Data-driven multi-target control method, device and system based on model prediction

Abstract

The invention provides a data-driven multi-target control method, device and system based on model prediction, is applied to a three-phase three-level neutral point clamped converter, and aims to realize dual accurate control of switching frequency and neutral point voltage. According to the method, a switch weight factor and a neutral point voltage weight factor are used as inputs, and an average switch frequency in a fixed sliding window time and a neutral point voltage maximum value in another fixed sliding window time are used as outputs; and constructing an equivalent dynamic linearization data model under each dynamic working point of the closed-loop system by adopting a tight-format dynamic linearization method. Based on the data model, a data driving controller is elaborately designed to realize accurate regulation and control of the switching frequency and the maximum value of the neutral point voltage. The method has a high practical value and a wide application prospect in typical application scenes of power electronics and power transmission, such as grid connection and motor driving.

Description

TECHNICAL FIELD

[0001] The application belongs to the field of limited control set model prediction, and particularly relates to a data-driven multi-target quantitative control method and device based on model prediction control. BACKGROUND

[0002] Limited control set model prediction is widely studied due to its advantages in multi-target optimization and nonlinear control. Its discrete period being shorter than the switching period improves low-frequency accuracy and avoids narrow pulses, making it very suitable for the control of low-switching-frequency and high-power converters.

[0003] Although limited control set model prediction can achieve multi-target optimization, it is difficult to establish a clear relationship between the weight factor and the optimization target. The mutual coupling of multiple optimization targets often requires a large amount of offline adjustment of the weight factor. Fixed weight factors will cause performance deviation when the operating point changes, and offline parameter setting methods cannot achieve ideal multi-target optimization at different operating points. To solve the control problem of unknown model nonlinear systems, an adaptive data-driven control method using online data is proposed in the prior art, which performs well in the average switching frequency regulation of single-phase three-level neutral-point-clamped converters. However, this method is limited to online adjustment of a single weight factor, and can only achieve quantitative control of a single target.

[0004] Therefore, it is particularly important to achieve precise quantitative control of multiple targets for multi-level converter control systems with multiple control requirements. SUMMARY

[0005] The application aims to provide a data-driven multi-target control method and device based on model prediction, which can dynamically adjust the real-time optimal performance of the control algorithm with multiple target and multiple weight coefficients under all operating conditions. The application is applied to three-phase three-level neutral-point-clamped converters, aiming to achieve precise control of switching frequency and neutral-point voltage.

[0006] In a first aspect, the application provides a data-driven multi-target control method based on model prediction, which comprises:

[0007] By mathematically modeling a three-phase three-level neutral-point-clamped converter, a limited control set model prediction cost function is obtained. The constraint term of the cost function includes active power, reactive power, switching frequency and midpoint voltage.

[0008] During the operation of the three-phase three-level neutral-point-clamped converter, the average switching frequency and the absolute value of the maximum midpoint voltage deviation are calculated quantitatively in real time.

[0009] Based on real-time quantitative calculation of average switching frequency and maximum midpoint voltage deviation absolute value, a multi-input multi-output nonlinear system is established; the input of the multi-input multi-output nonlinear system is the weight coefficient of the average switching frequency and the maximum midpoint voltage deviation absolute value in the limited control set model prediction cost function, and the output is the average switching frequency and the maximum midpoint voltage deviation absolute value;

[0010] The dynamic time-varying relationship between the input increment and the output increment of the multi-input multi-output nonlinear system is described by using the pseudo-Jacobian matrix, and a dynamic linearization model of the multi-input multi-output nonlinear system is constructed.

[0011] A control input criterion function is established to update the pseudo-Jacobian matrix in real time, and a functional relationship between the input and the output of the multi-input multi-output nonlinear system is obtained.

[0012] Based on the dynamic linearization model, a pseudo-Jacobian matrix estimation criterion function is constructed, and a functional relationship between the pseudo-Jacobian matrix and the input and output of the multi-input multi-output nonlinear system is obtained.

[0013] Based on the real-time storage and update of the input and output data set of the three-phase three-level neutral point clamped converter control system, the functional relationship between the input and the output of the multi-input multi-output nonlinear system, the functional relationship between the pseudo-Jacobian matrix and the input and output of the multi-input multi-output nonlinear system, and the reference value of the average switching frequency and the maximum midpoint voltage deviation absolute value are used to calculate the weight coefficient of the average switching frequency and the maximum midpoint voltage deviation absolute value, so as to realize data-driven multi-objective quantitative control based on limited control set model prediction.

[0014] In one embodiment, the real-time quantitative calculation of the average switching frequency and the maximum midpoint voltage deviation absolute value during the operation of the three-phase three-level neutral point clamped converter includes:

[0015] During the operation of the three-phase three-level neutral point clamped converter, the average switching frequency is quantitatively calculated within a predefined fixed sliding time window.

