A control method for a converter and a power grid system

Abstract

The present application discloses a control method for a converter and a power grid system. The control method is applied to the control of the converter in the power grid system, and the control method includes the following steps: constructing an adaptive superlocal model, where the adaptive superlocal model includes an adaptive input gain and a superlocal model, and the superlocal model includes physical parameters within the controlled system and disturbance factors in the power grid system; using an online neural network estimator to estimate the superlocal model and the adaptive input gain to eliminate the dependence of the superlocal model and the adaptive input gain on system parameters; predicting the output current of the converter based on the superlocal model and the adaptive input gain, and generating a voltage vector of the converter based on the predicted current of the converter, and using the voltage vector to control the converter. The present application can achieve current tracking of the converter and neutral point voltage balance, and realize parameter-free operation of the control system.

Description

Technical Field

[0001] This application relates to the technical field of commutation equipment, and particularly to a control method for a converter and a power grid system. Background Art

[0002] In the fields of new energy power generation and smart power grids, traditional converter control strategies rely on preset mathematical models, which are often difficult to cope with in the face of complex and changing external environments. Existing predictive control methods still rely on a variety of physical parameters of the power grid system (including capacitance, inductance, and resistance parameters). When the inductance parameters are significantly mismatched, the control performance of existing predictive control methods will significantly decline. When the inductance parameters are unknown, existing predictive control methods are difficult to apply. Summary of the Invention

[0003] To solve the deficiencies of the existing technology, this application adopts the following technical solutions:

[0004] A control method for a converter provided by this application, the control method is applied to the control of the converter in a power grid system, and the control method includes the following steps:

[0005] Construct an adaptive super-local model, the adaptive super-local model includes an adaptive input gain and a super-local model, and the super-local model includes physical parameters in the controlled system and disturbance factors in the power grid system;

[0006] Use an online neural network estimator to estimate the super-local model and the adaptive input gain, and eliminate the dependence of the super-local model and the adaptive input gain on system parameters;

[0007] Predict the output current of the converter based on the super-local model and the adaptive input gain, and generate a voltage vector of the converter based on the predicted current of the converter, and use the voltage vector to control the converter.

[0008] In summary, according to the above description, a control method for a converter provided by this application uses an online neural network estimator to achieve online estimation of the super-local model and the adaptive input gain, realizes real-time update of the parameters in the super-local model, and eliminates the dependence of the super-local model and the adaptive input gain on system parameters, realizing parameter-free operation of the control system; it also obtains a voltage vector for controlling the converter based on the predicted current of the converter, and realizes the balance of the neutral point voltage of the converter.

[0009] Further, the control method further includes:

[0010] The super-local model is represented by the following formula:

[0011] ;

[0012] Among them, represents the weight coefficient matrix, is the activation function, is the estimation error, represents the system output current in the

[0013] The weight coefficient matrix is updated online, and its update rate is expressed by the following formula:

[0014] ;

[0015] Among them, represents the estimated value of the weight coefficient matrix, is the estimated value of the system output current information, , and are all adjustable weight coefficient matrices, represents the transpose matrix of the matrix represents the system input voltage in the represents the adaptive input gain;

[0016] The weight coefficient matrix is discretized to obtain a current prediction equation.

[0017] Furthermore, the first-order Euler discretization method is used to discretize the weight coefficient matrix, and the obtained current prediction equation is expressed by the following formula:

[0018] ;

[0019] In the formula, represents the control period.

[0020] Furthermore, eliminating the dependence of the adaptive input gain on system parameters includes:

[0021] Design the update law of the adaptive input gain, and the update law of the adaptive input gain is expressed by the following formula:

[0022] ;

[0023] In the formula, the parameter represents the adaptive input gain, the parameter represents the estimated value of the parameter , represents the output current information of the axis, Predicted output current information of the shaft, Indicates Voltage information of the shaft, Indicates the control period.

[0024] Furthermore, the control method further includes:

[0025] Introduce a step factor, and the update law of the adaptive input gain is expressed by the following formula:

[0026] ;

[0027] In the formula, p represents the step factor.

