Battery management system for maximizing battery life and method thereof

Abstract

The present invention relates to a battery management system for maximizing the service life of batteries in a BESS connected between a power supply device and a load, the battery management system comprising: a data collection unit for collecting measurement data on at least one of voltage, current, and temperature for a plurality of cells connected in parallel to a power line in a battery pack at each time interval; an estimation unit for estimating SOC and SOH values of each cell by using the measurement data to acquire an SOC data set and an SOH data set; and a control unit for controlling, on the basis of a state vector including the SOC data set and the SOH data set, the cell switching of the battery pack in real time by scheduling, for each time interval, the on / off operation of each cell switch installed between each cell and the power line by means of a deep learning-based scheduling model. According to the present invention, the on / off operation of the cell switch connected to each cell is scheduled on the basis of the estimated SOC and SOH values for each cell in the battery pack in a BESS so that the performance of the BESS is optimized and the battery life is extended.

Description

Battery management system and method for maximizing battery life

[0001] The present invention relates to a battery management system and method for maximizing battery life, and more particularly, to a battery management system and method for maximizing the battery life of a battery energy storage system (BESS) utilizing waste lithium ion batteries.

[0002] Lithium-ion batteries have become an essential component of modern life, powering everything from smartphones to electric vehicles (EVs). Lithium-ion batteries are widely used compared to other battery types due to their minimal memory effect, extended lifespan, and low self-discharge rate. However, lithium-ion batteries for electric vehicles have a limited lifespan for safety reasons, requiring replacement when their lifespan drops below 80%.

[0003] Research predicts that the total power output of end-of-life battery packs will reach 120 GWh globally by 2030. Furthermore, the continued growth in demand for lithium-ion batteries is expected to generate significant amounts of waste and a significant financial burden. Therefore, solutions for reusing discarded lithium-ion batteries with less than 80% of their lifespan are becoming increasingly urgent.

[0004] Promising applications for discarded lithium-ion batteries include battery energy storage systems (BESS) that can be used as backup power in homes, EV charging stations, or telecommunications and data center systems.

[0005] BESS has the potential to significantly reduce the demand for new batteries and help reduce the environmental impact of battery production.

[0006] A BESS consists of a battery pack with multiple battery cells connected in parallel or series to increase the pack's capacity or voltage. Switches are added to the battery cells, allowing each cell to be electrically connected or disconnected from the pack.

[0007] However, each cell within a battery pack has different capacity levels (e.g., lifespan), which reduces the efficiency of the battery pack and ultimately significantly affects the performance of the battery pack.

[0008] Therefore, a scheduling policy for cell switching is required to extend the life of the battery pack in the BESS and reduce the unbalanced capacity of the battery cells.

[0009] For optimal scheduling policy in BESS, accurate identification of battery characteristics, including State of Charge (SOC) and State of Health (SOH), is important.

[0010] Figure 1 is a diagram illustrating the relationship between a battery's state of charge (SOC) and its lifespan (SOH). A battery's SOC represents the charge level (state of charge) relative to its capacity, while SOH represents the ratio of maximum battery charge to rated capacity, indicating its lifespan.

[0011] In scheduling policies, information on SOC and SOH protects battery cells by preventing overcharging or discharging, thereby increasing BESS capacity. While SOC and SOH parameters are not directly measured for each cell, they can be estimated using measurable parameters such as voltage, current, and temperature for each cell.

[0012] By efficiently scheduling the on / off switching of the cell switches connected to each cell in the battery pack based on the state (SOC, SOH) of the battery cells in the battery pack so that the states of all battery cells are balanced, the life of the battery pack can be extended.

[0013] Therefore, to optimize the performance of BESS, accurate estimation of SOC and SOH for each cell is required along with efficient scheduling of switching operations of cell switching.

[0014] The technology underlying the present invention is disclosed in Korean Patent Publication No. 10-2023-0034546 (published on March 10, 2023).

