Battery dynamic balance control method, terminal equipment and storage medium

Abstract

The invention relates to a battery dynamic balance control method, terminal equipment and a storage medium, and the method comprises the steps: constructing a heterogeneous graph G = (V, E, W) used for representing a battery dynamic topology network, and V represents a node in the graph and represents a connection point in a battery topology structure; e represents the edges in the graph and comprises two types, one type is the edge where the battery is located, and the other type is the edge where the switch is located; w represents the weight parameter of the edge in the graph, the weight of the edge occupied by the battery is represented by the SOC of the battery, and the weight of the edge where the switch is located is represented by the switch state; constructing a graph convolutional neural network-based reinforcement learning model for battery dynamic balance control, and training the model through historical data; the state space of the model is set as a heterogeneous graph, and the action space is the switching action of each switch; and performing battery dynamic balance control through the trained model. The performance of the battery in all aspects can be balanced, and the discharging efficiency is improved to the maximum extent.

Description

Technical Field

[0001] The present invention relates to the field of battery balancing control, and in particular to a battery dynamic balancing control method, terminal equipment and storage medium. Background Art

[0002] Performance differences within the battery network and environmental impacts can cause individual battery differences to grow over time, affecting battery life and safety. Current passive balancing methods suffer from energy loss and low efficiency, while some of the active balancing methods currently proposed require more complex expert knowledge and control algorithms to implement.

[0003] Battery energy storage systems are widely used in various industrial and commercial fields due to their safety, cleanliness and high energy density. It provides the system with the required output voltage and increases the capacity by connecting multiple batteries in series and parallel. However, due to inconsistent manufacturing processes, temperature changes and fluctuations in SOC, the effective discharge time of each battery is not consistent. In traditional systems, the fixed battery topology used leads to inefficiency and even safety issues due to the inability to handle battery differences. With the emergence of high-performance power electronic switches, dynamically reconfigurable battery networks (DRBs) came into being. A DRB can transform a variety of different battery connections. Such as Figure 1 As shown in Figure 1, the performance of the battery pack is maximized by the switches in the system. Specifically, the battery network needs to be modeled and analyzed, and then the switches in the circuit are adjusted in real time according to the operating status of the battery in the system. Since the charge and discharge performance of the battery is affected by many factors such as temperature, discharge current, SOC, etc., it is very difficult to build a control strategy with such a complex relationship. In addition, this also increases the difficulty of strategy selection because the switch arrangements will increase exponentially when combined. Summary of the invention

[0004] In order to solve the above problems, the present invention proposes a battery dynamic balancing control method, a terminal device and a storage medium.

[0005] The specific plan is as follows:

[0006] A battery dynamic balancing control method comprises the following steps:

[0007] S1: Construct a heterogeneous graph G = (V, E, W) for representing a battery dynamic topology network, where V represents a node in the graph, referring to a connection point in the battery topology structure; E represents an edge in the graph, which includes two types, one is an edge where the battery is located, and the other is an edge where the switch is located; W represents a weight parameter of the edge in the graph, where the weight of the edge occupied by the battery is represented by the SOC of the battery, and the weight of the edge where the switch is located is represented by the switch state;

[0008] S2: Construct a reinforcement learning model based on graph convolutional neural network for battery dynamic balancing control, and train the model through the historical data of battery dynamic topology network during battery dynamic balancing control; set the state space of the model to the heterogeneous graph corresponding to the battery dynamic topology network, and the action space to the switching action of each switch in the battery dynamic topology network. The reward function is expressed by the following formula:

[0009]

[0010] Among them, r represents reward; Indicates whether the SOC error value of the battery at the current moment has a decreasing trend. The SOC error value is the difference between the maximum and minimum values ​​of the battery SOC in the battery dynamic topology network; when When When max(s ab ) represents the maximum number of switch actions among all actions, |s t -s t-1 | represents the number of switch actions between the previous moment and the current moment;

[0011] S3: Perform dynamic balancing control of batteries in the battery dynamic topology network through the trained model.

[0012] Furthermore, the calculation formula of the battery SOC is:

[0013]

[0014] Where SOC(t) represents the SOC at time t; t s represents a given time before time t; τ represents the time between t and t s The time difference between the two; i(τ) represents the current curve within the time τ; Q(t s ) indicates that the battery is s The charge value at the moment.

