LCL type grid-connected inverter stability analysis method based on neural network

Abstract

The invention discloses an LCL type grid-connected inverter stability analysis method based on a neural network, and the method comprises the following steps: S1, building an LCL type single-phase grid-connected inverter state space model through employing a capacitive current feedback active damping control strategy based on an inverter control principle and a zero static error current tracking control requirement; s2, determining a parameter distribution condition obeyed by key parameters in the inverter system, and obtaining a neural network data set based on parameter distribution; and S3, building a neural network based on the parameter model, and proposing a neural network stability judgment method based on feature values. The analysis method is suitable for a single-phase LCL type grid-connected inverter, and a neural network model is built according to potential characteristics of parameters to carry out stability research on an inverter system; the conclusion obtained by the scheme can be used for online analysis and evaluation of the stability of the LCL type single-phase grid-connected inverter system.

Description

Technical Field

[0001] The invention relates to the technical field of power electronics, and more specifically to a stability analysis method of an LCL type grid-connected inverter based on a neural network. Background Art

[0002] With the continuous increase in the penetration rate of new energy in new power systems, a large number of power electronic grid-connected inverters are connected to the power grid as interface units between distributed new energy and power systems. Due to its good filtering effect, LCL grid-connected inverters are more suitable for renewable energy distributed power generation grid-connected systems. As an interface unit of a distributed power generation system, the control performance of LCL grid-connected inverters determines the output power quality of the system to a certain extent. The inverter inevitably operates under uncertain parameter conditions. Due to different working positions and conditions, the grid-side impedance parameters will be uncertain; due to the aging of components caused by long-term operation, the inverter component parameters may deviate from the nominal value, affecting the stability margin of the system and bringing certain risks of operational instability.

[0003] At present, the stability analysis methods of inverter systems are mostly based on traditional mathematical models or impedance models, which have large calculation amount and slow analysis speed. With the development of smart grids and new power systems, artificial intelligence and data-driven methods have been widely used in modeling, dynamic analysis and control of power systems. Therefore, the present invention proposes a stability analysis method for LCL type single-phase grid-connected inverter based on neural network, and studies the voltage and current tracking control problem of inverter under input and output constraints and parameter disturbance conditions. A neural network is built, and the objective function is designed based on the first Lyapunov method. Stability analysis is performed, and the dynamic, anti-disturbance and steady-state performance of the inverter under different parameter disturbances are observed. Summary of the invention

[0004] In order to overcome the above-mentioned defects of the prior art, the present invention provides a stability analysis method of an LCL type grid-connected inverter based on a neural network.

[0005] The technical solution of the present invention is as follows:

[0006] A Stability Analysis Method for LCL Grid-connected Inverter Based on Neural Network

[0007] Step S1: Based on the inverter control principle and the requirement of zero-static-error current tracking control, a state space model of an LCL-type single-phase grid-connected inverter is established;

[0008] Step S2: clarify the parameter distribution of the key parameters in the inverter system, and obtain a neural network data set based on the parameter distribution;

[0009] Step S3: Build a neural network based on the parameter model and propose a neural network stability judgment method based on eigenvalues.

[0010] Furthermore, the step S1 includes the following sub-steps:

[0011] Step S11: Based on the zero-static-error current tracking control requirement, a quasi-PR regulator is selected as a current tracking controller, and a state space equation of the quasi-PR regulator is established:

[0012]

[0013] Where: x1 and x2 are defined as the state variables of the quasi-PR controller; ω0 is the operating angular frequency of the AC system, that is, the resonant frequency; ω c is the resonant cutoff frequency; K c is the capacitor current feedback coefficient; K r is the controller resonance coefficient;

[0014] Step S12: Considering the controller delay, construct the state space equation of the LCL type single-phase grid-connected inverter:

[0015]

[0016] Where: x is the state vector of the system, denoted as x = [u c ,i1,i g ,x1,x2,v ref ] T , where u c is the voltage across the capacitor, i1 is the current flowing through the inductor on the inverter side, i g is the current flowing through the grid-side inductor, v ref is the output of the controller delay module; u is the input vector of the system, denoted as u = [u g ,i ref ] T , where u g is the AC power supply voltage; the system output y is the state variable i g .

