Battery management control method based on joint state hash rate estimation
Patent Information
- Application Number
- CN202611081331.1
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2026-07-21
- Publication Date
- 2026-10-16
AI Technical Summary
[0002]现有电池管理系统中，均衡控制任务与热管理控制任务的执行周期通常采用固定参数配置，中央处理器调度策略未考虑电池实时状态变化对算力需求的动态影响
本发明将电压信号提取的电压波动特征量、温度信号提取的温度峰值特征量与中央处理器占用率一并输入联合估计模型，输出均衡任务和热管理任务在未来控制周期的执行时间序列。根据执行时间序列与可用计算资源的比较结果生成任务周期缩放因子，在负载峰值到来前提前放大任务周期以平抑算力冲击，在负载回落后恢复短周期以维持控制及时性。热管理任务被设定为数值较小的抢占优先级，均衡任务被设定为数值较大的抢占优先级，热管理任务在算力竞争中对均衡任务形成抢占关系。联合估计模型以电压、温度、占用率三类信号的历史序列为输入，捕捉电压离散化趋势、温升速率与算力消耗之间的关联模式，使调度参数调整先于算力超限发生。
Smart Images
Figure CN122931161A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of battery management technology, and more specifically to a battery management control method based on joint state computational power estimation. Background Technology
[0002] In existing battery management systems, the execution cycles of equalization control and thermal management control tasks are typically configured with fixed parameters, and the CPU scheduling strategy does not consider the dynamic impact of real-time battery status changes on computing power requirements. When cell voltage dispersion increases or temperature rises abnormally, the computational load of the control tasks increases accordingly. Fixed-cycle scheduling causes CPU utilization to exceed limits during peak load periods, resulting in dropped frames or delayed responses. During off-peak load periods, fixed high-frequency scheduling causes unnecessary CPU resource consumption. Some solutions reduce task frequency based solely on CPU utilization, without considering voltage fluctuations and temperature conditions to jointly determine future computing power demand trends. The trade-off between the frequency reduction magnitude and control response timeliness lacks quantitative basis. Summary of the Invention
[0003] The purpose of this invention is to provide a battery management control method based on joint state computing power estimation to solve the above-mentioned technical problems.
[0004] The objective of this invention can be achieved through the following technical solutions: The battery management control method based on joint state computational power estimation includes the following steps: Collects battery voltage and temperature signals, and manages the utilization rate of the controller's central processing unit; Feature extraction is performed on voltage and temperature signals to obtain voltage fluctuation features and temperature peak features; Input voltage fluctuation characteristics, temperature peak characteristics, and occupancy rate into the joint estimation model, and output the task execution time series; Based on the task execution time sequence, the current task cycle, and available computing resources, a task cycle scaling factor is generated, and a preemption priority is set. Battery balancing and thermal management tasks are executed based on task cycle scaling factors and preemption priorities.
[0005] As a further aspect of the present invention, the specific process for acquiring voltage signals, temperature signals, and occupancy rates is as follows: The voltage at the end of each cell and the surface temperature of each cell are synchronously collected by a distributed sensor array in the battery pack. After passing through an anti-aliasing low-pass filter, the data are stored as voltage time series and temperature time series at a fixed sampling frequency. The idle task cycle count is read from the real-time operating system kernel interface of the management controller. The difference between the total cycle count and the idle task cycle count is divided by the total cycle count to obtain the utilization rate of the management controller's central processing unit.
[0006] As a further aspect of the present invention, the specific process for obtaining voltage fluctuation characteristic quantities and temperature peak characteristic quantities is as follows: Within a sliding time window of a preset length, the maximum and minimum values of the voltage time series are taken, and the difference between the maximum and minimum values is used as the voltage fluctuation characteristic. Within the same sliding time window, the maximum value of the temperature time series is taken as the temperature peak feature.
