BMS battery intelligent management system

Abstract

The invention discloses a BMS battery intelligent management system, and belongs to the technical field of battery management. The method is used for solving the technical problems of passive equalization energy waste and active equalization efficiency bottleneck in the existing scheme. The long-term trend and real-time dynamic characteristics of battery aging are captured through a time sequence attention mechanism, and dynamic equilibrium threshold calculation is implemented, so that the dynamic equilibrium threshold can be dynamically adjusted along with SOH and temperature; magnetic coupling resonance topology is combined with impedance self-matching, a transmission strategy is adjusted in real time through SOC gradient and temperature data, temperature is corrected in real time, so that efficiency fluctuation is controlled within a preset range, uploading of original data is avoided through a federated learning architecture, battery state privacy is protected, the cycle is updated based on the strategy of the cycle index, and the reliability of the battery is improved. The strategy optimization precision is kept in the full life cycle of the battery, and the equalization time can be effectively shortened compared with a traditional fixed strategy; through SOH weighted priority ranking, the problem of energy transmission channel competition can be solved.

Description

Technical Field

[0001] This invention relates to the field of battery management technology, and more specifically to a BMS (Battery Management System). Background Technology

[0002] The Battery Management System (BMS) is an electronic system specifically designed to manage and monitor the operating status of battery packs, ensuring that batteries operate safely, efficiently, and reliably, and maximizing their lifespan.

[0003] When implementing existing BMS battery management solutions, the efficiency of traditional resistive balancing is less than 30%, resulting in a loss of range and wasted energy due to passive balancing. At the same time, the multi-stage energy transfer between adjacent batteries takes too long and the conversion efficiency is only 50%-60%, which is a bottleneck in the efficiency of active balancing. Summary of the Invention

[0004] The purpose of this invention is to provide a BMS (Battery Management System) to solve the technical problems of passive balancing energy waste and active balancing efficiency bottlenecks in existing solutions.

[0005] The objective of this invention can be achieved through the following technical solutions: A battery intelligent management system (BMS) includes: The battery multi-source data acquisition and processing module collects and preprocesses the battery pack's temperature distribution data, charge / discharge rate sequence, and cycle count. It then performs data analysis on the preprocessed multi-source data through a pre-built battery state coupling model, outputting the state of charge, state of health, and dynamic equilibrium threshold. The battery energy monitoring and intelligent control module, based on the output state vector, designs a magnetically coupled resonant dynamic energy routing network that includes a distributed multi-coil array and an impedance self-matching unit. It triggers energy transfer by calculating the SOC gradient. When the voltage difference of a single cell exceeds the dynamic equilibrium threshold, it activates a non-contact energy transmission channel and corrects the channel efficiency in real time using temperature distribution data. The optimized energy allocation module introduces federated reinforcement learning to optimize the balancing priority based on the energy routing network status. Each battery cell is treated as an agent, and the optimal energy allocation strategy is solved through multi-agent collaboration. The strategy update cycle is adaptively adjusted by the number of iterations.

[0006] Preferably, the expression involved in calculating the state of charge (SOC) is as follows: ;in, This is the initial SOC; This represents the charging and discharging current; a positive value indicates discharging, and a negative value indicates charging. Coulomb efficiency; Rated capacity; t represents the correction term output by the LSTM network; t is the time variable.

[0007] Preferably, the expression involved in calculating the State of Health (SOH) is as follows: ;in, This represents the current actual capacity. , The health factor is the output of the LSTM network, with a value ranging from 0 to 1.

[0008] Preferably, the expression involved in calculating the dynamic equilibrium threshold is: ;in, The dynamic equilibrium threshold; This represents the minimum SOH value in the battery pack. is the average temperature; e is a mathematical constant.

[0009] Preferably, based on the output State of Charge (SOC) value, the SOC gradient between the i-th battery cell and its neighboring cells is calculated, involving the following expression: ;in, Let SOC gradient be the SOC gradient between the i-th battery cell and its neighboring cells; The state of charge of the i-th battery cell; The state of charge of the (i-1)th adjacent battery cell; The state of charge of the (i+1)th adjacent battery cell; Energy transfer is initiated when any energy transfer condition is met.