[0015] The purpose of the present invention is to provide a battery management system for maximizing battery life, which can extend the life of a battery pack by efficiently scheduling cell switching operations for each cell using a deep learning-based scheduling model based on SOC and SOH estimation values ​​for each cell in a battery pack.

[0016] The present invention relates to a battery management system for maximizing the battery life of a BESS connected between a power supply unit and a load, comprising: a data collection unit that collects measurement data on at least one of voltage, current, and temperature for a plurality of cells connected in parallel to a power line within a battery pack every hour; an estimation unit that estimates SOC and SOH values ​​of each cell using the measurement data to obtain an SOC dataset and an SOH dataset; and a control unit that controls cell switching of the battery pack in real time by scheduling on / off operations of each of a plurality of cell switches installed between each cell and a power line every hour through a deep learning-based scheduling model based on a state vector including the SOC dataset and the SOH dataset.

[0017] Additionally, the estimation unit can estimate SOC and SOH values ​​from the measurement data of each cell using an Extended Kalman Filter (EKF).

[0018] Additionally, the state vector may further include the current value of the battery pack and the power demand of the load.

[0019] Additionally, the above scheduling model can be pre-learned based on the DQN (Deep Q-Network) algorithm.

[0020] In addition, the control unit may set the state vector observed at the current time as a state value, set the on / off operation information of each cell switch, which is an operation selected in the state, as an action value, and perform a reward according to the operation selected in the state, thereby pre-learning the scheduling model based on the DQN algorithm.

[0021] In addition, the control unit, when it is determined that the battery pack needs to be charged based on the current value of the battery pack calculated based on the measurement data of each cell and the power demand of the load, connects the battery pack to the power supply device and executes a charging process, and when it is determined that the battery pack needs to be discharged, connects the battery pack to the load and executes a discharging process, and can perform the charging or discharging process while controlling cell switching for each cell switch according to the scheduling information.

[0022] Additionally, the control unit can execute a charging or discharging process for the battery pack in the next cycle based on whether the battery pack needs to be charged or discharged, which is determined based on the current value of the battery pack measured in the current cycle and the power demand of the load.

[0023] And, the present invention provides a battery management method performed by a battery management system of a BESS connected between a power supply unit and a load, comprising the steps of: collecting measurement data on at least one of voltage, current, and temperature for a plurality of cells connected in parallel to a power line within a battery pack every hour; estimating SOC and SOH values ​​of each cell using the measurement data to obtain an SOC dataset and an SOH dataset, and executing a charging or discharging process by connecting the battery pack to the power supply unit or the load based on a current value of the battery pack and a power demand of the load; and controlling cell switching of the battery pack in real time by scheduling on / off operations of each of a plurality of cell switches installed between each cell and a power line every hour through a deep learning-based scheduling model based on a state vector including the SOC dataset, the SOH dataset, the current value of the battery pack, and the power demand.

[0024] According to the present invention, the performance of the BESS can be optimized and the life of the battery can be extended by efficiently scheduling the on / off operations of cell switches connected to each cell using a deep learning-based scheduling model based on the SOC and SOH estimation values ​​for each cell of a battery pack in the BESS.

[0025] Figure 1 is a diagram showing the relationship between the state of charge and life of a battery.

[0026] FIG. 2 is a diagram showing the configuration of a battery management system according to an embodiment of the present invention.

[0027] FIG. 3 is a drawing illustrating how a BESS including the battery management system of FIG. 2 operates between a power supply unit and a battery.

[0028] FIG. 4 is a diagram illustrating a training process of a scheduling algorithm implemented with a DQN algorithm in an embodiment of the present invention.

[0029] FIG. 5 is a drawing illustrating a battery management method according to an embodiment of the present invention.

