Permanent magnet synchronous generator parameter-free model predictive control method based on dynamic mode decomposition

Abstract

The application relates to a permanent magnet synchronous generator parameter-free model predictive control method based on dynamic mode decomposition, which comprises the following steps: providing a target permanent magnet synchronous generator, and performing system identification on the target permanent magnet synchronous generator to determine model predictive state parameters for model predictive control of the target permanent magnet synchronous generator, wherein the model predictive state parameters at least include a state matrix A and a control matrix B; configuring the identified model predictive state parameters in a generator predictive control model in a closed-loop controller to configure the closed-loop controller to perform closed-loop control on a bus terminal voltage output by the target permanent magnet synchronous generator; and in the process of closed-loop control, the model predictive control is performed by using the generator predictive control model. The application can effectively control the bus terminal voltage output by the PMSG under the condition that the initial parameters are unknown or the parameters are mismatched, and has the characteristics of strong robustness, data-driven adaptability and high dynamic response.

Description

Technical Field

[0001] This invention relates to a model predictive control method, and more particularly to a parameterless model predictive control method for permanent magnet synchronous generators based on dynamic mode decomposition. Background Technology

[0002] Permanent magnet synchronous generators (PMSGs) are commonly used power generation devices, widely applied in renewable energy fields, especially wind power and ocean power generation. PMSGs are characterized by high efficiency, low maintenance requirements, and strong reliability. Their working principle is based on the interaction between the magnetic field generated by permanent magnets and the stator windings, converting mechanical energy into electrical energy through electromagnetic induction.

[0003] Currently, PMSG control typically relies on traditional control methods such as vector control and direct power control, which usually require accurate modeling of the PMSG system. However, when faced with the nonlinearity, dynamic changes, and external disturbances of the PMSG system, traditional methods often struggle to achieve optimal dynamic performance. Therefore, traditional control methods are insufficient for optimal control of PMSGs.

[0004] Dynamic Mode Decomposition (DMD) is a data-driven dimensionality reduction method that extracts key modal information from high-dimensional dynamic system time-series data, identifying the system's inherent dynamic characteristics. Traditional DMD methods are primarily applied in fluid dynamics, structural health monitoring, and signal processing. Specifically, DMD effectively analyzes the dynamic behavior of a system and captures its temporal evolution. DMDc is a method that combines dynamic mode decomposition with control; it not only extracts dynamic modal information from real-time system data but also enables the design of effective control strategies based on this information.

[0005] Unlike traditional control methods, DMDc uses a data-driven approach, modeling the system using its time-series data rather than relying on precise mathematical models. This allows it to adapt to the time-varying and uncertainties of the system. However, effectively controlling the PMSG output bus voltage accurately when the initial parameters of the PMSG are unknown or when PMSG parameters are mismatched remains a pressing technical challenge. Summary of the Invention

[0006] The purpose of this invention is to overcome the shortcomings of the existing technology and provide a parameterless model predictive control method for permanent magnet synchronous generators based on dynamic mode decomposition. This method can effectively control the bus terminal voltage output by the PMSG under unknown initial parameters or parameter mismatch, and has the characteristics of strong robustness, data-driven adaptability and high dynamic response.

[0007] According to the technical solution provided by this invention, a parameterless model predictive control method for permanent magnet synchronous generators based on dynamic mode decomposition is provided, the parameterless model predictive control method for permanent magnet synchronous generators comprising:

[0008] A target permanent magnet synchronous generator is provided, and system identification is performed on the target permanent magnet synchronous generator to determine the model predictive state parameters for model predictive control of the target permanent magnet synchronous generator after system identification.

[0009] The model predicts state parameters that include at least a state matrix A and a control matrix B.

[0010] During system identification, the target permanent magnet synchronous generator is configured to generate an observable dynamic response. Subsequently, the model prediction state parameters corresponding to the target permanent magnet synchronous generator are identified and determined based on the DMDc method. During the identification process, the target permanent magnet synchronous generator is configured to be in the open-loop state of the bus voltage.

[0011] The identified model-predicted state parameters are configured within the generator predictive control model of the closed-loop controller. This configures the closed-loop controller to perform closed-loop control on the bus terminal voltage output by the target permanent magnet synchronous generator. During the closed-loop control process, model predictive control is performed using the generator predictive control model.

[0012] In the kth sampling control cycle of the closed-loop control, the power generation state information of the target permanent magnet synchronous generator at time k and the power generation control information at time k are loaded into the generator predictive control model so as to perform model predictive control using the generator predictive control model and generate the power generation state information at time k+1. The power generation state information includes the q-axis current and the d-axis current.

[0013] Based on the power generation state information at time k+1 and the power generation state reference information at time k, the closed-loop control value function representing the corresponding error states of the q-axis current and d-axis current at time k+1 is calculated.

[0014] Based on the optimal state of the closed-loop control value function at time k+1, the target rectifier switch control information for controlling the rectifier's rectification state is determined. Then, based on the target rectifier switch control information, the bus terminal voltage, adapted to the bus reference voltage and output by the rectifier, is configured.

[0015] At least based on the bus reference voltage and the bus sample voltage at time k, the power generation state reference information at time k is generated, wherein the bus sample voltage is generated by sampling the bus terminal voltage of the rectifier output.

[0016] Closed-loop control also includes parameter mismatch adjustment processing, among which,

[0017] When performing parameter mismatch control, the following are included:

[0018] The target for parameter mismatch monitoring should be determined, and the target for parameter mismatch monitoring should include at least the q-axis current.

[0019] The parameter mismatch state is determined based on the parameter mismatch monitoring target. When the parameter mismatch state is that the parameters are mismatched, the model prediction state parameters are updated online, and model prediction state parameters adapted to the target permanent magnet synchronous generator are generated online. Then, the model prediction state parameters are configured in the generator predictive control model so that model predictive control can be performed in the closed-loop control.

[0020] When determining the parameter mismatch state based on the parameter mismatch monitoring target, if multiple consecutive q-axis current residuals match the parameter mismatch monitoring threshold, the parameter mismatch state is configured as parameter mismatched. For any q-axis current residual, the q-axis current residual is generated based on the difference between the corresponding q-axis currents at two adjacent times.

[0021] For the q-axis current at time m, the q-axis current residual is:

[0022]

[0023] Among them, i q(m) Let i be the q-axis current at time m. q(m-1) Let be the q-axis current at time m-1. Let be the q-axis current residual corresponding to the q-axis current at time m;

[0024] With q-axis current residual The corresponding parameter mismatch monitoring threshold is:

[0025]

[0026] in, This is the parameter mismatch monitoring threshold corresponding to the q-axis current at time m. This is the average value calculated based on the q-axis current corresponding to the n sampling control cycles prior to time m. Let λ be the standard deviation calculated based on the q-axis current corresponding to the n sampling control cycles prior to time m. m This is the adjustment coefficient corresponding to the q-axis current at time m;

[0027] when When, then the q-axis current residual Monitoring threshold for parameter mismatch match.

[0028] Adjustment coefficient λ m The adjustment coefficient λ is determined using a fuzzy control method.m Then:

[0029] Determine the residual of the q-axis current Corresponding residual rate of change Among them, the rate of change of residuals Then we have: Let T be the q-axis current residual corresponding to the q-axis current at time m-1. s The sampling control period;

[0030] q-axis current residual and residual change rate The variable is loaded into the constructed fuzzy controller so that the corresponding adjustment coefficient increment Δλ can be calculated by the fuzzy controller. m ,

[0031] Based on the adjustment coefficient increment Δλ m Generate adjustment coefficient λ m Then we have: λ m =λ m-1 +Δλ m , where λ m-1 This is the adjustment coefficient corresponding to the q-axis current at time m-1.