[0010] Preferably, when dynamically switching non-contact energy transfer channels, the SOC gradient and voltage difference of all coils are scanned to generate candidate transfer pairs; the comprehensive score of each candidate pair is calculated, and the channel with the highest score is selected for activation.

[0011] Preferably, when adjusting the resonant frequency in real time, the target resonant frequency is obtained. And calculate the actual frequency. Relationship with temperature; when At that time, the impedance self-matching unit corrects the capacitance by switching the capacitor array. This makes the actual frequency Return to target resonant frequency .

[0012] Preferably, a two-tier architecture of local intelligent agent and central server is constructed, in which each individual battery unit acts as an independent intelligent agent, equipped with a lightweight training module, responsible for local state perception, policy iteration, and encrypted data uploading.

[0013] Preferably, the multi-agent cooperation strategy includes local training and global aggregation; During local training, each agent uses a proximate policy optimization algorithm to iterate its local policy. During global aggregation, each agent encrypts and uploads its local policy network parameters to the central server; the central server calculates the aggregation weight based on the SOH value and balance contribution of each agent, and generates global policy parameters. The central server distributes global policy parameters to each agent, which then initializes its local policy network using the global parameters and begins the next round of local training.

[0014] Preferably, when adaptively adjusting the strategy update cycle, the update cycle is determined by the number of iterations. Dynamic adjustment involves the following expressions: ;in, For update cycle; This is the error factor, with a value of 100; When multiple agents simultaneously request to send or receive energy, the central server prioritizes them based on the SOH value and SOC gradient.

[0015] Compared to existing solutions, the beneficial effects achieved by this invention are: This invention captures the long-term trend and real-time dynamic characteristics of battery aging through a temporal attention mechanism, achieving high-precision state estimation. Both SOC and SOH errors are reduced, and the SOC and SOH estimation accuracy can be effectively improved compared to traditional Kalman filtering. By implementing dynamic equalization threshold calculation, the dynamic equalization threshold can be dynamically adjusted according to SOH and temperature, avoiding over-equalization or under-equalization problems caused by traditional fixed thresholds, and effectively reducing equalization energy loss.

[0016] This invention utilizes a magnetically coupled resonant topology combined with impedance self-matching, which effectively improves efficiency compared to traditional wired active balancing and wireless inductive balancing. By adjusting the transmission strategy in real-time using SOC gradient and temperature data, it effectively shortens balancing time in scenarios with inconsistent battery packs. Non-contact transmission avoids wire wear and contact resistance issues, and the distributed coil array supports redundant design, automatically switching to a backup channel in case of a single coil failure, effectively increasing the system's mean time between failures (MTBF). Real-time temperature correction keeps efficiency fluctuations within a preset range, effectively reducing balancing energy loss compared to fixed-parameter designs.

[0017] This invention utilizes a federated learning architecture to avoid uploading raw data, protecting battery state privacy. Simultaneously, lightweight training of local agents reduces the computational burden on the central server, minimizing response latency. A reward function and priority ranking weighted by State of Health (SOH) ensure that healthy batteries with SOH ≥ 0.9 undertake over 60% of energy transfer tasks, while the energy transfer load of aging batteries with SOH ≤ 0.7 is reduced to below 20%, effectively extending the overall cycle life of the battery pack. The policy update cycle based on the number of cycles maintains policy optimization accuracy throughout the battery's lifespan, effectively shortening the equalization time compared to traditional fixed policies. Furthermore, the SOH-weighted priority ranking resolves energy transfer channel contention, effectively improving channel utilization and preventing equalization stagnation caused by conflicts. Attached Figure Description

[0018] The invention will now be further described with reference to the accompanying drawings.

[0019] Figure 1 This is a flowchart illustrating the operation of a BMS (Battery Management System) intelligent management system according to the present invention. Detailed Implementation

[0020] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and 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.

[0021] like Figure 1 As shown, the present invention is a BMS (Battery Management System) intelligent management system, comprising: The battery multi-source data acquisition and processing module collects and preprocesses temperature distribution data, charge / discharge rate sequences, and cycle counts of the battery pack. It then analyzes the preprocessed multi-source data using a pre-built battery state coupling model, outputting state of charge, state of health, and dynamic balancing thresholds. Specific steps include: The surface temperature of the battery pack is collected by eight distributed NTC temperature sensors and combined to obtain temperature distribution data. The sensor locations are: two positive tabs of individual cells, two negative tabs, and four in the middle of the casing. The collection range is -20℃ to 60℃.