[0028] Furthermore, the control method further includes:

[0029] Design a neutral point voltage update law, and the neutral point voltage update law is expressed by the following formula:

[0030] ;

[0031] In the formula, p represents the step factor, Represents the estimated value of , , C dc Represents the bus capacitor on the DC side;

[0032] Predict the neutral point voltage of the power grid system through the following formula:

[0033] ;

[0034] ;

[0035] In the formula, Represents the estimated value of the neutral point voltage, Represents the neutral point voltage, Represents the neutral point voltage update law, Represents the three-switch expressions of the voltage vectors used; to achieve the control goal of neutral point voltage balance and realize the delay compensation of the converter.

[0036] Furthermore, the control method further includes:

[0037] Design a sequential value function, and the sequential value function is expressed by the following formula:

[0038] ;

[0039] In the formula, Represents the requirement for current tracking quality, Represents the requirement for the balance of the neutral point voltage, while represents the requirement for the switching frequency, which is the reference value, and is the predicted value obtained after using the voltage vector where is the sign function, representing the number of switching changes;

[0040] Based on the predicted current of the converter and the predicted value of the neutral point voltage, and based on the sequential value function, the voltage vector of the converter is screened to select the optimal voltage vector, and the converter is controlled using the optimal voltage vector.

[0041] Further, the control method further includes:

[0042] When the predicted value of the neutral point voltage is less than the neutral point voltage threshold, the cost function in the sequential value function is used to select the optimal vector. If the optimal vector is a long vector or a medium vector, it is directly applied to the converter;

[0043] If the optimal vector is a small vector, the cost function in the sequential value function is used to select the optimal switching state that makes the midpoint voltage closer to zero;

[0044] If it is a zero vector, the cost function in the sequential value function is used to select the optimal switching state that makes the switching frequency minimum.

[0045] Further, the control method further includes:

[0046] When the predicted value of the neutral point voltage exceeds the neutral point voltage threshold, the cost function in the sequential value function and are used to select the optimal switching state and apply it to the converter.

[0047] In a second aspect, the present application further provides a power grid system, the power grid system includes a converter, and the power grid system applies the control method of the above converter. BRIEF DESCRIPTION OF THE DRAWINGS

[0048] Figure 1 is a flowchart of the steps of the control method of the converter provided by an embodiment of the present application;

[0049] Figure 2 is an operating block diagram of the control method of the converter provided by an embodiment of the present application in practical application;

[0050] Figure 3Schematic topology diagram of the control method of the converter provided by an embodiment of the present application applied to a three-level NPC inverter;

[0051] Figure 4 Schematic diagram of the spatial distribution of voltage vectors in the control method of the converter provided by an embodiment of the present application;

[0052] Figure 5 Schematic flow diagram of the control method of the converter provided by an embodiment of the present application in specific applications;

[0053] Figure 6 Schematic diagram of the performance of the control method of the converter provided by an embodiment of the present application, the traditional finite set model predictive control method, the model-free predictive control method based on an extended state observer, and the model-free control method based on an integral sliding mode observer under the condition of matching simulation parameters;

[0054] Figure 7 Schematic diagram of the performance of the control method of the converter provided by an embodiment of the present application, the traditional finite set model predictive control method, the model-free predictive control method based on an extended state observer, and the model-free control method based on an integral sliding mode observer under the condition of input gain mismatch;

[0055] Figure 8 Schematic diagram of the performance of the control method of the converter provided by an embodiment of the present application, the traditional finite set model predictive control method, the model-free predictive control method based on an extended state observer, and the model-free control method based on an integral sliding mode observer under the condition of inductor parameter mismatch;

[0056] Figure 9 Schematic diagram of the comparison results of four control methods under different degrees of parameter mismatch conditions;

[0057] Figure 10 Schematic diagram of the online adjustment of the input gain when the control method of a converter provided by the present application is adapted by 50% up and down based on a true inductance value of 10 mH;

[0058] Figure 11a Schematic diagram of the performance of the traditional finite set model predictive control method on the 3L-NPC based power conversion experimental platform under the condition of parameter balancing;

[0059] Figure 11b Schematic diagram of the performance of the model-free predictive control method based on an extended state observer on the 3L-NPC based power conversion experimental platform under the condition of parameter balancing;

[0060] Figure 11cSchematic diagram of the performance of the model-free control method based on the integral sliding mode observer for the 3L-NPC-based power conversion experimental platform under parameter balancing;

[0061] Figure 11d Schematic diagram of the performance of a control method for a converter provided in this application for the 3L-NPC-based power conversion experimental platform under parameter balancing;