[0007] As a further aspect of the present invention, the specific process of outputting the task execution time sequence is as follows: The joint estimation model is a recurrent neural network containing gated recurrent units. The input vector is composed of the voltage fluctuation features, temperature peak features, and occupancy rate at the current time, as well as the voltage fluctuation features, temperature peak features, and occupancy rate at the previous K sampling times. The output layer of the recurrent neural network outputs a vector of length 2Q. The first Q components correspond to the predicted execution time of the equilibrium task in the next Q control cycles, and the last Q components correspond to the predicted execution time of the thermal management task in the next Q control cycles, thus forming a task execution time sequence.
[0008] As a further aspect of the present invention: the specific process for generating the task cycle scaling factor is as follows: Add the predicted execution time of the equilibrium task to the predicted execution time of the thermal management task for each future control cycle, divide the sum by the fixed duration of a single control cycle to obtain the total occupancy rate for each cycle, and extract the maximum value of the total occupancy rate for all future control cycles as the predicted peak total occupancy rate. The maximum utilization rate of the central processing unit minus the security reserved utilization rate is used as the upper limit of available computing resources; When the predicted peak total utilization rate exceeds the upper limit of available computing resources, the task cycle scaling factor is the predicted peak total utilization rate divided by the upper limit of available computing resources. When the predicted peak total utilization rate does not exceed the upper limit of available computing resources, the task cycle scaling factor is 1.
[0009] As a further aspect of the present invention: the process of setting the preemption priority is as follows: Set the preemption priority of the balancing task to the second priority level, and set the preemption priority of the thermal management task to the first priority level, with the first priority level being lower than the second priority level.
[0010] As a further aspect of the present invention, the process of performing battery balancing and thermal management tasks is as follows: Locate the Balanced Task Control Block and the Thermal Management Task Control Block in the Real-Time Operating System Task Control Block List. Multiply the original values of the Balanced Task Control Block's periodic field and the Thermal Management Task Control Block's periodic field by the task period scaling factor to obtain the new periodic values for the Balanced Task and the Thermal Management Task, respectively.
[0011] As a further aspect of the present invention, the process of performing battery balancing and thermal management tasks further includes: The software timer for the balancing task is configured based on the global system tick of the operating system, with the new cycle value of the balancing task as the timing period. After the timer expires, the balancing task is added to the ready queue in the kernel critical section. The thermal management task software timer is configured with a new cycle value as the timing period and is based on the global system tick of the operating system. After the timer expires, the thermal management task is added to the ready queue in the kernel critical section.
[0012] As a further aspect of the present invention, the process of performing battery balancing and thermal management tasks further includes: Both balancing tasks and thermal management tasks have pre-set basic priorities. When a balancing task or thermal management task is added to the ready queue, its position in the ready queue is determined according to its basic priority. Starting with the first task in the ready queue, tasks in the ready queue enter the running state in sequence. When another task is added to the ready queue and its base priority is lower than the preemption priority of the currently running task, the CPU execution right is allocated to the newly added task.
[0013] As a further aspect of the present invention, the process of performing battery balancing and thermal management tasks further includes: When the balancing task is in operation, a balancing current command is sent to the battery balancing circuit. When the thermal management task is in operation, a coolant flow command is sent to the cooling system.
[0014] The beneficial effects of this invention compared to the prior art are as follows: This invention inputs voltage fluctuation features extracted from voltage signals, temperature peak features extracted from temperature signals, and CPU utilization rate into a joint estimation model, outputting the execution time series of balancing tasks and thermal management tasks for future control cycles. Based on the comparison between the execution time series and available computing resources, a task cycle scaling factor is generated. The task cycle is extended before load peaks arrive to mitigate computing power surges, and shortened again after load declines to maintain control timeliness. Thermal management tasks are assigned a lower preemption priority, while balancing tasks are assigned a higher preemption priority, creating a preemptive relationship between thermal management tasks and balancing tasks in computing power competition. The joint estimation model uses historical sequences of voltage, temperature, and utilization rate signals as input to capture the correlation patterns between voltage discretization trends, temperature rise rates, and computing power consumption, ensuring that scheduling parameter adjustments precede computing power overruns. Attached Figure Description
[0015] The invention will now be further described with reference to the accompanying drawings.