[0022] The charging and discharging current is collected by a Hall current sensor connected in series in the main circuit. The real-time rate is calculated based on the rated capacity of the battery. The calculated real-time rates are arranged and combined in chronological order to obtain a charging and discharging rate sequence, which is pre-divided into 4 levels: 0~0.5C, 0.5~1C, 1~2C, and >2C.

[0023] The number of charge-discharge cycles of the battery pack is recorded cumulatively by the BMS. Each completion of a 0.2C deep discharge is counted as one cycle. For example, when the SOC changes from 100% to 20%, four intervals are pre-divided: 0~500 cycles, 501~1000 cycles, 1001~2000 cycles, and >2000 cycles. When preprocessing the collected temperature distribution data, charge / discharge rate sequence, and cycle count, the preprocessing includes, but is not limited to, normalizing the temperature distribution data; performing one-hot encoding on the charge / discharge rate sequence, with four levels corresponding to four-dimensional binary vectors; and numerically mapping the cycle count range: 0-500 cycles correspond to 0.1, 501-1000 cycles to 0.3, 1001-2000 cycles to 0.6, and >2000 cycles to 0.9. All preprocessing methods used are existing conventional techniques, and the specific implementation steps are not detailed here. When constructing a battery state coupling model based on a temporal attention mechanism, the model structure includes an input layer, a hidden layer, an attention layer, and an output layer. The input layer has a dimension of 16, including 8 normalized values ​​from temperature sensors, 4 unique thermal codes for magnification levels, and 4 loop number range mapping values. The hidden layer contains three bidirectional LSTM layers, or Bi-LSTM, with 128, 64, and 32 neurons per layer, respectively. The ReLU activation function is used, and the dropout rate is 0.2 to prevent overfitting. The attention layer employs a temporal attention mechanism, integrating features from three time windows: short-term (10 seconds), medium-term (1 minute), and long-term (10 minutes). The specific implementation is as follows: Sliding time window settings: window size is 50 sampling periods, sampling period is 1 second, window duration is 50 seconds, step size is 10 sampling periods, sliding interval is 10 seconds, and 16-dimensional feature vectors are extracted for each window; When calculating attention weights, define the query vector. Key vector Value vector , All are dimensions. The expression for calculating the attention weight matrix A is: ;in, For the mask matrix, This is a scaling factor to prevent gradient vanishing; This is the transpose of the key vector; It is the probability distribution function; The influence weights of real-time voltage signals and historical decay data are dynamically allocated through the attention weight matrix A. The voltage signal corresponds to the recent window, and the historical decay data corresponds to the long-term window. For example, for batteries with more than 1,000 cycles, the weight of historical decay data is increased by 20%; for windows with temperature fluctuations of more than ±3℃, the weight of real-time voltage signals is increased by 15%.

[0024] The output layer outputs a state vector containing the state of charge (SOC), state of health (SOH), and dynamic equilibrium threshold through a fully connected layer, using a linear activation function, with an output dimension of 3. The battery state coupling model is trained and optimized, which is a conventional technical solution. The specific implementation steps are not described here. When performing data analysis on the preprocessed data using a battery state-coupled model, the expression involved in calculating the State of Charge (SOC) is as follows: ;in, For the initial SOC, calibration is performed using the OCV-SOC curve; This represents the charging and discharging current; a positive value indicates discharging, and a negative value indicates charging. The coulomb efficiency is related to temperature and rate of increase, and is obtained by fitting experimental data. For example, η=0.99 at 0.5C and η=0.96 at 2C. The rated capacity decreases with increasing SOH. This is the correction term for the LSTM network output, compensating for accumulated errors, ranging from -2% to +2%; t is the time variable. The expression involved in calculating the State of Health (SOH) is as follows: ;in, This represents the current actual capacity. , The health factor output by the LSTM network has a value range of 0 to 1 and is obtained through training with cycle number and temperature aging data. The expression involved in calculating the dynamic equilibrium threshold is as follows: ;in, The dynamic equilibrium threshold; This represents the minimum SOH value in the battery pack. The average temperature is calculated by averaging data from eight temperature sensors; e is a mathematical constant.

