A high-precision data acquisition and processing system for a battery module

Abstract

The application discloses a kind of data high-precision acquisition processing systems of battery module, belong to acquisition processing system technical field, comprising: distributed intelligent acquisition module, digital communication connection line, main control and collaborative computing module and fast self-checking and protection circuit, the distributed intelligent acquisition module is used to connect and acquire the analog signal of each electric core or electric core module electric parameter and environmental parameter, and it is converted into original digital data, then original data is carried out online adaptive weighted data fusion, sparse sampling compression and feature extraction, generates situation data package, and according to time series prediction model dynamic adjustment data acquisition cycle to optimize energy consumption, simultaneously according to network state and data priority dynamic selection communication protocol and transmission path.The application can also be distributed intelligent acquisition on the basis of realizing high-precision acquisition processing to battery module, and can realize the intelligent closed loop of sensory transmission calculation protection depth cooperation.

Description

Technical Field

[0001] This invention relates to the field of data acquisition and processing systems, and more specifically, to a high-precision data acquisition and processing system for battery modules. Background Technology

[0002] A battery module typically consists of a housing and battery cells housed within the housing. To monitor the battery module, a sampling device is usually placed between the housing and the battery cells to collect signals from the cells, such as voltage and temperature signals. To ensure the normal operation of the battery module, structural components, such as insulating plates, can also be placed between the housing and the battery cells.

[0003] Existing processing systems still employ a centralized acquisition architecture, which uses long wiring harnesses to collect analog voltage and temperature signals from each battery cell onto one or several centralized acquisition boards (AFEs) for processing. This architecture suffers from the vulnerability of weak analog signals transmitted over long distances to electromagnetic interference generated by high-power charging and discharging within the battery pack, leading to decreased sampling accuracy. Sampling accuracy directly determines the quality of SOC estimation and equalization control. Furthermore, the complex wiring harnesses not only increase weight, cost, and assembly difficulty but also become a reliability bottleneck, with connector failures and cable wear being common problems. Finally, the number of acquisition channels in a centralized AFE chip is fixed, resulting in poor system scalability. Different module sizes often require redesign, leading to limited versatility. Summary of the Invention

[0004] To address the problems existing in the prior art, the purpose of this invention is to provide a high-precision data acquisition and processing system for battery modules. In addition to achieving high-precision acquisition and processing of battery modules, this invention can also perform distributed intelligent acquisition and realize a smart closed loop of deep collaboration between sensing, computing, and protection.

[0005] To solve the above problems, the present invention adopts the following technical solution: A high-precision data acquisition and processing system for a battery module includes: multiple distributed intelligent acquisition modules, digital communication connection lines, a main control and collaborative computing module, and a fast self-test and protection circuit; Multiple distributed intelligent acquisition modules are respectively set on each cell or cell module of the battery module. The distributed intelligent acquisition modules are used to connect and acquire analog signals of electrical parameters and environmental parameters, convert them into raw digital data, and then perform online adaptive weighted data fusion, sparse sampling compression based on compressed sensing and feature extraction on the raw data to generate a concise situation data packet. The data acquisition cycle is dynamically adjusted according to the time series prediction model to optimize energy consumption. At the same time, the communication protocol and transmission path are dynamically selected according to the network status and data priority, and the multi-hop transmission energy efficiency is optimized based on the asynchronous distributed aggregation layout construction algorithm. The situation data packet is uploaded, and the policy update event of the configuration center is monitored, the new acquisition strategy is dynamically loaded and immediately takes effect, and the working sequence and energy consumption are coordinated. The digital communication connection base supports fully digital, bidirectional, and multi-station communication, enabling real-time information interaction and command transmission between the distributed intelligent acquisition module, the main control and collaborative computing module, and the fast self-test and protection circuit. It also employs a mechanism for rapid retransmission of change information and timed transmission of no change information to ensure real-time communication. Simultaneously, it actively queries and alarms the status of adjacent nodes through a mirror reflection echo location method to ensure network reliability. The main control and collaborative computing are used to receive global computing tasks, compile them into a directed acyclic graph of tasks through a task graph compiler, obtain a globally consistent heterogeneous computing resource status, divide and allocate the task graph according to the critical path dynamic perception and reconstruction algorithm, execute sub-tasks after allocation, achieve efficient data access, and coordinate the iterative computing process by the superstep iterative controller. The rapid self-test and protection circuit continuously performs automatic checks on its own hardware, software, input / output channels, communication system, and key components during device power-on and operation. When the self-test detects a fault or a line fault is sensed through the distributed intelligent acquisition and communication network, it can quickly identify the fault type and scope, delay trip to disconnect the corresponding fault area, achieve rapid fault isolation and power restoration to non-faulty areas, and generate a fault report containing complete information such as the fault time, action behavior, and sampling data after the fault. As a preferred embodiment of the present invention, the distributed intelligent acquisition module includes a multimodal sensing interface unit, an intelligent processing unit, an adaptive communication unit, a collaborative management unit, a hub, a host computer, and a configuration center. The multimodal sensing interface unit is used to connect to and acquire analog signals from various types of sensors through different interface protocols, including electrical parameter sensors and environmental parameter sensors, and converts the analog signals into raw digital data. The intelligent processing unit is used to perform online data fusion, compression, and feature extraction on the raw digital data to generate a concise situational data package; The adaptive communication unit is used to dynamically select the communication protocol and transmission path according to the network status and data priority, and upload the situation data packet to the hub or host computer. The collaborative management unit is connected to the multimodal sensing interface unit, the intelligent processing unit, and the adaptive communication unit respectively, and is used to coordinate the working sequence of each unit, manage the energy consumption of the module, and respond to configuration and query commands from the host computer. The hub connects multiple adaptive communication units through a star topology and polls to receive data uploaded by each module. The host computer is connected to the hub and is used to configure the acquisition parameters of each module, receive and process global data, and realize data visualization and analysis. The configuration center is connected to the host computer and stores global policy configuration parameters; The collaborative management unit of the distributed intelligent acquisition module listens to the update events of the configuration center and dynamically loads new acquisition strategies to make them take effect immediately. The data fusion adaptive weighted fusion algorithm executed by the intelligent processing unit is given by the formula: , The fused output value has the same units as the original physical quantities from the sensor. For the raw observation value of the i-th sensor, Let be the adaptive weighting coefficient of the i-th sensor, satisfying The weighting coefficient represents the number of sensors participating in the fusion. Based on the sensor's real-time confidence level and historical error Dynamic calculation: .