[0016] Figure 1 This is a flowchart illustrating the battery management control method based on joint state computational power estimation according to the present invention. Detailed Implementation
[0017] 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.
[0018] Please see Figure 1 As shown, this invention is a battery management control method based on joint state computational power estimation, comprising the following steps: Collects battery voltage and temperature signals, and manages the utilization rate of the controller's central processing unit; In a preferred embodiment of the present invention, the specific process of acquiring voltage signals, temperature signals, and occupancy rates is as follows: The voltage at the end of each cell and the surface temperature of each cell are synchronously collected by a distributed sensor array in the battery pack. After passing through an anti-aliasing low-pass filter, the data are stored as voltage time series and temperature time series at a fixed sampling frequency. The idle task cycle count is read from the real-time operating system kernel interface of the management controller. The difference between the total cycle count and the idle task cycle count is divided by the total cycle count to obtain the utilization rate of the management controller's central processing unit.
[0019] Feature extraction is performed on voltage and temperature signals to obtain voltage fluctuation features and temperature peak features; In another preferred embodiment of the present invention, the specific process for obtaining voltage fluctuation characteristic and temperature peak characteristic is as follows: Within a sliding time window of a preset length, the maximum and minimum values of the voltage time series are taken, and the difference between the maximum and minimum values is used as the voltage fluctuation characteristic. Within the same sliding time window, the maximum value of the temperature time series is taken as the temperature peak feature.
[0020] Input voltage fluctuation characteristics, temperature peak characteristics, and occupancy rate into the joint estimation model, and output the task execution time series; In a preferred embodiment of the present invention, the specific process of outputting the task execution time sequence is as follows: A recurrent neural network (RNN) containing gated recurrent unit (ROU) layers is pre-constructed as the joint estimation model. The RNN's structure consists of an input layer, a first gated RNU layer, a second gated RNU layer, and a fully connected output layer. The number of hidden units in the first gated RNU layer is set to a first preset value, and the number of hidden units in the second gated RNU layer is set to a second preset value. These first and second preset values are pre-set by the designer based on the available storage space and computing power constraints of the management controller. The fully connected output layer contains twice the number of prediction steps, which is a preset integer value.
[0021] Voltage fluctuation characteristics, temperature peak characteristics, and occupancy rates generated by the management controller during historical operation are used as training samples. For each sampling, the actual execution time series of the balancing task and the thermal management task for multiple consecutive control cycles after that sampling are recorded as training labels. Mean squared error is used as the loss function, and the weight and bias parameters of each layer in the recurrent neural network are updated using a stochastic gradient descent algorithm, gradually reducing the mean squared error between the predicted value output by the recurrent neural network and the training labels. After training, all weight and bias parameters of the recurrent neural network are permanently stored in the non-volatile memory of the management controller.
[0022] During the operation of the management controller, the voltage fluctuation characteristics, temperature peak characteristics, and occupancy rate at the current moment are acquired at a fixed sampling period. The corresponding voltage fluctuation characteristics, temperature peak characteristics, and occupancy rates for the previous K sampling moments are read from the circular buffer, where K is a preset integer value. The three types of values at the current moment are concatenated with the three types of values from the previous K moments in chronological order to form a one-dimensional input vector. This one-dimensional input vector is then fed into the input layer of a recurrent neural network with pre-loaded fixed parameters.
[0023] The input vector first enters the first gated recurrent unit (ROU) layer. The first GU layer encodes the temporal dependencies in the input vector and outputs the first hidden state sequence. The first hidden state sequence then enters the second GU layer, which further encodes the temporal dependencies in the first hidden state sequence and outputs the second hidden state sequence. The hidden state vector corresponding to the last time step in the second hidden state sequence is then taken and fed into the fully connected output layer.