[0016] The formula for quantitatively calculating the average switching frequency within the fixed sliding time window is:

[0017] ,

[0018] ,

[0019] In the formula, is the average switching frequency; is the duration of the fixed sliding time window; is the total number of switches of the three-phase three-level neutral point clamped converter within the updated sliding time window at the kth moment. total number of switches of the three-phase three-level neutral point clamped converter in the sliding time window updated at the k-1th moment; total number of switches of the three-phase three-level neutral point clamped converter at the kth moment; total number of switches of the three-phase three-level neutral point clamped converter at the k-nth moment, 12 represents 12 insulated gate bipolar transistors of the three-phase three-level neutral point clamped converter;

[0020] In another independent fixed sliding time window, the maximum midpoint voltage deviation absolute value in the sliding time window is realized in real time based on a step-by-step comparison method for the absolute value of the input signal.

[0021] In an embodiment, the step-by-step comparison method for the absolute value of the input signal is used to realize real-time extraction of the maximum midpoint voltage deviation absolute value in the sliding time window, which includes:

[0022] The absolute value of the input signal uo is stored in a persistent variable max0 with a dimension of [2x1], and the absolute values of the last two input signals are stored in the persistent variable max0;

[0023] A loop counter i0 is used to realize the alternative update of the two storage positions of max0, thereby completing the cyclic storage of the input value;

[0024] A persistent variable and a loop counter are used to realize the hierarchical calculation of the maximum value;

[0025] In each level, by comparing the two values of the previous level, the larger value is stored in the variable of the current level, and then the index counter is incremented and reset to 1 when it exceeds 2, thereby realizing the alternative update of the storage unit;

[0026] An output variable uomax is generated, which corresponds to the maximum value of the level.

[0027] In an embodiment, the multi-input multi-output nonlinear system is represented as:

[0028] ,

[0029] is the input of the system at the kth moment, which is composed of the weight coefficient of the average switching frequency and the weight coefficient of the maximum midpoint voltage deviation absolute value, wherein is the weight coefficient of the average switching frequency at the kth moment, is the weight coefficient of the maximum midpoint voltage deviation absolute value at the kth moment; is the input of the system at the kth moment, which is composed of the average switching frequency and the maximum midpoint voltage deviation absolute value, wherein is the average switching frequency at the kth moment; is the maximum midpoint voltage deviation absolute value at time k; and represents two positive integers, represents a nonlinear function, where m is a positive integer, represents a real vector with dimension 2.

[0030] In an embodiment, the dynamic linearization model of the multi-input multi-output nonlinear system is:

[0031] ,

[0032] where, is the output increment between the kth time and the (k+1)th time, is the input increment between the kth time and the (k-1)th time, is the pseudo-Jacobian matrix of the nonlinear system at the kth time, which contains time-varying parameters.

[0033] In an embodiment, the control input criterion function is:

[0034] ,

[0035] where, λ>0 is a weight factor for suppressing excessive changes in the control input, is the reference output at the (k+1)th time, is the input of the system at the (k+1)th time; is the input of the system at the (k-1)th time;

[0036] Minimizing the control input criterion function gives the function relationship between the input and output of the multi-input multi-output nonlinear system:

[0037] ,

[0038] where, the parameter λ is used to suppress the change of the control input u, and this constraint aims to ensure the necessary smoothness of the control signal in the control system design, ∈(0,1] is a step factor introduced; is an identity matrix with the same dimension as the pseudo-Jacobian matrix, is the transpose matrix of the pseudo-Jacobian matrix at the kth time, is the actual output at the kth time, and is the actual input at the kth time and the (k-1)th time, is the reference output at the (k+1)th time.

[0039] In an embodiment, the pseudo-Jacobian matrix estimation criterion function is:

[0040] ,

[0041] wherein μ>0 is a weight coefficient for restraining excessive change of the estimated pseudo-Jacobian matrix, is an output increment between the kth time instant and the (k-1)th time instant, is an input increment between the (k-1)th time instant and the (k-2)th time instant, is a pseudo-Jacobian matrix of a nonlinear system containing time-varying parameters at the kth time instant, is an estimated value of the pseudo-Jacobian matrix of the nonlinear system containing time-varying parameters at the (k-1)th time instant;

[0042] The function relationship between the pseudo-Jacobian matrix and the input and output of the multi-input and multi-output nonlinear system is obtained by minimizing the pseudo-Jacobian matrix estimation criterion function as follows:

[0043] ,

[0044] wherein is an estimated value of the pseudo-Jacobian matrix; parameter η∈(0,2] represents a search step of the pseudo-Jacobian matrix, and reflects a search speed of the pseudo-Jacobian matrix; μ>0 is a weight coefficient for restraining excessive change of the estimated pseudo-Jacobian matrix, is a transpose matrix of the input increment between the (k-1)th time instant and the (k-2)th time instant.

[0045] In an embodiment, the method further comprises introducing an algorithm resetting mechanism to make the pseudo-Jacobian matrix estimation algorithm track the time-varying parameters faster, and the algorithm resetting mechanism is as follows:

[0046] ,

[0047] ,

[0048] wherein is an initial value of a non-diagonal element of the pseudo-Jacobian matrix . is an initial value of a diagonal element of the pseudo-Jacobian matrix ; λ>0, μ>0, η∈(0,2], ∈(0,1]; is a sufficiently small positive number, is a coefficient related to system autocorrelation, is a coefficient related to system coupling, is a sign function.