[0017] Furthermore, step S2 comprises the following sub-steps:

[0018] Step S21: Based on the common parameter distribution, analyze the core uncertain parameters that affect the system state, and determine the distribution they obey according to their characteristics. Among them, the filter inductor L1 and the grid-side inductor L2 obey the normal distribution, and the filter capacitor C obeys the Weibull distribution. The probability density function of the filter capacitor obeying the Weibull distribution is:

[0019]

[0020] Step S22: Analyze the perturbation range of system parameters according to the distribution obeyed by the parameters to obtain a parameter distribution data set.

[0021] Furthermore, step S3 includes the following sub-steps:

[0022] Step S31: using the system core parameters determined in step S2 as neural network input, using the dichotomy principle, taking the stability and instability as outputs respectively, and establishing a neural network with a single hidden layer;

[0023] Step S32: Based on the principle of the eigenvalue method and the Lyapunov stability criterion, a parameter distribution data set for neural network training is obtained based on the parameter distribution data set obtained in step S22;

[0024] Step S33: Divide the data set in step S32 into a training data set and a test data set to judge the stability of the LCL type single-phase grid-connected inverter.

[0025] The beneficial effects of the present invention include: in the process of establishing the LCL type single-phase grid-connected inverter model, an active damping method based on a current feedforward control strategy is used, and no physical resistance is introduced, which can improve the inverter control performance and work efficiency; a parameter model is applied to construct a neural network, and a binary classification method is used to optimize the output during the neural network design process, and the stability of the grid-connected inverter system is intuitively obtained through changes in system parameters; a smaller data set is used for neural network training, which saves training time and has good stability analysis results. BRIEF DESCRIPTION OF THE DRAWINGS

[0026] The drawings herein are incorporated into and constitute a part of the specification, showing embodiments consistent with the present invention, and together with the specification, are used to explain the principles of the present invention. Obviously, the drawings described below are only some embodiments of the present invention, and for those of ordinary skill in the art, other drawings can be obtained based on these drawings without creative work. In the drawings:

[0027] Figure 1 A schematic diagram of the process flow of the LCL type single-phase grid-connected inverter stability assessment method provided in this application;

[0028] Figure 2 The control circuit topology of the LCL type single-phase grid-connected inverter provided in the embodiment of the present application;

[0029] Figure 3 A design diagram of a feedforward neural network based on eigenvalues ​​provided in an embodiment of the present application;

[0030] Figure 4A schematic diagram of a grid-connected inverter stability assessment process based on a neural network provided in an embodiment of the present application;

[0031] Figure 5 The grid-connected current i provided in the embodiment of the present application g Simulation image of

[0032] Figure 6 The AC voltage u provided in the embodiment of the present application c Simulation image of . DETAILED DESCRIPTION

[0033] The technical solutions in the embodiments of the present invention are described clearly and completely below. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of them. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0034] Embodiment 1

[0035] The specific implementation steps of the present invention are as follows Figure 1 As shown, the main circuit topology of the LCL type single-phase grid-connected inverter is as follows Figure 2 The present invention can quantitatively analyze the stability of the LCL type single-phase grid-connected inverter and qualitatively analyze the influence of different system parameters on the inverter stability. The analysis result has guiding significance for the parameter design of the LCL type single-phase grid-connected inverter.

[0036] For details, please refer to Figure 1 , this assessment method includes:

[0037] Step S1: Based on the zero-error tracking control requirement of the inverter, a capacitor current feedback active damping control strategy is adopted to establish a state space model of the LCL type single-phase grid-connected inverter.