[0032] For the closed-loop control value function, we have:

[0033]

[0034] in, Let be the closed-loop control value function at time k+1. Let q be the current at time k+1. The given value for the q-axis current at time k is... Let q be the current at time k+1. The given value for the d-axis current at time k;

[0035] When determining the optimal state of the closed control value function at time k+1, we have:

[0036] Iterate through the rectification states of the rectifier and the rectifier switch state information for each rectification state, and obtain the bus sampling voltage for each rectification state. Then, based on the bus sampling voltage, generate at least k q-axis current setpoints.

[0037] Based on the generated q-axis current setpoint at time k The closed-loop control value function value at time k+1 is calculated, and the smallest closed-loop control value function value at time k+1 is configured as the optimal state of the closed-loop control value function value at time k+1. The rectifier switch state information corresponding to the optimal state of the closed-loop control value function value at time k+1 is configured as the target rectifier switch control information.

[0038] The q-axis current given at time k At least the bus reference voltage and the bus sampled voltage at time k are generated by PI calculation;

[0039] The d-axis current at time k is given by... Set to 0.

[0040] When performing system identification on the target permanent magnet synchronous generator, the following steps are included:

[0041] Construct the DMDc state-space equation corresponding to the target permanent magnet synchronous generator and configure the identification reference voltage information, wherein the identification reference voltage information includes the d-axis reference voltage and q-axis reference voltage of multiple reference segments;

[0042] After configuring the target permanent magnet synchronous generator to generate an observable dynamic response, the basic voltage vector of the target permanent magnet synchronous generator under the dynamic response is determined, wherein the basic voltage vector includes the d-axis response voltage and the q-axis response voltage.

[0043] The open-loop identification value function values ​​representing the d-axis response voltage and q-axis response voltage, respectively, are calculated and compared with the d-axis reference voltage and q-axis reference voltage under each reference segment. The optimal value of the open-loop identification value function value under each reference segment is determined, so as to determine the target identification rectification state of the rectifier based on the optimal value of the open-loop identification value function value under each reference segment. Subsequently, data sampling processing is performed under each reference segment to collect and generate corresponding identification sampling data. For each identification sampling data, the identification sampling data includes identification state quantity and identification control quantity.

[0044] The identified state quantities include the identified d-axis current and the identified q-axis current;

[0045] The identification control quantities include the identification of d-axis voltage, the identification of q-axis voltage, and the angular velocity of the target permanent magnet synchronous generator;

[0046] The state-space equations of the DMDc constructed above are solved using all the identified sampling data to obtain the model's predicted state parameters.

[0047] When configuring the target permanent magnet synchronous generator to produce an observable dynamic response, the target permanent magnet synchronous generator should be driven to rotate at a low speed by the engine at least.

[0048] For the d-axis reference voltage with multiple reference segments, we have:

[0049]

[0050] For a q-axis reference voltage with multiple reference segments, then:

[0051]

[0052] in, This is the bus reference voltage. The d-axis reference voltage. f is the q-axis reference voltage. N The fundamental frequency of the target permanent magnet synchronous generator.

[0053] During data sampling, periodic downsampling is performed within each reference segment of the d-axis reference voltage, and transient full sampling is performed when the d-axis reference voltage jumps from one reference segment to another adjacent reference segment.

[0054] For the open-loop identification value function, we have:

[0055]

[0056] Among them, J e To identify the value function in an open-loop manner, The d-axis response voltage. This is the q-axis response voltage;

[0057] When determining the optimal value of the open-loop identification value function for each reference segment, the d-axis response voltage q-axis response voltage The same reference segment corresponding to the d-axis reference voltage and the q-axis reference voltage.

[0058] The advantages of this invention are as follows: For the target permanent magnet synchronous generator, the model prediction state parameters for model predictive control of the target permanent magnet synchronous generator are determined during the system identification stage. Subsequently, the model prediction state parameters are configured in the generator predictive control model of the closed-loop controller so that model predictive control can be performed using the generator predictive control model during the closed-loop control process. This enables effective control of the bus terminal voltage output by the PMSG under unknown initial parameters or parameter mismatch, and has the characteristics of strong robustness, data-driven adaptability and high dynamic response. Attached Figure Description

[0059] Figure 1 This is a schematic flowchart of an embodiment of the parameterless model predictive control method for permanent magnet synchronous generators of the present invention.

[0060] Figure 2This is a structural block diagram of one embodiment of the present invention for controlling a target permanent magnet synchronous generator.

[0061] Figure 3 This is a schematic diagram of one embodiment of the bus terminal voltage identification stage of the present invention.

[0062] Figure 4 This is a waveform diagram illustrating one embodiment of the bus terminal voltage and torque during the system identification and closed-loop control switching process of the present invention.

[0063] Figure 5 This is a waveform diagram illustrating an embodiment of the bus terminal voltage and torque response under varying operating conditions according to the present invention. Detailed Implementation

[0064] The present invention will be further described below with reference to specific accompanying drawings and embodiments.

[0065] To effectively control the bus terminal voltage output by a PMSG with unknown initial parameters, this invention provides a parameterless model predictive control method for permanent magnet synchronous generators based on dynamic mode decomposition. Specifically, the parameterless model predictive control method for permanent magnet synchronous generators includes:

[0066] A target permanent magnet synchronous generator is provided, and system identification is performed on the target permanent magnet synchronous generator to determine the model predictive state parameters for model predictive control of the target permanent magnet synchronous generator after system identification.

[0067] The model predicts state parameters that include at least a state matrix A and a control matrix B.

[0068] During system identification, the target permanent magnet synchronous generator is configured to generate an observable dynamic response. Subsequently, the model prediction state parameters corresponding to the target permanent magnet synchronous generator are identified and determined based on the DMDc method. During the identification process, the target permanent magnet synchronous generator is configured to be in the open-loop state of the bus voltage.

[0069] The identified model-predicted state parameters are configured within the generator predictive control model of the closed-loop controller. This configures the closed-loop controller to perform closed-loop control on the bus terminal voltage output by the target permanent magnet synchronous generator. During the closed-loop control process, model predictive control is performed using the generator predictive control model.

[0070] In the kth sampling control cycle of the closed-loop control, the power generation state information of the target permanent magnet synchronous generator at time k and the power generation control information at time k are loaded into the generator predictive control model so as to perform model predictive control using the generator predictive control model and generate the power generation state information at time k+1. The power generation state information includes the q-axis current and the d-axis current.

[0071] Based on the power generation state information at time k+1 and the power generation state reference information at time k, the closed-loop control value function representing the corresponding error states of the q-axis current and d-axis current at time k+1 is calculated.

[0072] Based on the optimal state of the closed-loop control value function at time k+1, the target rectifier switch control information for controlling the rectifier's rectification state is determined. Then, based on the target rectifier switch control information, the bus terminal voltage, adapted to the bus reference voltage and output by the rectifier, is configured.

[0073] At least based on the bus reference voltage and the bus sample voltage at time k, the power generation state reference information at time k is generated, wherein the bus sample voltage is generated by sampling the bus terminal voltage of the rectifier output.

[0074] Figure 1 The diagram shows a flowchart of an embodiment of the present invention for parametric model predictive control of a permanent magnet synchronous generator. As can be seen from the diagram, when performing parametric model predictive control, a target permanent magnet synchronous generator should be provided. That is, the target permanent magnet synchronous generator and the corresponding connected rectifier are the control objects of the parametric model predictive control of the present invention. The target permanent magnet synchronous generator can be a commonly used permanent magnet synchronous generator, that is, the type of target permanent magnet synchronous generator can be selected as needed.