[0049] In a second aspect, the present application provides a model prediction-based data-driven multi-objective control device, which comprises:

[0050] The limited control set model prediction cost function construction module is configured to derive a limited control set model prediction cost function by mathematically modeling the three-phase three-level neutral point clamped converter; the constraint term of the cost function includes active power, reactive power, switching frequency and neutral point voltage;

[0051] The quantitative calculation module is configured to quantitatively calculate the average switching frequency and the maximum neutral point voltage deviation absolute value in real time during operation of the three-phase three-level neutral point clamped converter.

[0052] The multi-input multi-output nonlinear system construction module is configured to establish a multi-input multi-output nonlinear system based on the average switching frequency and the maximum neutral point voltage deviation absolute value quantitatively calculated in real time; the input of the multi-input multi-output nonlinear system is the weight coefficient of the average switching frequency and the maximum neutral point voltage deviation absolute value in the limited control set model prediction cost function, and the output is the average switching frequency and the maximum neutral point voltage deviation absolute value.

[0053] The dynamic linearization model module is configured to describe the dynamic time-varying relationship between the input increment and the output increment of the multi-input multi-output nonlinear system by using a pseudo-Jacobian matrix, and to construct a dynamic linearization model of the multi-input multi-output nonlinear system.

[0054] The function relationship one module is configured to establish a control input criterion function for updating the pseudo-Jacobian matrix in real time, and to derive the function relationship between the input and the output of the multi-input multi-output nonlinear system.

[0055] The function relationship two module is configured to construct a pseudo-Jacobian matrix estimation criterion function based on the dynamic linearization model, and to derive the function relationship between the pseudo-Jacobian matrix and the input and output of the multi-input multi-output nonlinear system.

[0056] The weight coefficient calculation module is configured to update the input and output data set in real time based on the three-phase three-level neutral point clamped converter control system, and to calculate the weight coefficient of the average switching frequency and the maximum neutral point voltage deviation absolute value by using the function relationship between the input and the output of the multi-input multi-output nonlinear system, the function relationship between the pseudo-Jacobian matrix and the input and output of the multi-input multi-output nonlinear system, and the reference value of the average switching frequency and the maximum neutral point voltage deviation absolute value, so as to realize data-driven multi-objective quantitative control based on limited control set model prediction.

[0057] In a third aspect, the application provides a multi-level converter control system, which comprises the above-mentioned model prediction-based data-driven multi-objective control device.

[0058] Compared with the prior art, the application has the following advantages:

[0059] (1) The method realizes quantitative control of multiple optimization objectives through online self-tuning of the weight factor.

[0060] (2) The method can realize online self-tuning of the weight factor under all working conditions, so that the control algorithm can operate with optimal performance under all working conditions.

[0061] (3) The method does not depend on system models and parameter information.

[0062] (4) The method can be applied to different power electronic topologies or control of different optimization objectives under a limited control set model prediction framework. BRIEF DESCRIPTION OF DRAWINGS

[0063] The accompanying drawings, which are part of the present application, serve to provide a further understanding of the present application, and the schematic embodiments of the present application and their descriptions serve to explain the present application, but do not constitute an improper limitation on the present application.

[0064] Figure 1 is a three-phase three-level neutral point clamped converter topology diagram;

[0065] Figure 2 is a switch state vector diagram of the three-phase three-level neutral point clamped converter;

[0066] Figure 3 is a flowchart of a data-driven multi-objective control method based on model prediction provided by an embodiment of the present application;

[0067] Figure 4 is a working principle diagram for calculating an average switching frequency based on a fixed sliding time window provided in an embodiment of the present application;

[0068] Figure 5 is a schematic diagram for solving the maximum midpoint voltage deviation absolute value based on a fixed sliding time window provided in an embodiment of the present application;

[0069] Figure 6 is an experimental result of the midpoint voltage, average switching frequency and grid-side current of the three-phase three-level neutral point clamped converter provided by an embodiment of the present application.

[0070] It should be noted that these drawings and written descriptions are not intended to limit the scope of the concept of the present application in any way, but to illustrate the concept of the present application to those skilled in the art by referring to specific embodiments. DETAILED DESCRIPTION

[0071] The present invention will now be described in detail with reference to the accompanying drawings and specific embodiments. These embodiments are based on the technical solution of the present invention, providing detailed implementation methods and specific operating procedures, but are not intended to limit its application, and the present invention is not limited to the following embodiments.

[0072] To enhance understanding of the present invention, the technical solution of the present invention will be further described below in conjunction with the accompanying drawings and specific embodiments.

[0073] The implementation focuses on a three-phase three-level neutral point clamping converter. The circuit structure of the three-phase three-level neutral point clamping converter is as follows: Figure 1 As shown. The three-phase three-level neutral point clamping converter generates three effective switching states per phase arm, where S... x ∈{-1, 0, 1}(x=a,b,c). Assuming that the voltage of each capacitor is equal to half of the DC bus voltage, the output voltage generated by these three switching states (relative to) Figure 1 Midpoint o) is u xo ∈{Udc / 2, 0, -Udc / 2}(x=a,b,c). The table below summarizes the switching states generated by this inverter. This power inverter generates a total of 3^3=27 switching states, such as... Figure 2 As shown.

[0074]

[0075] Reference Figure 3 As shown in the figure, the data-driven multi-objective control method based on model prediction provided in this disclosure specifically includes the following steps:

[0076] Step S100: By mathematically modeling the three-phase three-level neutral point clamp converter, the finite control set model prediction cost function is obtained.