[0038] Figure 2 This is the control circuit schematic diagram of the single-phase full-bridge LCL grid-connected inverter. The filter capacitor voltage is recorded as u c The currents through L1, L2, and C are i1, i2, and i3 respectively. g 、i c The DC voltage is U dc , the DC side current is i dc , the inverter output voltage is u inv The grid is considered to consist of a voltage source u g With grid impedance Z g The phase angle of the detected grid voltage is θ, I ref is the reference current amplitude, and multiplying the two together gives the reference current i ref Track the grid current and compare it with the reference current to get i* Input current tracking controller. K c is the capacitor current feedback control coefficient, v ref is a voltage modulation signal.

[0039] Depend on Figure 2 The system differential equations are shown in equations (1)-(3):

[0040]

[0041] The transfer function of the quasi-PR regulator used is:

[0042]

[0043] Where: ω0 is the resonant frequency; ω c is the resonant cutoff frequency; K p and K r are the proportional gain coefficient and the controller resonance coefficient respectively.

[0044] The state variables of the controller are defined as x1 and x2, and the state space equation and output of the quasi-PR regulator can be expressed as:

[0045]

[0046] u f =K p i * +x1 (6)

[0047] Assume the controller delay is G d (s), according to the first-order Taylor expansion formula, the controller delay can be written as:

[0048]

[0049] Controller delay module G d The output of (s) is v ref The system state vector x = [u c ,i1,i g ,x1,x2,v ref ] T , input vector u=[u g ,i ref ] T , output state variable i g Therefore, the state space model of the system at the nominal parameters is constructed as shown in equation (8):

[0050]

[0051] The expressions of matrices A, B, and C are shown in equations (9)-(11), where matrix A is the characteristic matrix of the system.

[0052]

[0053] C=[0 0 1 0 0 0] (11)

[0054] Step S2: clarify the system parameters that affect the stable state of the inverter, and analyze the parameter distribution it obeys according to the parameter characteristics; calculate the characteristic values ​​of each group of parameters under different parameter values ​​to obtain a neural network data set based on the parameter distribution.

[0055] Through the modeling process of the LCL type single-phase grid-connected inverter control system, it can be seen that the inverter system state space model is a nonlinear description of the system under certain working conditions. The grid-connected inverter can work under certain operating conditions. If the system parameter model covering the operating condition range can be quickly estimated with sufficient accuracy, the efficiency of the system stability analysis can be significantly improved.

[0056] The parameter model of the grid-connected inverter system can take different forms, depending on the modeling assumptions and the modeling reference framework. By analyzing equations (9)-(11), among all the perturbed parameters, the controller parameters can be set manually, so the stable value state of the inverter depends on the three parameters L1, L2, and C. The stability expression of the system can be written as:

[0057] f s =f(L1,L2,C) (12)

[0058] Equation (12) shows that the system parameter model is essentially a nonlinear system containing multiple variables, so the neural network can be used to learn the approximate value of the model.

[0059] Since the parameter value of the filter inductor L1 is affected by many small and independent factors, such as the operating environment temperature, current size, etc., it is believed that the distribution of the filter inductor L1 belongs to the normal distribution. The filter capacitor C will age due to the influence of reactive power, and its aging process can be characterized by Weibull distribution. According to the grid impedance measurement results in existing research, the grid-side inductor L2 conforms to the normal distribution.

[0060] With the nominal value of the system parameter as the center, three sets of working condition sets containing 50 data are randomly generated according to the parameter distribution they obey, and the system eigenvalues ​​under the corresponding working conditions are calculated. Through the above process, a training data set with a data volume of 150 is obtained. Since this study only considers the change in system stability when a single parameter is perturbed, the remaining parameters remain unchanged at the perturbed boundary value during a parameter perturbation process. Three sets of data are taken in each type of parameter perturbation process as examples of training data sets. The training data set examples are shown in Table 1.

[0061] Table 1. Examples of training datasets

[0062]

[0063] Step S3: Build a neural network based on the parameter model. The input of the neural network is set as the key parameters of the system in step S2. The stability analysis results are classified by a binary classification method. The digital quantity is used as the output of the neural network, and a neural network stability judgment method based on eigenvalues ​​is proposed.

