Switching matrix battery active equalization system and method based on dab converter

Abstract

The application relates to a switch matrix type battery active balancing system and method based on a DAB converter, and belongs to the technical field of battery management and battery active balancing. The system comprises a battery pack, an equalization module, a DC bus, a DAB converter, a switch matrix and a driving module, a BMU and a main control MCU. The first side of the DAB converter is connected with the DC bus, and the other side is connected with any single battery through the switch matrix, so that isolated bidirectional energy transmission is realized. The SOC is collected and the deviation is calculated, the equalization is triggered or stopped according to the SOC deviation, the maximum single battery is selected as the target after safety constraint, the equalization current and time are calculated, and the equalization is executed in the order of discharging first and then charging. Further, the equalization decision can be output by a deep reinforcement learning strategy network based on a confidence domain strategy optimization algorithm and optimized through a conjugate gradient method and a backtracking line search. The application has the advantages of single-converter serving multiple single batteries, large equalization current, small number of devices, easy modular expansion and the like, and is suitable for large-capacity battery pack fast consistency maintenance scenes.

Description

Technical Field

[0001] This invention belongs to the field of battery management and active battery balancing technology, specifically relating to a switching matrix active battery balancing system and method based on a DAB converter. Background Technology

[0002] With the widespread application of power batteries and energy storage battery packs, individual cells in series-connected battery packs are prone to inconsistencies in capacity, internal resistance, and state of charge (SOC) due to differences in manufacturing, aging, and operating environments. This leads to a decrease in usable battery pack capacity, a shortened cycle life, and even safety risks such as overcharging, over-discharging, and overheating. Therefore, active battery balancing is of great significance for improving battery pack consistency, safety, and lifespan. Existing balancing methods mainly include passive balancing and active balancing. Passive balancing has a simple structure but suffers from significant energy waste and low balancing efficiency. Active balancing, while achieving energy transfer and higher efficiency, has drawbacks such as a large number of components, complex control, poor scalability, and limited balancing speed in some solutions.

[0003] The dual active bridge (DAB) converter is a typical isolated bidirectional DC / DC converter with advantages such as bidirectional energy flow, ease of soft switching, high power density, and suitability for medium to high power applications. In recent years, it has been widely used in energy storage interfaces and battery energy management. However, if DAB is directly used for balancing multiple series of individual cells, issues still need to be addressed, such as how to flexibly select the target cell among multiple cells, how to reduce hardware redundancy, and how to balance balancing speed and efficiency while ensuring safety.

[0004] Against this backdrop, there is an urgent need for a new type of equalization device and control strategy that can fully leverage the advantages of DAB bidirectional isolation power transfer, reduce the number of power devices, facilitate modular expansion, and enable rapid active equalization of any target unit. Summary of the Invention

[0005] The purpose of this invention is to overcome the problems of low balancing efficiency, large number of devices, complex control, poor scalability, and difficulty in balancing rapid balancing and safe and stable operation of large-capacity battery packs in existing active battery balancing systems. This invention provides a switch-matrix active battery balancing system and method based on a DAB converter, which enables a single DAB converter to select any target single cell as needed with the cooperation of a switch matrix, and realizes isolated bidirectional energy transfer between the target single cell and the system through the DC bus, thereby achieving active balancing of high SOC cells discharging back to the bus and the bus charging low SOC cells.

[0006] To achieve the above objectives, the present invention adopts the following technical solution: In a first aspect, the present invention provides a switch matrix battery active balancing system based on a DAB converter, comprising a battery pack, a balancing module, a switch matrix, a DAB converter, a DC bus, a battery management slave unit (BMU), a master control MCU, and a switch matrix drive module connected by wires; wherein the battery pack comprises a plurality of individual cells connected in series; One side of the DAB converter is connected to the DC bus, and isolated bidirectional energy transmission is achieved through phase shift modulation to complete the back-feed discharge equalization from high SOC cells to the DC bus, and the replenishment charging equalization from the DC bus to low SOC cells; the other side is connected to the selected target cell battery in the equalization module through a switch matrix. The battery management slave unit (BMU) is used to collect voltage, temperature, and current information of each individual battery cell. The main control MCU is used to receive information collected by the battery management slave control unit (BMU), control the DAB converter to perform equalization control based on the collected information, and select the working individual batteries through the switch matrix drive module.

