Power decoupling method and system for three-port dc converter

Abstract

This invention discloses a power decoupling method and system for a three-port DC-DC converter. The method includes transforming the TAB converter topology into a type equivalent circuit, and training a neural network with the operating currents of the two ports to be decoupled in the type equivalent circuit as inputs and the corresponding two phase shift angles as outputs. The method calculates the error between the phase shift angles obtained from the trained neural network and the actual phase shift angles, and determines the boundary and intermediate regions of the power transmission range of each port based on the error. When controlling the operating currents of the two ports, a local lookup table method is used for power decoupling in the boundary regions of the power transmission range of each port, and the trained neural network is used for power decoupling in the intermediate regions of the power transmission range of each port. This invention can effectively reduce the required storage costs while ensuring decoupling accuracy.

Description

Technical Field

[0001] This invention relates to the field of power electronics technology, specifically to a power decoupling method and system for a three-port DC-DC converter. Background Technology

[0002] Since its inception, the three-port isolated bidirectional DC-DC converter (Triple Active Bridge, TAB) has been applied in various fields, such as photovoltaic-energy storage power generation systems, DC microgrids, and fuel cell-energy storage electric vehicle systems. Compared with the traditional method of using multiple two-port bidirectional converters, the use of a three-port DC-DC converter improves the overall system operating efficiency and converter power density, reduces system design costs and size, and enables bidirectional power flow between any two ports according to actual needs.

[0003] The fully isolated three-port DC-DC converter connects all ports through a three-winding high-frequency transformer, achieving electrical isolation for all ports. However, the power transmission between ports relies on an inductance network composed of transformer leakage inductance or external inductance. This causes the power of each port to be coupled after the magnetic flux of the windings is superimposed, which in turn creates a coupling constraint relationship between control variables. A change in the power of a single port will cause fluctuations in the power of other ports. This is a key problem that must be solved in its practical application.

[0004] Among related technologies, the paper "Research on Frequency Selective Decoupling Control of Dual Series Resonant Bidirectional Three-Port DC / DC Converter, Sun Xiaofeng, Xue Li, Meng Yufei, et al. Acta Energiae Solaris Sinica" proposes introducing two resonant capacitors based on the TAB converter and achieving frequency selective decoupling by selecting the ratio of the switching frequency to the resonant frequency. However, this strategy does not achieve complete decoupling, but only weakens the effect of coupling. The paper "Decoupling Control Strategy of Three-Port DC-DC Converter Based on Neural Network, Song Jiakang, Peng Yonggang, Wang Xiaoming, et al. New Technologies in Electrical Engineering and Power" proposes obtaining the decoupling matrix online using an offline-trained neural network and achieving decoupling using the diagonal matrix decoupling method. This method suffers from poor accuracy at the boundaries of the operating region. The patent application CN110912167A proposes an improved decoupling control method for hybrid energy storage systems. It solves the improved decoupling model using the quantum particle swarm optimization algorithm, obtaining a table showing the correspondence between the current value and phase angle at the stable operating point. By inputting a reference current value, the steady-state operating point is obtained from the table and then calculated using bilinear interpolation, improving control accuracy in low-resolution scenarios. However, this scheme requires embedding a large number of tables to achieve power decoupling over a wide operating current range, consuming significant storage space and making it unusable in existing control units. Summary of the Invention

[0005] The technical problem to be solved by this invention is how to effectively reduce the required storage cost while ensuring decoupling accuracy.

[0006] The present invention solves the above-mentioned technical problems through the following technical means:

[0007] On one hand, this invention proposes a power decoupling method for a three-port DC-DC converter, the method comprising:

[0008] The TAB converter topology is transformed into a Δ-type equivalent circuit, and the working currents of the two ports that need to be decoupled in the Δ-type equivalent circuit are used as inputs, and the corresponding two phase shift angles are used as outputs to train the neural network.

[0009] The error between the phase shift angle obtained from the trained neural network and the actual phase shift angle is calculated, and based on the error and error limit, the boundary region and intermediate region of the power transmission range of each port are determined.

[0010] When controlling the operating current of the two ports, a local lookup table method is used for power decoupling in the boundary region of the power transmission range of each port, and a trained neural network is used for power decoupling in the middle region of the power transmission range of each port.

[0011] Furthermore, the step of transforming the TAB converter topology into a delta-type equivalent circuit, and using the operating currents of the two ports that need to be decoupled in the delta-type equivalent circuit as inputs and the corresponding two phase shift angles as outputs to train the neural network includes:

[0012] Ignoring the influence of the transformer magnetizing inductance and applying the power transfer inductance and port electrical parameters to a specific port of the TAB converter topology, the Y-type equivalent circuit of the converter is obtained.

[0013] Based on the Y-Δ transformation principle, the Y-type equivalent circuit is transformed into a Δ-type equivalent circuit;

[0014] When the two phase angles associated with a certain port change, calculate the operating current of the two ports that need to be decoupled in the Δ-type equivalent circuit.

[0015] The neural network is trained using the operating current of the two ports as input and the corresponding two phase shift angles as output.

[0016] Furthermore, the error between the phase shift angle obtained from the trained neural network and the actual phase shift angle is calculated, and based on the error and error limit, the boundary region and intermediate region of the power transmission range of each port are determined, including:

[0017] The predicted phase shift angle is calculated using a pre-trained neural network, and the error between the predicted phase shift angle and the actual phase shift angle is calculated.