[0062] Figure 12a Schematic diagram of the performance of the traditional finite set model predictive control method for the 3L-NPC-based power conversion experimental platform under parameter mismatch;

[0063] Figure 12b Schematic diagram of the performance of the model-free predictive control method based on the extended state observer for the 3L-NPC-based power conversion experimental platform under parameter mismatch;

[0064] Figure 12c Schematic diagram of the performance of the model-free control method based on the integral sliding mode observer for the 3L-NPC-based power conversion experimental platform under parameter mismatch;

[0065] Figure 12d Schematic diagram of the performance of a control method for a converter provided in this application for the 3L-NPC-based power conversion experimental platform under parameter mismatch. Specific embodiments

[0066] The following will describe this application in detail in conjunction with the specific embodiments shown in the drawings, but these embodiments do not limit this application, and any structural, method, or functional transformation made by those of ordinary skill in the art based on these embodiments is included within the protection scope of this application.

[0067] To solve the deficiencies of the prior art, this application provides a control method for a converter applied to the control of the converter in the power grid system, as Figure 1 shown, the control method includes the following steps:

[0068] Step S11, constructing an adaptive superlocal model, the adaptive superlocal model includes an adaptive input gain and a superlocal model, and the superlocal model includes physical parameters in the controlled system and disturbance factors in the power grid system;

[0069] Step S12, using an online neural network estimator to estimate the superlocal model and the adaptive input gain to eliminate the dependence of the superlocal model and the adaptive input gain on system parameters;

[0070] Step S13: Predict the output current of the converter based on the hyperlocal model and the adaptive input gain, generate the voltage vector of the converter based on the predicted current of the converter, and use the voltage vector to control the converter.

[0071] Specifically, taking the neutral-point clamped three-level converter topology as an example, the model-based predictive control equation can be expressed as:

[0072] (1);

[0073] In the formula, R represents the converter resistance, L represents the converter inductance, and respectively represent the input voltage and output current of the power grid system in the

[0074] coordinate system. The resistance, inductance and other disturbance factors in the power grid system are equivalent to a hyperlocal model. The predictive control equation of the adaptive hyperlocal model can be expressed as:

[0075] (2);

[0076] Among them, F represents the equivalent hyperlocal model of the system, is the adaptive input gain. By constructing a predictive control framework based on the adaptive hyperlocal model, the dependence on system parameters is eliminated.

[0077] Under the framework of the adaptive hyperlocal model, the equivalent hyperlocal model F of the system and the adaptive input gain will change with the change of parameters in the power grid system. Observe and update the changes of the equivalent hyperlocal model of the system and the adaptive input gain to improve the predictive control performance. An online neural network estimator is used to estimate the hyperlocal model and the adaptive input gain, realize the real-time update of the parameters within the hyperlocal model framework, improve the anti-interference ability of the control system, eliminate the dependence of the hyperlocal model and the adaptive input gain on system parameters, and realize the parameter-free operation of the control system.

[0078] Based on the hyperlocal model and the adaptive input gain, the output current of the converter is obtained through the current prediction equation. Moreover, based on the predicted output current of the converter, the voltage vector of the converter is generated according to the switching state of the converter arm, and the optimal voltage vector is obtained according to the optimal voltage vector screening mechanism to control the converter, so as to realize the neutral-point voltage balance of the converter.

[0079] In summary, as Figure 2As shown, the on-line neural network estimator calculates based on the input voltage and output current of the power grid system, combines the reference voltage calculation, generates the voltage vector of the converter based on the predicted current of the converter, and realizes the optimal control of the converter through vector screening based on the prediction.

[0080] According to the above description, a control method for a converter provided by the present application uses an on-line neural network estimator to realize on-line estimation of the hyperlocal model and the adaptive input gain, realizes real-time update of the parameters in the hyperlocal model, eliminates the dependence of the hyperlocal model and the adaptive input gain on the system parameters, and realizes the parameter-free operation of the control system; it also obtains the voltage vector for controlling the converter based on the predicted current of the converter, and realizes the balance of the neutral point voltage of the converter.

[0081] As an implementation manner, the hyperlocal model is represented by the following formula:

[0082] (3);

[0083] Wherein, represents the weight coefficient matrix, is the activation function, represents the estimation error, represents the system output current in the coordinate system.

