

[0025] S2. Reconstruct the neural network based on the functional relationship between the inverter side admittance, grid system impedance and hysteresis matrix and the preset steady-state value and frequency of the new energy grid-connected system; The neural network is reconstructed based on the functional relationship between the inverter-side admittance, grid system impedance and hysteresis matrix, and the steady-state value and frequency of the general grid-connected system. Figure 5 As shown, the reconstructed neural network consists of three hidden layers, one input layer, and one output layer. The input layer takes the steady-state value of the system as input. and frequency f The neural network input is divided into two parts and trained simultaneously. The first part contains two interconnected hidden layers, with the first hidden layer corresponding to a function. R 1( s ), R 2( s ), R 3( s ), R 4( s ), R 5( s ), s = j2πf The function corresponding to the second hidden layer G Y The second part includes the corresponding functions of a hidden layer. G Z The output of the second hidden layer of the first part yields the admittance model of the offline grid-connected inverter. The output of the second hidden layer is obtained The output layer calculates the outputs of the two parts, corresponding to the function. G LThe final output return matrix Meanwhile, the offline grid-connected inverter admittance model As the second output result.

[0026] S3. Establish an offline simulation model of the target new energy grid-connected system based on the preset new energy grid-connected system, and obtain a dataset for offline stability analysis of the target new energy grid-connected system through offline simulation; Specifically, an offline simulation model of the target renewable energy grid-connected system can be established in Matlab / Simulink. After establishing the offline simulation model, the dataset for offline stability analysis of the target renewable energy grid-connected system obtained from the offline simulation can be achieved through the following steps: In the offline simulation model of the target new energy grid-connected system, the frequency is injected at PCC. fp The current disturbance is 5% of the steady-state rated current; the inverter-side response voltage and current at the PCC are measured, and the disturbance frequency component is decomposed by Fourier transform. The inverter-side response voltage and current are then subjected to Park transform to... dq coordinate system, to obtain dq Offline grid-connected inverter admittance in coordinate system; measurement PCC The grid-side response voltage and current are analyzed, and a Fourier transform is performed to decompose the disturbance frequency components. Then, a Park transform is applied to the grid-side response voltage and current. dq coordinate system, to obtain dq The offline grid impedance in the coordinate system is used to establish the offline hysteresis matrix using the offline grid-connected inverter admittance and the offline grid impedance. Stability analysis is conducted in the 0-200Hz frequency band, and the injected disturbance frequency is selected. fp The above process is repeated, changing the frequency range from 0 to 200Hz at 1Hz intervals, while simultaneously changing the operating point of the grid-connected system, to establish an offline grid-connected inverter admittance dataset and an offline return rate matrix dataset. The obtained offline grid-connected inverter admittance dataset and offline return rate matrix dataset constitute the dataset for the offline stability analysis of the new energy grid-connected system.

[0027] S4. Input the offline stability analysis dataset of the target new energy grid-connected system into the reconstructed neural network for training to obtain the offline stability analysis model of the target new energy grid-connected system; When the offline stability analysis dataset of the new energy grid-connected system is input into the reconstructed neural network, 80% of the data is randomly selected as the training dataset and the remaining 20% ​​of the dataset is used as the test dataset. The model is trained and tested to finally obtain the offline stability analysis model of the new energy grid-connected system.