[0018] The boundary region is determined by the operating current points of the two ports corresponding to when the error exceeds the error limit.

[0019] The intermediate region is determined by the operating current points of the two ports corresponding to the error not exceeding the error limit.

[0020] Furthermore, the calculation process for the error limit includes:

[0021] The objective function is solved, and the minimum value of the objective function is found under the constraints. The ε that satisfies the minimum value is taken as the error limit.

[0022] The objective function and constraints are as follows:

[0023] (x1, x2 ∈ R)

[0024] stg(x1) = x1 - θ th1 <0

[0025] h(x2) = x2 - θ th2 <0

[0026] In the formula: constraint g(x1) indicates that the current fluctuation does not exceed the threshold θ th1 The constraint h(x2) indicates that the ratio of the amount of data exceeding the error limit to the total amount of data does not exceed the threshold θ. th2 ; I1 and I3 are the currents at ports 1 and 3, respectively. 1max I 3max Let a1 be the maximum current at port 1 and the maximum current at port 3, a3 be the ratio of the absolute difference between the actual current at port 1 and the reference value to the error limit, and R be the ratio of the absolute difference between the actual current at port 3 and the reference value to the error limit.

[0027] Furthermore, when controlling the operating current of the two ports, a local lookup table method is used for power decoupling in the boundary region of the power transmission range of each port, and a trained neural network is used for power decoupling in the middle region of the power transmission range of each port, including:

[0028] The disturbance variables corresponding to the operating currents of the two ports that need to be decoupled are converted into corresponding virtual phase shift angle variables after passing through the PI controller.

[0029] The two virtual phase shift angle variables are used to calculate the corresponding two phase shift angle variables through the decoupling matrix H;

[0030] When the operating currents of the two ports that need to be decoupled are in the boundary region, the reference phase shift angles corresponding to the two phase shift angle variables are determined by looking up the decoupling parameter table.

[0031] When the operating currents of the two ports that need to be decoupled are in the intermediate region, a pre-trained neural network is used to determine the reference phase shift angles corresponding to the two phase shift angle variables;

[0032] Based on the two phase shift angle variables and the corresponding reference phase shift angle, the output current of the two ports is controlled to achieve power decoupling.

[0033] Furthermore, the decoupling parameter table stores the correspondence between the port operating current and the corresponding phase shift angle, which is pre-calculated based on the transfer function matrix. The transfer function matrix is:

[0034]

[0035] In the formula: Δi is the current variable matrix, and Δi1 and Δi3 are the operating current variables of the two ports that need to be decoupled. The phase shift angle variable matrix, Let G be the two phase shift angles, and G be the gain matrix. 11 G 12 G 21 G 22 These are elements in the gain matrix.

[0036] Furthermore, this invention also proposes a power decoupling system for a three-port DC-DC converter, the system comprising:

[0037] The training module is used to transform the topology of the TAB converter into a delta-type equivalent circuit, and to train the neural network with the operating currents of the two ports that need to be decoupled in the delta-type equivalent circuit as inputs and the corresponding two phase shift angles as outputs.

[0038] The region division module is used to calculate the error between the phase shift angle obtained by the trained neural network and the actual phase shift angle, and to analyze based on the error and error limit to determine the boundary region and intermediate region of the power transmission range of each port;

[0039] The power decoupling module is used to decouple power in the boundary region of the power transmission range of each port by using a local lookup table method when controlling the operating current of the two ports, and to decouple power in the middle region of the power transmission range of each port by using a trained neural network.

[0040] Furthermore, the training module includes:

[0041] The first circuit equivalent unit is used to ignore the influence of the transformer magnetizing inductance and to convert the power transfer inductance and port electrical parameters to a certain port of the TAB converter topology, thus obtaining the Y-type equivalent circuit of the converter.

[0042] The second circuit equivalent unit is used to transform the Y-type equivalent circuit into a Δ-type equivalent circuit according to the Y-Δ conversion principle;

[0043] The training data calculation unit is used to calculate the operating current of the corresponding two ports in the Δ-type equivalent circuit when the two phase shift angles associated with a certain port change.

[0044] The training unit is used to train the neural network by taking the operating current of the two ports as input and the corresponding two phase shift angles as output.

[0045] Furthermore, the region division module includes:

[0046] The error calculation unit is used to calculate the predicted phase shift angle using a pre-trained neural network, and to calculate the error between the predicted phase shift angle and the actual phase shift angle.

[0047] A boundary region determination unit is used to determine the boundary region based on the operating current points of the two ports corresponding to when the error exceeds the error limit.

[0048] The intermediate region determination unit is used to determine the intermediate region based on the operating current points of the two ports corresponding to when the error does not exceed the error limit.

[0049] Furthermore, the system also includes an error limit calculation module, used for:

[0050] The objective function is solved, and the minimum value of the objective function is found under the constraints. The ε that satisfies the minimum value is taken as the error limit.

[0051] The objective function and constraints are as follows:

[0052] (x1, x2 ∈ R)

[0053] stg(x1) = x1 - θ th1 <0

[0054] h(x2) = x2 - θ th2 <0

[0055] In the formula: constraint g(x1) indicates that the current fluctuation does not exceed the threshold θ th1 The constraint h(x2) indicates that the ratio of the amount of data exceeding the error limit to the total amount of data does not exceed the threshold θ. th2 ; I1 and I3 are the currents at ports 1 and 3, respectively. 1max I 3maxLet a1 be the maximum current at port 1 and the maximum current at port 3, a3 be the ratio of the absolute difference between the actual current at port 1 and the reference value to the error limit, and R be the ratio of the absolute difference between the actual current at port 3 and the reference value to the error limit.