[0077] Specifically, the voltage equation based on the three-phase three-level neutral point clamping converter in the αβ stationary coordinate system is as follows:

[0078]

[0079] In the formula, For the grid-side current in the αβ coordinate system, For the grid voltage in the αβ coordinate system, Let L be the αβ-axis voltage output from the AC side of the inverter, L be the filter inductance, and R be its equivalent resistance. The variables in the αβ stationary coordinate system can be obtained using the Clarke transformation matrix. , , , , , , The Clarke transformation matrix is ​​as follows:

[0080]

[0081] The discrete form of the above voltage equation is as follows:

[0082]

[0083] In the formula, The sampling period is Let α be the β-axis network current at time k. Let α be the grid voltage along the β axis at time k. Let be the AC output voltage of the αβ-axis rectifier at time k. It can be determined from the current DC bus voltage. and the switching state at time k , , The conclusion is as follows.

[0084]

[0085] From the voltage equation above, it can be seen that the rate of change of current is affected by system parameters, grid voltage, AC output voltage, and instantaneous current value. The control system regulates the current by adjusting the on / off state and duration of the inverter switching devices to change the AC output voltage. The grid-side voltage of the αβ axis at time k+1 can be expressed as:

[0086]

[0087] Active power of grid-side voltage and reactive power The predicted value at time k+1 is:

[0088]

[0089] The midpoint voltage is defined as:

[0090]

[0091] The predicted value of the midpoint voltage at time k+1 is:

[0092]

[0093] The switching rules for a three-phase three-level neutral point clamping converter are as follows: Figure 2 As shown. For the candidate switching states generated by the three-phase three-level neutral point clamping converter, the grid-side current is predicted using equations (3), (5), (6), and (8). Grid-side voltage Active power reactive power and midpoint voltage of future values. The switching state that minimizes the cost function is selected and applied to the next control period. The finite control set model predictive cost function is:

[0094]

[0095] wherein, is the reference active power, is the reference reactive power, is the midpoint voltage weight coefficient, is the average switching frequency weight coefficient, is the midpoint voltage prediction value at the k+1 moment, is the switching times.

[0096] Step S200: In the operation process of the three-phase three-level neutral point clamped converter, the average switching frequency and the maximum midpoint voltage deviation absolute value are calculated in real time.

[0097] In the embodiments of the present application, in the operation process of the three-phase three-level neutral point clamped converter, the average switching frequency and the maximum midpoint voltage deviation absolute value are calculated in real time, including:

[0098] Step S210: In the operation process of the three-phase three-level neutral point clamped converter, the average switching frequency is quantitatively calculated in a pre-defined fixed sliding time window.

[0099] Figure 4 is the working principle diagram for calculating the average switching frequency based on the fixed sliding time window. The formula for quantitatively calculating the average switching frequency in the fixed sliding time window is:

[0100]

[0101] wherein, is the average switching frequency; is the duration of the fixed sliding time window; is the total switching times of the three-phase three-level neutral point clamped converter in the sliding time window updated at the k moment; is the total switching times of the three-phase three-level neutral point clamped converter in the sliding time window updated at the k-1 moment; is the switching times of the three-phase three-level neutral point clamped converter at the k moment; is the switching times of the three-phase three-level neutral point clamped converter at the k-n moment, 12 indicates 12 insulated gate bipolar transistors of the three-phase three-level neutral point clamped converter.

[0102] Step S220: Real-time extraction of the maximum midpoint voltage deviation absolute value in the fixed sliding time window is achieved based on the method of step-by-step comparison of the absolute values of the input signal.

[0103] It should be noted that the above two sliding time window settings do not repeat.

[0104] Figure 5 The flowchart of the quantitative calculation method for the maximum midpoint voltage deviation absolute value in the fixed sliding time window.

[0105] In a preferred embodiment, real-time extraction of the maximum midpoint voltage deviation absolute value in the fixed sliding time window is achieved based on the method of step-by-step comparison of the absolute values of the input signal, including:

[0106] Step S221: The absolute values of the input signal uo are stored in the persistent variable max0 with a dimension of [2x1], and the absolute values of the last two input signals are stored in the persistent variable max0.

[0107] Step S222: A loop counter i0 is used to realize the alternative update of the two storage locations of max0, thereby completing the cyclic storage of the input values.

[0108] Step S223: The persistent variables (such as max1, max2, max3) and the loop counters (such as i1, i2, i3) are used to realize the hierarchical calculation of the maximum values.

[0109] Step S224: In each level, the larger value is stored in the current level variable (such as max2[i2]) by comparing the two values of the previous level (such as max1[1] and max1[2]), and then the index counter is incremented and reset to 1 when it exceeds 2, thereby realizing the alternative update of the storage units.

[0110] Step S225: The output variable uomax is generated, which corresponds to the maximum value of the current level. That is, the value of the output variable uomax is determined by selecting the larger value after comparing the two values of the current level.