[0084] For on-line update of the weight coefficient matrix, substituting formula (3) into formula (2), its update rate is represented by the following formula:

[0085] (4);

[0086] Wherein, represents the estimated value of the weight coefficient matrix, is the estimated value of the system output current information, , and are all adjustable weight coefficient matrices, represents the matrix the transpose matrix of, represents the system input voltage in the coordinate system, represents the adaptive input gain;

[0087] Through the above update law formula, the convergence and stability of the weight coefficient matrix update process are ensured, the weight coefficient matrix can be updated and adjusted on-line, the hyperlocal model can dynamically follow the changes of the actual power grid system, and effectively cope with the influence of parameter changes on the control performance.

[0088] As an implementation, the first-order Euler discretization method is used to discretize the weight coefficient matrix to obtain the current prediction equation.

[0089] The first-order Euler discretization method is represented by the following formula:

[0090] (5);

[0091] In the formula, represents the current, represents the time, represents the control period.

[0092] By discretizing the weight coefficient matrix, the obtained current prediction equation is represented by the following formula:

[0093] (6);

[0094] In the formula, T s represents the control period, that is, the control algorithm is executed once every T s .

[0095] As an implementation, eliminating the dependence of the adaptive input gain on the system parameters includes:

[0096] Considering the current information of the

[0097] axis and discretizing the current information, the following current equation can be obtained:

[0098] Regarding the estimated value of the parameter as , then there is , and at this time the current prediction formula can be expressed as follows:

[0099] (8);

[0100] Taking the difference between formula (7) and formula (8) and designing the update law of the adaptive input gain, the update law of the adaptive input gain is represented by the following formula:

[0101] (9);

[0102] In the formula, the parameter represents the adaptive input gain, the parameter represents the estimated value of the parameter , represents the output current information of the axis, The predicted output current information of the shaft, denotes the voltage information of the shaft, denotes the control period.

[0103] Furthermore, a step-size factor is introduced and the case where the denominator in formula (9) is zero is considered to improve the rigor of the parameter update law. The update law of the adaptive input gain can be expressed by the following formula:

[0104] (10);

[0105] where p represents the step-size factor.

[0106] By designing the weight coefficient matrix and the update law of the input gain, the current prediction equation is obtained and the estimation of the hyper-local model and the adaptive input gain is completed. Moreover, the dependence of the hyper-local model and the adaptive input gain on the system parameters is eliminated, realizing the parameter-free predictive control of the power grid system.

[0107] To illustrate the reliability of the online neural network estimator, the stability analysis of the online neural network estimator will be carried out according to the Lyapunov stability principle as follows:

[0108] First, the estimation error of the weight coefficient matrix is defined as , then the error equation of the online neural network estimator can be expressed as follows:

[0109] (11);

[0110] In a control period , when the control period is too short, the change of the input gain can be ignored, that is , and its observation error can be defined as , then the error equation can be constructed as follows:

[0111] (12);

[0112] From the above formula, it can be obtained that by analyzing the stability of the online neural network estimator when , the Lyapunov equation of the online neural network estimator can be constructed as:

[0113] (13);

[0114] Define the parameters , , then the derivative form of the Lyapunov equation can be expressed as follows:

[0115] (14);

[0116] By analyzing the stability of the online neural network estimator through scaling, it can be obtained that:

[0117] (15);

[0118] Define the step size factor satisfying , in addition, define the step size factor p as follows:

[0119] (16);

[0120] At this time, the derivative form of Lyapunov satisfies:

[0121] (17);

[0122] Further, define the function as follows:

[0123] (18);

[0124] Adopt equation , then there is:

[0125] (19);

[0126] Through the above description, it is proved that the online neural network estimator maintains good stability in estimating the hyperlocal model and the adaptive input gain.

[0127] As an implementation method, the control method further includes: designing a neutral point voltage update law, and the neutral point voltage update law is expressed by the following formula:

[0128] (20);

[0129] Wherein, p represents the step size factor, represents the estimated value of , , C dc represents the bus capacitor on the DC side.

[0130] Predict the neutral point voltage of the power grid system through the following formula:

[0131] (21);

[0132] (22);

[0133] Wherein, represents the estimated value of the neutral point voltage, denotes the neutral point voltage, represents the neutral point voltage update law, represents the three switch expressions of the voltage vectors used.