[0075] It should be noted that "parameterless" in this invention specifically refers to not knowing or not relying on the initial parameters of the target permanent magnet synchronous generator. Furthermore, when performing parameterless model predictive control on the target permanent magnet synchronous generator, the control objective should be to ensure that the output bus terminal voltage of the target permanent magnet synchronous generator is matched with the bus reference voltage. Matching the bus terminal voltage with the bus reference voltage specifically means that the bus terminal voltage is consistent with the bus reference voltage, or that the difference between the bus terminal voltage and the bus reference voltage is within an allowable numerical range. The allowable numerical range of the difference is generally related to the control accuracy of the output bus terminal voltage of the target permanent magnet synchronous generator, and can be selected and determined according to the control accuracy.

[0076] Specifically, the bus terminal voltage output by the target permanent magnet synchronous generator refers to the DC voltage obtained by rectifying the three-phase voltage output by the target permanent magnet synchronous generator through a rectifier and then through a bus capacitor. Figure 2 The figure shows one embodiment of obtaining the bus terminal voltage. Figure 2In the diagram, ZS is the rectifier connected to the target permanent magnet synchronous generator, and C1 is the bus capacitor connected to the rectifier. In specific implementation, the rectifier can be a commonly used three-phase rectifier. Generally, the rectifier can include three bridge arms, each bridge arm including two bridge arm switching transistors. The bridge arm switching transistors can be commonly used switching transistors, such as IGBT devices. The method of using IGBT devices as bridge arm switching transistors to form the rectifier and the rectification method can be consistent with existing technology, and will not be elaborated here.

[0077] To control the bus terminal voltage output by the target permanent magnet synchronous generator (PMSG), system identification of the PMSG is first required. System identification determines the model predictive state parameters (MPCs) used for model predictive control of the PMSG. In other words, after system identification, the MPCs are obtained, and then model predictive control (MPC) can be performed using these parameters. To meet the requirements of MPC, the MPCs of this invention include at least a state matrix A and a control matrix B. Generally, when performing MPC, only state matrix A and control matrix B can be used. Of course, the MPC state parameters are determined based on what satisfies the requirements of MPC. When the model predictive state parameters are state matrix A and control matrix B, the method of performing MPC can be referred to the corresponding description below.

[0078] In order to meet the requirements of system identification, a target permanent magnet synchronous generator should be configured to generate an observable dynamic response. For example, the target permanent magnet synchronous generator can be driven to rotate at low speed by the engine. That is, when the target permanent magnet synchronous generator is driven to rotate at low speed, it can generate an observable dynamic response. The dynamic response is observable, specifically, the dynamic response of the target permanent magnet synchronous generator can be measured. The dynamic response can be referred to in the following description.

[0079] Specifically, once the target permanent magnet synchronous generator (PMSG) generates an observable dynamic response, the model predicted state parameters can be identified and determined using the DMDc method. It is understood that the model predicted state parameters should correspond to the target PMSG, meaning they should reflect the main modal information of the target PMSG. The method and process of identifying and determining the model predicted state parameters using the DMDc method will be explained in detail below. It should be noted that during the identification process, the target PMSG is configured in a bus voltage open-loop state. Specifically, the bus voltage open-loop state means that the bus terminal voltage is not collected during the identification process; that is, the aforementioned bus terminal voltage is not used to form an open-loop control state.

[0080] It should be understood that in order to accurately control the bus terminal voltage and achieve the matching of the bus terminal voltage with the bus reference voltage, the bus terminal voltage of the target permanent magnet synchronous generator should be subject to closed-loop control. The closed-loop control corresponds to the open-loop control state mentioned above. Therefore, during closed-loop control, the bus terminal voltage should be utilized, such as sampling the bus terminal voltage to obtain the bus sample voltage.

[0081] To achieve closed-loop control, a closed-loop controller should be constructed. Figure 2 The closed-loop control is the constructed closed-loop controller, which should include a generator predictive control model. Model predictive control can be executed using the generator predictive control model. Figure 2 The predicted current model in the model is the generator predictive control model, which primarily performs current prediction. It's understood that the predicted state parameters must first be configured within the generator predictive control model before it can execute model predictive control. Therefore, when using a closed-loop controller to perform closed-loop control on the bus terminal voltage of the target permanent magnet synchronous generator, and during this closed-loop control process, the generator predictive control model can be used for model predictive control. In other words, during closed-loop control, the generator predictive control model is required for model predictive control; thus, the model predictive control executed by the generator predictive control model is a control stage within the closed-loop control process.

[0082] It should be understood that when using a closed-loop controller for closed-loop control of the bus terminal voltage, it includes sequential sampling control cycles. The size of the sampling control cycle is related to the voltage control requirements of the bus terminal voltage and can generally be selected based on actual needs. In each sampling control cycle, sampling is performed to obtain closed-loop control sampling information. For example, in the k-th sampling control cycle, sampling yields the closed-loop control sampling information at time k. This information can be the generation status information and generation control information at time k. It can be understood that the time interval between two adjacent sampling control cycles is one sampling control cycle Ts.

[0083] Specifically, in the k-th sampling control cycle of the closed-loop control, the power generation state information of the target permanent magnet synchronous generator at time k and the power generation control information at time k are loaded into the generator predictive control model. Thereafter, the generator predictive control model can be used to perform model predictive control and generate the power generation state information at time k+1. The power generation state information includes the q-axis current and the d-axis current. Therefore, the power generation state information at time k+1 is the q-axis current and the d-axis current at time k+1. The power generation control information may include the d-axis voltage, the q-axis voltage, and the angular velocity of the target permanent magnet synchronous generator.

[0084] Specifically, when the generator predictive control model executes model predictive control, then:

[0085]

[0086] Among them, i q (k+1|k) represents the q-axis current at time k+1 predicted based on the generation state information and generation control information at time k. d (k+1|k) represents the d-axis current at time k+1 predicted based on the generation state information and generation control information at time k. q (k|k) represents the q-axis current at time k, i d (k|k) represents the d-axis current at time k. Let be the q-axis voltage at time k. Let be the d-axis voltage at time k. Let be the angular velocity at time k.

[0087] It should be noted that the q-axis current i at time k q (k|k), the d-axis current i at time k d (k|k) can be obtained by direct measurement and conversion. Figure 1 The figure shows the q-axis current i at time k. q (k|k), the d-axis current i at time k d In one embodiment of (k|k), specifically, when performing closed-loop control on the target permanent magnet synchronous generator, the operating current i of the target permanent magnet synchronous generator can be measured. s Then, by transforming from the abc coordinate system to the dq coordinate system, the q-axis current i at time k can be obtained. q (k|k), the d-axis current i at time k d (k|k), the specific method and process of transforming the abc coordinate system to the dq coordinate system can be consistent with the existing technology, and will not be elaborated here.

[0088] Specifically, the q-axis voltage at time k d-axis voltage at time k angular velocity at time k It can be calculated using commonly used computational methods in existing technologies, such as obtaining the electrical angle θ of the target permanent magnet synchronous generator at time k and the voltage at the rectifier output bus. Subsequently, the q-axis voltage at time k can be calculated. d-axis voltage at time k angular velocity at time k For specific calculation methods, commonly used existing methods can be adopted, such as referring to the relevant instructions in the book "Modern Permanent Magnet Motor Control Principles and MATLAB Simulation". Figure 2 In the middle, e a eb e c These are the phase a voltage, phase b voltage, and phase c voltage output by the target permanent magnet synchronous generator, respectively.

[0089] It should be understood that when controlling the bus terminal voltage of the target permanent magnet synchronous motor, the main focus is on controlling the rectification state of the rectifier, that is, controlling the rectification state of the rectifier so that the bus terminal voltage generated after rectification and through the bus capacitor C1 matches the bus reference voltage. Therefore, when the closed-loop controller performs closed-loop control, it mainly generates rectifier switch control information to control the rectification state of the rectifier. Figure 2 S in abc This refers to the rectifier switch control information that controls the rectifier's rectification state.