[0007] The switch matrix includes a cell selection switch group and a converter polarity selection switch group. The cell selection switch group is used to lead out the positive and negative terminals of any cell in the equalization module to the polarity selection terminal. The converter polarity selection switch group is used to connect the positive and negative terminals of the polarity selection terminal to the positive and negative terminals of the DAB converter battery side, respectively.

[0008] In the battery pack, the positive and negative switches of each individual cell are connected to a set of cell selection switches. Each set of cell selection switches consists of two MOSFETs connected back to back to form a bidirectional blocking switch, and the two MOSFETs connected back to back are synchronously driven by the same switch matrix drive signal.

[0009] The converter polarity selection switch group includes four power switches, which are used to accurately map the positive and negative terminals of the target single cell to the positive and negative terminals of the DAB converter battery side when different target single cells are selected.

[0010] The DAB converter includes a first full-bridge, an isolation transformer, and a second full-bridge. The main control MCU adjusts the phase difference between the first and second full-bridges through phase shift modulation to continuously control the transmission power. When the SOC of the target single cell is lower than the balancing target, the DAB converter operates in charging mode to transmit DC bus energy to the target single cell. When the SOC of the target single cell is higher than the balancing target, the DAB converter operates in discharging mode to feed the target single cell energy back to the DC bus.

[0011] Secondly, the present invention provides a switching matrix battery active balancing method based on a DAB converter, comprising the following steps: S1: Collect the voltage, temperature and current of each individual battery cell, calculate or obtain the SOC of each individual battery cell through the main control MCU, and calculate the average SOC of the battery pack and the SOC deviation of each individual battery cell. S2: Calculate the difference between the SOC of the largest and smallest single cells in the battery pack. D SOC, when D When SOC is greater than the start threshold, the system enters the equilibrium phase; when ΔSOC is less than the stop threshold, the system exits the equilibrium phase. S3: Sort the SOC of all individual cells in the battery pack, select the cell with the largest SOC as the target cell for discharge balancing, and select the cell with the smallest SOC as the target cell for charge balancing. S4: Calculate the discharge equalization current and the charging equalization current based on the deviation amplitude of the relative average SOC of the discharge equalization target cell and the charging equalization target cell, and limit the equalization current to not exceed the maximum equalization current allowed by the device. S5: Calculate the discharge equalization time and the charging equalization time based on the amount of electricity that needs to be transferred from the target single battery and the discharge equalization current and the charging equalization current. S6: Select the target cell for discharge equalization using a switch matrix and control the DAB to operate in discharge mode for a duration equal to the discharge equalization time; select the target cell for charge equalization using a switch matrix and control the DAB to operate in charge mode for a duration equal to the charge equalization time; within one equalization cycle, perform discharge equalization first, then perform charge equalization in sequence; after the equalization cycle ends, return to step S2 to continue judging until the stop condition is met.

[0012] S2 further includes: allowing equalization to be performed only when the voltage and temperature of each individual cell are within the allowable voltage window and the allowable temperature window, otherwise balancing is paused. The allowable voltage window is:

[0013] in, This refers to the voltage of a single cell. This is the discharge cutoff voltage. This is the charging cutoff voltage. This is the voltage margin.

[0014] The equalization current given in step S5 satisfies a function relationship that is monotonically correlated with the SOC deviation amplitude, and includes a deviation threshold. ,when When the equalization current is 0 or maintains its minimum value, According to the proportional coefficient k I Calculate the equalization current and limit it.