[0025] In this embodiment of the invention, a time-series attention mechanism is used to capture the long-term trend and real-time dynamic characteristics of battery aging, thereby achieving high-precision state estimation. Both SOC and SOH errors are reduced, and the SOC and SOH estimation accuracy can be effectively improved compared to traditional Kalman filtering. By implementing dynamic equalization threshold calculation, the dynamic equalization threshold can be dynamically adjusted according to SOH and temperature, avoiding over-equalization or under-equalization problems caused by traditional fixed thresholds, and effectively reducing equalization energy loss.

[0026] The battery energy monitoring and intelligent control module, based on the output state vector, designs a magnetically coupled resonant dynamic energy routing network including a distributed multi-coil array and an impedance self-matching unit. Energy transfer is triggered by the calculated SOC gradient. When the voltage difference between individual cells exceeds the dynamic equilibrium threshold, a non-contact energy transfer channel is activated, and the channel efficiency is corrected in real time using temperature distribution data. Specific steps include: The designed magnetically coupled resonant dynamic energy routing network has a hardware architecture that includes a distributed multi-coil array topology and an impedance self-matching unit structure. The distributed multi-coil array topology employs a 3×N three-dimensional coil matrix, where N is the number of cells connected in series (e.g., N=12 series). Each individual cell is configured with one transmitting coil and one receiving coil. The coils are wound with Litz wire, and the magnetic core is Mn-Zn ferrite. The array is divided into rows and columns. The row direction consists of 3 parallel groups, corresponding to 3 temperature zones, which are divided by temperature distribution data, including low temperature zone <15℃, normal temperature zone 15-45℃, and high temperature zone >45℃; The column consists of N series-connected cells, with each cell coil independently connected to an impedance self-matching unit. Impedance self-matching unit structure: Each coil integrates an adjustable capacitor array and a digital control module, which collects the phase difference of voltage and current across the coil in real time and dynamically adjusts the capacitor combination through a PID algorithm to make the system work at the resonant point. This is a conventional technical solution, and the specific implementation steps will not be elaborated here. The adjustable capacitor array contains 16 0.1μF ceramic capacitors, which are switched by relays, and the total capacitance adjustment range is 0.1-1.6μF. The digital control module includes an STM32L4 microcontroller. Based on the output State of Charge (SOC) value, the SOC gradient between the i-th cell and its neighboring cells is calculated, using the following expression: ;in, Let SOC gradient be the SOC gradient between the i-th battery cell and its neighboring cells; The state of charge of the i-th battery cell; The state of charge of the (i-1)th adjacent battery cell; The state of charge of the (i+1)th adjacent battery cell; Energy transfer is initiated when any energy transfer condition is met; Among them, the energy transfer conditions include: when SOH > 80%, When SOH≤80% ; In addition, the voltage difference between individual cells is greater than the dynamic equilibrium threshold; When energy transfer is initiated, if the SOH of a high SOC battery cell is ≥90%, it will act as the energy transmitting end. If the SOH of a low-SOC battery cell is ≤80%, it is used as an energy receiving end. When multiple battery groups meet the conditions simultaneously, the unit in the normal temperature zone is selected for priority transmission based on temperature distribution data. When dynamically switching non-contact energy transmission channels, a distance-efficiency dual-factor weighted scoring method is used to select the optimal transmission channel. The relevant expression is: ;in, For the first The overall score of the channel; specifically, the score of the channel. The transmitting coil and the first The overall performance score of the transmission channels between the receiving coils is used to quantify and compare the priorities of different channels; the higher the score, the better the channel. This is the distance weighting coefficient, with a value range of [0,1], and a typical value of 0.3; For the first Center distance between coil pairs; This refers to the temperature-corrected transmission efficiency. This is the current coil temperature; When switching channels, the SOC gradient and voltage difference of all coils are scanned to generate candidate transmission pairs; for example, <transmitting unit 3, receiving unit 7>. Calculate the candidate pairs Select the channel with the highest value to activate and generate a channel activation flag; the way the channel activation flag is represented is not limited. The system is re-evaluated every 100ms during transmission. If a higher-scoring channel is found or the triggering condition disappears, the channel is switched. The switching time is ≤5ms, and the transmission is uninterrupted. When performing real-time correction of channel efficiency, the coil inductance is calculated. and capacitor The relationship between temperature and temperature is expressed by the following expression: ; ;in, These are the inductance and capacitance values ​​at 25℃, respectively. This is the temperature coefficient of inductance, with a default value of -0.002; This is the temperature coefficient of the capacitor, with a default value of -0.001. The reference temperature is 25°C. When performing real-time adjustment of the resonant frequency, the target resonant frequency is obtained. And calculate the actual frequency. The relationship with temperature is expressed by the following expression: ; It is a mathematical constant; when At that time, the impedance self-matching unit corrects the capacitance by switching the capacitor array. This makes the actual frequency Return to target resonant frequency The corrected formula is: ;in, Adjust the capacitance value to the target. The specific working steps of the impedance self-matching unit include: Real-time measurement of coil voltage U, current I, and phase difference The sampling period is 1ms. When performing mismatch judgment, if If so, it is determined to be an impedance mismatch, and a matching scheme is initiated; Adjust the capacitance value according to the calculated target, and select the closest capacitance combination by looking up a table and adjust it. Measure the phase difference again after adjustment. If still Then repeat the above steps until... .