[0028] S5. Using transfer learning, obtaining an online stability analysis model of the target new energy grid-connected system based on the offline stability analysis model of the target new energy grid-connected system; Obtaining the online stability analysis model of the new energy grid-connected system through training by transfer learning, and obtaining online grid-connected inverter admittance and an online return ratio matrix, comprising: By utilizing the idea of transfer learning, domain data consists of input X feature space and marginal probability distribution P(X ). With D s = {(x s1 , y s1 )，…，(x sn , y sn )} representing the data set for offline stability analysis of said new energy grid-connected system marked in the source domain, wherein the input X s is the operating point and frequency of the system, that is, , y s is the corresponding return ratio matrix, that is, L dq . Similarly, the data set for online stability analysis of said new energy grid-connected system in the target domain is represented as D t = {(x t1 , y t1 )，…，(x tm , y tm ) } , wherein m < n. In the transfer learning setting, if the data of the source domain and the target domain have the same distribution, that is, X s ≠X t , P(X s ) = P(X t )When the online stability analysis model of the new energy grid-connected system is established, it can achieve satisfactory application accuracy. Therefore, the method can ensure that the distribution of the dataset for the online stability analysis of the new energy grid-connected system is consistent with that of the dataset for the offline stability analysis of the new energy grid-connected system. In other words, by utilizing transfer learning, the amount of data required for the online stability analysis dataset of the new energy grid-connected system can be reduced while maintaining the accuracy of the model, thereby accelerating the calculation speed and meeting the real-time requirements of online stability analysis.

[0029] The online stability analysis dataset of the new energy grid-connected system is input into the online stability analysis model of the new energy grid-connected system for training, and the admittance and online return rate matrix of the online grid-connected inverter are obtained.

[0030] S6. Input the dataset of the online stability analysis of the new energy grid-connected system to be analyzed obtained by measurement into the online stability analysis model of the target new energy grid-connected system to obtain the online grid-connected inverter admittance model and online return matrix of the new energy grid-connected system to be analyzed; The dataset for the online stability analysis of the new energy grid-connected system to be analyzed was obtained through the following methods, and the implementation steps include: In the new energy grid-connected system to be analyzed, PCC Injection frequency fs The current disturbance is 5% of the steady-state rated current; measurement PCC The inverter-side response voltage and current are analyzed, and a Fourier transform is performed to decompose the disturbance frequency components. Then, a Parker transform is applied to the inverter-side response voltage and current. dq coordinate system, to obtain dq Online grid-connected inverter admittance in coordinate system; measurement PCC The grid-side response voltage and current are analyzed, and a Fourier transform is performed to decompose the disturbance frequency components. Then, a Park transform is applied to the grid-side response voltage and current. dq coordinate system, to obtain dq The online grid impedance in the coordinate system is used; an online hysteresis matrix is ​​established using the online grid-connected inverter admittance and the online grid impedance. Stability analysis is considered to be performed in the 0-200Hz frequency band. To reduce measurement time and accelerate the calculation speed of online stability analysis, the injected disturbance frequency is selected. fs The above process is repeated at 10Hz intervals within the range of 0-200Hz to establish an online grid-connected inverter admittance dataset and an online hysteresis matrix dataset. The online grid-connected inverter admittance dataset and online hysteresis matrix dataset obtained from multiple repeated measurements constitute the dataset for the online stability analysis of the new energy grid-connected system.

[0031] S7. Perform online stability analysis on the online cyclicity matrix according to the generalized Nyquist criterion.

[0032] Specifically, when performing online stability analysis on the online return rate matrix, the phase margin of the online renewable energy grid-connected system to be analyzed at the current time is first calculated and output. If the phase margin is less than the threshold, the online inverter admittance model is output, frequency domain analysis is performed on the online inverter admittance model, the inverter control framework or control parameters are corrected, and the process returns to step S6 to re-perform the online stability analysis of the renewable energy grid-connected system to be analyzed. Otherwise, the process ends after outputting the phase margin of the online renewable energy grid-connected system to be analyzed.

[0033] The modified inverter framework and control parameters include: adjusting the proportional-integral parameters of the phase-locked loop (PLL) to reduce the PLL bandwidth; adjusting the proportional-integral parameters of the current loop to reduce the current loop bandwidth; and adding stability control.

[0034] The above description is only a preferred embodiment of the present invention. It should be noted that, for those skilled in the art, several improvements and modifications can be made without departing from the principle of the present invention, and these improvements and modifications should also be considered within the scope of protection of the present invention.