[0015] Furthermore, the weight of the edge where the switch is located is 0 or -1, wherein when the switch is turned on, the weight value of the edge is -1, otherwise it is 0.

[0016] Furthermore, the model consists of two heterogeneous graph convolutional layers for information transfer, and then a fully connected layer integrates and processes the features.

[0017] A battery dynamic balancing control terminal device comprises a processor, a memory, and a computer program stored in the memory and executable on the processor, wherein the processor implements the steps of the above-mentioned method in an embodiment of the present invention when executing the computer program.

[0018] A computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the steps of the method described above in an embodiment of the present invention are implemented.

[0019] The present invention adopts the above technical solution, which can effectively learn the battery network structure information, thereby balancing various aspects of battery performance and maximizing discharge efficiency. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] Figure 1 Shown is a schematic diagram of a battery network.

[0021] Figure 2 FIG. 1 is a schematic diagram of a dynamically reconfigurable battery network in Embodiment 1 of the present invention.

[0022] Figure 3 The figure shows the operating framework diagram of reinforcement learning in battery topology in the first embodiment of the present invention.

[0023] Figure 4 FIG. 1 is a schematic diagram showing the discharge condition of the fixed connection structure in the first embodiment of the present invention.

[0024] Figure 5 The figure is a schematic diagram of the discharge status of the reinforcement learning model based on the heterogeneous graph convolutional neural network in the first embodiment of the present invention.

[0025] Figure 6 The figure shows the performance comparison between heterogeneous graph reinforcement learning and traditional reinforcement learning in the first embodiment of the present invention.

[0026] Figure 7 The figure shows the performance comparison of different backup batteries in the first embodiment of the present invention. DETAILED DESCRIPTION

[0027] To further illustrate various embodiments, the present invention provides drawings. These drawings are part of the disclosure of the present invention, which are mainly used to illustrate the embodiments and can be used in conjunction with the relevant descriptions in the specification to explain the operating principles of the embodiments. With reference to these contents, ordinary technicians in this field should be able to understand other possible implementations and advantages of the present invention.

[0028] The present invention will now be further described with reference to the accompanying drawings and specific implementation methods.

[0029] Embodiment 1:

[0030] The embodiment of the present invention provides a battery dynamic balancing control method, which uses a graph neural network to represent a battery dynamic topology network for the first time. The method fully considers the influence of spatial information constraints and coupling relationships through the information transmitted by the graph, thereby obtaining the most appropriate switch configuration.

[0031] Graph convolutional neural networks (GCNs) can capture information about graph structures and features in data. In order to apply standard convolution to graph structures, this embodiment uses Fourier transform to apply convolution to the spectral domain and uses GCN to extract data from the graph at the current moment, thereby preserving the topological features of the battery network.

[0032] The purpose of this example is to find a solution to fully discharge the battery power, which can reduce the battery temperature and extend the overall life of the battery pack. The battery topology graph summarizes information into the network. The agent uses the information summarized in the graph to make appropriate decisions. For a neural network with a battery pack, its design is as follows Figure 2 shown.

[0033] In this embodiment, a heterogeneous graph is used to represent the dynamic topology network of the battery, that is, G = (V, E, W), where V = {v1, v2, ..., v n} represents n nodes in the graph, and its physical meaning is the connection point in the battery topology. The data finally obtained will be summarized at the node for training and testing of the prediction model. a and v b , there are two types of edges: e1 (a，b) or e2 (a，b) , which together form E, where e1 (a，b) represents the side where the battery is located, e2 (a，b) Indicates the edge where the switch is located.

[0034] In e1 (a，b) In the example, the SOC value is used as the edge weight. When the battery is connected to the system, the weight is the SOC parameter, otherwise it is 0. The calculation formula of SOC is as follows:

[0035]

[0036] Where SOC(t) represents the SOC at time t; t s represents a given time before time t; τ represents the time between t and t s The time difference between the two; i(τ) represents the current curve within the time τ; Q(t s ) indicates that the battery is s The charge value at the moment, in Ah.

[0037] Since SOC needs to be calculated based on current, and the battery temperature rise in the battery system is determined by the heat released and is closely related to the current, the calculation formula is as follows:

[0038]

[0039] Where Q hrepresents the heat released by the battery, and r represents the internal resistance of the battery.