[0027] In this embodiment of the invention, by combining magnetically coupled resonant topology with impedance self-matching, efficiency can be effectively improved compared to traditional wired active balancing and wireless inductive balancing. By adjusting the transmission strategy in real time using SOC gradient and temperature data, the balancing time can be effectively shortened in scenarios with inconsistent battery packs. Non-contact transmission avoids wire wear and contact resistance issues. The distributed coil array supports redundant design, automatically switching to a backup channel when a single coil fails, which can effectively improve the system's mean time between failures (MTBF). Real-time temperature correction keeps efficiency fluctuations within a preset range, effectively reducing balancing energy loss compared to fixed parameter designs.

[0028] The optimized energy allocation module, based on the energy routing network state, introduces federated reinforcement learning to optimize the allocation priority. Each battery cell is treated as an agent, with the State of Health (SOH) value as the reward function weight. The optimal energy allocation strategy is solved through multi-agent collaboration, and the strategy update cycle is adaptively adjusted by the number of iterations. Specific steps include: Construct a two-tier architecture of local intelligent agent and central server, wherein: Local agent: Each individual battery cell acts as an independent agent, equipped with a lightweight training module, responsible for local state perception, policy iteration, and encrypted data upload; the state input of the local agent includes SOH value, SOC value, number of cycles, and energy routing network status. The energy routing network status includes channel activation flags, temperature-corrected transmission efficiency, and individual cell voltage differences. Central server: Deployed on the BMS main controller, responsible for global policy aggregation, conflict coordination and model distribution, and communicates with agents via CAN bus; the central server's aggregation is based on the local policy gradient of each agent, with the goal of maximizing the combined benefits of global balancing efficiency and battery life; global balancing efficiency corresponds to the shortest balancing time, and battery life corresponds to the minimum SOH decay; When designing the state space, action space, and reward function of the local agent, the state vector of the i-th agent at time t is defined as follows: ;in, This is the output of the i-th battery health status, with a value range of 0 to 1, where 0 indicates failure and 1 indicates normal. This represents the i-th battery state of charge output, with a value ranging from 0% to 100%. For SOC gradient; This represents the difference between the voltage of the i-th battery and the average voltage. Let be the real-time transmission efficiency of the i-th battery-associated channel; This represents the number of loop iterations. The action space of the local agent is defined as a combination of the direction and power of energy transfer, and the design expression is: Where (0,0) indicates that it does not participate in energy transmission; (1,P) indicates that it acts as a transmitter, outputting power P, and the power P is determined by the real-time transmission efficiency of the channel associated with the i-th battery. Decide: The power P is 3W; The power P is 2W in the case of receiving; the power P is 1W in other cases; (2,P) represents the receiving power P when acting as a receiver; (3,P) represents the forwarding power P when acting as a relay. Define the reward function as a weighted sum of the two objectives of equilibrium efficiency and health protection, with the SOH value serving as the weight for health protection. The relevant expression is as follows: ;in, To influence the weighting coefficients, by Dynamic adjustment ; When the values ​​are 0.8, 0.5, and 0.2, they correspond to priority protection for battery aging, balance protection and equalization for semi-aging, and priority equalization for healthy batteries, respectively. For health protection items, ; For the balanced effect item, , The maximum allowed gradient; This is the energy loss term. λ is the penalty coefficient, with a value of 0.3, which balances the priority between energy loss and efficiency. Multi-agent collaborative strategies include local training and global aggregation; During local training, each agent uses a proximate policy optimization algorithm to iterate its local policy, with the goal of maximizing the cumulative reward. , The discount