Claims

1. A method for online stability analysis of renewable energy grid-connected systems based on neural networks, characterized in that, Including the following steps: S1. Using the preset new energy grid-connected system under offline data, derive the mathematical model for determining the stability of the grid-connected system, and obtain the functional relationship between the inverter side admittance, grid system impedance and hysteresis matrix and the preset new energy grid-connected system steady-state value and frequency; S2. Reconstruct the neural network based on the functional relationship between the inverter side admittance, grid system impedance and hysteresis matrix and the preset steady-state value and frequency of the new energy grid-connected system; S3. Establish an offline simulation model of the target new energy grid-connected system based on the preset new energy grid-connected system, and obtain a dataset for offline stability analysis of the target new energy grid-connected system through offline simulation; S4. Input the offline stability analysis dataset of the target new energy grid-connected system into the reconstructed neural network for training to obtain the offline stability analysis model of the target new energy grid-connected system; S5. Using transfer learning, an online stability analysis model for the target new energy grid-connected system is trained based on the offline stability analysis model of the target new energy grid-connected system; S6. Input the dataset of the online stability analysis of the new energy grid-connected system to be analyzed obtained by measurement into the online stability analysis model of the target new energy grid-connected system to obtain the online grid-connected inverter admittance model and online return matrix of the new energy grid-connected system to be analyzed; S7. Perform online stability analysis on the online retracement matrix according to the generalized Nyquist criterion, including: Calculate the phase margin of the online renewable energy grid-connected system to be analyzed at the current moment; If the phase margin is less than the threshold, the online inverter admittance model is output, frequency domain analysis is performed on the online inverter admittance model, the inverter control framework or control parameters are corrected, and the process returns to step S6 to re-perform the online stability analysis of the new energy grid-connected system to be analyzed; otherwise, the process ends after outputting the phase margin of the online new energy grid-connected system to be analyzed.

2. The online stability analysis method for new energy grid-connected systems based on neural networks according to claim 1, characterized in that, The inverter-side admittance is expressed as a function of the steady-state value, and is defined as follows: ； in, R 1( s ), R 2( s ), R 3( s ), R 4( s ), R 5( s () represents the frequency domain component of the inverter-side admittance that is independent of the steady-state value; Indicates the inverter-side admittance. G Y The function representing the steady-state value and the inverter-side admittance. The voltages at point PCC in the system coordinate system are respectively represented as follows: d, q Axial components and inverter output current in system coordinates d, q Axis components: The impedance of the power grid system is expressed as a function of its steady-state value, and is defined as follows: ， Indicates the impedance of the power grid system. G Z This represents the steady-state value and the impedance function of the power grid system. Represents the mains voltage d, q Axial components; The hysteresis matrix is ​​the product of the inverter-side admittance and the grid system impedance, expressed as a function of the steady-state value, and defined as follows: , Represents the return matrix, G L A function representing the steady-state value and the hysteresis matrix.

3. The online stability analysis method for new energy grid-connected systems based on neural networks according to claim 2, characterized in that, The neural network consists of three hidden layers, one input layer, and one output layer. The input layer takes the steady-state value of the system as input. and frequency f The neural network input is divided into two parts and trained simultaneously; the first part contains two connected hidden layers, and the first hidden layer corresponds to a function. R 1( s ), R 2( s ), R 3( s ), R 4( s ), R 5( s ), s = j2πf The function corresponding to the second hidden layer G Y The second part includes the function corresponding to one hidden layer. G Z The output of the second hidden layer of the first part yields the admittance model of the offline grid-connected inverter. The output of the second hidden layer is obtained The output layer calculates the outputs of the two parts, corresponding to the function. G L The final output return matrix Meanwhile, the offline grid-connected inverter admittance model As the second output result.