[0030] Hereinafter, embodiments of the present invention will be described in detail with reference to the attached drawings so that those skilled in the art can easily practice the present invention. However, the present invention may be implemented in various different forms and is not limited to the embodiments described herein. In addition, in the drawings, parts irrelevant to the description have been omitted to clearly explain the present invention, and similar parts have been designated with similar reference numerals throughout the specification.

[0031] Throughout the specification, when a part is said to be "connected" to another part, this includes not only the cases where the parts are "directly connected" but also the cases where the parts are "electrically connected" with other elements intervening. Furthermore, when a part is said to "include" a component, this does not exclude other components, but rather includes other components, unless otherwise stated.

[0032] The present invention relates to a battery management system for maximizing battery life, wherein the state of charge and life of all battery cells (hereinafter, “cells”) in a battery pack included in a battery energy storage system (BESS) are estimated and applied to a scheduling algorithm based on deep reinforcement learning (DRL) to schedule efficient cell switching for maximizing battery life, thereby optimizing the performance of the BESS and balancing the states of all cells in the battery pack, thereby extending the battery life.

[0033] FIG. 2 is a drawing showing the configuration of a battery management system according to an embodiment of the present invention, and FIG. 3 is a drawing explaining how a BESS including the battery management system of FIG. 2 operates between a power supply unit and a battery.

[0034] Referring to FIGS. 2 and 3, a battery management system (110) is included in a BESS (100) and monitors the status of a battery pack (120) to control the charging and discharging operation of the battery pack (120). The BESS (100) is connected between a power supply unit (200) and a load (300) and is configured to include a battery management system (110) and a battery pack (120).

[0035] The battery pack (120) has a structure in which a plurality of cells (121) (Cells; Cell 1 to N) are connected in parallel to a power line. Here, the plurality of cells (121) may correspond to waste lithium ion battery cells whose lifespan has expired by 80% or less.

[0036] As shown in Fig. 3, a cell switch (122) is individually installed between each cell (121) and the power line to perform cell switching. When the cell switch (122) is turned on, the corresponding cell (121) is electrically connected to the power line, and when turned off, the connection is disconnected. The cell switch (122) performs cell switching according to a control signal (scheduling signal) of the battery management system (110).

[0037] The state of charge (SOC) or state of health (SOH) of each cell (121) included in the battery pack (120) may be different, and the battery management system (110) can control the cell switching operation for each cell (121) by efficiently planning the on / off timing of the cell switch (122) connected to each cell (121) to balance the states of each cell and maximize the battery life through this.

[0038] The battery management system (110) measures the voltage, current, and temperature of each cell (121) included in the battery pack (120). ) is used to estimate the SOC and SOH values, and by generating scheduling information (scheduling signal) that plans the on / off operation of each cell switch (122) by time based on the estimated SOC and SOH values, the status of each cell can be uniformized and the performance of the BESS can be optimized.

[0039] Below, the configuration of the battery management system (110) is described in more detail.

[0040] As illustrated in FIG. 2, the battery management system (110) includes a data collection unit (111), an estimation unit (112), and a control unit (113). Here, the operation of each unit (110, 120) and the data flow between each unit can be controlled by the control unit (113).

[0041] The data collection unit (111) collects measurement data on at least one of voltage, current, and temperature for a plurality of cells (121) connected in parallel to a power line within a battery pack (120) every hour (time slot). The data collection unit (111) can transmit the collected data to the estimation unit (112).

[0042] The estimation unit (112) can estimate the SOC value and SOH value for each cell using the measurement data, thereby generating an SOC data set corresponding to the SOC value of the entire cell and an SOH data set corresponding to the SOH value of the entire cell. The estimation unit (112) can first estimate the SOC value of the corresponding cell using the measurement data of the cell, and then estimate the SOH value based on the estimated SOC value.