[0040] It can be seen that when the battery is discharged, the battery temperature will increase due to heat. When the battery is disconnected, the heat generated by a single battery will decrease, which helps to reduce the temperature. Therefore, the battery dynamic topology network can reduce the overall average temperature of the battery network. It can be seen that the change in SOC indirectly reflects the change in temperature to a certain extent, so in this embodiment, the calculated SOC is used as the feature of the input graph.

[0041] In e2 ( a, b ) -1 and 0 are used as edge weights. When the switch is turned on, the edge weight value is -1, otherwise it is 0. ( a, b ) For different types of edges in the graph, negative values ​​are selected. From the perspective of energy, the aggregation of different types of edges on the node is the aggregation of different forms of energy, and the SOC value of the battery and the on / off of the switch are both considered as representations of energy. From the perspective of energy, the aggregation of different types of edges on the node is the aggregation between different forms of energy, and the SOC value of the battery and the on / off of the switch are both considered as forms of energy. Therefore, negative numbers are used here to define the closure of the switch. Although these values ​​do not represent the actual energy consumption, reasonable output values ​​can be obtained from the changes in the parameters W of each layer of the network (that is, the weight parameters of each edge). By summarizing the information in the graph, the battery can be discharged as completely as possible, the temperature rise can be slowed down, and the loss of equipment in the battery connection topology can be reduced. Finally, the corresponding modules in the obtained information are summarized at the node. As shown in the formula:

[0042]

[0043] Among them, 1 represents the number of layers of the network, h (l+1) The information collected at layer l+1, R composed of different r represents different relationships, agg is the aggregation function used, and h represents the information to be obtained. When the unit topology runs to time t and its connections change, the information aggregated by each node will also change, so the changes of different switches can be determined in the generated dynamic heterogeneous graph neural network. In this embodiment, the current feature information of N nodes is used to predict the strategy to be taken in the next time step, as shown in the formula:

[0044] Y t+1 = argmaxP{Y t+1 |X t}

[0045] Among them, X t Indicates the graph information at the current input time, Y t+1 Indicates the switch decision of the output at the next moment, P{Yt+1 |X t} indicates that from the input X t To output Y t+1 The transition probability.

[0046] In order to obtain multi-hop information (neighbors of neighbors), the model in this embodiment uses two heterogeneous graph convolution layers to transfer information, and then a fully connected layer integrates and processes the acquired features to generate a prediction result.

[0047] Through the graph convolutional neural network, we get a graph that aggregates various information. Next, we implement this network as a decision network for reinforcement learning. The switching decision task in the dynamic topology network is modeled as a Markov decision process, in which there are five tuples (S, A, P sa , R, γ), where S and A represent the state space and action space respectively, P sa represents the state transition probability, which means the probability of being able to transfer to other states after taking action a in the current state S, R represents the reward function, and γ represents the discount factor. Figure 3 The following is a flowchart of reinforcement learning decision making.

[0048] State: The state represents the intelligent agent’s perception of the overall condition of the battery module and is represented by a heterogeneous graph.

[0049] Action: In the action of the model, the switch action is provided according to the voltage required by the current load. In this embodiment, the output voltage is considered to be constant. When the load requires V volts, the battery that provides the load voltage needs to meet v1+v2+......+v m > V. If one battery is used as a regulating battery, the total number of batteries required is m+1, so the number of cases where all battery combinations meet the voltage requirement is 2m+1.

[0050] Reward: Reward refers to the benefit obtained by the agent after taking a certain action in a certain state, which can reflect whether the agent's behavior meets the task goal. The reward function designed in this embodiment is shown in the formula:

[0051]

[0052] Among them, r represents reward; Indicates whether the SOC error value of the battery at the current moment has a decreasing trend (the decreasing trend means that the value at the current moment is less than the value at the previous moment), and the SOC error value is the difference between the maximum and minimum values ​​of the battery SOC in the battery dynamic topology network; when When When there is no shrinking trend, a certain penalty is given; max(s ab) represents the maximum number of switch actions among all actions, |s t -s t-1 | represents the number of switching actions between the previous moment and the current moment; the constant 1 is added to ensure that the reward obtained when the SOC difference is reduced must be greater than the reward when it is not reduced.

[0053] The reason why the SOC difference is not considered as a reward in this embodiment, but only the reduction trend is considered, is that due to the trade-off with other objectives, setting a looser restriction helps improve the performance of other objectives, while requiring to reduce the SOC difference will affect the convergence effect. In battery balancing, the input is a heterogeneous graph rather than specific parameters. The complex information in the graph enables the intelligent system to learn a control strategy that takes all indicators into account. In summary, the heterogeneous graph reinforcement learning framework used in this embodiment is divided into two parts: aggregation of battery topology information and switching control using reinforcement learning algorithms.