factor is 0.95; the local policy network is a 2-layer fully connected neural network, consisting of an input layer, a hidden layer, and an output layer. Input layer: 6-dimensional state vector ; Hidden layer: 32 neurons, containing the ReLU activation function; Output layer: 4-dimensional action probability distribution, corresponding to the action space. Softmax activation function; The training parameters include a learning rate of 0.001, a batch size of 32, and 5 local iterations per aggregation cycle. During global aggregation, each agent encrypts and uploads its local policy network parameters to the central server. These local policy network parameters include weights and biases. The upload is triggered by the following conditions: Local cumulative rewards A decrease of more than 10%; And, to reach the preset aggregation period; The central server bases its decisions on the SOH value and the balance contribution of each agent. Calculate aggregate weights Generate global policy parameters The expression involved is: ; ;in, , This is the equilibrium effect term for the i-th agent; The total number of agents; The aggregate weights are for the i-th agent; These are the local policy parameters for the i-th agent; The central server will use global policy parameters The data is distributed to each agent, which initializes its local policy network with global parameters and begins the next round of local training. When adaptively adjusting the policy update cycle, the update cycle is determined by the number of iterations. Dynamic adjustment involves the following expressions: ;in, For update cycle; This is an error factor, set to a value of 100, to avoid... Division by zero error; When multiple agents simultaneously request to send or receive energy, the central server prioritizes them based on the SOH value and SOC gradient. Among them, receiving priority ; Sending priority is ; Energy transmission channels are allocated in descending order of priority. Only one pair of agents is allowed to occupy the same channel, and one pair of agents corresponds to the transmitter and receiver. Agents not assigned a channel enter a waiting queue; The timeout period is 0.5 seconds. Request again after the timeout.

[0029] In this embodiment of the invention, a federated learning architecture is used to avoid uploading raw data, protecting battery state privacy. Simultaneously, lightweight training of local agents reduces the computational burden on the central server, thus reducing response latency. A reward function and priority ranking weighted by State of Health (SOH) ensure that healthy batteries with SOH ≥ 0.9 undertake more than 60% of the energy transfer tasks, while the energy transfer load of aging batteries with SOH ≤ 0.7 is reduced to below 20%, effectively extending the overall cycle life of the battery pack. A policy update cycle based on the number of cycles maintains policy optimization accuracy throughout the battery's entire lifespan, effectively shortening the equalization time compared to traditional fixed policies. The SOH-weighted priority ranking solves the energy transfer channel contention problem, effectively improving channel utilization and preventing equalization stagnation caused by conflicts.

[0030] In the several embodiments provided by this invention, it should be understood that the disclosed system can be implemented in other ways. For example, the embodiments of the invention described above are merely illustrative; for example, the division of modules is only a logical functional division, and there may be other division methods in actual implementation.

[0031] The modules described as separate components may or may not be physically separate. The components shown as modules may or may not be physical modules; they may be located in one place or distributed across multiple network modules. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs.

[0032] Furthermore, the functional modules in the various embodiments of the present invention can be integrated into one processing module, or each module can exist physically separately, or two or more modules can be integrated into one module. The integrated module can be implemented in hardware or in the form of hardware plus software functional modules.

[0033] It will be apparent to those skilled in the art that the present invention is not limited to the details of the exemplary embodiments described above, and that the present invention can be implemented in other specific forms without departing from the essential characteristics of the present invention.

[0034] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention.