[0134] Substituting formula (20) into formula (21) can realize the prediction of the neutral point voltage. By applying a delay compensation strategy and adopting an adaptive architecture similar to that in the online neural network estimator, the prediction of the neutral point voltage can be achieved without increasing the algorithm complexity, and the delay compensation of the converter can be completed.

[0135] To further illustrate the control method of the converter provided by this application, a three-level NPC (Neutral Point Clamped) inverter will be taken as an example for illustration below. The topology diagram of the three-level NPC inverter is as Figure 3 shown. This inverter has a total of three arms. According to the different switching states of the switching tubes, each arm has three working states: 1, 0, and -1. The three arms can form a total of 27 switching states. According to the action characteristics of different voltage vectors, the voltage vectors are divided into four types: large, medium, small, and zero vectors. Among the 27 switching states of the three-level NPC inverter, there are 19 effective vectors, and the remaining 8 are redundant vectors.

[0136] The spatial distribution of the voltage vectors is as Figure 4 shown. Among the 19 effective vectors, the large vectors are: V7(1, -1, -1), V9(1, 1, -1), V11(-1, 1, -1), V13(-1, 1, 1), V15(-1, -1, 1), V17(1, -1, 1); the medium vectors are: V8(1, 0, -1), V10(0, 1, -1), V12(-1, 1, 0), V14(-1, 0, 1), V16(0, -1, 1), V18(1, -1, 0); the small vectors are: V1(1, 0, 0), (0, -1, -1), V2(1, 1, 0), (0, 0, -1), V3(0, 1, 0), (-1, 0, -1), V4(0, 1, 1), (-1, 0, 0), V5(0, 0, 1), (-1, -1, 0), V6(1, 0, 1), (-1, -1, 0); the zero vectors: V0(-1, -1, -1), (0, 0, 0), (1, 1, 1). Among the 8 redundant vectors, there are 6 small vectors and 2 zero vectors. The small vectors among the 8 redundant vectors will generate the same voltage magnitude and phase and generate opposite neutral point voltages.

[0137] As an implementation method, the control method further includes: designing a sequential value function to achieve the control objectives of current tracking and neutral point voltage balance.

[0138] The sequential value function is expressed by the following formula:

[0139] (23);

[0140] In the formula, represents the requirement for the current tracking quality, represents the requirement for the neutral point voltage balance, and represents the requirement for the switching frequency, is the reference value, is the predicted value obtained after using the voltage vector , is the sign function, represents the number of switching changes.

[0141] As an implementation method, according to the predicted current of the converter and the predicted value of the neutral point voltage, based on the sequential value function, the voltage vectors of the converter are screened, the optimal voltage vector is selected, and the converter is controlled using the optimal voltage vector.

[0142] The predicted value of the neutral point voltage is the absolute value of the predicted value obtained by using the voltage vector. For the convenience of explanation, the predicted value of the neutral point voltage is denoted as , and the neutral point voltage threshold is denoted as . Specifically, when the predicted value of the neutral point voltage is less than the neutral point voltage threshold , at this time, current tracking is given priority, and the optimal vector is selected through the cost function in the sequential value function. If the optimal vector is a long vector or a medium vector, it is directly applied to the converter; if the optimal vector is a small vector, the cost function in the sequential value function is used to select the optimal switching state that makes the midpoint voltage closer to zero among the two switching states corresponding to the small vector; if it is a zero vector, the cost function in the sequential value function is used to select the optimal switching state that makes the switching frequency the smallest among the three switching states corresponding to the zero vector.

[0143] Furthermore, when the predicted value of the neutral point voltage exceeds the neutral point voltage threshold , at this time, the neutral point voltage is given priority, and the optimal switching state is selected through the cost functions and in the sequential value function and applied to the converter.

[0144] Through the above voltage vector screening strategy, the selection of the optimal voltage vector can be achieved without the participation of capacitance parameters and weighting factors, avoiding excessive neutral point voltage and further eliminating the dependence of the control method on system parameters.

[0145] In summary, as Figure 5 shown, a control method for a converter provided by this application first performs current sampling, estimates the hyperlocal model F and the adaptive input gain through an online neural network estimator , realizes the prediction of current and neutral point voltage, and screens the optimal voltage vector through a sequential cost function to achieve the optimal control of the converter.