[0024] The fully connected output layer performs a linear transformation on the hidden state vector, outputting an output vector with a length equal to twice the number of prediction steps. The first few predicted steps of the output vector are sequentially mapped to the predicted execution times of the balanced task from the first control cycle to the [number]th predicted step control cycle. The last few predicted steps of the output vector are sequentially mapped to the predicted execution times of the thermal management task from the first control cycle to the [number]th predicted step control cycle. The first few predicted steps are arranged in chronological order according to the control cycle to form the balanced task execution time series, and the last few predicted steps are arranged in chronological order according to the control cycle to form the thermal management task execution time series. The balanced task execution time series and the thermal management task execution time series together constitute the task execution time series.
[0025] Based on the task execution time sequence, the current task cycle, and available computing resources, a task cycle scaling factor is generated, and a preemption priority is set. In a preferred embodiment of the present invention, the process of generating a task cycle scaling factor, setting preemption priority, and executing battery balancing and thermal management tasks is as follows: After obtaining the task execution time series corresponding to multiple future control cycles, the management controller calculates the computational resource requirements for each control cycle. Specifically, for any future control cycle, it reads the predicted execution time of the balanced task and the predicted execution time of the thermal management task for that cycle, adds the two predicted values to obtain the total CPU time expected to be occupied by the two tasks within that control cycle, and then divides this total time by the fixed duration of a single control cycle to obtain the total utilization rate of that future control cycle. For example, if the predicted execution time of the balanced task in a future control cycle is a first duration and the predicted execution time of the thermal management task is a second duration, then the sum of the first and second durations reflects the expected processor time consumed by the two control tasks within that control cycle, and its ratio to the fixed duration of the control cycle reflects the expected utilization of computational resources by the two tasks within that control cycle. The total utilization rate for each future control cycle is calculated in the same way, and the maximum value is selected as the predicted peak total utilization rate, so that task scheduling is configured according to the control cycle with the highest future computational load.
[0026] The management controller pre-stores the maximum CPU utilization and the safety reserved utilization. The maximum CPU utilization represents the upper limit of processor resources allowed by the management controller, while the safety reserved utilization reserves processor resources for communication processing, fault diagnosis, system interrupts, and operating system kernel tasks. Subtracting the safety reserved utilization from the maximum CPU utilization yields the upper limit of available computing resources that can be allocated to balancing tasks and hot management tasks. The management controller compares the predicted peak total utilization with the upper limit of available computing resources. When the predicted peak total utilization does not exceed the upper limit, it means that running the balancing and hot management tasks according to their original execution cycles will not consume all the processor resources beyond the reserved ones. In this case, the task cycle scaling factor is set to one, keeping the original cycles of the two tasks unchanged. When the predicted peak total utilization exceeds the upper limit of available computing resources, the predicted peak total utilization is divided by the upper limit, and the resulting ratio is used as the task cycle scaling factor. Since this ratio is greater than one, multiplying the original task cycle by this ratio subsequently extends the task trigger interval, reduces the number of triggers per unit time for the two tasks, and adapts the predicted computing resource demand to the available computing resources.
[0027] While determining the task cycle scaling factor, the management controller sets preemption priorities for balancing tasks and thermal management tasks. Thermal management tasks directly affect battery temperature regulation, and their preemption priority is set to the first priority level; balancing tasks are used to adjust the charge difference between battery cells, and their preemption priority is set to the second priority level. The real-time operating system uses a priority representation where lower values indicate higher priority; therefore, the first priority level is lower than the second priority level. This preemption priority is used to determine whether a currently running task can be interrupted after it enters the ready queue.
[0028] When performing task cycle adjustments, the management controller searches the task control block list of the real-time operating system for the balancing task control block and the thermal management task control block according to the task identifier. It reads the original value of the cycle field stored in the balancing task control block and multiplies it by the task cycle scaling factor to obtain the new cycle value for the balancing task. Similarly, it reads the original value of the cycle field stored in the thermal management task control block and multiplies it by the task cycle scaling factor to obtain the new cycle value for the thermal management task. When the task cycle scaling factor is one, the new cycle value is the same as the original cycle value; when the task cycle scaling factor is greater than one, the new cycle values for both tasks are extended proportionally. The cycle adjustment only changes the time interval between the periodic wake-ups of the tasks and does not alter the internal control calculation processes of the balancing task and the thermal management task.