Claims

1. A battery intelligent management system (BMS), characterized in that, include: The battery multi-source data acquisition and processing module collects and preprocesses the battery pack's temperature distribution data, charge / discharge rate sequence, and cycle count. It then performs data analysis on the preprocessed multi-source data through a pre-built battery state coupling model, outputting the state of charge, state of health, and dynamic equilibrium threshold. The battery energy monitoring and intelligent control module, based on the output state vector, designs a magnetically coupled resonant dynamic energy routing network that includes a distributed multi-coil array and an impedance self-matching unit. It triggers energy transfer by calculating the SOC gradient. When the voltage difference of a single cell exceeds the dynamic equilibrium threshold, it activates a non-contact energy transmission channel and corrects the channel efficiency in real time using temperature distribution data. The optimized energy allocation module introduces federated reinforcement learning to optimize the balancing priority based on the energy routing network status. Each battery cell is treated as an agent, and the optimal energy allocation strategy is solved through multi-agent collaboration. The strategy update cycle is adaptively adjusted by the number of iterations.

2. The BMS (Battery Management System) intelligent management system according to claim 1, characterized in that, The expression involved in calculating the state of charge (SOC) is as follows: ;in, This is the initial SOC; This represents the charging and discharging current; a positive value indicates discharging, and a negative value indicates charging. Coulomb efficiency; Rated capacity; t represents the correction term output by the LSTM network; t is the time variable.

3. The BMS (Battery Management System) intelligent management system according to claim 2, characterized in that, The expression involved in calculating the State of Health (SOH) is as follows: ;in, This represents the current actual capacity. , The health factor is the output of the LSTM network, with a value ranging from 0 to 1.

4. The BMS (Battery Management System) intelligent management system according to claim 3, characterized in that, The expression involved in calculating the dynamic equilibrium threshold is as follows: ;in, The dynamic equilibrium threshold; This represents the minimum SOH value in the battery pack. is the average temperature; e is a mathematical constant.

5. A BMS (Battery Management System) intelligent management system according to claim 4, characterized in that, Based on the output State of Charge (SOC) value, the SOC gradient between the i-th cell and its neighboring cells is calculated, using the following expression: ;in, Let SOC gradient be the SOC gradient between the i-th battery cell and its neighboring cells; The state of charge of the i-th battery cell; The state of charge of the (i-1)th adjacent battery cell; The state of charge of the (i+1)th adjacent battery cell; Energy transfer is initiated when any energy transfer condition is met.

6. The BMS (Battery Management System) intelligent management system according to claim 5, characterized in that, When performing dynamic switching of non-contact energy transfer channels, the SOC gradient and voltage difference of all coils are scanned to generate candidate transfer pairs; the comprehensive score of each candidate pair is calculated, and the channel with the highest score is selected for activation.

7. A BMS (Battery Management System) intelligent management system according to claim 6, characterized in that, When performing real-time adjustment of the resonant frequency, the target resonant frequency is obtained. And calculate the actual frequency. Relationship with temperature; when At that time, the impedance self-matching unit corrects the capacitance by switching the capacitor array. This makes the actual frequency Return to target resonant frequency .

8. A BMS (Battery Management System) intelligent management system according to claim 7, characterized in that, A two-tier architecture of local intelligent agent and central server is constructed, in which each individual battery unit acts as an independent intelligent agent, equipped with a lightweight training module, responsible for local state perception, policy iteration, and encrypted data upload.

9. A BMS (Battery Management System) intelligent management system according to claim 8, characterized in that, Multi-agent collaborative strategies include local training and global aggregation; During local training, each agent uses a proximate policy optimization algorithm to iterate its local policy. During global aggregation, each agent encrypts and uploads its local policy network parameters to the central server. The central server calculates the aggregate weights based on the SOH values ​​and balanced contributions of each agent, and generates global policy parameters. The central server distributes global policy parameters to each agent, which then initializes its local policy network using the global parameters and begins the next round of local training.

10. A BMS (Battery Management System) intelligent management system according to claim 9, characterized in that, When adaptively adjusting the policy update cycle, the update cycle is determined by the number of iterations. Dynamic adjustment involves the following expressions: ;in, For update cycle; This is the error factor, with a value of 100; When multiple agents simultaneously request to send or receive energy, the central server prioritizes them based on the SOH value and SOC gradient.

Images