Photovoltaic grid-connected control method and control system based on data driving

Abstract

The invention discloses a photovoltaic grid-connected control method based on data driving in the technical field of new energy power generation control, and the method comprises the following steps: S1, collecting the current and network voltage on an alpha-beta axis under a two-phase static alpha-beta coordinate system at the AC side of an inversion module at a moment k, and initializing parameters; s2, based on the collected current and voltage, establishing a local prediction model of the grid-connected inverter at the current moment by adopting a real-time learning method; s3, predicting the current at the k + 1 moment according to the local prediction model and the eight different voltage vectors, and selecting the voltage vector corresponding to the minimized target function by using a model prediction control method; s4, the switching state corresponding to the obtained voltage vector acts on the inverter to output alternating current, the problem that a traditional control system depends on an accurate system mechanism model is solved, and when the external environment changes, set parameter values can be tracked.

Description

Technical Field

[0001] The present invention relates to the technical field of new energy power generation control, and particularly to a photovoltaic grid-connected control method and control system. Background Art

[0002] In recent years, with the rapid development of new energy technologies, solar energy, as a typical new energy, meets the requirements of people for clean, pollution-free and green concepts. After new energy power generation is grid-connected, due to internal and external disturbances such as environmental changes, the system parameters change, which will affect the performance of the power grid. As an important part of the new energy power generation field, the grid-connected inverter system is connected to the power grid and is responsible for converting the direct current generated by the photovoltaic modules into alternating current that meets the requirements of the power grid. Therefore, the photovoltaic inverter system is the key to energy conversion, and its control performance will directly affect the power quality, efficiency and stability of the entire power generation system. The performance of the system mainly depends on the structure and control algorithm of the grid-connected inverter system. The research on the inverter system model and control method provides important support for the design and operation of the photovoltaic system. Establishing an accurate control system model is the basis for effective control, and the performance of the system depends to a large extent on its model.

[0003] At present, most of the inverter systems adopt the mechanism modeling method. However, the mechanism modeling depends to a large extent on people's understanding of the equipment and system, and it is necessary to clearly understand the operation process of the system. Therefore, a large amount of time and energy are required for modeling and control. In addition, due to environmental changes, the system parameters will be affected, resulting in uncertain changes in the system model and unstable system operation. With the development of information technology, a large amount of production and operation data of the system can be stored, and these data reflect the operation status of the system. How to make full use of the offline and online data of the system is of great significance for the optimal control and stable operation of the grid-connected process. Summary of the Invention

[0004] In view of the deficiencies in the prior art, the present invention provides a data-driven photovoltaic grid-connected control method and control system, which makes full use of offline and online data information, and forms a new grid-connected control system on the basis of establishing a local prediction model; it solves the problem that the traditional control system depends on an accurate system mechanism model, and can track the set parameter values when the external environment changes.

[0005] The object of the present invention is achieved as follows: A data-driven photovoltaic grid-connected control method includes the following steps:

[0006] Step 1: At time k, collect the currents i α (k), i β (k) on the α and β axes in the two-phase stationary αβ coordinate system on the AC side of the inverter module, and the grid voltages e α (k), eβ (k), and initialize the parameters;

[0007] Step 2: Based on the collected current i α (k), i β (k) and voltage e α (k), e β (k), use the instant learning method to establish the local prediction model of the grid-connected inverter at the current moment;

[0008] Step 3: According to the local prediction model and 8 different voltage vectors, predict the current i at the k+1 moment α (k+1), i β (k+1), and use the model predictive control method to select the voltage vector corresponding to the minimized objective function;

[0009] Step 4: Apply the switching state corresponding to the obtained voltage vector to the AC output of the inverter.