[0054] Simulation experiment

[0055] Most of the current experimental environments are physical tests or semi-physical simulation test experiments. However, the high cost and safety of the test have brought difficulties to the research. Therefore, we designed an interactive environment based on Simulink during the experiment to generate a battery topology model with the help of the provided modules. In the simulation, we set up six battery connections and added a constant current source as an external load, and returned the test parameters related to the battery. Before each discharge, the initial SOC value of the battery was randomly set to 80%-100% to ensure the robustness of the model. During the simulation, a decision is made every ten seconds to generate different topologies by controlling the switches in the environment to obtain different connections.

[0056] (1)Battery pack balancing

[0057] In order to characterize the performance of the proposed framework, the battery is operated in the traditional fixed battery topology and the proposed battery dynamic topology. In comparison, the battery dynamic topology system has a higher cost performance, and its battery can consider the discharge power of the entire battery. In addition, due to the flexibility of the battery dynamic topology system, the faulty battery can be isolated to ensure stability. At the same time, based on the idle time of the battery, it also helps to cool the battery temperature.

[0058] Since the batteries are connected in series and the current of each battery is the same, it can be seen that there is always a difference in the SOC and voltage of the battery, regardless of other conditions. This difference does not disappear as the discharge process proceeds. The total energy discharged is about 27.86Wh, and the total energy is calculated as follows:

[0059]

[0060] Among them, Q(t) represents the energy released by the battery up to time t, and i and v represent the corresponding current and voltage at different times, respectively. The reason for using energy measurement is to ensure that the total number of batteries in the system remains unchanged, which will lead to inconsistent voltages provided by different schemes during the test, and if only the battery current and time are used, objective evaluation cannot be performed, so energy is selected to evaluate battery performance. In the calculation process, the cumulative sum of current and voltage is multiplied by the decision control interval at the time of decision making as the total energy, thereby simplifying the calculation formula.

[0061] Under fixed connection, when the SOC of the discharge process reaches the preset lower limit, the SOC will have a large difference, which indicates that the battery is unbalanced during the discharge process. In addition, since the battery is continuously discharged during operation, the battery temperature will continue to rise. After adopting the balancing solution proposed in this embodiment, the battery balancing effect is greatly improved.

[0062] The discharge of fixed connection structure is as follows Figure 4 As shown in Figure 2, the discharge of the reinforcement learning battery balancing model based on heterogeneous graph convolutional neural network is shown in Figure 2. Figure 5 As shown, the left figure in the figure represents the battery SOC, the middle figure represents the battery voltage, and the right figure represents the battery temperature. It can be seen from the figure that after adopting the heterogeneous graph reinforcement learning model, the difference in the SOC value of the battery gradually decreases from 12% at the beginning of discharge to 0.4% at the end of discharge. The use of the proposed control method allows more batteries to be almost fully discharged. In addition, the discharge time at the end of each cycle discharge is also extended. This increases the available capacity and improves overall safety. The total discharge power of the battery is 30.15Wh. The battery temperature is also reduced by letting the battery stand to reduce heat accumulation and improve heating efficiency. The average maximum temperature of the battery drops from 25.1438℃ to 25.1310℃. Since the 2Ah battery capacity is small and the battery is discharged at room temperature of 25℃, the temperature does not rise significantly. However, we can still see that the maximum temperature rise of the battery is reduced by about 10% due to the existence of the standing time. The experimental results show that the intelligent agent trained by the scheme proposed in this embodiment has a good balancing effect on the battery topology system.

[0063] (2) Performance comparison

[0064] In order to verify whether the performance of the heterogeneous graph reinforcement learning model is better than that of the traditional reinforcement learning model, we compared the two. In ten experiments, we used the same random SOC initial value and analyzed it from two aspects: energy index and the number of system switching during the connection process. The comparison results are shown in Figure 2. Figure 6 shown.