[0029] The management controller establishes software timers based on the new cycle values for the load balancing task and the hot management task, respectively. Both software timers use the global system clock cycle as their timing reference. Each time the system generates a global clock cycle, the operating system updates the remaining time values of the software timers. When the load balancing task's software timer expires, the operating system enters the kernel critical section, transitions the load balancing task from a waiting state to a ready state, and adds it to the ready queue. When the hot management task's software timer expires, it is added to the ready queue in the same way. After the task is added, the system exits the kernel critical section and restarts the corresponding software timer to begin the next round of timing with the new cycle value.
[0030] Both load balancing and thermal management tasks are pre-configured with base priorities. When a task enters the ready queue, the real-time operating system determines its position according to its base priority, with tasks having higher base priorities preceding those with lower base priorities. The scheduler selects a task from the head of the ready queue and puts it into the running state. During task execution, if another task's software timer expires and causes that task to enter the ready queue, the real-time operating system compares the base priority of the newly added task with the preemption priority of the currently running task. If the base priority of the newly added task is lower than the preemption priority of the currently running task, the processor context of the currently running task is saved, the currently running task is put back into the ready state, and the CPU execution right is allocated to the newly added task; if this condition is not met, the currently running task continues to execute, and the newly added task remains in the ready queue awaiting scheduling.
[0031] After the balancing task gains execution control from the central processing unit, it generates balancing control results based on the current cell voltage state and sends balancing current commands to the battery balancing circuit, causing the corresponding cell to enter charging balancing, discharging balancing, or balancing stop state. After the thermal management task gains execution control from the central processing unit, it generates thermal management control results based on the current cell temperature state and sends coolant flow commands to the cooling system to adjust the coolant delivery rate. After completing the sending of this control command, the task exits the running state, and the real-time operating system continues to select subsequent tasks from the ready queue for execution. Thus, battery balancing control and thermal management control continue to be executed according to the adjusted task cycle and the set preemption relationship.
[0032] It should be noted that the basic priority of the thermal management task is set to 1, and the preemption priority is set to 1; the basic priority of the battery balancing task is set to 3, and the preemption priority is set to 2. The priorities from highest to lowest are: basic priority 1 and preemption priority 1 for the thermal management task, preemption priority 2 for the battery balancing task, and basic priority 3 for the battery balancing task.
[0033] The foregoing has provided a detailed description of one embodiment of the present invention, but this description is merely a preferred embodiment and should not be construed as limiting the scope of the invention. All equivalent variations and modifications made within the scope of the present invention should still fall within the scope of the present invention.
Claims
1. A battery management control method based on joint state computational power estimation, characterized in that, Includes the following steps: Collects battery voltage and temperature signals, and manages the utilization rate of the controller's central processing unit; Feature extraction is performed on voltage and temperature signals to obtain voltage fluctuation features and temperature peak features; Input voltage fluctuation characteristics, temperature peak characteristics, and occupancy rate into the joint estimation model, and output the task execution time series; Based on the task execution time sequence, the current task cycle, and available computing resources, a task cycle scaling factor is generated, and a preemption priority is set. Battery balancing and thermal management tasks are executed based on task cycle scaling factors and preemption priorities.
2. The battery management control method based on joint state computational power estimation according to claim 1, characterized in that, The specific process for acquiring voltage signals, temperature signals, and occupancy rate is as follows: The voltage at the end of each cell and the surface temperature of each cell are synchronously collected by a distributed sensor array in the battery pack. After passing through an anti-aliasing low-pass filter, the data are stored as voltage time series and temperature time series at a fixed sampling frequency. The idle task cycle count is read from the real-time operating system kernel interface of the management controller. The difference between the total cycle count and the idle task cycle count is divided by the total cycle count to obtain the utilization rate of the management controller's central processing unit.