[0111] Step S300: Based on the real-time quantitative calculation of the average switching frequency and the maximum midpoint voltage deviation absolute value, a multi-input multi-output nonlinear system is established; the inputs of the multi-input multi-output nonlinear system are the weight coefficients of the average switching frequency and the maximum midpoint voltage deviation absolute value in the finite control set model predictive cost function, and the outputs are the average switching frequency and the maximum midpoint voltage deviation absolute value.

[0112] Further, the multi-input multi-output nonlinear system is represented as:

[0113]

[0114] is the input of the system at time k, consisting of the weight coefficient of the average switching frequency and the weight coefficient of the maximum absolute value of the midpoint voltage deviation, where is the weight coefficient of the average switching frequency at time k, is the weight coefficient of the maximum absolute value of the midpoint voltage deviation at time k; is the input of the system at time k, consisting of the average switching frequency and the maximum absolute value of the midpoint voltage deviation, where is the average switching frequency at time k; is the maximum absolute value of the midpoint voltage deviation at time k; and represent two positive integers, represents a nonlinear function, where m is a positive integer, represents a real number vector with a dimension of 2.

[0115] Step S400: The dynamic time-varying relationship between the input increment and the output increment of the multi-input multi-output nonlinear system is described using the pseudo-Jacobian matrix, and a dynamic linearization model of the multi-input multi-output nonlinear system is constructed.

[0116] Further, the dynamic linearization model of the multi-input multi-output nonlinear system is constructed as:

[0117]

[0118] In the formula, is the output increment of the two adjacent time points of the kth time point and the k+1th time point, is the input increment of the two adjacent time points of the kth time point and the k-1th time point, is the pseudo-Jacobian matrix of the nonlinear system at the kth time point containing time-varying parameters.

[0119] Step S500: A control input criterion function is established to update the pseudo-Jacobian matrix in real time, and a functional relationship between the input and the output of the multi-input multi-output nonlinear system is obtained.

[0120] Further, the control input criterion function is:

[0121]

[0122] In the formula, λ>0 is a weight factor for inhibiting excessive changes in the control input, is the reference output at the k+1th time point, is the input of the system at the k+1th time point; is the input of the system at the k-1th time point.

[0123] To improve the algorithm's versatility, a step size factor is introduced. ∈(0,1], minimizing the control input criterion function yields the functional relationship between the input and output of a multi-input multi-output nonlinear system:

[0124]

[0125] In the formula, the parameter λ is used to suppress changes in the control input u. This constraint aims to ensure that the control signal has the necessary smoothness in the control system design. ∈(0,1] introduces a step size factor; It is an identity matrix with the same dimension as the pseudo-Jacobi matrix. Let be the transpose of the pseudo-Jacobi matrix at time k. This represents the actual output at time k. and The actual input values ​​at time k and time k-1 are... This is the reference output at time k+1.

[0126] Step S600: Construct a pseudo-Jacobi matrix estimation criterion function based on the dynamic linearization model, and derive the functional relationship between the pseudo-Jacobi matrix and the input and output of the multi-input multi-output nonlinear system.

[0127] Furthermore, the criterion function for estimating the pseudo-Jacobi matrix is:

[0128]

[0129] In the formula, μ>0 represents the weighting coefficient used to suppress excessive changes in the estimated pseudo-Jacobi matrix. The output increments at time k and time (k-1) are the two nearest times. The input increments at two adjacent time points, k-1 and k-2, are given. Let be the pseudo-Jacobi matrix containing time-varying parameters for the nonlinear system at time k. This is an estimate of the pseudo-Jacobi matrix containing time-varying parameters for the nonlinear system at time k-1.

[0130] Furthermore, to improve the algorithm's flexibility and versatility, a step size factor η∈(0,2) is introduced. By minimizing the pseudo-Jacobi matrix estimation criterion function, the functional relationship between the pseudo-Jacobi matrix and the input-output of a multi-input multi-output nonlinear system is derived as follows:

[0131]

[0132] In the formula, η is a parameter representing a search step of the pseudo-Jacobian matrix, reflecting a search speed of the pseudo-Jacobian matrix, μ>0 is a weight coefficient for suppressing excessive change of the estimated pseudo-Jacobian matrix, is an input increment of the two adjacent time instants of the k-1 time instant and the k-2 time instant, is a transpose matrix of the input increment of the two adjacent time instants of the k-1 time instant and the k-2 time instant, is an output increment of the two adjacent time instants of the k time instant and the k-1 time instant.

[0133] Step S700: Based on the three-phase three-level neutral point clamped converter control system, the updated input and output data set is stored in real time, the function relationship between the multi-input multi-output nonlinear system input and output, the function relationship between the pseudo-Jacobian matrix and the multi-input multi-output nonlinear system input and output, and the reference values of the average switching frequency and the maximum midpoint voltage deviation absolute value are used to calculate the weight coefficients of the average switching frequency and the maximum midpoint voltage deviation absolute value, so as to realize the data-driven multi-objective quantitative control based on the limited control set model prediction.

[0134] Based on the pseudo-Jacobian matrix estimation criterion function and the function relationship among the pseudo-Jacobian matrix, the input and the output, a data-driven multi-objective quantitative control strategy can be obtained.