[0065] Figure 6The two straight lines in the middle left figure represent the average number of times the switch needs to be switched when the system makes a decision. When the discharge amount is the same, the number of battery switches varies greatly. The number of system switching actions is reduced from an average of 2.00 switching actions to 1.78 times, which is about 12.4%, and some of these switching actions are necessary because they need to be balanced as much as possible. Therefore, we believe that this reduction in proportion has greatly reduced the number of system switching operations. The total number of operations for a complete discharge process is reduced from an average of 577.4 times to 513.8 times. When the decision time interval is shortened, the impact on the number of system switching times becomes more obvious. The frequent switching of the reconfigurable battery network system will generate large transient currents and transient voltages, which will cause the temperature of the internal components of the switch to rise rapidly and shorten the system life. In addition, each switch will cause energy loss as well as voltage fluctuations and battery interference, which will affect the entire device. Therefore, it is necessary to consider the number of actions of each switch to maintain the stability of the system. In the experiment, when the switch action is required, the battery balancing can still proceed smoothly, and the system discharges all the electric energy as much as possible. The experiment compares the soc difference in ten discharge processes, and the detailed parameters are shown in Table 1.

[0066] Table 1

[0067]

[0068]

[0069] During the balancing process, the number of system switching operations is considered, so the decision is not necessarily made in the most favorable way to reduce the difference, but the final SOC difference will not be too large during the entire discharge process. As can be seen from the table, the model used in the experiment is able to significantly reduce the switching on and off in the system while performing the balancing task. Overall, the solution of this embodiment is superior to the traditional method, not only in balancing the differences between batteries, but also in considering the overall performance of the model and reducing the switching actions of the system.

[0070] (3) Comparison of the number of backup batteries

[0071] The experiments compared the performance when one backup battery and two backup batteries were connected. Increasing the number of backup batteries improves the overall flexibility of the battery and provides more options during the action selection process.

[0072] like Figure 7In , the battery action calculation shows that the maximum number of switches between two actions in all actions using one backup battery is 4 times, while the maximum number of switches between two actions in all actions using two backup batteries is 7 times. It can be seen that the cost of each action of choosing between two backup batteries is higher than the cost of choosing between one backup battery. The experiment also shows that the average number of actions required for each decision is 3.08 times when choosing two backup batteries, while it is 1.78 times when choosing one backup battery. In terms of the dischargeable capacity determined by the training effect of the training process, there is no significant difference between the two schemes. According to the above analysis, the backup battery with constant current discharge is the best choice in terms of cost and performance.

[0073] The embodiment of the present invention studies reinforcement learning based on heterogeneous graphs, and develops a model that can balance the energy of a reconfigurable battery network from the perspective of system stability. The model used in the experiment can make the most appropriate decision in a short time without recalculation like the model-based method. It overcomes the problem of unbalanced battery packs, effectively balances the SOC of the battery, discharges as much as possible, and has a suppressive effect on the temperature rise of the system. Compared with traditional reinforcement learning, due to the combination of the advantages of heterogeneous graph neural networks, the experimental model can effectively learn the information in the structure and grasp the distribution of switches in the system. Frequent operation of the switch will generate large instantaneous current and instantaneous voltage, increase the temperature of the switch, and shorten its service life. In addition, the action of the switch will generate back electromotive force, leading to problems such as energy loss and electromagnetic interference energy fluctuation. Therefore, the information of the battery network can be used to suppress the switching action of the system to a certain extent, reducing the switching action while meeting the balance requirements. The model used in the experiment reduced the number of switches by 12.4%. Finally, we analyzed the number of backup batteries, and the results showed that the flexibility of the system is not the more the better, and a single backup battery is the best choice.

[0074] Embodiment 2:

[0075] The present invention also provides a battery dynamic balancing control terminal device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the steps in the above method embodiment of embodiment 1 of the present invention when executing the computer program.

[0076] Further, as an executable solution, the battery dynamic balancing control terminal device can be a computing device such as a desktop computer, a notebook, a PDA, and a cloud server. The battery dynamic balancing control terminal device may include, but is not limited to, a processor and a memory. Those skilled in the art can understand that the composition structure of the above-mentioned battery dynamic balancing control terminal device is only an example of a battery dynamic balancing control terminal device, and does not constitute a limitation on the battery dynamic balancing control terminal device. It may include more or less components than the above, or a combination of certain components, or different components. For example, the battery dynamic balancing control terminal device may also include input and output devices, network access devices, buses, etc., and the embodiments of the present invention do not limit this.