[0146] Next, the control method of the converter provided by this application will be compared with the traditional finite set model predictive control method, the model-free predictive control method based on the extended state observer, and the model-free control method based on the integral sliding mode observer to verify the performance of the control method of the converter provided by this application.

[0147] First, simulate the above methods, and the simulation parameters are shown in Table 1.

[0148]

[0149] Table 1

[0150] Under the condition of parameter matching, the performance of the four control methods is as Figure 6 shown. In the figure, part (a) is the traditional finite set model predictive control method, part (b) is the model-free predictive control method based on the extended state observer, part (c) is the model-free control method based on the integral sliding mode observer, and part (d) is the control method provided by this application. Under the condition of parameter matching, the THD (Total Harmonic Distortion) value of the control method provided by this application is 1.09%, which is lower than the other three control methods.

[0151] Under the condition of input gain mismatch (-70%), the performance of the four control methods is as Figure 7 shown. Under the condition of input gain mismatch (-70%), the THD value of the control method provided by this application is 1.05%, which is lower than the other three control methods. Further, under the condition of inductor parameter mismatch (-40%), the performance of the four control methods is as Figure 8 shown. Under the condition of inductor parameter mismatch (-40%), the THD value of the control method provided by this application is 1.79%, which is lower than the other three control methods.

[0152] Under the above parameter matching and parameter mismatch conditions, the control method provided by this application has obtained lower current harmonic content, and the control performance of the control method provided by this application is better than that of other control methods.

[0153] Design different degrees of parameter adaptation conditions to verify the robust range of the above four control methods. As Figure 9 shown, it can be seen that the robust range of the control method provided by this application is wider, and the control method of the converter provided by this application has better robustness. Further, as Figure 10 shown, taking the true inductance value of 10 mH as the reference, with a 50% mismatch up and down, verify the estimation effect of the online neural network estimator on the input gain, and realize the online adjustment of the input gain.

[0154] Further, a power conversion experimental platform based on 3L-NPC was designed. The specific parameters of the platform are the same as the above simulation parameters. Comparative experiments were conducted under two conditions of parameter matching and parameter mismatch. The results of the above four control methods under the parameter matching condition are as Figure 11a 、 Figure 11b 、 Figure 11c 、 Figure 11d shown; under the parameter mismatch condition (L - 36%), the results of the four methods are as Figure 12a 、 Figure 12b 、 Figure 12c 、 Figure 12d shown.

[0155] Under the condition of parameter mismatch, the control method provided by this application has improved the current quality by 80.34% compared with the traditional finite set model predictive control method, and has improved by 54.94% and 56.29% respectively compared with the existing two methods based on the hyperlocal model. The robustness of the control method provided by this application is better than the other three control methods, and the control system provided by this application can operate with high current quality under the condition of severe parameter mismatch.

[0156] According to the above description, a control method for a converter provided by this application uses an online neural network estimator to realize the online estimation of the hyperlocal model and the adaptive input gain, realizes the real-time update of the parameters in the hyperlocal model, eliminates the dependence of the hyperlocal model and the adaptive input gain on the system parameters, and realizes the parameter-free operation of the control system; also obtains the voltage vector for controlling the converter based on the predicted current of the converter, and realizes the balance of the neutral point voltage of the converter; and uses an adaptive architecture similar to that in the online neural network estimator to realize the prediction of the neutral point voltage, realizes the prediction of the neutral point voltage without increasing the algorithm complexity, and realizes the delay compensation of the converter.

[0157] In a second aspect, the present application further provides a power grid system. The power grid system includes a converter. The power grid system applies the control method of the converter described above. The converter can achieve current tracking of the converter and neutral point voltage balance, and realize parameterless operation of the control system.

[0158] It can be understood that the term "exemplary" used herein means "as an example, illustration, or instance". Any embodiment described as "exemplary" is not necessarily preferred or superior to other embodiments and / or does not exclude combining the features of other embodiments. It should be understood that certain features of the present application described in the context of separate embodiments may also be provided in combination in a single embodiment. Conversely, the various features of the present application described in the context of a single embodiment may also be provided separately or in any suitable combination or as any other described embodiment of the present application.