[0135] In a preferred embodiment, the data-driven multi-objective control method based on model prediction further comprises: introducing an algorithm reset mechanism to make the pseudo-Jacobian matrix estimation algorithm track time-varying parameters faster, and the algorithm reset mechanism is:

[0136]

[0137]

[0138] In the formula, is a non-diagonal element of the pseudo-Jacobian matrix is an initial value of the non-diagonal element of the pseudo-Jacobian matrix is a diagonal element of the pseudo-Jacobian matrix is an initial value of the diagonal element of the pseudo-Jacobian matrix; λ>0, μ>0, η∈(0,2], ∈(0,1]; is a sufficiently small positive number, is a coefficient related to system autocorrelation, is a coefficient related to system coupling, is a sign function. In the data-driven multi-objective quantitative control strategy, the algorithm reset mechanism in the above formula is introduced to make the pseudo-Jacobian matrix estimation algorithm track time-varying parameters faster.

[0139] In an embodiment, a model prediction based data-driven multi-objective control device is shown, the device comprising:

[0140] A limited control set model prediction cost function construction module is configured to derive a limited control set model prediction cost function by mathematically modeling a three-phase three-level neutral point clamped converter; the constraint terms of the cost function include: active power, reactive power, switching frequency and neutral point voltage;

[0141] A quantitative calculation module is configured to quantitatively calculate the average switching frequency and the maximum neutral point voltage deviation absolute value in real time during the operation of the three-phase three-level neutral point clamped converter;

[0142] A multi-input multi-output nonlinear system construction module is configured to establish a multi-input multi-output nonlinear system based on the real-time quantitative calculation of the average switching frequency and the maximum neutral point voltage deviation absolute value; the input of the multi-input multi-output nonlinear system is the weight coefficient of the average switching frequency and the maximum neutral point voltage deviation absolute value in the limited control set model prediction cost function, and the output is the average switching frequency and the maximum neutral point voltage deviation absolute value;

[0143] A dynamic linearization model module is configured to describe the dynamic time-varying relationship between the input increment and the output increment of the multi-input multi-output nonlinear system by using a pseudo-Jacobian matrix, and to construct a dynamic linearization model of the multi-input multi-output nonlinear system;

[0144] A function relationship one module is configured to establish a control input criterion function for real-time updating of the pseudo-Jacobian matrix, and to derive the function relationship between the input and the output of the multi-input multi-output nonlinear system;

[0145] A function relationship two module is configured to construct a pseudo-Jacobian matrix estimation criterion function based on the dynamic linearization model, and to derive the function relationship between the pseudo-Jacobian matrix and the input and output of the multi-input multi-output nonlinear system;

[0146] A weight coefficient calculation module is configured to update the input and output data set in real time based on the three-phase three-level neutral point clamped converter control system, and to calculate the weight coefficient of the average switching frequency and the maximum neutral point voltage deviation absolute value by using the function relationship between the input and the output of the multi-input multi-output nonlinear system, the function relationship between the pseudo-Jacobian matrix and the input and output of the multi-input multi-output nonlinear system, and the reference value of the average switching frequency and the maximum neutral point voltage deviation absolute value, thereby realizing the data-driven multi-objective quantitative control based on the limited control set model prediction.

[0147] It should be noted that the model prediction based data driven multi-objective control provided in the above embodiment is only exemplified by the above division of functional modules when performing the model prediction based data driven multi-objective control method. In actual application, the above functions can be completed by different functional modules according to needs, that is, the internal structure of the device is divided into different functional modules to complete all or part of the functions described above. In addition, the model prediction based data driven multi-objective control device provided in the above embodiment and the model prediction based data driven multi-objective control method embodiment belong to the same concept, and the implementation process is described in detail in the model prediction based data driven multi-objective control method embodiment. Here, it is not repeated.

[0148] In one embodiment, a multi-level converter control system is proposed, which includes the above model prediction based data driven multi-objective control device.

[0149] The description of the model prediction based data driven multi-objective control device is described in the same or similar parts described above, and is not repeated here.

[0150] In a specific embodiment, the grid voltage is 90 V, the DC bus voltage is 160 V, the grid frequency is 50 Hz, the sampling period =100 μs、 = =2200 μF. The reactive power reference value is zero, and the grid-side filter inductance parameter is L=3.3 mH, R=0.1 Ω. The method is realized based on a 300-MIPS 32-bit TMS320C28346 type DSP control board, and the power device uses FGY75N60SMD type IGBT.

[0151] Experimental setting: the absolute value of the midpoint voltage deviation is not more than 10V, at 10s, the load is increased from 0.8pu to 1.2pu. In =15s, the switching frequency is reduced from 800Hz to 600Hz.

[0152] (1) Midpoint voltage accurately controlled

[0153] It can be seen from Figure 6 that the method proposed by the invention can control the absolute value of the midpoint voltage deviation to be within 10V in steady state and dynamic conditions.

[0154] (2) Switching frequency accurately controlled

[0155] from Figure 6It can be known that the method can accurately control the switching frequency at 800Hz in the steady state, and can accurately and quickly control the switching frequency from 800Hz to 600Hz in the dynamic state.

[0156] (3) reference current accurate tracking

[0157] by Figure 6 It can be known that the method can accurately control the grid-side current at 10A in the steady state, and can accurately and quickly adjust the grid-side current to meet the load jump in the dynamic state. When the switching frequency changes, the grid-side current is almost not affected, which proves that the control method has strong robustness.