[0064] In order to achieve the learning tasks and goals of the neural network, the categories of input and output quantities should be determined first. From formula (12), it can be seen that the system parameter model is related to L1, L2, and C, so these three system parameters are selected as input quantities.

[0065] The output should be able to intuitively judge the stability of the system. In the present invention, the stability judgment result is obtained by training and predicting the system eigenvalue. Therefore, the output can be selected as the system eigenvalue, and the system stability can be judged by the position relationship between the real part of the eigenvalue and the imaginary axis. According to the Lyapunov stability criterion, when the real part of the system eigenvalue is located on the left side of the imaginary axis, the system is stable; conversely, if the real part of the eigenvalue is located on the right side of the imaginary axis, the system is unstable.

[0066] In order to make the system output more intuitive, the idea of ​​binary classification is applied in the neural network training process, and two specific categories are distinguished by constructing a classifier model. In the present invention, the system is judged to be "stable" or "unstable" under different working conditions through the training results of different input parameters in the neural network. The two conditions of stability and instability are digitized, and the stable condition is set to 0 and the unstable condition is set to 1, which constitutes a binary classification modeling method. When the output is close to 0, the system is stable; when the output is close to 1, the system is unstable. Therefore, the output of the neural network is the result of the binary classification calculation. An example of the expected output of the training data set is shown in the fifth column of Table 1.

[0067] In the process of neural network design, the basic structure of feedforward neural network is adopted, a single hidden layer is designed, the training accuracy is improved by increasing the number of neurons, and the Sigmoid function is selected as the activation function to map the calculation results of neurons and weight matrix. The design diagram of feedforward neural network is as follows: Figure 3 shown.

[0068] The stability evaluation process of grid-connected inverter based on neural network is as follows Figure 4As shown. First, the data is preprocessed to calculate the preliminary eigenvalues ​​under the corresponding parameters, then the neurons are assembled, the neural network is constructed, and the threshold weights are initialized. The loss function is calculated using the variance, and the loss is minimized based on the gradient descent method. If the training round does not meet the standard or the loss does not reach the specified value, the threshold and weight are updated again based on the back propagation algorithm to obtain the output to determine whether the system is stable.

[0069] The training data set is input into the neural network for training. If the training results can reduce the training loss, it proves that the system characteristics can be obtained through neural network training, and then the stability of the test data set is judged. Then the test data set is input into the network, and the stability judgment results are shown in Table 2.

[0070] Table 2 Neural network output and stability judgment results

[0071]

[0072]

[0073] The present disclosure uses the MATLAB / Simulink simulation platform to verify the effectiveness of the inverter stability analysis method of Example 1. Specifically, in order to verify the effectiveness of the neural network method in judging the inverter stability, the present application uses the MATLAB / Simulink simulation platform. The main parameters of the system are listed in Table 3.

[0074] Table 3 Inverter parameter nominal values

[0075]

[0076] Change the system operating conditions and observe the grid-connected current i g And AC voltage u c Changes in the waveform. When the waveform oscillates, it indicates that the system is in an unstable state.

[0077] Figure 5 (a) is the grid-connected current i when all system parameters are at nominal values g The image, i g When other parameters are at nominal values ​​and the capacitance and inductance are increased, the simulation results are as follows: Figure 5 (b) and (c). In the process of parameter increase, when C = 12μF and L1 = 1.3mH, the system begins to oscillate and become unstable. This change trend is the same as the change trend of system stability predicted by the neural network. The critical stability value is within a reasonable error range, which can prove the accuracy of the neural network stability judgment result.

[0078] Figure 6(a) is the AC voltage u across the filter capacitor when the system parameters are at the nominal value c The image of u c Similarly, when the other parameters are at the nominal values, and the capacitance and inductance are increased, the simulation results are as follows: Figure 6 When C = 12μF, L1 = 1.3mH, the system begins to oscillate and become unstable, which also proves the accuracy of the neural network's stability judgment result.