[0010] Furthermore, the process of identifying the model parameters using the instant learning method in Step 2 is as follows:

[0011] Step 2-1: Store the system historical operation data information in the database;

[0012] Step 2-2: Considering both the angle and distance between the data, select similar data from the database as the modeling data according to the following similarity criterion:

[0013] ;

[0014] where d i represents the i-th historical data in the database and the input data at the current k moment The distance between them, θ i represents the i-th historical data in the database and the input data at the current k moment The included angle between them, is the weighting coefficient, and select the l data with the largest similarity as the modeling data set at the current moment;

[0015] After selecting the similar data, assign different weights to different data with the following weighted values

[0016] ;

[0017] where h is the bandwidth of the kernel function K(d), X j is the j-th data in the modeling data set, and the weighting matrix is W l =diag(w1,w2,…,wl ) and then use the locally weighted sum of least squares method to establish a local prediction model for the grid-connected inverter;

[0018] Step 2-4: Define , the weighted data set , , and the leave-one-out cross-error value is:

[0019] ;

[0020] where y j is the j-th element in Y l , and are respectively and of the j-th row vector. The optimal value of l is:

[0021] ;

[0022] Step 2-5: When the system input data arrives at the next moment, re-select a similar data set for modeling.

[0023] Furthermore, the model predictive control method in Step 3 specifically includes:

[0024] S3-1: Substitute the 8 basic voltage vectors corresponding to the following formula into the local prediction model to obtain the predicted voltage value

[0025] ;

[0026] where the output voltage of each phase

[0027] ,

[0028]

[0029] S3-2: According to the predicted value of the system, establish the following objective function:

[0030] ;

[0031] where and are the reference current values in the αβ coordinate system at the k + 1 moment, and q and r are weighting coefficients;

[0032] S3-3: Find the predicted current value that minimizes the objective function g and apply the corresponding switching state to the switching devices in the three-phase inverter.

[0033] A data-driven photovoltaic grid-connected control system for implementing the above control method, including an MPPT controller and a three-phase full-bridge inverter module, further including:

[0034] A data-driven module is used to collect the currents \(i_{\alpha}(k)\) and \(i_{\beta}(k)\) on the \(\alpha\)-axis and \(\beta\)-axis in the two-phase stationary \(\alpha\beta\) coordinate system on the AC side of the inverter module at time \(k\), and the grid voltages \(e_{\alpha}(k)\) and \(e_{\beta}(k)\), and initialize the parameters; based on the collected currents \(i_{\alpha}(k)\), \(i_{\beta}(k)\) and voltages \(e_{\alpha}(k)\), \(e_{\beta}(k)\), an instantaneous learning method is adopted to establish a local prediction model of the grid-connected inverter at the current moment. α (k), i β (k) and grid voltage e α (k), e β (k), and initialize the parameters; based on the collected current i α (k), i β (k) and voltage e α (k), e β (k), an instantaneous learning method is used to establish a local prediction model of the grid-connected inverter at the current moment;

[0035] A model predictive controller is used to predict the currents \(i_{\alpha}(k + 1)\) and \(i_{\beta}(k + 1)\) at time \(k + 1\) according to the local prediction model and 8 different voltage vectors, and use the model predictive control method to select the voltage vector corresponding to the minimized objective function; the switching state corresponding to the obtained voltage vector is applied to the output alternating current of the inverter. α (k+1), i β (k+1), and use the model predictive control method to select the voltage vector corresponding to the minimized objective function; the switching state corresponding to the obtained voltage vector is applied to the output alternating current of the inverter.

[0036] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0037] (1) The present invention can establish a local prediction model based on the process data of the system, and realize accurate tracking of the set parameter values by the photovoltaic grid-connected inverter under the condition that the system parameters are unknown and uncertain;

[0038] (2) When there are external disturbances, the present invention can continuously update the system model based on the changes of the input real-time data, showing good stability. BRIEF DESCRIPTION OF THE DRAWINGS

[0039] In order to more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the following will briefly introduce the drawings required for the description of the embodiments or the prior art. Obviously, the drawings in the following description are only the embodiments of the present invention. For those of ordinary skill in the art, other drawings can be obtained according to the provided drawings without creative efforts.