[0056] Furthermore, the power decoupling module includes:

[0057] The control unit is used to convert the disturbance variables corresponding to the operating currents of the two ports that need to be decoupled into corresponding virtual phase shift angle variables after passing through the PI controller.

[0058] The decoupling unit is used to calculate the two corresponding phase shift angle variables from the two virtual phase shift angle variables through the decoupling matrix H;

[0059] The lookup table unit is used to determine the reference phase shift angles corresponding to the two phase shift angle variables by looking up the decoupling parameter table when the operating currents of the two ports that need to be decoupled are in the boundary region.

[0060] The fitting unit is used to determine the reference phase shift angles corresponding to the two phase shift angle variables using a pre-trained neural network when the operating currents of the two ports that need to be decoupled are in the middle region.

[0061] The current control unit is used to control the output current of the two ports based on the two phase shift angle variables and the corresponding reference phase shift angle, so as to achieve power decoupling.

[0062] Furthermore, the decoupling parameter table stores the correspondence between the port operating current and the corresponding phase shift angle, which is pre-calculated based on the transfer function matrix. The transfer function matrix is:

[0063]

[0064] In the formula: Δi is the current variable matrix, and Δi1 and Δi3 are the operating current variables of the two ports that need to be decoupled. The phase shift angle variable matrix, Let G be the two phase shift angles, and G be the gain matrix. 11 G 12 G 21 G 22 These are elements in the gain matrix.

[0065] The advantages of this invention are:

[0066] (1) This invention trains a neural network by using the port current as the input and the phase shift angle as the output, and obtains a trained neural network for calculating the decoupling matrix. It also calculates the error between the phase shift angle predicted by the trained neural network and the actual phase shift angle, and analyzes the error and error limit to divide the operating region of the converter into a boundary region and an intermediate region. For the operating current point in the boundary region, a local lookup table method is used to obtain the decoupling matrix for power decoupling. For the operating current point in the intermediate region, a pre-trained neural network is used to calculate the decoupling parameters and determine the decoupling matrix to achieve power decoupling. The power decoupling method of "global fitting-local lookup table" proposed in this invention avoids the problem of excessive table dimension occupying a lot of memory when only the lookup table method is used, and also avoids the problem of poor decoupling effect at the boundary point when only the neural network is used. It ensures the power decoupling performance in the full power range, and at the same time greatly reduces the hardware requirements for data storage. It effectively solves the contradiction between high precision and low data volume and has high engineering value.

[0067] (2) By setting the objective function and multiple constraints, the error limit that meets the conditions is obtained and compared with the error calculated by the neural network. This can effectively determine the boundary and intermediate regions of the port power transmission range, so as to achieve the effective integration of the lookup table method and the neural network method.

[0068] Additional aspects and advantages of the invention will be set forth in part in the description which follows, and in part will be obvious from the description, or may be learned by practice of the invention. Attached Figure Description

[0069] Figure 1 This is a flowchart illustrating a power decoupling method for a three-port DC-DC converter proposed in an embodiment of the present invention.

[0070] Figure 2 This is a power decoupling block diagram of "global fitting-local table lookup" in an embodiment of the present invention;

[0071] Figure 3 This is a topology diagram of an isolated three-port DC-DC converter in an embodiment of the present invention;

[0072] Figure 4 This is an equivalent circuit diagram of the isolated three-port DC-DC converter topology in an embodiment of the present invention. (a) is the Y-type equivalent circuit, and (b) is the Δ-type equivalent circuit.

[0073] Figure 5 This is a flowchart illustrating the selection of the lookup table method and the neural network in this embodiment of the invention.

[0074] Figure 6This is a control block diagram of the present invention after adding a decoupling element at a certain steady-state operating point;

[0075] Figure 7 This is a diagram of the decoupling structure based on a 3-layer neural network in an embodiment of the present invention;

[0076] Figure 8 This is a control block diagram after adding a decoupling step within the global working scope in an embodiment of the present invention;

[0077] Figure 9 This is a block diagram example of the decoupling method used in different operating regions of the TAB converter in this embodiment of the invention;

[0078] Figure 10 This is embodiment D of the present invention. 12 Error graph;

[0079] Figure 11 This is embodiment D of the present invention. 13 Error graph;

[0080] Figure 12 This is a block diagram of the decoupling method used in different operating regions of the TAB converter in this embodiment of the invention;

[0081] Figure 13 This is a waveform diagram of load mutation in the diagonal matrix decoupling control based on neural network in an embodiment of the present invention;

[0082] Figure 14 This is a waveform diagram of load mutation based on diagonal matrix decoupling control using a combination of lookup table method and neural network in an embodiment of the present invention.

[0083] Figure 15 This is a schematic diagram of the power decoupling system of a three-port DC-DC converter proposed in an embodiment of the present invention. Detailed Implementation

[0084] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below in conjunction with the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0085] like Figures 1 to 2 As shown, the first embodiment of the present invention proposes a power decoupling method for a three-port DC-DC converter, the method comprising the following steps:

[0086] S10. Transform the TAB converter topology into a Δ-type equivalent circuit, and use the operating currents of the two ports that need to be decoupled in the Δ-type equivalent circuit as inputs and the corresponding two phase shift angles as outputs to train the neural network.