[0090] It is understandable that the rectifier switch control information includes multiple configurable switch control signals for the corresponding bridge arm switch states within the rectifier. The switch control signals can be digital signals. For example, when the switch control signal is "1", the corresponding connected bridge arm switch can be driven into the conducting state. When the switch control signal is "0", the corresponding connected bridge arm switch should be in the disconnected state. Specifically, the rectification state of the rectifier refers to the rectifier rectifying the three-phase voltage generated by the target permanent magnet synchronous generator.

[0091] When a generator predictive control model is used for model predictive control within the closed-loop controller, in order to generate rectifier switch control information, in one embodiment of the present invention, a closed-loop control value function unit should be provided. Figure 2 The value function J in bh The location is the closed-loop control value function unit, which can perform the calculation and judgment of the closed-loop control value function value. In order to generate rectifier switch control information, the generation state information at time k+1 should be loaded into the closed-loop control value function unit, such as... Figure 2 As shown in the figure, i q (k+1),i d (k+1) represent the q-axis current and d-axis current at time k+1, respectively. To calculate the closed-loop control value function, the generation state reference information at time k should also be loaded into the closed-loop control value function unit. This generation state reference information should include both the q-axis reference current and the d-axis reference current. Figure 2 middle, This refers to the q-axis reference current, which can be understood as the d-axis reference current. Not in Figure 2 As shown in the image.

[0092] In practical implementation, a closed-loop control value function should be set at least within the closed-loop control value function unit. Calculating the closed-loop control value function value involves calculating the error states representing the q-axis current and d-axis current at time k+1, which corresponds to the error states of the q-axis current at time k+1 and the q-axis reference current at time k, as well as the error states of the d-axis current at time k+1 and the d-axis reference current at time k. Therefore, for the closed-loop control value function, we have:

[0093]

[0094] in, Let be the closed-loop control value function at time k+1. Let q be the current at time k+1. The given value for the q-axis current at time k is... Let q be the current at time k+1. The given value for the d-axis current at time k.

[0095] In one embodiment of the present invention, the q-axis current at time k is given a value. At least the bus reference voltage and the bus sampled voltage at time k are generated by PI calculation;

[0096] The d-axis current at time k is given by... Set to 0.

[0097] Figure 2 The figure shows the q-axis current setpoint generated at time k. One embodiment, Figure 2 Middle,U dc For bus sampling voltage, This is the bus reference voltage. This refers to setting the output voltage via the bus capacitor; therefore, the bus reference voltage... The value of can be selected and determined according to the actual application requirements. When performing PI calculations, the bus voltage difference between the bus sampling voltage and the bus reference voltage should be calculated first. Then, the PI calculation should be performed, and the q-axis current setpoint at time k can be obtained from the PI calculation. The method and process for calculating the bus voltage difference using PI control can be consistent with existing PI control, and will not be elaborated here. In specific implementation, the d-axis current setpoint at time k can be used... Set to 0.

[0098] Generally, the bus sampling voltage can be consistent with the bus terminal voltage, that is, the sampling ratio is 1:1. Since the bus terminal voltage may be different at different control cycles, the q-axis current setpoint at time k is calculated. The bus terminal voltage should be sampled during the kth sampling control cycle to obtain the bus sample voltage.

[0099] As can be seen from the above explanation, the q-axis current setpoint at time k is... The value of the closed-loop control function is related to the bus terminal voltage at time k, and the bus terminal voltage is related to the rectification state of the rectifier. Therefore, after calculating the closed-loop control value function value, the optimal state of the closed-loop control value function value at time k+1 can be determined based on the rectification state of the rectifier. Furthermore, the optimal state of the closed-loop control value function value at time k+1 should correspond to the rectification state of the rectifier, thus determining the target rectifier switch control information for the rectifier's rectification state. After determining the target rectifier switch control information, the aforementioned... Figure 2 S in abc The target value.

[0100] It should be understood that after determining the target rectifier switch control information, the rectifier's rectification state should be configured using the target rectifier switch control information. At this time, the rectifier is in the target rectification state, so that the rectifier based on the target rectification state can make the bus terminal voltage match the bus reference voltage.

[0101] In one embodiment of the present invention, when determining the optimal state of the closed control value function at time k+1, the following is true:

[0102] Iterate through the rectification states of the rectifier and the rectifier switch state information for each rectification state, and obtain the bus sampling voltage for each rectification state. Then, based on the bus sampling voltage, generate at least k q-axis current setpoints.

[0103] Based on the generated q-axis current setpoint at time k The closed-loop control value function value at time k+1 is calculated, and the smallest closed-loop control value function value at time k+1 is configured as the optimal state of the closed-loop control value function value at time k+1. The rectifier switch state information corresponding to the optimal state of the closed-loop control value function value at time k+1 is configured as the target rectifier switch control information.

[0104] It is understandable that the rectification state of the rectifier is the rectification state of the three-phase voltage generated by the rectifier on the target permanent magnet synchronous generator. As explained above, the rectification state of the rectifier is related to the switching state of the bridge arm switch tubes. Therefore, according to the type of rectifier used, all rectification states of the rectifier and the rectifier switch state information in each rectification state can be traversed. The rectifier switch state information is the control information for controlling each bridge arm switch tube.

[0105] Specifically, in each rectification state, a corresponding bus sampling voltage is obtained, and then the q-axis current setpoint at time k can be calculated. Subsequently, the corresponding closed-loop control value function value can be calculated for the above-mentioned closed-loop control value function. Among the multiple calculated closed-loop control value function values, the smallest closed-loop control value function value is taken as the optimal state of the closed-loop control value function value. Then, the rectifier switch state information corresponding to the optimal state of the closed-loop control value function value at time k+1 is configured as the target rectifier switch control information, which can be obtained. Therefore, within k+1 sampling control cycles, the rectifier is configured to be in the corresponding target rectification state using the target rectifier switch control information. At this time, the bus terminal voltage can be matched with the bus reference voltage within k+1 sampling control cycles.

[0106] In specific implementation, during closed-loop control, the above control process should be repeated. For example, after entering the k+1 sampling control cycle, the target rectifier switch control information for the k+2 sampling control cycles can be determined, and the rectifier can be configured to be in the target rectification state according to the determined target rectifier switch control information. The control method and process of the target permanent magnet synchronous generator in subsequent sampling control cycles can be referred to here. Examples will not be given here.

[0107] As explained above, the closed-loop controller should include a PI controller, a generator predictive control model, and a closed-loop control value function unit. In each sampling control cycle, the generator predictive control model can be used to perform model predictive control, the PI controller can be used to perform PI calculation, and the closed-loop control value function unit can be used to perform the calculation and judgment of the closed-loop control value function value. The corresponding operation of the PI controller, generator predictive control model, and closed-loop control value function unit can be referred to the above explanation, and will not be repeated here.

[0108] In one embodiment of the present invention, closed-loop control further includes parameter mismatch adjustment processing, wherein...

[0109] When performing parameter mismatch control, the following are included:

[0110] The target for parameter mismatch monitoring should be determined, and the target for parameter mismatch monitoring should include at least the q-axis current.

[0111] The parameter mismatch state is determined based on the parameter mismatch monitoring target. When the parameter mismatch state is that the parameters are mismatched, the model prediction state parameters are updated online, and model prediction state parameters adapted to the target permanent magnet synchronous generator are generated online. Then, the model prediction state parameters are configured in the generator predictive control model so that model predictive control can be performed in the closed-loop control.

[0112] When determining the parameter mismatch state based on the parameter mismatch monitoring target, if multiple consecutive q-axis current residuals match the parameter mismatch monitoring threshold, the parameter mismatch state is configured as parameter mismatched. For any q-axis current residual, the q-axis current residual is generated based on the difference between the corresponding q-axis currents at two adjacent times.