[0077] Further, as an executable solution, the processor may be a central processing unit (CPU), or other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), field-programmable gate arrays (FPGA) or other programmable logic devices, etc. The general-purpose processor may be a microprocessor or any conventional processor, etc. The processor is the control center of the battery dynamic balancing control terminal device, and uses various interfaces and lines to connect various parts of the entire battery dynamic balancing control terminal device.

[0078] The memory can be used to store the computer program and / or module, and the processor realizes various functions of the battery dynamic balancing control terminal device by running or executing the computer program and / or module stored in the memory, and calling the data stored in the memory. The memory can mainly include a program storage area and a data storage area, wherein the program storage area can store an operating system and an application required for at least one function; the data storage area can store data created according to the use of the mobile phone, etc. In addition, the memory can include a high-speed random access memory, and can also include a non-volatile memory, such as a hard disk, a memory, a plug-in hard disk, a smart memory card (Smart Media Card, SMC), a secure digital (Secure Digital, SD) card, a flash card (Flash Card), at least one disk storage device, a flash memory device, or other volatile solid-state storage devices.

[0079] The present invention also provides a computer-readable storage medium, wherein the computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the steps of the above method in the embodiment of the present invention are implemented.

[0080] If the module / unit integrated in the battery dynamic balancing control terminal device is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the present invention implements all or part of the processes in the above-mentioned embodiment method, and can also be completed by instructing the relevant hardware through a computer program. The computer program can be stored in a computer-readable storage medium, and the computer program can implement the steps of the above-mentioned various method embodiments when executed by the processor. Among them, the computer program includes computer program code, and the computer program code can be in source code form, object code form, executable file or some intermediate form. The computer-readable medium may include: any entity or device capable of carrying the computer program code, recording medium, U disk, mobile hard disk, disk, optical disk, computer memory, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory) and software distribution medium, etc.

[0081] Although the present invention has been specifically shown and described in conjunction with the preferred embodiments, it should be understood by those skilled in the art that various changes may be made to the present invention in form and details without departing from the spirit and scope of the present invention as defined by the appended claims, all of which are within the scope of protection of the present invention.

Claims

1. A battery dynamic balancing control method, characterized in that: The following steps are involved: S1: Construct a heterogeneous graph G = (V, E, W) for representing a battery dynamic topology network, where V represents a node in the graph, referring to a connection point in the battery topology structure; E represents an edge in the graph, which includes two types, one is an edge where the battery is located, and the other is an edge where the switch is located; W represents the weight parameter of the edge in the graph, where the weight of the edge occupied by the battery is represented by the SOC of the battery, and the weight of the edge where the switch is located is represented by the switch state; S2: Construct a reinforcement learning model based on graph convolutional neural network for battery dynamic balancing control, and train the model through the historical data of battery dynamic topology network during battery dynamic balancing control; set the state space of the model to the heterogeneous graph corresponding to the battery dynamic topology network, and the action space to the switching action of each switch in the battery dynamic topology network. The reward function is expressed by the following formula: Among them, r represents reward; Indicates whether the SOC error value of the battery at the current moment has a decreasing trend. The SOC error value is the difference between the maximum and minimum values ​​of the battery SOC in the battery dynamic topology network; when When When max(s ab ) represents the maximum number of switch actions among all actions, |s t -s t-1 | represents the number of switch actions between the previous moment and the current moment; S3: Perform dynamic balancing control of batteries in the battery dynamic topology network through the trained model.

2. The battery dynamic balancing control method according to claim 1, characterized in that: The calculation formula of battery SOC is: Where SOC(t) represents the SOC at time t; t s represents a given time before time t; τ represents the time between t and t s The time difference between the two; i(τ) represents the current curve within the time τ; Q(t s ) indicates that the battery is s The charge value at the moment.

3. The battery dynamic balancing control method according to claim 1, characterized in that: The weight of the edge where the switch is located is 0 or -1, where the weight value of the edge is -1 when the switch is turned on, and 0 otherwise.

4. The battery dynamic balancing control method according to claim 1, characterized in that: The model consists of two heterogeneous graph convolutional layers for information transfer, and then a fully connected layer integrates and processes the features.

5. A battery dynamic balancing control terminal device, characterized in that: The method comprises a processor, a memory and a computer program stored in the memory and running on the processor, wherein the processor implements the steps of the method according to any one of claims 1 to 4 when executing the computer program.

6. A computer-readable storage medium storing a computer program, characterized in that: When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 4 are implemented.

Images