[0159] The above-disclosed are only the preferred embodiments of the present application, but they are not intended to limit the scope of the rights of the present application. Those of ordinary skill in the art can understand that within the spirit and scope of the present application and the appended claims, changes, modifications, substitutions, combinations, and simplifications should all be equivalent replacement methods and still fall within the scope covered by the invention.

Claims

1. A control method for a converter, the control method being applied to control the converter in a power grid system, characterized in that: The control method comprises the following steps: Constructing an adaptive hyperlocal model, the adaptive hyperlocal model including an adaptive input gain and a hyperlocal model, the hyperlocal model including physical parameters in the controlled system and disturbance factors in the power grid system; An online neural network estimator is used to estimate the hyperlocal model and the adaptive input gain, thereby eliminating the dependence of the hyperlocal model and the adaptive input gain on system parameters; Predicting the output current of the converter based on the hyperlocal model and the adaptive input gain, generating a voltage vector of the converter based on the predicted current of the converter, and controlling the converter using the voltage vector; The control method further comprises: The hyperlocal model is expressed by the following formula: ； in, represents the weight coefficient matrix, is the activation function, is the estimation error, represent The system output current in the coordinate system; The weight coefficient matrix is ​​updated online, and its update rate is expressed by the following formula: ； in, represents the estimated value of the weight coefficient matrix, is the estimated value of the system output current information, , and are all adjustable weight coefficient matrices, Representative Matrix The transposed matrix of represent The system input voltage in the coordinate system, represents the adaptive input gain; The weight coefficient matrix is ​​discretized to obtain a current prediction equation.

2. The control method of the converter according to claim 1, characterized in that: The weight coefficient matrix is ​​discretized using the first-order Euler discretization method, and the current prediction equation obtained is expressed by the following formula: ； In the formula, Indicates the control period.

3. The control method of the converter according to claim 1, characterized in that: Eliminating the dependence of the adaptive input gain on system parameters includes: The update law of the adaptive input gain is designed, and the update law of the adaptive input gain is expressed by the following formula: ； In the formula, the parameters represents the adaptive input gain, parameter Representation parameters The estimated value of express Output current information of the axis, express The predicted output current information of the axis, express Shaft voltage information, Indicates the control period.

4. The control method of the converter according to claim 3, characterized in that: The control method further comprises: Introducing the step size factor, the update law of the adaptive input gain is expressed by the following formula: ； Wherein, p represents the step size factor.

5. The control method of the converter according to claim 1, characterized in that: The control method further comprises: Design a neutral point voltage update law, which is expressed by the following formula: ； In the formula, p represents the step size factor, Representatives The estimated value of , C dc Indicates the bus capacitance on the DC side; The neutral point voltage of the power grid system is predicted by the following formula: ； ； In the formula, represents the estimated value of the neutral point voltage, Indicates the neutral point voltage, represents the neutral point voltage update law, Three switch expressions representing the voltage vector used; delay compensation of the converter is achieved to achieve the control target of neutral point voltage balance.

6. The control method of the converter according to claim 5, characterized in that: The control method further comprises: Design a sequential value function, which is expressed by the following formula: ； In the formula, It represents the requirements for current tracking quality. represents the balance requirement for the neutral point voltage, and It represents the requirement for switching frequency. is the reference value, That is, using the voltage vector The predicted value obtained later is is the symbolic function, Represents the number of switch changes; According to the predicted current and the predicted neutral point voltage of the converter, based on the sequential value function, the voltage vector of the converter is screened, the optimal voltage vector is selected, and the converter is controlled using the optimal voltage vector.

7. The control method of the converter according to claim 6, characterized in that: The control method further comprises: When the predicted neutral point voltage value is less than the neutral point voltage threshold, the cost function in the sequential value function is used to Selecting an optimal vector, and if the optimal vector is a long vector or a medium vector, directly applying it to the converter; If the optimal vector is a small vector, the cost function in the sequential value function is used Select the optimal switching state that makes the midpoint voltage closer to zero; If it is a zero vector, the cost function in the sequential value function is used Select the optimal switching state that minimizes the switching frequency.

8. The control method of the converter according to claim 7, characterized in that: The control method further comprises: When the predicted neutral point voltage value exceeds the neutral point voltage threshold, the cost function in the sequential value function is used. and The optimal switching state is selected and applied to the converter.

9. A power grid system, comprising a converter, characterized in that: The power grid system applies the converter control method according to any one of claims 1 to 8.

Images