[0015] The equalization execution in step S6 is performed through the output of the deep reinforcement learning policy network, and the specific method is as follows: Obtain the SOC deviation of each cell in the battery pack at the current decision moment, as the current state for executing the Markov decision process; The current state is input into the deep reinforcement learning policy network, and the deep reinforcement learning policy network outputs the equilibrium control action at the current decision moment. The equilibrium control action includes at least: the selection of the target single cell, the energy transmission direction, and the equilibrium current setting or the DAB converter phase shift setting. The equalization control action controls the DAB converter to perform the corresponding energy transfer operation, and after the operation is completed, the updated SOC deviation of each cell in the battery pack is re-acquired. The updated SOC bias is used as the next state of the Markov decision process, and the next state is re-inputted into the deep reinforcement learning policy network for training and updating, and the equilibrium control action at the next moment is output, thereby realizing iterative closed loop. The deep reinforcement learning policy network is trained and updated using a trust region policy optimization method. During training, the substitution advantage function is used as the optimization objective, and the average KL divergence of the new and old policy action distributions does not exceed a preset threshold δ as a constraint to limit the policy update magnitude and improve training stability.

[0016] The trust region strategy optimization method uses a second-order approximation and the conjugate gradient method to solve for the natural gradient direction, and uses a backtracking search to adjust the step size until the KL divergence constraint is satisfied.

[0017] Compared with the prior art, the present invention has the following beneficial effects: This invention provides a switch-matrix battery active balancing system and method based on a DAB converter. The DAB converter serves as the core of energy conversion, enabling isolated bidirectional energy transfer between individual cells and the DC bus. The DAB converter features high efficiency, high power density, and wide-range soft-switching characteristics, effectively reducing energy loss during the balancing process. Simultaneously, the electrical isolation design ensures system safety, preventing battery-side faults from affecting the high-voltage bus. By controlling the cell selection switch group and the converter polarity selection switch group through the switch matrix drive module, any individual cell within the balancing module can be selected at any time, and polarity matching can be automatically completed.

[0018] Furthermore, the cell selection switch adopts a dual MOSFET back-to-back structure, which is synchronously driven by the same drive signal. This completely suppresses the risk of parasitic diode mis-conduction and fundamentally prevents battery pack short circuit accidents, greatly improving system reliability.

[0019] Furthermore, the balancing device uses balancing modules as the basic unit. Each module can independently manage M series-connected individual cells. By increasing the number of modules, it can adapt to battery packs with different numbers of series. It has good versatility and scalability and is suitable for various application scenarios such as electric vehicles and energy storage power stations.

[0020] Furthermore, the main control MCU, based on multi-dimensional information such as SOC deviation, voltage, and temperature, combined with safety constraint judgments, achieves precise equalization triggering, target selection, and current setting. In particular, by adopting a "discharge first, charge later" equalization cycle sequence, the energy generated and absorbed by the DC bus within one cycle tends to be balanced, significantly reducing bus voltage fluctuations, reducing the number of switching operations, and extending device lifespan.

[0021] Furthermore, a deep reinforcement learning framework based on the trust-region policy optimization algorithm is introduced, modeling the equilibrium decision as a Markov decision process. The optimal action (target unit, energy flow direction, current magnitude) is automatically output through the policy network. The trust-region policy optimization algorithm, through KL divergence constraints and second-order approximation optimization, ensures training stability and convergence, enabling the system to adapt to complex and changing operating conditions and achieve end-to-end intelligent equilibrium control. Its equilibrium speed and accuracy are superior to traditional rule-based methods.

[0022] Furthermore, the system is equipped with bus voltage detection and protection logic, allowable voltage window and temperature window constraints. When any parameter exceeds the limit, the equalization current is automatically reduced or equalization is paused, ensuring the safe operation of the battery pack and circuit in all aspects. Attached Figure Description

[0023] Figure 1 This is a system structure diagram of the present invention. Detailed Implementation

[0024] To further understand the content of this invention, the invention will be described in detail below with reference to the accompanying drawings and specific embodiments. It should be understood that the embodiments are merely illustrative and not limiting of the invention.

[0025] Example 1 like Figure 1 As shown, a switching matrix battery active balancing system based on a DAB (Dual Active Bridge) converter is composed of the following components: The system includes a battery pack, an equalization module, a DC bus, a DAB converter, a switch matrix drive module, a battery management slave unit (BMU), and a master control MCU.