[0079] Therefore, the accuracy of the LCL type grid-connected inverter stability analysis method based on neural network proposed in this application is verified through the simulation results of the grid-connected current and grid-connected voltage in the grid-connected inverter.

[0080] According to the potential characteristics of the parameters, this method builds a neural network model to study the stability of the LCL type grid-connected inverter system, improves the calculation speed, and obtains good learning performance; the conclusions obtained by the present invention can be applied to the online stability analysis and evaluation of the LCL type single-phase grid-connected inverter system.

[0081] It should be understood that parts not elaborated in detail in this specification belong to the prior art.

[0082] The present disclosure has been described by the above-mentioned relevant embodiments, but the above-mentioned embodiments are only examples for implementing the present disclosure. It must be pointed out that the disclosed embodiments do not limit the scope of the present disclosure. On the contrary, changes and modifications made without departing from the spirit and scope of the present disclosure are all within the scope of patent protection of the present disclosure.

Claims

1. A stability analysis method for LCL type grid-connected inverter based on neural network, characterized in that: The following steps are involved: Based on the inverter control principle and the requirements of zero-static current tracking control, the state space model of LCL single-phase grid-connected inverter is established. Clarify the parameter distribution of key parameters in the inverter system and obtain a neural network data set based on the parameter distribution; A neural network is built based on the parameter model, and a neural network stability judgment method based on eigenvalues ​​is proposed.

2. The method for analyzing the stability of an LCL type grid-connected inverter based on a neural network according to claim 1, characterized in that: Based on the inverter control principle and the requirements of zero-static current tracking control, the state space model of the LCL single-phase grid-connected inverter is established, including the following sub-steps: Based on the requirement of zero-static-error current tracking control, the quasi-PR regulator is selected as the current tracking controller, and the state space equation of the quasi-PR regulator is established: Where: x1 and x2 are defined as the state variables of the quasi-PR controller; ω0 is the operating angular frequency of the AC system, that is, the resonant frequency; ω c is the resonant cutoff frequency; K c is the capacitor current feedback coefficient; K r is the controller resonance coefficient; Considering the controller delay, the state space equation of the LCL type single-phase grid-connected inverter is constructed: Where: x is the state vector of the system, denoted as x = [u c ,i1,i g ,x1,x2,v ref ] T , where u c is the voltage across the capacitor, i1 is the current flowing through the inductor on the inverter side, i g is the current flowing through the grid-side inductor, v ref is the output of the controller delay module; u is the input vector of the system, denoted as u = [u g ,i ref ] T , where u g is the AC power supply voltage; the output y of the system is the state variable i g .

3. The method for analyzing the stability of an LCL type grid-connected inverter based on a neural network according to claim 1, characterized in that: The step of clarifying the parameter distribution of the key parameters in the inverter system and obtaining a neural network data set based on the parameter distribution includes the following sub-steps: Based on the common parameter distribution, the core uncertain parameters that affect the system state are analyzed, and their respective distributions are determined according to their characteristics; among them, the filter inductor L1 and the grid-side inductor L2 obey the normal distribution, and the filter capacitor C obeys the Weibull distribution; the probability density function of the filter capacitor obeying the Weibull distribution is: According to the distribution that the parameters obey, the perturbation range of the system parameters is analyzed to obtain the parameter distribution data set.

4. The method for analyzing the stability of an LCL type grid-connected inverter based on a neural network according to claim 3, characterized in that: Based on the parameter model, a neural network is built and a neural network stability judgment method based on eigenvalues ​​is proposed, which includes the following sub-steps: The key parameters in the determined system are used as the neural network input, and the stable and unstable conditions are converted into digital quantities as outputs using the dichotomy principle, and a feedforward neural network with a single hidden layer is established. Based on the principle of eigenvalue method and Lyapunov stability criterion, the parameter distribution data set for neural network training is obtained based on the parameter distribution data set. The parameter distribution data set used for neural network training is divided into a training data set and a test data set to judge the stability of the LCL type single-phase grid-connected inverter.

Images