[0113] As explained above, when controlling a target permanent magnet synchronous generator (PMSG) with any unknown initial parameters, a system identification operation should be performed first. Afterward, closed-loop control can be implemented on the bus terminal voltage generated by the PMSG. It is understandable that during the closed-loop control operation of the PMSG, parameter mismatch may occur. That is, the state matrix A and control matrix B determined by the system identification may not be suitable for the current control of the output bus terminal voltage of the PMSG. In this case, parameter mismatch adjustment should be performed to improve the control accuracy of the bus terminal voltage. Specifically, when parameter mismatch occurs, the bus terminal voltage cannot be matched with the bus reference voltage.

[0114] It should be understood that when performing parameter mismatch control, it is necessary to first determine whether a parameter mismatch exists. If a parameter mismatch exists, the model's predicted state parameters can be updated online, and model predicted state parameters adapted to the target permanent magnet synchronous generator can be generated online. Specifically, when determining whether a parameter mismatch exists, the parameter mismatch monitoring target should be identified. In other words, the status of the parameter mismatch monitoring target is used to determine whether a parameter mismatch exists. In practice, the parameter mismatch monitoring target should at least include the q-axis current, that is, the status of the q-axis current in different sampling control cycles is used to determine whether a parameter mismatch exists.

[0115] In specific implementation, when multiple consecutive q-axis current residuals match the parameter mismatch monitoring threshold, the parameter mismatch state is configured as parameter mismatched. Specifically, parameter mismatch means that the parameter is already in a mismatched state. For any q-axis current residual, the q-axis current residual is generated based on the difference between the corresponding q-axis currents at two adjacent times. For example, for the q-axis current at time k, the difference between the q-axis current at time k and the q-axis current at time k-1 forms the q-axis current residual at time k. Multiple consecutive q-axis current residuals can specifically be the q-axis current residual at time k-2, the q-axis current residual at time k-1, the q-axis current residual at time k, and the q-axis current residual at time k+1. Of course, there are other cases for multiple consecutive q-axis current residuals, which will not be illustrated here.

[0116] Figure 2 The image shows an embodiment where the parameter mismatch monitoring target uses the q-axis current and is updated online. Figure 2In this model, the event-triggered mechanism is triggered when multiple consecutive q-axis current residuals match the parameter mismatch monitoring threshold. Generally, when three consecutive q-axis current residuals match the parameter mismatch monitoring threshold, the online update mechanism is triggered to update the model's predicted state parameters, i.e., to update the state matrix A and control matrix B. After updating state matrix A and control matrix B, they should be reconfigured within the generator predictive control model to perform model predictive control in closed-loop control. The method and process of performing model predictive control using the generator predictive control model in closed-loop control can be found in the corresponding descriptions above.

[0117] Understandably, the method of updating state matrix A and control matrix B can be consistent with the state matrix A and control matrix B identified by the system. The following explanation will focus on the process of identifying state matrix A and control matrix B by the system. The updated state matrix A and control matrix B are better adapted to the current operating conditions of the target permanent magnet synchronous generator, thereby improving the closed-loop control accuracy of the bus terminal voltage of the target permanent magnet synchronous generator, and thus ensuring that the bus terminal voltage matches the bus reference voltage.

[0118] In one embodiment of the present invention, for the q-axis current at time m, the q-axis current residual is:

[0119]

[0120] Among them, i q(m) Let i be the q-axis current at time m. q(m-1) Let be the q-axis current at time m-1. Let be the q-axis current residual corresponding to the q-axis current at time m;

[0121] With q-axis current residual The corresponding parameter mismatch monitoring threshold is:

[0122]

[0123] in, This is the parameter mismatch monitoring threshold corresponding to the q-axis current at time m. This is the average value calculated based on the q-axis current corresponding to the n sampling control cycles prior to time m. Let λ be the standard deviation calculated based on the q-axis current corresponding to the n sampling control cycles prior to time m. m This is the adjustment coefficient corresponding to the q-axis current at time m;

[0124] when When, then the q-axis current residual Monitoring threshold for parameter mismatch match.

[0125] As explained above, the q-axis current at time m specifically refers to the q-axis current sampled within the m-th sampling control cycle. The q-axis current in other cases can be found in the explanation provided here. To improve the stability and reliability of the event triggering mechanism and avoid false triggering, in one embodiment of this invention, the q-axis current corresponding to the first n sampling control cycles can be used. Therefore, when executing the event triggering mechanism, the target permanent magnet synchronous generator should be run under closed-loop control for at least n sampling control cycles. Specifically, n can be 50. In this case, the event triggering mechanism can be executed in the 51st (m=51st) sampling control cycle, and the average q-axis current can be calculated based on the 50 sampling control cycles. and standard deviation mean and standard deviation It can be calculated using existing and commonly used methods, such as by statistically calculating the average value from the corresponding q-axis currents over 50 control cycles. and standard deviation

[0126] It is understandable that when m is 52, the corresponding mean value can be calculated using the q-axis current of 2 to 51 sampling control cycles. and standard deviation When m takes other values, please refer to the explanation here. Additionally, the adjustment coefficient λ... m The settings can be referenced in the following explanations. As can be seen from the above explanations, for each q-axis current residual, the corresponding parameter mismatch monitoring threshold can be calculated. In this case, the q-axis current residual should be determined. Monitoring threshold for parameter mismatch match.

[0127] In one embodiment of the present invention, the adjustment coefficient λ m The adjustment coefficient λ is determined using a fuzzy control method. m Then:

[0128] Determine the residual of the q-axis current Corresponding residual rate of change Among them, the rate of change of residuals Then we have: Let T be the q-axis current residual corresponding to the q-axis current at time m-1. s The sampling control period;

[0129] q-axis current residual and residual change rate The variable is loaded into the constructed fuzzy controller so that the corresponding adjustment coefficient increment Δλ can be calculated by the fuzzy controller. m ,

[0130] Based on the adjustment coefficient increment Δλ m Generate adjustment coefficient λ m Then we have: λ m =λ m-1 +Δλ m , where λ m-1 This is the adjustment coefficient corresponding to the q-axis current at time m-1.

[0131] To further improve the accuracy of parameter mismatch monitoring thresholds, the adjustment coefficient λ of this invention... m The adjustment coefficient λ is determined using fuzzy control. m At that time, the residual current along the q-axis should be determined. Corresponding residual rate of change Subsequently, the q-axis current residual and residual change rate Loaded into the constructed fuzzy controller, it can be understood that the fuzzy controller should be built within an event-triggered mechanism, and the fuzzy controller can be based on the q-axis current residual. and residual change rate The adjustment coefficient increment Δλ is obtained by calculation. m Subsequently, the adjustment coefficient increment Δλ can be used. m The adjustment coefficient λ is obtained. m The following section describes the calculated adjustment coefficient increment Δλ. m Examples will be given to illustrate this situation.

[0132] In practical implementation, the q-axis current residual and residual change rate Fuzzy sets are created and divided into five fuzzy sets (NB / NS / ZO / PS / PB). For fuzzy control, triangular membership functions are used, and the q-axis current residual can be calculated using these functions. residual change rate The corresponding membership degrees are defined. A 5x5 rule matrix is ​​constructed, where the row index corresponds to the residual fuzzy set, and the column index corresponds to the parametric residual change rate fuzzy set, through the q-axis current residual. residual change rate The activation intensity is calculated based on the corresponding membership degrees, and then the calculated activation intensity is defuzzified using the centroid method. The centroid method formula is as follows:

[0133]

[0134] Where N is the number of activated rules, Δλ iThis represents the output increment corresponding to the i-th fuzzy rule. Let be the activation strength of the i-th rule.

[0135] In practice, for a given q-axis current residual and residual change rate The number of activated rules N and the output increment Δλ corresponding to the i-th fuzzy rule can be determined. i and the activation strength of the i-th rule The specific method for determination can be consistent with existing technologies, and will not be elaborated here.