[0040] Figure 1 FIG. is a schematic structural diagram of the photovoltaic grid-connected inverter control system of the present invention.

[0041] Figure 2 FIG. is a flowchart of the photovoltaic grid-connected inverter control method of the present invention.

[0042] Figure 3 FIG. is a flowchart of the inverter modeling of the present invention. DETAILED DESCRIPTION OF THE INVENTION

[0043] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0044] As Figure 1 shown, a novel grid-connected inverter control system based on data driving includes an MPPT controller, a model predictive controller, a data driving module, and a three-phase full-bridge inverter module; the photovoltaic maximum power point is obtained through MPPT control and direct current is obtained through a Boost converter; the i α (k), i β (k), e α (k), e β (k) values at the k-th moment of the alternating current side current in the two-phase stationary αβ coordinate system are measured and input into the system database, and then the system local prediction model is obtained by using the least squares method; the model predictive controller obtains the optimal voltage vector by minimizing the objective function; the switching state corresponding to the optimal voltage vector is applied to the switching devices on the bridge arms of the three-phase full-bridge inverter control module, and finally the control of the grid-connected current is realized.

[0045] As Figure 2 shown, a novel grid-connected inverter control method based on data driving includes the following steps:

[0046] Step 1: Initialize the parameters and measure the i α (k), i β (k), e α (k), e β (k) values at the k-th moment in the two-phase stationary αβ coordinate system. The PLL is a three-phase phase-locked loop. The output voltage and current of the three-phase inverter are determined by the state of the switching devices on the bridge arms. The voltage vector is expressed as

[0047] x = a, b, c;

[0048] The output voltage of each phase can be expressed as

[0049] ;

[0050] The inverter output voltage vector is expressed as

[0051] ;

[0052] The three-phase grid-connected inverter has 8 switching states, and 8 output voltage vectors u i(i = 1, 2, 3, …, 8), the mathematical model of the three-phase grid-connected inverter in the two-phase orthogonal coordinate system can be expressed as

[0053] ;

[0054] where

[0055] , ,

[0056] , , ;

[0057] are the components of the inverter output voltage vector at the k-th moment in the αβ coordinate system;

[0058] Step 2: Based on the real-time sampling data, an instant learning method is used to identify the model parameters, so as to establish a local prediction model of the system;

[0059] As Figure 3 shown, the process of identifying model parameters by the sampling instant learning method is as follows:

[0060] 2-1) Store the historical operation data information of the system in the database;

[0061] 2-2) Calculate the distance and angle q between the data X at the current k-th moment and the data X i in the database according to the following formula

[0062] ;

[0063] If calculate the similarity as follows

[0064] ;

[0065] where is the weighting coefficient.

[0066] If , the corresponding historical data in the database will not be added to the modeling dataset;

[0067] 2-3) Sort the values of s i in descending order, and select the l data with the largest similarity as the modeling dataset at the current moment , for the first variable of the state vector, its modeling dataset is constructed as follows:

[0068] ;

[0069] Let the weighting coefficient , where h is the bandwidth of the kernel function K(d), and X j is the j-th data in the modeling dataset; the weighting matrix is as follows:

[0070] W l =diag(w1,w2,…,w l );

[0071] The weighted modeling dataset is

[0072] ;

[0073] The model parameters corresponding to the first variable of the system state are obtained by the least squares method, and the calculation formula is as follows:

[0074] ;

[0075] To obtain the optimal value of l and the best local prediction model. Define , and the leave-one-out cross-error value is:

[0076] ;

[0077] where y j is the j-th element in Y l , and are respectively and 's j-th row vector. The optimal value of l is:

[0078] ;

[0079] Calculate the model parameters corresponding to the variables (k), (k), (k) in the same way, and thus obtain the system parameters A k and B k , and then establish a local prediction model;

[0080] Step 3: Predict the currents and at time k + 1 according to 8 different voltage vectors and the local prediction model;

[0081] Step 4: Substitute the predicted currents at time k + 1 into the following objective function

[0082] ;

[0083] Select the switching state of the voltage vector corresponding to the predicted current that minimizes the objective function, and apply it to the switching devices on the three-phase inverter bridge arm.