[0087] It should be noted that the relationship between the current and phase shift angle at the two ports can be obtained from the expression for transmission power.

[0088] S20. Calculate the error between the phase shift angle obtained by the trained neural network and the actual phase shift angle, and analyze based on the error and error limit to determine the boundary region and intermediate region of the power transmission range of each port.

[0089] S30. When controlling the operating current of the two ports, a local lookup table method is used for power decoupling in the boundary region of the power transmission range of each port, and a trained neural network is used for power decoupling in the middle region of the power transmission range of each port.

[0090] The power decoupling method proposed in this embodiment, which is "global fitting-local lookup table", uses a local lookup table method to obtain the decoupling matrix for power decoupling at the operating current point in the boundary region, and uses a pre-trained neural network to calculate the decoupling parameters and determine the decoupling matrix for power decoupling at the operating current point in the middle region. This avoids the problem of excessive memory consumption caused by the large table dimension when using only the lookup table method, and also avoids the problem of poor decoupling effect at the boundary point when using only the neural network.

[0091] In one embodiment, step S10: transforming the TAB converter topology into a Δ-type equivalent circuit, and training the neural network with the operating currents of the two ports that need to be decoupled in the Δ-type equivalent circuit as inputs and the corresponding two phase shift angles as outputs, includes the following steps:

[0092] S11. Ignoring the influence of the transformer magnetizing inductance and applying the power transfer inductance and port electrical parameters to a specific port of the TAB converter topology, we obtain the Y-type equivalent circuit of the converter.

[0093] S12. Based on the Y-Δ conversion principle, the Y-type equivalent circuit is transformed into a Δ-type equivalent circuit;

[0094] S13. When the phase angles of the two phase shifts associated with a certain port change, calculate the operating current of the two ports that need to be decoupled in the Δ-type equivalent circuit.

[0095] S14. The neural network is trained using the operating current of the two ports as input and the corresponding two phase shift angles as output.

[0096] Specifically, the topology of the three-port active full-bridge converter is as follows: Figure 3 As shown, by changing the phase shift angle and The magnitude and direction of the power can be adjusted. Ignoring the influence of the transformer's magnetizing inductance and converting the power transmission inductance and port electrical parameters to terminal 1, the Y-type equivalent circuit of the converter can be obtained. Based on the Y-Δ conversion principle, the Δ-type equivalent circuit can be obtained, such as... Figure 4 As shown.

[0097] In one embodiment, such as Figure 5 As shown, step S20 involves calculating the error between the phase shift angle obtained from the trained neural network and the actual phase shift angle, and analyzing the error and error limit to determine the boundary and intermediate regions of the power transmission range at each port. This includes the following steps:

[0098] S21. Calculate the predicted phase shift angle using a pre-trained neural network, and calculate the error between the predicted phase shift angle and the actual phase shift angle;

[0099] S22. Determine the boundary region based on the operating current points of the two ports corresponding to when the error exceeds the error limit;

[0100] S23. Determine the intermediate region based on the operating current points of the two ports corresponding to when the error does not exceed the error limit.

[0101] In one embodiment, the calculation process of the error limit includes:

[0102] The objective function is solved, and the minimum value of the objective function is found under the constraints. The ε that satisfies the minimum value is taken as the error limit.

[0103] The objective function and constraints are as follows:

[0104] (x1, x2 ∈ R)

[0105] stg(x1) = x1 - θ th1 <0

[0106] h(x2) = x2 - θ th2 <0

[0107] In the formula: constraint g(x1) indicates that the current fluctuation does not exceed the threshold θ th1 The constraint h(x2) indicates that the ratio of the amount of data exceeding the error limit to the total amount of data does not exceed the threshold θ. th2 ; I1 and I3 are the currents at ports 1 and 3, respectively. 1max I 3maxLet a1 be the maximum current at port 1 and the maximum current at port 3, a3 be the ratio of the absolute difference between the actual current at port 1 and the reference value to the error limit, and R be the ratio of the absolute difference between the actual current at port 3 and the reference value to the error limit.

[0108] Specifically, based on the delta-type equivalent circuit, the expression for the absolute value of the difference between the actual current value and the reference value at port 1 is as follows:

[0109]

[0110] make,

[0111] We can obtain:

[0112] |I 1R -I1|≤a1·ε

[0113] Similarly, we can conclude that:

[0114] |I 3R -I3|≤a3·ε

[0115] We can obtain: therefore

[0116] In the formula: I1 and I3 are the currents at ports 1 and 3, respectively. 1R I 3R For port 1, the reference current is I; for port 3, the reference current is I. 1max I 3max V1 represents the maximum current at port 1, and V2 represents the maximum current at port 3; V2 represents the voltage at port 2; n1 represents the transformer turns ratio between port 1 and port 3; n2 represents the transformer turns ratio between port 2 and port 3; f represents the operating frequency; L represents the voltage at port 2. 12 D is the inductance between port 1 and port 2 in the delta-type equivalent circuit. 12R For the shift compared to D 12 Reference value, L 13 D is the inductance between port 1 and port 3 in the delta-type equivalent circuit. 13R For the shift compared to D 12 The reference value, D 12 Compared to the shift between port 1 and port 2, D 13 For the shift ratio between port 1 and port 3, a1 is the ratio of the absolute value of the difference between the actual current and the reference value of port 1 to the error limit, and a3 is the ratio of the absolute value of the difference between the actual current and the reference value of port 3 to the error limit.