[0136] In one embodiment of the present invention, system identification of the target permanent magnet synchronous generator includes:

[0137] Construct the DMDc state-space equation corresponding to the target permanent magnet synchronous generator and configure the identification reference voltage information, wherein the identification reference voltage information includes the d-axis reference voltage and q-axis reference voltage of multiple reference segments;

[0138] After configuring the target permanent magnet synchronous generator to generate an observable dynamic response, the basic voltage vector of the target permanent magnet synchronous generator under the dynamic response is determined, wherein the basic voltage vector includes the d-axis response voltage and the q-axis response voltage.

[0139] The open-loop identification value function values ​​representing the d-axis response voltage and q-axis response voltage, respectively, are calculated and compared with the d-axis reference voltage and q-axis reference voltage under each reference segment. The optimal value of the open-loop identification value function value under each reference segment is determined, so as to determine the target identification rectification state of the rectifier based on the optimal value of the open-loop identification value function value under each reference segment. Subsequently, data sampling processing is performed under each reference segment to collect and generate corresponding identification sampling data. For each identification sampling data, the identification sampling data includes identification state quantity and identification control quantity.

[0140] The identified state quantities include the identified d-axis current and the identified q-axis current;

[0141] The identification control quantities include the identification of d-axis voltage, the identification of q-axis voltage, and the angular velocity of the target permanent magnet synchronous generator;

[0142] The state-space equations of the DMDc constructed above are solved using all the identified sampling data to obtain the model's predicted state parameters.

[0143] Specifically, when constructing the state-space equations of the DMDc, one can start with the mathematical model of the target permanent magnet synchronous generator (PMSG) to obtain its state equations. Based on these state equations, the control matrix of the target PMSG can be obtained, and the state-space equations of the DMD can then be constructed. According to the operating characteristics of the target PMSG, the mathematical model of the PMSG in the rotating coordinate system can be obtained, resulting in:

[0144]

[0145] Among them, R s L is the stator resistance. q For q-axis inductance, L d For the d-axis inductance, ω e It is the electrical angular velocity of the target permanent magnet synchronous generator, ψ f u is the flux linkage amplitude of the permanent magnet in the target permanent magnet synchronous generator. d and u q It is the stator voltage component along the dq axis, i d and i q It is the stator current component along the dq axis, i.e., u d For the d-axis voltage, u q Let i be the q-axis voltage. d Let i be the d-axis current. q Let ψ be the q-axis current. q Let ψ be the q-axis flux linkage. d denoted as d-axis flux linkage.

[0146] Combining equations (1) and (2) above, we obtain the differential equation for the current, which is:

[0147]

[0148] In equation (3), the system parameters of the target permanent magnet synchronous generator are time-varying. Since the rotational speed is usually kept constant under the operating conditions of the target permanent magnet synchronous generator, the angular velocity ω is assumed to be constant. e If it remains unchanged, then four time variables u are extracted from the coefficient matrix of the target permanent magnet synchronous generator. d u q i d i q This allows us to obtain a fixed coefficient matrix, and then incorporate the error terms into the input to form the standard DMD state-space equations:

[0149]

[0150] Discretizing equation (4) yields a mathematical model that can predict the current value at the next moment. The discretized model is written as:

[0151]

[0152] in:

[0153]

[0154] In equation (5) above, the expression for model predictive control is executed. In equation (6), A is the state matrix, B is the control matrix, and T... s This is the sampling control period.

[0155] Based on equations (5) and (6) above, the state-space equation of DMDc can be obtained, and then:

[0156] X2 = AX1 + BU (7)

[0157] in,

[0158] The above describes the construction of the DMDc state-space equations corresponding to the target permanent magnet synchronous generator. After obtaining the DMDc state-space equations, identification reference voltage information should be configured before system identification. This identification reference voltage information includes d-axis and q-axis reference voltages for multiple reference segments. Specifically, the d-axis and q-axis reference voltages for multiple reference segments enhance the information content of the identification reference voltages, thereby enabling a more comprehensive and accurate extraction of the dynamic modes and state-space model of the target permanent magnet synchronous generator.

[0159] In one embodiment of the present invention, for the d-axis reference voltage of multiple reference segments, the following applies:

[0160]

[0161] For a q-axis reference voltage with multiple reference segments, then:

[0162]

[0163] in, This is the bus reference voltage. The d-axis reference voltage. f is the q-axis reference voltage. N The fundamental frequency of the target permanent magnet synchronous generator.

[0164] Specifically, in order to effectively extract the dynamic modes of the target permanent magnet synchronous generator, this invention superimposes a high-frequency AC signal onto the bus reference voltage. The frequency of the high-frequency AC signal is 80 times the fundamental frequency. The AC component amplitude of the d-axis reference voltage is 2% of the bus reference voltage, and the AC component amplitude of the q-axis reference voltage is 1% of the bus reference voltage. The fundamental frequency f of the target permanent magnet synchronous generator... N The fundamental frequency f of the target permanent magnet synchronous generator can be calculated from its rotational speed characteristics.N The situation is consistent with existing technology, and will not be elaborated here.

[0165] In the above d-axis and q-axis reference voltages, t represents time, and the unit of time is seconds (s). As can be seen from the corresponding explanations of the d-axis and q-axis reference voltages, each time interval forms a reference segment. For example, for the d-axis reference voltage, 0 ≤ t < 0.5 is the first reference segment, 0.5 ≤ t < 1 is the second reference segment, and the situations for other reference segments can be found in the explanation here.

[0166] As explained above, a low-speed engine can be applied to the target permanent magnet synchronous generator (PMSG) to induce an observable dynamic response. This observable dynamic response includes a fundamental voltage vector and a fundamental stator current. The fundamental voltage vector comprises the d-axis response voltage and the q-axis response voltage. The determination of the d-axis and q-axis response voltages can be referenced from the determination methods for the d-axis and q-axis voltages within the aforementioned generator control information. The fundamental stator current typically consists of the q-axis current and the d-axis current.

[0167] Since model predictive control is used in closed-loop control, similar to closed-loop control, an open-loop identification value function should also be set in system identification. To enable the calculation and judgment of the open-loop identification value function value, an open-loop identification value function unit should be set up. Figure 2 In the middle, the value function J e The location is the open-loop identification value function unit. As shown in the figure, after obtaining the basic voltage vector, the basic voltage vector should be loaded into the open-loop identification value function unit so that the open-loop identification value function value can be calculated by the open-loop identification value function unit. It can be understood that the above-mentioned multi-reference segment d-axis reference voltage and multi-reference segment q-axis reference voltage should also be set in the open-loop identification value function unit.

[0168] In one embodiment of the present invention, for the open-loop identification value function, the following is true:

[0169]

[0170] Among them, J e To identify the value function in an open-loop manner, The d-axis response voltage. This is the q-axis response voltage;

[0171] When determining the optimal value of the open-loop identification value function for each reference segment, the d-axis response voltage q-axis response voltage The same reference segment corresponding to the d-axis reference voltage and the q-axis reference voltage.

[0172] It should be noted that the method for calculating the open-loop identification value function can refer to the above explanation of the calculation process for the closed-loop control value function. The difference is that when calculating the open-loop identification value function, the open-loop identification value function corresponding to each reference segment should be calculated separately. For example, the d-axis reference voltage and q-axis reference voltage of the multi-reference segment shown above should have their open-loop identification value function calculated separately for the corresponding reference segments such as 0≤t<0.5 and 0.5≤t<1.

[0173] As explained above, when calculating the open-loop identification value function, the d-axis response voltage and q-axis response voltage should be obtained. Specifically, to obtain the d-axis and q-axis response voltages, the rectification states of the rectifier should be traversed until the optimal value of the open-loop identification value function for each reference segment is achieved. The optimal value of the open-loop identification value function for each reference segment is the minimum value of the open-loop identification value function for that reference segment. At this point, the rectification state of the rectifier is the target identification rectification state. After determining the target identification rectification state, the open-loop identification value function unit should load the reference segment identification rectification switch control information into the rectifier. Figure 2 Q in abc This is the reference segment for identifying rectifier switch control information to control the rectifier's rectification state.