[0158] The above-mentioned embodiments only express several embodiments of the present application, and the description is more specific and detailed, but it cannot be understood as the limitation of the patent scope of the present application. It should be pointed out that for ordinary skilled in the art, without departing from the concept of the present application, a number of modifications and improvements can be made, which belong to the protection scope of the present application. Therefore, the protection scope of the patent of the present application should be subject to the appended claims.

Claims

1. A data-driven multi-objective control method based on model prediction, characterized in that, The method includes: By mathematically modeling a three-phase three-level neutral point clamping converter, a finite control set model prediction cost function is obtained; the constraints of the cost function include: active power, reactive power, switching frequency, and neutral point voltage. During the operation of the three-phase three-level neutral point clamping converter, the average switching frequency and the absolute value of the maximum neutral point voltage deviation are calculated in real time. A multi-input multi-output nonlinear system is established based on the average switching frequency and the absolute value of the maximum midpoint voltage deviation calculated in real time. The input of the multi-input multi-output nonlinear system is the weighting coefficient of the average switching frequency and the absolute value of the maximum midpoint voltage deviation in the prediction cost function of the finite control set model, and the output is the average switching frequency and the absolute value of the maximum midpoint voltage deviation. A pseudo-Jacobi matrix is ​​used to describe the dynamic time-varying relationship between the input increment and the output increment of a multi-input multi-output nonlinear system, and a dynamic linearization model of the multi-input multi-output nonlinear system is constructed. A control input criterion function is established to update the pseudo-Jacobi matrix in real time, and the functional relationship between the input and output of a multi-input multi-output nonlinear system is derived. Based on the dynamic linearization model, a pseudo-Jacobi matrix estimation criterion function is constructed, and the functional relationship between the pseudo-Jacobi matrix and the input and output of a multi-input multi-output nonlinear system is obtained. Based on the real-time storage and updating of the input-output dataset of the three-phase three-level neutral point clamping converter control system, the weighting coefficients of the average switching frequency and the absolute value of the maximum neutral point voltage deviation are calculated using the functional relationship between the input and output of the multi-input multi-output nonlinear system, the pseudo-Jacobi matrix and the functional relationship between the input and output of the multi-input multi-output nonlinear system, as well as the reference values ​​of the average switching frequency and the absolute value of the maximum neutral point voltage deviation. This enables data-driven multi-objective quantitative control based on finite control set model prediction.

2. The data-driven multi-objective control method based on model prediction according to claim 1, characterized in that, The real-time quantitative calculation of the average switching frequency and the absolute value of the maximum neutral point voltage deviation during the operation of the three-phase three-level neutral point clamping converter includes: The average switching frequency is quantitatively calculated within a predefined fixed sliding time window during the operation of a three-phase three-level neutral point clamping converter. The formula for quantitatively calculating the average switching frequency within a fixed sliding time window is: ， ， In the formula, This represents the average switching frequency. The duration of the fixed sliding time window; The total number of switching operations of the three-phase three-level neutral point clamp converter within the sliding time window updated at time k; The total number of switching operations of the three-phase three-level neutral point clamp converter within the sliding time window updated at time k-1; The switching count of the three-phase three-level neutral point clamping converter at time k; The number of switching operations of the three-phase three-level neutral point clamping converter at time kn is given, and 12 represents the 12 insulated-gate bipolar transistors of the three-phase three-level neutral point clamping converter. Within another independent fixed sliding time window, the absolute value of the maximum midpoint voltage deviation within the sliding time window is extracted in real time based on a method of stepwise comparison of the absolute value of the input signal.

3. The data-driven multi-objective control method based on model prediction according to claim 2, characterized in that, The method based on stepwise comparison of the absolute value of the input signal, which enables real-time extraction of the absolute value of the maximum midpoint voltage deviation within a sliding time window, includes: The absolute value of the input signal uo is stored in a persistent variable max0 with dimension [2×1]. The persistent variable max0 stores the absolute values ​​of the two most recent input signals. A loop counter i0 is used to alternately update the two storage locations of max0, thereby completing the cyclic storage of the input value. Hierarchical calculation of maximum value is implemented using persistent variables and loop counters; In each level, by comparing the two values ​​of the previous level, the larger value is stored in the current level variable, and then the index counter is incremented and reset to 1 when it exceeds 2, thereby realizing the alternating update of storage units; Generate an output variable uomax, which is the maximum value of the corresponding level.

4. The data-driven multi-objective control method based on model prediction according to claim 3, characterized in that, The construction of a multi-input multi-output nonlinear system is represented as follows: ， In the formula, The input to the system at time k consists of a weighting factor for the average switching frequency and a weighting factor for the absolute value of the maximum midpoint voltage deviation, where... The weighting coefficients for the average switching frequency at time k are: The weighting coefficient for the absolute value of the maximum midpoint voltage deviation at time k; The input to the system at time k consists of the average switching frequency and the absolute value of the maximum midpoint voltage deviation, where Let k be the average switching frequency at time k; The absolute value of the maximum midpoint voltage deviation at time k; and Represents two positive integers. Represents a nonlinear function, where m is a positive integer. This represents a real vector with dimension 2.