[0117] Furthermore, since a larger error limit results in a smaller amount of data exceeding the error limit, and consequently a smaller x2, x2 and ε are negatively correlated. This embodiment sets a current fluctuation constraint condition to ensure that the current fluctuation does not exceed the threshold θ. th2 Otherwise, operating at points with large errors will severely affect the decoupling effect; the ratio of data exceeding the error limit to the total data volume must not exceed the threshold θ. th1 Otherwise, the controller would still need to store a large amount of data. It should be understood that θ th1 θ th2 The value of should be determined based on the specific parameters in the actual application and the requirements for decoupling accuracy.

[0118] This embodiment filters out the larger error points generated during neural network training and replaces them with more accurate lookup table data. Simultaneously, based on the precise selection of error limits, it achieves an optimal trade-off between ensuring accuracy and reducing data volume.

[0119] In one embodiment, since the amount of data exceeding the error limit is fixed after the error limit is determined, the constraint h(x2) becomes ineffective. This embodiment can also use a penalty function to determine whether to use a neural network or a lookup table for decoupling. Let the penalty function be:

[0120] Φ(x) = f(x) - r{max[0, g(x)]} 2

[0121] Here, r is a relatively large positive number. If the operating point is within the feasible region (the region where the error does not exceed the error limit), g(x) is less than zero, the penalty term is 0, and the penalty function Φ(x) is equal to the original function f(x). If the operating point is within the infeasible region (the region where the error exceeds the error limit), g(x) is greater than zero, the penalty term is greater than 0, and the penalty function Φ(x) is less than the original function f(x). Therefore, the comparison between the penalty function and the original function value can be used to determine whether to use a neural network or a lookup table method.

[0122] Specifically, based on the delta-type equivalent circuit, the power expressions for ports 1 and 3 are as follows:

[0123]

[0124] The current expressions for ports 1 and 3 are as follows:

[0125]

[0126] The partial derivative of the current at port 1 with respect to the phase shift angle is:

[0127]

[0128] The partial derivative of the current at port 3 with respect to the phase shift angle is:

[0129]

[0130] It can be seen that: when and The smaller the value of I1, the faster its rate of change, and the greater the impact of phase shift angle error. Similarly, when... and The difference between The smaller the value of I3, the faster its rate of change, and the greater the impact of phase shift angle error.

[0131] Therefore, at different operating points, the magnitude of current fluctuation may differ even when the shift ratio error is the same. Once the error limit is determined, the maximum current fluctuation when the error equals the error limit can be calculated. The magnitude of the current fluctuation directly affects decoupling, and the penalty function directly uses the magnitude of the current fluctuation to determine whether the operating point is within the feasible region.

[0132] In one embodiment, step S30: when controlling the operating current of the two ports, a local lookup table method is used for power decoupling in the boundary region of the power transmission range of each port, and a trained neural network is used for power decoupling in the middle region of the power transmission range of each port, including the following steps:

[0133] S31. The disturbance variables corresponding to the operating currents of the two ports that need to be decoupled are converted into corresponding virtual phase shift angle variables after passing through the PI controller.

[0134] S32. The two virtual phase shift angle variables are used to calculate the corresponding two phase shift angle variables through the decoupling matrix H;

[0135] S33. When the operating currents of the two ports that need to be decoupled are in the boundary region, find the decoupling parameter table to determine the reference phase shift angles corresponding to the two phase shift angle variables;

[0136] S33. When the operating current of the two ports that need to be decoupled is in the middle region, a pre-trained neural network is used to determine the reference phase shift angles corresponding to the two phase shift angle variables.

[0137] S34. Based on the two phase shift angle variables and the corresponding reference phase shift angle, control the output current of the two ports to achieve power decoupling.

[0138] Furthermore, the transmission power P between each port can be derived using the fundamental frequency analysis method. 12 P 13 P 23 They are respectively:

[0139]

[0140] The transfer function matrix of the small signal can be obtained by performing a Taylor expansion at a certain operating point A:

[0141]

[0142] In the formula: Δi is the current variable matrix, and Δi1 and Δi3 are the operating current variables of the two ports. The phase shift angle variable matrix, Let G be the two phase shift angles, and G be the gain matrix. 11 G 12 G 21 G 22 These are elements in the gain matrix.

[0143] Furthermore, the formula for calculating the elements in the gain matrix is ​​as follows:

[0144]

[0145] According to the small-signal model expression of the DC-side current of the TAB converter, there is a coupling relationship between the current variables Δi1 and Δi3: through the gain matrix element G. 11 and G 22 The current variables Δi1 and Δi3 are affected by the phase shift angle variable. The combined effects of these factors mean that the converter current variable cannot be controlled independently, which is detrimental to system stability. Therefore, decoupling control methods are needed to eliminate the coupling between current variables.

[0146] This embodiment employs a diagonal matrix decoupling method. A decoupling matrix network H is added to the front end of the original TAB converter small-signal model. The phase shift angle variable is:

[0147]

[0148] In the formula: and This is a virtual perturbation variable with no actual physical meaning. The decoupling matrix H is designed as the inverse of the gain matrix G.

[0149]

[0150] The current variables Δi1, Δi3, and Δi3 can be derived from the combined gain matrix. The expression:

[0151]

[0152] As a result, the coupling relationship between currents is eliminated, and the control system of the TAB converter can use dummy variables. and This directly achieves independent control of current variables Δi1 and Δi3. The input current variable Δi1 at port 1 and the output current variable Δi3 at port 3 of the TAB converter are selected as the control targets. Combining this with the diagonal matrix decoupling method, a control scheme is designed as follows: Figure 6 The control system shown.