[0084] The description of the above embodiments is only used to help understand the method of the present invention and its core idea. It should be noted that for those of ordinary skill in the art, without departing from the principle of the present invention, several improvements and modifications can be made to the present invention, and these improvements and modifications also fall within the protection scope of the claims of the present invention.

Claims

1. A photovoltaic grid-connected control method based on data drive, characterized in that: The following steps are involved: Step 1: At time k, collect the current i on the αβ axis of the two-phase stationary αβ coordinate system on the AC side of the inverter module α (k), i β (k) and grid voltage e α (k), e β (k), and initialize the parameters; Step 2: Based on the collected current i α (k), i β (k) and voltage e α (k), e β (k), using a just-in-time learning method to establish a local prediction model of the grid-connected inverter at the current moment; Step 3: Based on the local prediction model and 8 different voltage vectors, predict the current i at time k+1 α (k+1), i β (k+1), and use the model predictive control method to select the voltage vector corresponding to the minimized objective function; Step 4: Apply the switch state corresponding to the obtained voltage vector to the inverter output AC power.

2. A photovoltaic grid-connected control method based on data drive according to claim 1, characterized in that: The process of identifying model parameters using the real-time learning method in step 2 is as follows: Step 2-1: Store the system historical operation data information in the database; Step 2-2: Considering the angle and distance between data, similar data are selected from the database as modeling data according to the following similarity criteria: ； where d i Represents the i-th historical data in the database With the current k-time input data The distance between i Represents the i-th historical data in the database With the current k-time input data The angle between is the weighted coefficient, and the l data with the largest similarity are selected As the modeling dataset at the current moment; Step 2-3: After similar data is selected, assign different weights to different data using the following weighted values: ； Where h is the bandwidth of the kernel function K(d), X j For the jth data in the modeling data set, the weight matrix is ​​W l =diag(w1,w2,…,w l ), and then establish a local prediction model of the grid-connected inverter using local weighting and least squares method; Step 2-4: Definition , weighted dataset , , the leave-one-out crossover error is: ； where y j Y l The jth element in and They are and The jth row vector of . The optimal value of l is: ； Step 2-5: When the system input data arrives at the next moment, reselect a similar data set for modeling.

3. A data-driven photovoltaic grid-connected control method according to claim 2, characterized in that: The model predictive control method in step 3 specifically includes: S3-1: Substitute the 8 basic voltage vectors corresponding to the following formula into the local prediction model to obtain the predicted voltage value ； The output voltage of each phase ， ； S3-2: According to the predicted value of the system, the following objective function is established: ； in and is the reference current value in the αβ coordinate system at time k+1, q and r are weighting coefficients; S3-3: Find the predicted current value that minimizes the objective function g, and apply the corresponding switching state to the switching devices in the three-phase inverter.

4. A photovoltaic grid-connected control system based on data drive, used to implement the control method as described in any one of claims 1 to 3, comprising an MPPT controller and a three-phase full-bridge inverter module, characterized in that: Also includes: The data drive module is used to collect the current i on the αβ axis of the two-phase stationary αβ coordinate system on the AC side of the inverter module at time k. α (k), i β (k) and grid voltage e α (k), e β (k), and initialize the parameters; based on the collected current i α (k), i β (k) and voltage e α (k), e β (k), using a just-in-time learning method to establish a local prediction model of the grid-connected inverter at the current moment; Model predictive controller, used to predict the current i at time k+1 based on the local prediction model and 8 different voltage vectors α (k+1), i β (k+1), and use the model predictive control method to select the voltage vector corresponding to the minimized objective function; the switch state corresponding to the obtained voltage vector is applied to the inverter to output AC power.

Images