3. The battery management control method based on joint state computational power estimation according to claim 1, characterized in that, The specific process for obtaining voltage fluctuation characteristics and temperature peak characteristics is as follows: Within a sliding time window of a preset length, the maximum and minimum values of the voltage time series are taken, and the difference between the maximum and minimum values is used as the voltage fluctuation characteristic. Within the same sliding time window, the maximum value of the temperature time series is taken as the temperature peak feature.
4. The battery management control method based on joint state computational power estimation according to claim 1, characterized in that, The specific process of outputting the task execution time series is as follows: The joint estimation model is a recurrent neural network containing gated recurrent units. The input vector is composed of the voltage fluctuation features, temperature peak features, and occupancy rate at the current time, as well as the voltage fluctuation features, temperature peak features, and occupancy rate at the previous K sampling times. The output layer of the recurrent neural network outputs a vector of length 2Q. The first Q components correspond to the predicted execution time of the equilibrium task in the next Q control cycles, and the last Q components correspond to the predicted execution time of the thermal management task in the next Q control cycles, thus forming a task execution time sequence.
5. The battery management control method based on joint state computational power estimation according to claim 1, characterized in that, The specific process for generating the task cycle scaling factor is as follows: Add the predicted execution time of the equilibrium task to the predicted execution time of the thermal management task for each future control cycle, divide the sum by the fixed duration of a single control cycle to obtain the total occupancy rate for each cycle, and extract the maximum value of the total occupancy rate for all future control cycles as the predicted peak total occupancy rate. The maximum utilization rate of the central processing unit minus the security reserved utilization rate is used as the upper limit of available computing resources; When the predicted peak total utilization rate exceeds the upper limit of available computing resources, the task cycle scaling factor is the predicted peak total utilization rate divided by the upper limit of available computing resources. When the predicted peak total utilization rate does not exceed the upper limit of available computing resources, the task cycle scaling factor is 1.
6. The battery management control method based on joint state computational power estimation according to claim 1, characterized in that, The process of setting preemption priority is as follows: Set the preemption priority of the balancing task to the second priority level, and set the preemption priority of the thermal management task to the first priority level, with the first priority level being lower than the second priority level.
7. The battery management control method based on joint state computational power estimation according to claim 1, characterized in that, The process of performing battery balancing and thermal management tasks is as follows: Locate the Balanced Task Control Block and the Thermal Management Task Control Block in the Real-Time Operating System Task Control Block List. Multiply the original values of the Balanced Task Control Block's periodic field and the Thermal Management Task Control Block's periodic field by the task period scaling factor to obtain the new periodic values for the Balanced Task and the Thermal Management Task, respectively.
8. The battery management control method based on joint state computational power estimation according to claim 7, characterized in that, The process of performing battery balancing and thermal management tasks also includes: The software timer for the balancing task is configured based on the global system tick of the operating system, with the new cycle value of the balancing task as the timing period. After the timer expires, the balancing task is added to the ready queue in the kernel critical section. The thermal management task software timer is configured with a new cycle value as the timing period and is based on the global system tick of the operating system. After the timer expires, the thermal management task is added to the ready queue in the kernel critical section.
9. The battery management control method based on joint state computational power estimation according to claim 8, characterized in that, The process of performing battery balancing and thermal management tasks also includes: Both balancing tasks and thermal management tasks have pre-set basic priorities. When a balancing task or thermal management task is added to the ready queue, its position in the ready queue is determined according to its basic priority. Starting with the first task in the ready queue, tasks in the ready queue enter the running state in sequence. When another task is added to the ready queue and its base priority is lower than the preemption priority of the currently running task, the CPU execution right is allocated to the newly added task.
10. The battery management control method based on joint state computational power estimation according to claim 9, characterized in that, The process of performing battery balancing and thermal management tasks also includes: When the balancing task is in operation, a balancing current command is sent to the battery balancing circuit. When the thermal management task is in operation, a coolant flow command is sent to the cooling system.