[0153] However, since the value of the decoupling matrix H is determined based on the TAB converter gain matrix G, and matrix G is determined based on the steady-state operating point, therefore... Figure 5 The control system shown is only applicable to converter power control under a certain steady state. When the voltage or current at a certain port changes significantly, the converter's operating point will also change significantly, and the decoupling control will fail. Therefore, when the operating point changes, it is necessary to calculate the decoupling matrix of the new operating point to achieve decoupling control.

[0154] To address this, this embodiment pre-calculates a decoupling parameter table and pre-trains a neural network, with the decoupling structure diagram based on a 3-layer neural network as shown below. Figure 7 As shown, the control block diagram when using the diagonal matrix decoupling method based on a lookup table and a neural network is as follows. Figure 8 As shown, the TAB converter control system uses port currents i1 and i3 as control targets. The disturbance variables of currents i1 and i3 are converted into virtual phase shift angle variables after passing through the PI controller. and Then, using the decoupling matrix H, the phase shift angle variable is calculated. and The reference phase shift angle, calculated using a lookup table or neural network, is adjusted according to the region where the port operating current point is located. and This changes the output currents i1 and i3 at ports 1 and 3, respectively.

[0155] In one embodiment, the decoupling parameter table stores the correspondence between the port operating current and the corresponding phase shift angle, which is calculated in advance from the transfer function matrix.

[0156] Furthermore, Figure 9 Images (a), (b), (c), (d), and (e) illustrate the decoupling methods used in different operating regions of the converter and provide some examples. The decoupling methods used by the converter in different operating regions will vary under different operating conditions. However, actual operating conditions are much more diverse, involving more than just the scenarios listed in the examples above. Therefore, it is necessary to analyze the specific situation and extrapolate from the examples above.

[0157] To verify the feasibility and effectiveness of the decoupling method described above, a simulation circuit was built in MATLAB in this embodiment. The simulation parameters are: V1 = 15V, V2 = 30V, V3 = 15V; switching frequency 50kHz; L1, L2, and L3 are 7.5μH, 14μH, and 7.5μH, respectively; C r =0.7μF; the transformer turns ratio is 1:2:1.

[0158] Using operating currents i1 and i3 as inputs, and a shift ratio D 12 and D 13 For output, the neural network is trained, where the resolution of the training data is (-2.5:0.02:2.5), and the operating currents i1 and i3 both range from [-2.5, 2.5]. The error graph obtained by calculating the error from the trained neural network is shown below. Figure 10 and Figure 11 As shown, the vertical axis represents the difference between the phase shift angles calculated by the trained neural network for i1 and i3 and the original data used for training. The error map is divided into four regions with relatively large errors. Then, the error limit is selected based on the objective function and constraints.

[0159] Since the amount of data exceeding the error limit is fixed regardless of the error limit value, x2 cannot be arbitrarily chosen during calculation; this acts as a constraint. The final calculated error limit ε is 0.02. Points with errors greater than 0.02 are stored in the controller using a lookup table. This reduces the required data size to only 6149 sets, compared to 63001 sets if only the lookup table method is used, significantly decreasing the required data size. Since error points are mainly concentrated in the boundary region, the converter operating area diagram using a combination of lookup table and neural network is approximately as follows: Figure 12 As shown.

[0160] Figure 13 and Figure 14 The simulation waveforms of the converter show that at 0.003s, the load on port 3 abruptly changes from 80% to full load, and at 0.005s, the load on port 3 abruptly changes from full load to 80% of full load. The waveforms show that the diagonal matrix decoupling method combining neural networks and lookup tables significantly improves the decoupling effect at the boundaries compared to the diagonal matrix decoupling method based solely on neural networks.

[0161] In addition, such as Figure 15 As shown, the second embodiment of the present invention also proposes a power decoupling system for a three-port DC-DC converter, the system comprising:

[0162] Training module 10 is used to transform the topology of the TAB converter into a Δ-type equivalent circuit, and to train the neural network with the operating currents of the two ports that need to be decoupled in the Δ-type equivalent circuit as inputs and the corresponding two phase shift angles as outputs.

[0163] The region division module 20 is used to calculate the error between the phase shift angle obtained by the trained neural network and the actual phase shift angle, and to analyze based on the error and error limit to determine the boundary region and intermediate region of the power transmission range of each port.

[0164] The power decoupling module 30 is used to perform power decoupling in the boundary region of the power transmission range of each port by using a local lookup table method when controlling the operating current of the two ports, and to perform power decoupling in the middle region of the power transmission range of each port by using a trained neural network.

[0165] In this embodiment, a local lookup table method is used to obtain the decoupling matrix for power decoupling at the operating current point in the boundary region. For the operating current point in the middle region, a pre-trained neural network is used to calculate the decoupling parameters and determine the decoupling matrix to achieve power decoupling. This avoids the problem of excessive table dimension occupying a lot of memory when only the lookup table method is used, and also avoids the problem of poor decoupling effect at the boundary point when only the neural network is used.

[0166] In one embodiment, the training module 10 includes:

[0167] The first circuit equivalent unit is used to ignore the influence of the transformer magnetizing inductance and to convert the power transfer inductance and port electrical parameters to a certain port of the TAB converter topology, thus obtaining the Y-type equivalent circuit of the converter.