[0026] The battery pack consists of N individual cells connected in series, and each balancing module corresponds to one of the cells in the battery pack. M A series-connected single cell, and M≥2, the balancing module is a local unit that performs specific balancing actions.

[0027] The DC bus is responsible for supplying power to the active balancing system and is also the energy exchange medium for active battery balancing. The DC bus is equipped with voltage detection and protection logic. When the bus voltage is abnormal, the main control MCU can intervene to adjust it.

[0028] One side of the DAB converter is connected to the DC bus, and isolated bidirectional energy transfer is achieved through phase shift modulation to complete the back-feed discharge balancing of high SOC cells to the DC bus, and the replenishment charging balancing of the DC bus to low SOC cells; the other side is connected to the selected target cell in the balancing module through a switch matrix.

[0029] The DAB converter includes a first full-bridge, an isolation transformer, and a second full-bridge. The main control MCU adjusts the phase difference between the first and second full-bridges through phase shift modulation to continuously control the transmission power. When the SOC of the target single cell is lower than the balancing target, the DAB converter operates in charging mode to transmit DC bus energy to the target single cell. When the SOC of the target single cell is higher than the balancing target, the DAB converter operates in discharging mode to feed the target single cell energy back to the DC bus.

[0030] The switch matrix is ​​a key component for achieving flexible balancing, physically consisting of a two-layer structure: a cell selection switch group and a converter polarity selection switch group. The cell selection switch group (S01, S02; S11, S12; ...; Sn1, Sn2) corresponds to each balancing module, used to lead the positive and negative terminals of any single cell within the balancing module to the polarity selection terminal, and consists of multiple bidirectional blocking switches. The positive and negative terminals of each cell are led out through a bidirectional blocking switch consisting of two MOSFETs connected back-to-back (S01, S02), controlled by the same drive signal to prevent short circuits within the battery pack and suppress parasitic diode conduction. The converter polarity selection switch group consists of four power switches (SA, SB, SC, SD). This group of switches can flexibly change the mapping relationship between the positive and negative terminals of the target cell and the battery-side interface of the DAB converter according to the physical location of the target cell, thereby achieving polarity matching.

[0031] The battery management slave unit (BMU) is located close to the battery pack and is used to collect the voltage, temperature and current information of each individual battery cell in real time and transmit it to the main control MCU.

[0032] The main control MCU receives information collected by the battery management slave control unit (BMU), calculates the SOC of each individual battery cell based on the collected information, controls the DAB converter to perform equalization control according to the equalization strategy, and selects the working individual batteries through the switch matrix drive module.

[0033] The switch matrix driver module receives instructions from the main control MCU and provides precise drive signals to the switch group, ensuring the reliability of gating and polarity matching operations.

[0034] Example 2 A switch-matrix battery active balancing system based on a DAB converter includes the following steps in its active balancing method: S1: After the system starts running, the battery management slave unit (BMU) periodically collects the voltage, temperature, and current data of each individual battery cell. The main control MCU calculates or obtains the SOC value of each individual battery cell based on the collected data, and records it as the first value. i The SOC of a single cell is SOC i ( k The average SOC of the battery pack and the SOC deviation of each individual cell were calculated. e i ; S2: The main control MCU calculates in real time the difference between the SOC of the largest single cell and the smallest single cell in the battery pack. D SOC. When D SOC is greater than the set startup threshold D When SOC_on, the system determines that equalization is needed and enters the working state; when D SOC decreases to below the stopping threshold D When SOC_off occurs, the battery pack is determined to have reached a balanced state, and the system exits the equalization process.

[0035] Furthermore, the system must perform a safety check before executing any equalization action. Equalization is only permitted when the voltage of all individual cells is within the permissible voltage window and the temperature is within the permissible range. If any individual cell exceeds its limit, equalization is immediately suspended to protect the battery.

[0036] The allowable voltage window is:

[0037] in, This refers to the voltage of a single cell. This is the discharge cutoff voltage. This is the charging cutoff voltage. This is the voltage margin.