[0043] The estimation unit (112) can estimate the SOC value and SOH value from the measurement data (e.g., voltage, current, temperature) of each cell using the Extended Kalman Filter (EKF). At this time, the estimation unit (112) can first estimate the SOC value using the Extended Kalman Filter based on the measurement information of the battery cell (121), and then estimate the SOH based on the SOC, and adjust the parameters of the Extended Kalman Filter based on the estimated SOH. The SOC and SOH value estimation algorithm used in the embodiment of the present invention is not necessarily limited to the above-described method, and more diverse estimation techniques can be applied.

[0044] In the following embodiment of the present invention, the SOC and SOH estimation algorithm is named Algorithm 1, and the battery pack charging / discharging process by the control unit (113) is named Algorithm 2. The estimation unit (112) estimates the SOC and SOH through the pre-stored Algorithm 1, and the control unit (113) can perform the charging process or the discharging process according to the current current value (load current) of the battery pack and the load demand through the pre-stored Algorithm 2.

[0045] The control unit (113) controls the current value of the battery pack (120) ) and the power demand of the load (300) ( ) can be connected to a battery pack (120) and a power supply (200) or a load (300) to execute a charging process or a discharging process.

[0046] Here, the current value of the battery pack (120) can be obtained using measurement data collected from each cell (121), and the power demand of the load (300) can be obtained based on the value transmitted from the load (300) or a preset value.

[0047] In the following embodiment of the present invention, the deep reinforcement learning-based scheduling algorithm is named Algorithm 3. The control unit (113) can optimally schedule the on / off operation of each cell switch based on the SOC and SOH values ​​for each cell through the previously stored Algorithm 3.

[0048] Specifically, the control unit (113) is a SOC dataset ( ) and SOH dataset ( ) is included, the on / off operation of each of the multiple cell switches (122) can be scheduled for each time slot through a deep learning-based scheduling model to control the cell switching operation of the battery pack in real time.

[0049] Here, the state vector ( ) is the SOC dataset as shown in the mathematical expression 1 below. , SOH dataset In addition, the current value of the battery pack (120) and power demand It can be composed of four factors including:

[0050]

[0051] Here, is the SOC dataset, is the SOH dataset, The current of the battery pack (120), represents the power demand.

[0052] In an embodiment of the present invention, the scheduling model may be implemented using a scheduling algorithm based on Deep Reinforcement Learning (DRL). Here, the embodiment of the present invention may use a scheduling algorithm based on a Deep Q-Network (DQN) for ON / OFF cell switching of a battery pack (120).

[0053] FIG. 4 is a diagram illustrating a training process of a scheduling algorithm implemented using the DQN algorithm in an embodiment of the present invention. As shown in FIG. 4, a scheduling model according to an embodiment of the present invention can be pre-trained by the control unit (113) based on the DQN (Deep Q-Network) algorithm.

[0054] The scheduling algorithm represents the current state of the battery pack (120) using a state vector , an action vector expressing the on or off operation state of multiple cell switches (122) , and a reward function that expresses the reward given according to the action (movement) selected in that state. It has three elements including:

[0055] Based on this, the control unit (113) sets the state vector observed at the current time as a state value (state), sets the on / off operation information of each cell switch, which is the operation selected in the state, as an action value (action), and immediately performs a reward according to the action selected in the state, thereby enabling pre-learning of a scheduling model based on the DQN algorithm.

[0056] The DQN scheduling model interacts with the environment (battery pack) to determine the state of the battery pack (120). Recognize and compensate while minimizing the decrease in the SOH value of the battery pack (120) Actions to maximize You can choose.

[0057] Additionally, the DQN-based scheduling model is state Best action for Knowledge acquired using deep reinforcement learning to schedule can utilize and update this knowledge. This knowledge may include scheduling policies for cell switches for a given battery state and corresponding compensation scheduling. The scheduling algorithm can update this acquired knowledge to explore actions based on past experience.

[0058] At this time, the DQN-based scheduling algorithm uses the state to utilize past experience. Ga knowledge Find out if it's inside, go If it exists, it acts based on the epsilon-greedy policy. can be selected. The epsilon-greedy policy is a widely used method for determining actions in reinforcement learning techniques.