[0168] The second circuit equivalent unit is used to transform the Y-type equivalent circuit into a Δ-type equivalent circuit according to the Y-Δ conversion principle;

[0169] The training data calculation unit is used to calculate the operating current of the corresponding two ports in the Δ-type equivalent circuit when the two phase shift angles associated with a certain port change.

[0170] The training unit is used to train the neural network by taking the operating current of the two ports as input and the corresponding two phase shift angles as output.

[0171] In one embodiment, the region division module 20 includes:

[0172] The error calculation unit is used to calculate the predicted phase shift angle using a pre-trained neural network, and to calculate the error between the predicted phase shift angle and the actual phase shift angle.

[0173] A boundary region determination unit is used to determine the boundary region based on the operating current points of the two ports corresponding to when the error exceeds the error limit.

[0174] The intermediate region determination unit is used to determine the intermediate region based on the operating current points of the two ports corresponding to when the error does not exceed the error limit.

[0175] In one embodiment, the system further includes an error limit calculation module, used for:

[0176] The objective function is solved, and the minimum value of the objective function is found under the constraints. The ε that satisfies the minimum value is taken as the error limit.

[0177] The objective function and constraints are as follows:

[0178] (x1, x2 ∈ R)

[0179] stg(x1) = x1 - θ th1 <0

[0180] h(x2) = x2 - θ th2 <0

[0181] In the formula: constraint g(x1) indicates that the current fluctuation does not exceed the threshold θ th1 The constraint h(x2) indicates that the ratio of the amount of data exceeding the error limit to the total amount of data does not exceed the threshold θ. th2 ; I1 and I3 are the currents at ports 1 and 3, respectively. 1max I 3max Let a1 be the maximum current at port 1 and the maximum current at port 3, a3 be the ratio of the absolute difference between the actual current at port 1 and the reference value to the error limit, and R be the ratio of the absolute difference between the actual current at port 3 and the reference value to the error limit.

[0182] In one embodiment, the power decoupling module 30 includes:

[0183] The control unit is used to convert the disturbance variables corresponding to the working currents of the two ports into corresponding virtual phase shift angle variables after passing through the PI controller.

[0184] The decoupling unit is used to calculate the two corresponding phase shift angle variables from the two virtual phase shift angle variables through the decoupling matrix H;

[0185] The lookup table unit is used to determine the reference phase shift angles corresponding to the two phase shift angle variables by looking up the decoupling parameter table when the operating currents of the two ports that need to be decoupled are in the boundary region.

[0186] The fitting unit is used to determine the reference phase shift angles corresponding to the two phase shift angle variables using a pre-trained neural network when the operating currents of the two ports that need to be decoupled are in the middle region.

[0187] The current control unit is used to control the output current of the two ports based on the two phase shift angle variables and the corresponding reference phase shift angle, so as to achieve power decoupling.

[0188] In one embodiment, the decoupling parameter table stores the correspondence between the port operating current and the corresponding phase shift angle, which is pre-calculated based on the transfer function matrix. The transfer function matrix is:

[0189]

[0190] In the formula: Δi is the current variable matrix, and Δi1 and Δi3 are the operating current variables of the two ports. The phase shift angle variable matrix, Let G be the two phase shift angles, and G be the gain matrix. 11 G 12 G 21 G 22 These are elements in the gain matrix.

[0191] It should be noted that other embodiments or implementation methods of the power decoupling system of the three-port DC-DC converter described in this invention can refer to the above-described method embodiments, and will not be repeated here.

[0192] In the description of this specification, references to terms such as "one embodiment," "some embodiments," "example," "specific example," or "some examples," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of the invention. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples.

[0193] Furthermore, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of technical features indicated. Thus, a feature defined as "first" or "second" may explicitly or implicitly include at least one of that feature. In the description of this invention, "a plurality of" means at least two, such as two, three, etc., unless otherwise explicitly specified.

[0194] Although embodiments of the present invention have been shown and described above, it is understood that the above embodiments are exemplary and should not be construed as limiting the present invention. Those skilled in the art can make changes, modifications, substitutions and variations to the above embodiments within the scope of the present invention.

Claims

1. A power decoupling method for a three-port DC-DC converter, characterized in that, The method includes: Transform the TAB converter topology to Type equivalent circuit, and with The neural network is trained by taking the operating currents of the two ports that need to be decoupled in the equivalent circuit as inputs and the corresponding two phase shift angles as outputs. The error between the phase shift angle obtained from the trained neural network and the actual phase shift angle is calculated, and based on the error and the error limit, the boundary region and intermediate region of the power transmission range of each port are determined. This includes using the pre-trained neural network to calculate the predicted phase shift angle and calculating the error between the predicted phase shift angle and the actual phase shift angle; determining the boundary region based on the operating current points of the two ports corresponding to when the error exceeds the error limit; and determining the intermediate region based on the operating current points of the two ports corresponding to when the error does not exceed the error limit. When controlling the operating current of the two ports, a local lookup table method is used for power decoupling in the boundary region of the power transmission range of each port, and a trained neural network is used for power decoupling in the middle region of the power transmission range of each port.