[0038] The allowable temperature window is:

[0039] S3: For the current balancing module, sort the SOC of its M individual cells. Select the cell with the highest SOC as the target cell for discharge balancing, and select the cell with the lowest SOC as the target cell for charge balancing.

[0040] S4: Calculate the discharge equalization current and the charging equalization current based on the deviation amplitude of the relative average SOC of the discharge equalization target cell and the charging equalization target cell, and limit the equalization current to not exceed the maximum equalization current allowed by the device.

[0041] The magnitude of the balancing current is monotonically correlated with the SOC deviation of a single cell. The system sets a threshold based on the deviation. ,when When the equalization current is set to 0 or maintained at its minimum value, the equalization current is set to 0; when At that time, according to the proportional coefficient Calculate the theoretical equilibrium current and limit it to the maximum equilibrium current allowed by the device.

[0042] The formula for calculating the equalization current is:

[0043] In the formula, C is the battery charge / discharge rate; This refers to the battery's rated capacity. To balance the current coefficient, which is proportional to the battery capacity, we take... = 833.3; e The threshold for the equalization deviation of a single cell is taken as follows: e = 0.1%.

[0044] S5: Calculate the required discharge equalization time and charge equalization time based on the amount of electricity to be transferred from the target single cell (determined by the SOC deviation and battery capacity) and the equalization current determined in step S4.

[0045] The formula for calculating the equilibrium time is:

[0046] In the formula, The coulomb efficiency is 0.98 during charging and 1 during discharging. C n This refers to the available battery capacity.

[0047] S6: Within a complete balancing cycle, the system operates in a "discharge first, charge later" sequence to ensure that the total energy emitted and absorbed by the DC bus within one cycle is close to zero, thereby reducing bus voltage fluctuations and minimizing unnecessary switching. The specific operation is as follows: Discharge equalization stage: The main control MCU selects the target cell for discharge through the switch matrix drive module and controls the DAB converter to work in discharge mode to feed the energy of the cell back to the DC bus. The duration of this process is the discharge equalization time.

[0048] Charging equalization phase: Subsequently, the target cell for charging is selected, and the DAB converter is controlled to operate in charging mode to transfer DC bus energy to the individual battery cell. The duration is the charging equalization time.

[0049] After the balancing cycle ends, the system returns to step S2 to continue the judgment, forming a closed-loop control.

[0050] Furthermore, the aforementioned balancing process is executed through the output of a deep reinforcement learning policy network, modeling the battery balancing problem as a standard Markov decision process. The specific method is as follows: Obtain the SOC deviation of each cell in the battery pack at the current decision moment, as the current state for executing the Markov decision process; The current state is input into a deep reinforcement learning policy network, which then outputs the equilibrium control action at the current decision moment. The equilibrium control action includes at least the following: Target cell selection: Specify the discharge target cell number and the charging target cell number that should be selected at the current moment; Energy transfer direction: Indicates whether the DAB converter is operating in charging or discharging mode; The equalization current is given or the phase shift of the DAB converter is given; the desired equalization current value is directly output, or the phase shift (duty cycle) of the DAB converter is output, the latter of which can directly control the transmission power.

[0051] The equalization control action controls the DAB converter to perform the corresponding energy transfer operation, and after the operation is completed, the updated SOC deviation of each cell in the battery pack is re-acquired. The updated SOC bias is used as the next state of the Markov decision process, and the next state is re-inputted into the deep reinforcement learning policy network for training and updating, outputting the equilibrium control action at the next moment, thereby realizing iterative closed loop.

[0052] Furthermore, to avoid performance crashes caused by excessive policy updates, the deep reinforcement learning policy network uses a trust region policy optimization method for training and updating. During training, the replacement advantage function is used as the optimization objective, and the average KL divergence of the new and old policy action distributions does not exceed a preset threshold δ as a constraint to limit the policy update magnitude and improve training stability.