[0059] The DQN-based scheduling algorithm can update knowledge K while minimizing the loss function using gradient descent. Loss function can be defined as in mathematical expression 2.

[0060]

[0061] Here, is a parameter of the DQN network (weight of the main network) and can be defined as in mathematical expression 3.

[0062]

[0063] Here, is the learning factor, is the main network ( ) is the selected action value, is a target network representing the maximum cumulative compensation to minimize the decrease in the SOH value of the battery pack (120). ) represents the target action value.

[0064] The control unit (113) The value A DQN-based scheduling model can be trained by minimizing the loss function within a set error to track the value. Minimizing the loss function means that the SOH of the battery pack (120) is optimized.

[0065] Meanwhile, if the control unit (113) determines that the battery pack (120) needs to be charged based on the current value of the battery pack calculated based on the measurement data of each cell (121) and the power demand of the load, the control unit (113) connects the battery pack (120) to the power supply unit (200) and executes a charging process, and if the control unit (113) determines that the battery pack (120) needs to be discharged, the control unit (113) connects the battery pack to the load (300) and executes a discharging process.

[0066] In addition, the control unit (113) performs a charging or discharging process while controlling cell switching for each cell switch (122) according to scheduling information derived from a DQN-based scheduling model.

[0067] For example, the control unit (113) controls the current value of the battery pack (120) ) is positive or negative, it determines whether the process in progress in the current cycle is in a discharging or charging state. The battery current is positive when discharging and negative when charging. In the following embodiments of the present invention, j represents a cycle index, k represents a time slot index, and one cycle consists of multiple time slots.

[0068] If the discharge process is in progress in the current cycle, the total power discharged ( ) and calculate the power demand of the load (300) ) is reached, the charging process can be switched to the next cycle. If the total discharged power has not reached the power demand, the discharging process is maintained. Conversely, if the charging process is in progress, the total charged power ( ) and calculate the corresponding value to obtain the load demand. Once reached, the discharge process can be switched to the next cycle.

[0069] In this way, the control unit (113) can execute a charging or discharging process for the battery pack in the next cycle based on whether the battery pack needs to be charged or discharged, which is determined based on the current value of the battery pack measured in the current cycle and the power demand of the load. At this time, the control unit (113) can perform the charging or discharging process while controlling the on / off state of each cell switch (122) for each time slot according to scheduling information derived from the DQN-based scheduling model.

[0070] FIG. 5 is a drawing illustrating a battery management method according to an embodiment of the present invention.

[0071] First, the battery management system (110) collects measurement data for at least one of voltage, current, and temperature for multiple cells (121) in the battery pack (120) of the BESS (100) every hour (S510).

[0072] At this time, time means time slot, and in step S510, the current kth time slot t k In the embodiment of the present invention, the operating time (W) is composed of multiple time slots. is expressed as . Here, since k=1,2,…,w, the last w means the total number of time slots that make up the operation time.

[0073] Next, the battery management system (110) estimates the SOC value and SOH value for each cell through the measurement data of each cell (121) based on the extended Kalman filter (algorithm 1) to create an SOC dataset. and SOH dataset Obtain (S520).

[0074] In addition, the battery management system (110) monitors the current value of the battery pack. Overload power demand A charging process or a discharging process is executed for the battery pack based on (S525). These steps S520 and S525 may be performed simultaneously.

[0075] Next, the battery management system (110) currently has a time slot t k Obtain the state vector in (S530). The state vector is as shown in mathematical expression 1. can be defined as

[0076] In this way, the battery management system (110) collects measurement data including voltage, current, and temperature of each cell in the battery pack (120), and then estimates the SOC and SOH of each cell using Algorithm 1, and can control the charging or discharging process of the battery pack (120) according to the load demand using Algorithm 2. These Algorithms 1 and 2 return the constituent elements of the state vector to Algorithm 3, thereby enabling deep learning-based battery scheduling.