4. The online stability analysis method for new energy grid-connected systems based on neural networks according to claim 1, characterized in that, The aforementioned offline simulation model of the target new energy grid-connected system is established based on the preset new energy grid-connected system, and the dataset for offline stability analysis of the target new energy grid-connected system is obtained through offline simulation, including: S21. In the established offline simulation model of the target new energy grid-connected system, inject frequency at PCC. fp The current disturbance is 5% of the steady-state rated current; S22. Measure the inverter-side response voltage and current at the PCC, and perform a Fourier transform to decompose the disturbance frequency components. Then, perform a Parker transform on the inverter-side response voltage and current to... dq coordinate system, obtain dq Offline grid-connected inverter admittance in coordinate system; S23. Measure the grid-side response voltage and current at PCC, and perform a Fourier transform to decompose the disturbance frequency components. Then, perform a Parker transform on the grid-side response voltage and current to... dq coordinate system, obtain dq Offline power grid impedance in a coordinate system; S24. Establish an offline hysteresis matrix using the offline grid-connected inverter admittance and the offline grid impedance; S25. Change the injection frequency in 1Hz increments within the range of 0-200Hz. fp Simultaneously, the operating point of the grid-connected system is changed, and S21-S24 above are repeated to establish the offline grid-connected inverter admittance dataset and offline return matrix dataset of the target new energy grid-connected system, which serve as the dataset for offline stability analysis of the target new energy grid-connected system obtained from offline simulation.

5. The online stability analysis method for new energy grid-connected systems based on neural networks according to claim 1, characterized in that, The process of inputting the offline stability analysis dataset of the target renewable energy grid-connected system into the reconstructed neural network for training to obtain the offline stability analysis model of the target renewable energy grid-connected system includes: The offline stability analysis dataset of the target new energy grid-connected system is input into the reconstructed neural network. 80% of the data is randomly selected as the training dataset and 20% of the data is selected as the test dataset. The model is trained and tested to obtain the offline stability analysis model of the target new energy grid-connected system.

6. The online stability analysis method for new energy grid-connected systems based on neural networks according to claim 1, characterized in that, The dataset for the online stability analysis of the new energy grid-connected system to be analyzed is obtained through the following steps: S31. In the new energy grid-connected system to be analyzed, inject frequency at PCC. fs The current disturbance is 5% of the steady-state rated current; S32. Measure the inverter-side response voltage and current at the PCC, and perform a Fourier transform to decompose the disturbance frequency components. Then, perform a Parker transform on the inverter-side response voltage and current to... dq coordinate system, obtain dq Inverter admittance in the online grid-connected system; S33. Measure the grid-side response voltage and current at PCC, and perform a Fourier transform to decompose the disturbance frequency components. Then, perform a Park transform on the grid-side response voltage and current to... dq coordinate system, obtain dq Online power grid impedance in a coordinate system; S34. Establish an online hysteresis matrix using the online grid-connected inverter admittance and the online grid impedance; S35. Change the injection frequency of the disturbance within the range of 0-200Hz at 10Hz intervals. fs Repeat steps S31-S34 to establish the online grid-connected inverter admittance dataset and online return matrix dataset of the new energy grid-connected system to be analyzed, which will serve as the dataset for online stability analysis of the new energy grid-connected system to be analyzed obtained from measurements.

7. The online stability analysis method for new energy grid-connected systems based on neural networks according to claim 1, characterized in that, The method of using transfer learning to train an online stability analysis model for the target renewable energy grid-connected system based on an offline stability analysis model includes: Using the concept of transfer learning, domain data is transformed from input... X Feature space and marginal probability distribution P(X Composed of, using D s = {(x s1 , y s1 )，…，(x sn , y sn )} The dataset representing the offline stability analysis of the new energy grid-connected system with source domain labeling. X sn For the system operating point and frequency, , y sn The dataset for the online stability analysis of the new energy grid-connected system in the target domain is represented as the corresponding return matrix. D t = {(x t1 , y t1 )，…，(x tm , y tm )}, where m < n, In transfer learning settings, if the data in the source and target domains have the same distribution, That is, X s ≠X t , P(X s ) = P(X t ) At that time, the online stability analysis model of the new energy grid-connected system reached the preset accuracy.

8. The online stability analysis method for new energy grid-connected systems based on neural networks according to claim 1, characterized in that, The modified inverter frame and / or control parameters include: adjusting the phase-locked loop proportional-integral (PLL) parameters to reduce the PLL bandwidth; adjusting the current loop proportional-integral (PLL) parameters to reduce the current loop bandwidth; and adding stability control.

Images