[0053] Since the KL divergence constraint is non-convex, direct solution is difficult. Therefore, a second-order approximation and the conjugate gradient method are used to solve for the natural gradient direction, and a backtracking search is employed to adjust the step size until the KL divergence constraint is satisfied. The specific process is as follows: In the k During the next policy update, let the current policy parameter be... The parameter increment is A first-order approximation is made to the substitution advantage function, and a second-order approximation is made to the average KL divergence:

[0054]

[0055] in, To replace the gradient of the advantage function, , Here is the Fisher information matrix corresponding to the KL divergence. is the damping coefficient.

[0056] Solving linear equations using the conjugate gradient method: To obtain the natural gradient direction The conjugate gradient iteration process is as follows:

[0057]

[0058] In the formula, the initial value is .

[0059] Based on the KL divergence threshold Scaling the natural gradient direction yields the initial update amount:

[0060] Then a backtracking search was used, the first... The parameters for the next candidate strategy are:

[0061] When candidate strategies simultaneously satisfy:

[0062]

[0063] make Otherwise, continue decreasing the step size. If the condition is still not met after reaching the maximum number of backtracking attempts, then maintain... .

[0064] After sufficient training, the policy network can directly output the optimal balancing action (target cell, energy flow direction, current magnitude) based on the current state of the battery pack, achieving end-to-end intelligent balancing control and significantly improving dynamic response capability and balancing efficiency.

[0065] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit it. Although the present invention has been described in detail with reference to the above embodiments, those skilled in the art should understand that modifications or equivalent substitutions can still be made to the specific implementation of the present invention. Any modifications or equivalent substitutions that do not depart from the spirit and scope of the present invention should be covered within the scope of protection of the claims of the present invention.

Claims

1. A switch-matrix battery active balancing system based on a DAB converter, characterized in that, It includes a battery pack, an equalization module, a switch matrix, a DAB converter, a DC bus, a battery management slave unit (BMU), a master control MCU, and a switch matrix drive module, all connected by wires; wherein the battery pack comprises several individual cells connected in series. One side of the DAB converter is connected to the DC bus, and isolated bidirectional energy transmission is achieved through phase shift modulation to complete the back-feed discharge equalization from high SOC cells to the DC bus, and the replenishment charging equalization from the DC bus to low SOC cells; the other side is connected to the selected target cell battery in the equalization module through a switch matrix. The battery management slave unit (BMU) is used to collect voltage, temperature, and current information of each individual battery cell. The main control MCU is used to receive information collected by the battery management slave control unit (BMU), control the DAB converter to perform equalization control based on the collected information, and select the working individual batteries through the switch matrix drive module.

2. The switching matrix battery active balancing system based on a DAB converter according to claim 1, characterized in that, The switch matrix includes a cell selection switch group and a converter polarity selection switch group. The cell selection switch group is used to lead out the positive and negative terminals of any cell in the equalization module to the polarity selection terminal. The converter polarity selection switch group is used to connect the positive and negative terminals of the polarity selection terminal to the positive and negative terminals of the DAB converter battery side, respectively.

3. The switching matrix battery active balancing system based on a DAB converter according to claim 2, characterized in that, In the battery pack, the positive and negative switches of each individual cell are connected to a set of cell selection switches. Each set of cell selection switches consists of two MOSFETs connected back to back to form a bidirectional blocking switch, and the two MOSFETs connected back to back are synchronously driven by the same switch matrix drive signal.

4. The switching matrix battery active balancing system based on a DAB converter according to claim 2, characterized in that, The converter polarity selection switch group includes four power switches, which are used to accurately map the positive and negative terminals of the target single cell to the positive and negative terminals of the DAB converter battery side when different target single cells are selected.

5. The switching matrix battery active balancing system based on a DAB converter according to claim 1, characterized in that, The DAB converter includes a first full-bridge, an isolation transformer, and a second full-bridge. The main control MCU adjusts the phase difference between the first and second full-bridges through phase shift modulation to continuously control the transmission power. When the SOC of the target single cell is lower than the equalization target, the DAB converter operates in charging mode to transmit DC bus energy to the target single cell. When the SOC of the target single cell is higher than the equilibrium target, the DAB converter operates in discharge mode to feed the energy of the target single cell back to the DC bus.