5. The data-driven multi-objective control method based on model prediction according to claim 4, characterized in that, The dynamic linearization model of the multi-input multi-output nonlinear system is as follows: ， In the formula, This represents the output increment between two adjacent time points, namely time k and time k+1. The input increments at time k and time (k-1) are the two nearest times. Let be the pseudo-Jacobi matrix containing time-varying parameters for the nonlinear system at time k.

6. The data-driven multi-objective control method based on model prediction according to claim 5, characterized in that, The control input criterion function is: ， In the formula, λ>0 is a weighting factor used to suppress excessive changes in the control input. Let be the reference output at time k+1. This is the system input at time k+1; This is the system input at time k-1; Minimizing the control input criterion function yields the functional relationship between the input and output of a multi-input multi-output nonlinear system: , In the formula, the parameter λ is used to suppress changes in the control input u. This constraint aims to ensure that the control signal has the necessary smoothness in the control system design. ∈(0,1] introduces a step size factor; It is an identity matrix with the same dimension as the pseudo-Jacobi matrix. Let be the transpose of the pseudo-Jacobi matrix at time k. This represents the actual output at time k. and The actual input values ​​at time k and time k-1 are... This is the reference output at time k+1.

7. The data-driven multi-objective control method based on model prediction according to claim 6, characterized in that, The pseudo-Jacobi matrix estimation criterion function is: ， In the formula, μ>0 represents the weighting coefficient used to suppress excessive changes in the estimated pseudo-Jacobi matrix. The output increments at time k and time (k-1) are the two nearest times. The input increments at two adjacent time points, k-1 and k-2, are given. Let be the pseudo-Jacobi matrix containing time-varying parameters for the nonlinear system at time k. This is an estimate of the pseudo-Jacobi matrix containing time-varying parameters for the nonlinear system at time k-1. By minimizing the pseudo-Jacobi matrix estimation criterion function, the functional relationship between the pseudo-Jacobi matrix and the input and output of a multi-input multi-output nonlinear system is obtained as follows: ， In the formula, The pseudo-Jacobi matrix is ​​an estimate; the parameter η∈(0,2] represents the search step size of the pseudo-Jacobi matrix, reflecting the search speed; μ>0 is a weighting coefficient used to suppress excessive changes in the estimated pseudo-Jacobi matrix. It is the transpose of the input increments of two adjacent time points, k-1 and k-2.

8. The data-driven multi-objective control method based on model prediction according to claim 7, characterized in that: The method further includes: introducing an algorithm reset mechanism to enable the pseudo-Jacobi matrix estimation algorithm to track time-varying parameters more quickly, wherein the algorithm reset mechanism is as follows: ， ， In the formula, Off-diagonal elements of a pseudo-Jacobi matrix The initial value; The diagonal elements of the pseudo-Jacobi matrix Initial values; λ>0, μ>0, η∈(0,2], ∈(0,1]; For sufficiently small positive numbers, The coefficient is the autocorrelation coefficient of the system. For coefficients related to system coupling, It is a symbolic function.

9. A data-driven multi-objective control device based on model prediction, characterized in that, The device includes: The finite control set model prediction cost function construction module is used to derive the finite control set model prediction cost function by mathematically modeling a three-phase three-level neutral point clamping converter; the constraints of the cost function include: active power, reactive power, switching frequency and neutral point voltage. The quantitative calculation module is used to perform real-time quantitative calculations of the average switching frequency and the absolute value of the maximum neutral point voltage deviation during the operation of a three-phase three-level neutral point clamping converter. A multi-input multi-output (MIMO) nonlinear system construction module is used to establish a multi-input multi-output (MIMO) nonlinear system based on the average switching frequency and the absolute value of the maximum midpoint voltage deviation calculated in real time. The input of the MIMO nonlinear system is the weighting coefficient of the average switching frequency and the absolute value of the maximum midpoint voltage deviation in the prediction cost function of the finite control set model, and the output is the average switching frequency and the absolute value of the maximum midpoint voltage deviation. The Dynamic Linearization Model module is used to describe the dynamic time-varying relationship between the input increment and the output increment of a multi-input multi-output nonlinear system using a pseudo-Jacobi matrix, and to construct a dynamic linearization model of the multi-input multi-output nonlinear system. The Function Relationship module is used to establish the control input criterion function for real-time updating of the pseudo-Jacobi matrix and to derive the functional relationship between the input and output of a multi-input multi-output nonlinear system. The second module of functional relationships is used to construct the pseudo-Jacobi matrix estimation criterion function based on the dynamic linearization model, and to derive the functional relationship between the pseudo-Jacobi matrix and the input and output of a multi-input multi-output nonlinear system. The weighting coefficient calculation module is used to calculate the weighting coefficients of the input-output dataset stored and updated in real time based on the three-phase three-level neutral point clamping converter control system. It utilizes the functional relationship between the input and output of the multi-input multi-output nonlinear system, the pseudo-Jacobi matrix and the functional relationship between the input and output of the multi-input multi-output nonlinear system, as well as the reference values ​​of the average switching frequency and the absolute value of the maximum neutral point voltage deviation. This enables data-driven multi-objective quantitative control based on finite control set model prediction.

10. A multilevel converter control system, characterized in that, The system includes the data-driven multi-objective control device based on model prediction as described in claim 9.

Images