2. The power decoupling method for a three-port DC-DC converter as described in claim 1, characterized in that, The topology transformation of the TAB converter is as follows: Type equivalent circuit, and with The neural network is trained using the operating currents of the two ports that need to be decoupled in the equivalent circuit as inputs and the corresponding two phase shift angles as outputs, including: Ignoring the influence of the transformer magnetizing inductance and applying the power transfer inductance and port electrical parameters to a specific port of the TAB converter topology, the Y-type equivalent circuit of the converter is obtained. According to Y- The conversion principle transforms the Y-type equivalent circuit into... Type equivalent circuit; When the two phase angles associated with a certain port change, calculate the following. The operating currents of the two ports that need to be decoupled in the equivalent circuit; The neural network is trained using the operating current of the two ports as input and the corresponding two phase shift angles as output.

3. The power decoupling method for a three-port DC-DC converter as described in claim 1, characterized in that, The calculation process for the error limit includes: Solve the objective function, find its minimum value under constraints, and then find the value that satisfies this minimum value. As the aforementioned error limit; The objective function and constraints are as follows: Where: constraints This indicates that the current fluctuation does not exceed the threshold. ;constraint This indicates that the ratio of the amount of data exceeding the error limit to the total amount of data does not exceed a threshold. ; , ; I 1, I 3 represents the current at ports 1 and 3. I 1max , I 3max These are the maximum current values ​​at port 1 and port 3. This is the ratio of the absolute value of the difference between the actual current at port 1 and the reference value to the error limit. This is the ratio of the absolute value of the difference between the actual current at port 3 and the reference value to the error limit. R It is the set of real numbers.

4. The power decoupling method for a three-port DC-DC converter as described in claim 1, characterized in that, When controlling the operating current of the two ports, a local lookup table method is used for power decoupling in the boundary region of the power transmission range of each port, and a trained neural network is used for power decoupling in the middle region of the power transmission range of each port, including: The disturbance variables corresponding to the operating currents of the two ports that need to be decoupled are converted into corresponding virtual phase shift angle variables after passing through the PI controller. The two virtual phase shift angle variables are decoupled by a decoupling matrix. H The two corresponding phase shift angle variables are calculated. When the operating currents of the two ports that need to be decoupled are in the boundary region, the reference phase shift angles corresponding to the two phase shift angle variables are determined by looking up the decoupling parameter table. When the operating currents of the two ports that need to be decoupled are in the intermediate region, a pre-trained neural network is used to determine the reference phase shift angles corresponding to the two phase shift angle variables; Based on the two phase shift angle variables and the corresponding reference phase shift angle, the output current of the two ports is controlled to achieve power decoupling.

5. The power decoupling method for a three-port DC-DC converter as described in claim 4, characterized in that, The decoupling parameter table stores the correspondence between the port operating current and the corresponding phase shift angle, which is pre-calculated based on the transfer function matrix. The transfer function matrix is: In the formula: For the current variable matrix, , For the operating current variables of the two ports that need to be decoupled, The phase shift angle variable matrix, , These are the variables of two phase shift angles. Here is the gain matrix. , , , These are elements in the gain matrix.

6. A power decoupling system for a three-port DC-DC converter, characterized in that, The system includes: The training module is used to transform the TAB transformer topology to Type equivalent circuit, and with The neural network is trained by taking the operating currents of the two ports that need to be decoupled in the equivalent circuit as inputs and the corresponding two phase shift angles as outputs. The region division module is used to calculate the error between the phase shift angle obtained by the trained neural network and the actual phase shift angle, and to analyze based on the error and error limit to determine the boundary region and intermediate region of the power transmission range of each port; The power decoupling module is used to decouple power in the boundary region of the power transmission range of each port by using a local lookup table method when controlling the operating current of the two ports, and to decouple power in the middle region of the power transmission range of each port by using a trained neural network. The region division module includes: The error calculation unit is used to calculate the predicted phase shift angle using a pre-trained neural network, and to calculate the error between the predicted phase shift angle and the actual phase shift angle. A boundary region determination unit is used to determine the boundary region based on the operating current points of the two ports corresponding to when the error exceeds the error limit. The intermediate region determination unit is used to determine the intermediate region based on the operating current points of the two ports corresponding to when the error does not exceed the error limit.

7. The power decoupling system for a three-port DC-DC converter as described in claim 6, characterized in that, The training module includes: The first circuit equivalent unit is used to ignore the influence of the transformer magnetizing inductance and to convert the power transfer inductance and port electrical parameters to a certain port of the TAB converter topology, thus obtaining the Y-type equivalent circuit of the converter. The second circuit equivalent unit is used to determine Y- The conversion principle transforms the Y-type equivalent circuit into... Type equivalent circuit; The training data calculation unit is used to calculate the data when the two phase angles associated with a certain port change. The operating current of the two corresponding ports in the equivalent circuit; The training unit is used to train the neural network by taking the operating current of the two ports as input and the corresponding two phase shift angles as output.

8. The power decoupling system for a three-port DC-DC converter as described in claim 6, characterized in that, The system also includes an error limit calculation module, used for: Solve the objective function, find its minimum value under constraints, and then find the value that satisfies this minimum value. As the aforementioned error limit; The objective function and constraints are as follows: Where: constraints This indicates that the current fluctuation does not exceed the threshold. ;constraint This indicates that the ratio of the amount of data exceeding the error limit to the total amount of data does not exceed a threshold. ; , ; I 1, I 3 represents the current at ports 1 and 3. I 1max , I 3max These are the maximum current values ​​at port 1 and port 3. This is the ratio of the absolute value of the difference between the actual current at port 1 and the reference value to the error limit. This is the ratio of the absolute value of the difference between the actual current at port 3 and the reference value to the error limit. R It is the set of real numbers.

Images