6. A switching matrix battery active balancing method based on a DAB converter, based on the switching matrix battery active balancing system based on a DAB converter as described in any one of claims 1 to 5, characterized in that, Includes the following steps: S1: Collect the voltage, temperature and current of each individual battery cell, calculate or obtain the SOC of each individual battery cell through the main control MCU, and calculate the average SOC of the battery pack and the SOC deviation of each individual battery cell. S2: Calculate the difference between the SOC of the largest and smallest single cells in the battery pack. Δ SOC, when Δ When SOC is greater than the start threshold, the system enters the equilibrium phase; when ΔSOC is less than the stop threshold, the system exits the equilibrium phase. S3: Sort the SOC of all individual cells in the battery pack, select the cell with the largest SOC as the target cell for discharge balancing, and select the cell with the smallest SOC as the target cell for charge balancing. S4: Calculate the discharge equalization current and the charging equalization current based on the deviation amplitude of the relative average SOC of the discharge equalization target cell and the charging equalization target cell, and limit the equalization current to not exceed the maximum equalization current allowed by the device. S5: Calculate the discharge equalization time and the charging equalization time based on the amount of electricity that needs to be transferred from the target single battery and the discharge equalization current and the charging equalization current. S6: Select the target cell for discharge equalization using a switch matrix and control the DAB to operate in discharge mode for a duration equal to the discharge equalization time; select the target cell for charge equalization using a switch matrix and control the DAB to operate in charge mode for a duration equal to the charge equalization time; within one equalization cycle, perform discharge equalization first, then perform charge equalization in sequence; after the equalization cycle ends, return to step S2 to continue judging until the stop condition is met.

7. The active battery balancing method based on a DAB converter according to claim 6, characterized in that, S2 further includes: allowing equalization to be performed only when the voltage and temperature of each individual cell are within the allowable voltage window and the allowable temperature window, otherwise balancing is paused. The allowable voltage window is: in, This refers to the voltage of a single cell. This is the discharge cutoff voltage. This is the charging cutoff voltage. This is the voltage margin.

8. The active battery balancing method based on a DAB converter according to claim 6, characterized in that, The equalization current given in step S5 satisfies a function that is monotonically correlated with the SOC deviation amplitude and includes a deviation threshold. ,when When the equalization current is 0 or maintains its minimum value, According to the proportional coefficient k I Calculate the equalization current and limit it.

9. The active battery balancing method based on a DAB converter according to claim 6, characterized in that, The equalization execution in step S6 is performed through the output of the deep reinforcement learning policy network, and the specific method is as follows: Obtain the SOC deviation of each cell in the battery pack at the current decision moment, as the current state for executing the Markov decision process; The current state is input into the deep reinforcement learning policy network, and the deep reinforcement learning policy network outputs the equilibrium control action at the current decision moment. The equilibrium control action includes at least: the selection of the target single cell, the energy transmission direction, and the equilibrium current setting or the DAB converter phase shift setting. The equalization control action controls the DAB converter to perform the corresponding energy transfer operation, and after the operation is completed, the updated SOC deviation of each cell in the battery pack is re-acquired. The updated SOC bias is used as the next state of the Markov decision process, and the next state is re-inputted into the deep reinforcement learning policy network for training and updating, and the equilibrium control action at the next moment is output, thereby realizing iterative closed loop. The deep reinforcement learning policy network is trained and updated using a trust region policy optimization method. During training, the substitution advantage function is used as the optimization objective, and the average KL divergence of the new and old policy action distributions does not exceed a preset threshold δ as a constraint to limit the policy update magnitude and improve training stability.

10. The active battery balancing method based on a DAB converter according to claim 9, characterized in that, The trust region strategy optimization method uses a second-order approximation and the conjugate gradient method to solve for the natural gradient direction, and uses a backtracking search to adjust the step size until the KL divergence constraint is satisfied.

Images