[0174] Once the target identification rectification state corresponding to each reference segment is determined, data sampling processing is performed at least under each reference segment to collect and generate corresponding identification sampling data. Each identification sampling data includes identification state quantities and identification control quantities. In one embodiment of the present invention, the identification state quantities include identification d-axis current and q-axis current; the identification control quantities include identification d-axis voltage, q-axis voltage, and the angular velocity of the target permanent magnet synchronous generator. Therefore, the identification state quantities correspond to the aforementioned power generation state information, and the identification control quantities correspond to the aforementioned power generation control information. It is understood that the identification state quantities should correspond directly to the identification control quantities, that is, the identification state quantities and control quantities correspond to the power generation state of the target permanent magnet synchronous generator under the same reference segment.

[0175] To effectively extract the dynamic modulus of the target permanent magnet synchronous generator, data sampling processing should be performed on the transition process between two adjacent reference segments. That is, during system identification, data sampling processing should be performed on the transition phase of each reference segment and adjacent reference segments. To meet the requirements of system identification while reducing the amount of data computation, in one embodiment of the present invention, periodic downsampling is performed within each reference segment of the d-axis reference voltage during data sampling processing, and full transient sampling is performed when the d-axis reference voltage jumps from one reference segment to another adjacent reference segment.

[0176] As explained above, during the system identification process, the data sampling process results in a mixed sampling state. Periodic downsampling within each reference segment reduces the amount of identification sampling data collected. During periodic downsampling, data sampling can be performed at intervals of multiple sampling control cycles, such as 100 sampling control cycles, to ensure that corresponding identification sampling data can be collected in each reference segment.

[0177] When moving from one reference segment to another between two adjacent reference segments, the DC amplitude of the d-axis reference voltage will jump. For example, when moving from the first reference segment (0≤t<0.5) to the second reference segment (0.5≤t<1), the DC amplitude of the d-axis reference voltage will jump. To effectively extract the dynamic modulus under this jump, full sampling of the transient process should be performed. For other situations where the DC amplitude of the d-axis reference voltage jumps, please refer to the explanation here.

[0178] Specifically, when performing full sampling of a transient process, data sampling is performed once within each sampling control cycle, such as performing full sampling of the transient process corresponding to the time window (0.05s) after the jump.

[0179] It should be understood that after performing the above mixed sampling state, the corresponding identification sampling data can be obtained. Based on all the identification sampling data, the following data matrix X1, data matrix X2, and data matrix U can be constructed, where X2 is X1 delayed by one time step, specifically:

[0180]

[0181] Specifically, after the data sampling and processing described above using mixed sampling, κ can be obtained. max +1 identification sampling data, This indicates the first data sampling process up to the κth. max The q-axis current is identified by the data sampling and processing. For other cases, please refer to the explanation here.

[0182] After constructing the data matrices X1, X2, and U, the constructed DMDc state space equations can be solved to obtain the model's predicted state parameters. The specific method and process for obtaining the model's predicted state parameters can be consistent with existing technologies and will not be elaborated here.

[0183] Figure 2In order to facilitate the execution of the above system identification process, a DMDc model unit and a matrix identification unit should be constructed. The DMDc model unit can complete the construction of the above DMDc state space equations, as well as the above data sampling processing and the construction of data matrices X1, X2, and U. The constructed data matrices X1, X2, and U are then loaded into the matrix identification unit, and the matrix identification unit is used to solve the above DMDc state space equations. The solved state matrix A and control matrix B are then loaded into the generator predictive control model.

[0184] Figure 3 The figure illustrates an embodiment of the bus terminal voltage generated by the target permanent magnet synchronous generator during the system identification process. As shown in the figure, the duration of the system identification process is 0 to 2.5 seconds, and the bus terminal voltage fluctuates within the range of 0 to 100V.

[0185] Figure 4 (a) illustrates an embodiment of the target permanent magnet synchronous generator's bus terminal voltage during the system identification phase and the closed-loop control phase. Figure 4 Open-loop excitation specifically refers to the system identification stage mentioned above, while closed-loop control is the stage of performing closed-loop control on the bus terminal voltage. In the following diagram, open-loop excitation and closed-loop control convey the same meaning. Figure 3 and Figure 4 (a) It can be seen that during the system identification phase, the bus terminal voltage is low and basically stable. During the closed-loop control phase, the bus terminal voltage first surges and eventually stabilizes at the bus reference voltage. Figure 4 One embodiment is shown when the bus reference voltage is 430V. Figure 4 (b) is an embodiment of the electromagnetic torque of the target permanent magnet synchronous generator in system identification and closed-loop control.

[0186] Figure 5 The figure shows an embodiment of the bus terminal voltage and electromagnetic torque of the target permanent magnet synchronous generator under varying operating conditions. As can be seen from the figure, the present invention can effectively achieve stable control of the bus terminal voltage under different operating conditions.

[0187] Depend on Figure 5As explained above, the target permanent magnet synchronous generator (PMSG) requires online updates to its model-predicted state parameters during operation. Specifically, the method for online updating of these parameters is largely consistent with the method used to obtain them during the system identification phase. The main difference lies in the data source. Specifically, during the system identification phase, the identification sampling data originates from the PMSG's system response under open-loop bus conditions; details regarding the identification sampling data can be found in the relevant explanations above. During the online update process, the data matrix is ​​taken from the PMSG's response data under specific actual operating conditions during the q-axis current residual pulse caused by the target PMSG's speed change. Therefore, collecting the relevant data and using the aforementioned method can update the model-predicted state parameters. It should be understood that the collected data should be sufficient to meet the requirements of the matrix identification unit in calculating the state matrix A and control matrix B.

[0188] Furthermore, as shown in the above drawings, during the system identification phase, the target permanent magnet synchronous generator has a low rotational speed, resulting in the bus terminal voltage being lower than the bus reference voltage. Therefore, it is necessary to increase the rotational speed of the target permanent magnet synchronous generator to make the bus terminal voltage match the bus reference voltage. In one embodiment of the present invention, a phased dynamic adjustment control method is adopted, that is, the bus reference voltage rises along with the engine speed until it reaches the target bus reference voltage, that is, the bus reference voltage is gradually adjusted until the bus reference voltage reaches the target bus reference voltage. Therefore, the bus reference voltage mentioned in the above closed-loop control is the final bus reference voltage that needs to be achieved.

[0189] It should be noted that during the process of gradually adjusting the engine speed until the target bus reference voltage is reached, since no change in operating conditions is involved, the aforementioned online update process will not be performed, meaning the model prediction state parameters obtained during the system identification phase will be maintained. Therefore, online updates of the model prediction state parameters mainly occur when parameter mismatches are caused by changes in operating conditions.

[0190] As explained above, this invention combines DMDc and MPC to identify the model-predicted state parameters of the target permanent magnet synchronous generator (PMSG), while reducing the MPC's dependence on model parameters. It enables real-time prediction of the system's future dynamic behavior and online optimization of control inputs, thereby achieving precise regulation of the bus terminal voltage. Therefore, the parameter-free model predictive control method for PMSG based on DMDc provides a new approach for the efficient and stable operation of PMSG. Under conditions of unknown initial parameters or PMSG parameter mismatch, the closed-loop control method for bus terminal voltage has unique advantages, exhibiting strong robustness, data-driven adaptability, and high dynamic response.