[0077] Thereafter, the battery management system (110) schedules cell switching through a deep learning-based scheduling model based on the acquired state vector to control the operation of each cell switch (S540).

[0078] In this S540 step, the battery management system (110) can control cell switching of the battery pack by scheduling the on / off operations of each of the multiple cell switches (122) installed between each cell (121) and the power line based on the state vector obtained from the current time slot. At this time, the scheduling model can be learned in advance by a DQN-based algorithm.

[0079] Additionally, the above-described process can be repeatedly performed for each time slot, and can be performed until the time slot index value k satisfies w.

[0080] According to the present invention as described above, the performance of the BESS can be optimized and the life of the battery can be extended by efficiently scheduling the on / off operation of the cell switch connected to each cell using a deep learning-based scheduling model based on the SOC and SOH estimation values ​​for each cell of the battery pack in the BESS.

[0081] While the present invention has been described with reference to the embodiments illustrated in the drawings, these are merely exemplary, and those skilled in the art will understand that various modifications and equivalent alternative embodiments are possible. Therefore, the true scope of technical protection of the present invention should be determined by the technical spirit of the appended claims.

Claims

1. In a battery management system for maximizing the battery life of a BESS connected between a power supply unit and a load, A data collection unit that collects measurement data on at least one of voltage, current, and temperature for a plurality of cells connected in parallel to a power line within a battery pack every hour; An estimation unit that estimates the SOC and SOH values ​​of each cell using the above measurement data to obtain a SOC dataset and a SOH dataset; and A battery management system including a control unit that controls cell switching of the battery pack in real time by scheduling the on / off operation of each of a plurality of cell switches installed between each cell and a power line for each hour through a deep learning-based scheduling model based on a state vector including the SOC dataset and the SOH dataset.

2. In claim 1, The above estimation part is, A battery management system that estimates SOC and SOH values ​​from measurement data of each cell using an Extended Kalman Filter (EKF).

3. In claim 1, A battery management system wherein the state vector further includes a current value of the battery pack and a power demand of a load.

4. In claim 1, The above scheduling model is a battery management system that has been trained based on the DQN (Deep Q-Network) algorithm.

5. In claim 4, The above control unit, A battery management system that sets the state vector observed at the current time as a state value, sets the on / off operation information of each cell switch, which is an operation selected from the state, as an action value, and performs compensation according to the action selected from the state while pre-training a scheduling model based on the DQN algorithm.

6. In claim 1, The above control unit, If it is determined that the battery pack needs to be charged based on the current value of the battery pack calculated based on the measurement data of each cell and the power demand of the load, the battery pack is connected to the power supply to execute a charging process, and if it is determined that the battery pack needs to be discharged, the battery pack is connected to the load to execute a discharging process. A battery management system that performs the charging or discharging process while controlling cell switching for each cell switch according to the above scheduling information.

7. In claim 1, The above control unit, A battery management system that executes a charging or discharging process for the battery pack in the next cycle based on whether the battery pack needs to be charged or discharged, which is determined based on the current value of the battery pack measured in the current cycle and the power demand of the load.

8. A battery management method performed by a battery management system of a BESS connected between a power supply device and a load, A step of collecting measurement data for at least one of voltage, current and temperature for a plurality of cells connected in parallel to a power line within a battery pack every hour; A step of estimating the SOC and SOH values ​​of each cell using the above measurement data to obtain an SOC dataset and an SOH dataset, and executing a charging or discharging process by connecting the battery pack to the power supply or load based on the current value of the battery pack and the power demand of the load; and A battery management method comprising a step of controlling cell switching of the battery pack in real time by scheduling the on / off operation of each of a plurality of cell switches installed between each cell and a power line for each hour through a deep learning-based scheduling model based on the state vector including the SOC data set, the SOH data set, the current value of the battery pack, and the power demand.

Images