Claims

1. A parameterless model predictive control method for permanent magnet synchronous generators based on dynamic mode decomposition, characterized in that, The parameterless model predictive control method for permanent magnet synchronous generators includes: A target permanent magnet synchronous generator is provided, and system identification is performed on the target permanent magnet synchronous generator to determine the model predictive state parameters for model predictive control of the target permanent magnet synchronous generator after system identification. The model predicts state parameters that include at least a state matrix A and a control matrix B. During system identification, the target permanent magnet synchronous generator is configured to generate an observable dynamic response. Subsequently, the model prediction state parameters corresponding to the target permanent magnet synchronous generator are identified and determined based on the DMDc method. During the identification process, the target permanent magnet synchronous generator is configured to be in the open-loop state of the bus voltage. The identified model-predicted state parameters are configured within the generator predictive control model of the closed-loop controller. This configures the closed-loop controller to perform closed-loop control on the bus terminal voltage output by the target permanent magnet synchronous generator. During the closed-loop control process, model predictive control is performed using the generator predictive control model. In the kth sampling control cycle of the closed-loop control, the power generation state information of the target permanent magnet synchronous generator at time k and the power generation control information at time k are loaded into the generator predictive control model so as to perform model predictive control using the generator predictive control model and generate the power generation state information at time k+1. The power generation state information includes the q-axis current and the d-axis current. Based on the power generation state information at time k+1 and the power generation state reference information at time k, the closed-loop control value function representing the corresponding error states of the q-axis current and d-axis current at time k+1 is calculated. Based on the optimal state of the closed-loop control value function at time k+1, the target rectifier switch control information for controlling the rectifier's rectification state is determined. Then, based on the target rectifier switch control information, the bus terminal voltage, adapted to the bus reference voltage via the rectifier output, is configured. At least based on the bus reference voltage and the bus sample voltage at time k, the power generation state reference information at time k is generated, wherein the bus sample voltage is generated by sampling the bus terminal voltage of the rectifier output.

2. The method according to claim 1, characterized in that, Closed-loop control also includes parameter mismatch adjustment processing, among which, When performing parameter mismatch control, the following are included: The target for parameter mismatch monitoring should be determined, and the target for parameter mismatch monitoring should include at least the q-axis current. The parameter mismatch state is determined based on the parameter mismatch monitoring target. When the parameter mismatch state is that the parameters are mismatched, the model prediction state parameters are updated online, and model prediction state parameters adapted to the target permanent magnet synchronous generator are generated online. Then, the model prediction state parameters are configured in the generator predictive control model so that model predictive control can be performed in the closed-loop control. When determining the parameter mismatch state based on the parameter mismatch monitoring target, if multiple consecutive q-axis current residuals match the parameter mismatch monitoring threshold, the parameter mismatch state is configured as parameter mismatched. For any q-axis current residual, the q-axis current residual is generated based on the difference between the corresponding q-axis currents at two adjacent times.

3. The method according to claim 2, characterized in that, For the q-axis current at time m, the q-axis current residual is: Among them, i q(m) Let i be the q-axis current at time m. q(m-1) Let be the q-axis current at time m-1. Let be the q-axis current residual corresponding to the q-axis current at time m; With q-axis current residual The corresponding parameter mismatch monitoring threshold is: in, This is the parameter mismatch monitoring threshold corresponding to the q-axis current at time m. This is the average value calculated based on the q-axis current corresponding to the n sampling control cycles prior to time m. Let λ be the standard deviation calculated based on the q-axis current corresponding to the n sampling control cycles prior to time m. m This is the adjustment coefficient corresponding to the q-axis current at time m; when When, then the q-axis current residual Monitoring threshold for parameter mismatch match.

4. The method according to claim 3, characterized in that, Adjustment coefficient λ m The adjustment coefficient λ is determined using a fuzzy control method. m Then: Determine the residual of the q-axis current Corresponding residual rate of change Among them, the rate of change of residuals Then we have: Let T be the q-axis current residual corresponding to the q-axis current at time m-1. s This is the sampling control period; q-axis current residual and residual change rate The variable is loaded into the constructed fuzzy controller so that the corresponding adjustment coefficient increment Δλ can be calculated by the fuzzy controller. m , Based on the adjustment coefficient increment Δλ m Generate adjustment coefficient λ m Then we have: λ m =λ m-1 +Δλ m , where λ m-1 This is the adjustment coefficient corresponding to the q-axis current at time m-1.

5. The method according to claim 1, characterized in that, For the closed-loop control value function, we have: in, Let be the closed-loop control value function at time k+1. Let q be the current at time k+1. The given value for the q-axis current at time k is... Let d be the current at time k+1. The given value for the d-axis current at time k; When determining the optimal state of the closed-loop control value function at time k+1, we have: Iterate through the rectification states of the rectifier and the rectifier switch state information for each rectification state, and obtain the bus sampling voltage for each rectification state. Then, based on the bus sampling voltage, generate at least k q-axis current setpoints. Based on the generated q-axis current setpoint at time k The closed-loop control value function value at time k+1 is calculated, and the smallest closed-loop control value function value at time k+1 is configured as the optimal state of the closed-loop control value function value at time k+1. The rectifier switch state information corresponding to the optimal state of the closed-loop control value function value at time k+1 is configured as the target rectifier switch control information.

6. The method according to claim 5, characterized in that, The q-axis current given at time k At least the bus reference voltage and the bus sampled voltage at time k are generated by PI calculation; The d-axis current at time k is given by... Set to 0.

7. The method according to any one of claims 1 to 6, characterized in that, When performing system identification on the target permanent magnet synchronous generator, the following steps are included: Construct the DMDc state-space equation corresponding to the target permanent magnet synchronous generator and configure the identification reference voltage information, wherein the identification reference voltage information includes the d-axis reference voltage and q-axis reference voltage of multiple reference segments; After configuring the target permanent magnet synchronous generator to generate an observable dynamic response, the basic voltage vector of the target permanent magnet synchronous generator under the dynamic response is determined, wherein the basic voltage vector includes the d-axis response voltage and the q-axis response voltage. The open-loop identification value function values ​​representing the d-axis response voltage and q-axis response voltage, respectively, are calculated and compared with the d-axis reference voltage and q-axis reference voltage under each reference segment. The optimal value of the open-loop identification value function value under each reference segment is determined, so as to determine the target identification rectification state of the rectifier based on the optimal value of the open-loop identification value function value under each reference segment. Subsequently, data sampling processing is performed under each reference segment to collect and generate corresponding identification sampling data. For each identification sampling data, the identification sampling data includes identification state quantity and identification control quantity. The identified state quantities include the identified d-axis current and the identified q-axis current; The identification control quantities include the identification of d-axis voltage, the identification of q-axis voltage, and the angular velocity of the target permanent magnet synchronous generator; The state-space equations of the DMDc constructed above are solved using all the identified sampling data to obtain the model's predicted state parameters.

8. The method according to claim 7, characterized in that, When configuring the target permanent magnet synchronous generator to produce an observable dynamic response, the target permanent magnet synchronous generator should be driven to rotate at a low speed by the engine at least. For the d-axis reference voltage with multiple reference segments, we have: For a q-axis reference voltage with multiple reference segments, then: in, This is the bus reference voltage. The d-axis reference voltage. f is the q-axis reference voltage. N The fundamental frequency of the target permanent magnet synchronous generator.

9. The method according to claim 8, characterized in that, in During data sampling, periodic downsampling is performed within each reference segment of the d-axis reference voltage, and transient full sampling is performed when the d-axis reference voltage jumps from one reference segment to another adjacent reference segment.

10. The method according to claim 8, characterized in that, For the open-loop identification value function, we have: Among them, J e To identify the value function in an open-loop manner, The d-axis response voltage. This is the q-axis response voltage; When determining the optimal value of the open-loop identification value function for each reference segment, the d-axis response voltage q-axis response voltage The same reference segment corresponding to the d-axis reference voltage and the q-axis reference voltage.

Images