[0006] As a preferred embodiment of the present invention, the multimodal sensing interface unit specifically includes multiple analog-to-digital conversion modules and a protocol adaptation subunit; Multiple analog-to-digital conversion modules constitute a multi-input channel for converting analog signals from different sensors into digital signals; The protocol adapter subunit supports SPI, IIC, RS485 and analog voltage / current interfaces to accommodate the differences in data format and sampling rate of different modal sensors.

[0007] As a preferred embodiment of the present invention, the data compression performed by the intelligent processing unit adopts a sparse sampling algorithm based on compression sensing, specifically: The intelligent processing unit has a built-in one OK Random perceptual matrix of columns ,in <6, in discrete time right The original data vector of each channel Compress to obtain Dimensional measurement The perception matrix Each row has only one non-zero element, and each column has at most one non-zero element, enabling distributed random data collection where only a portion of the node data is transmitted. The measured value The unit is a linear combination of the original data units, and its dimension is... It is determined by the preset compression ratio and reconstruction accuracy requirements.

[0008] As a preferred embodiment of the present invention, the intelligent processing unit also performs time series forecasting and dynamic scheduling, and the P-order autoregressive model AR(P) it employs is formulated as follows: ,in For the current moment The sensor prediction value, in the same unit as the sampled value, These are the regression coefficients of the model, dimensionless. For the past Historical observations at each moment, With a mean of 0 and a variance of A white noise sequence; The collaborative management unit compares the current actual sampled value. Compared with the predicted value ,like If the sampling period is not exceeded, the next sampling period for that channel will be automatically extended; otherwise, the sampling period will be shortened or data will be reported immediately. This is the preset error threshold.

[0009] As a preferred embodiment of the present invention, the main control and collaborative computing module includes a main control scheduling module and a collaborative computing agent module; The main control scheduling module is used to receive global computing tasks, parse, segment and schedule the tasks, and monitor the status of all collaborative computing resources. The collaborative computing agent module is used to receive the segmented subtasks, drive local or remote heterogeneous computing resources to perform calculations, and return the results; The main control scheduling module has a built-in heterogeneous computing resource abstract model, which is used to uniformly describe the computing power, memory and power consumption constraints of CPU, GPU, FPGA, DSP and edge computing nodes. The collaborative computing agent module supports task dependency parsing based on directed acyclic graphs to realize automatic data flow synchronization between subtasks.

[0010] As a preferred embodiment of the present invention, the main control scheduling module specifically includes a task graph compiler, a resource status monitor, an overstep iteration controller, and a dynamic partitioning and load balancer; The task graph compiler is used to compile user-submitted serial or parallel programs into a directed acyclic graph of tasks containing vertices, edges, and dependencies. The resource status monitor is used to periodically collect computing resource utilization, memory usage, network latency and power consumption data from each collaborative computing agent module, and update the heterogeneous computing resource abstract model. The superstep iteration controller is used to manage iterative computing tasks. Each iteration is defined as a superstep. The controller broadcasts a computing start instruction to all collaborative computing agent modules and collects completion signals to trigger the next superstep. The dynamic partitioning and load balancer adopts a chemometrics-inspired dynamic partitioning algorithm to divide the task directed acyclic graph into multiple subgraphs and distribute them to each collaborative computing agent module. Its objective function is to minimize the sum of global computation completion time and communication overhead.

[0011] As a preferred embodiment of the present invention, the collaborative computing agent module includes a primary and secondary data structure management unit, a user-defined function execution engine, and a hardware abstraction layer and task offloading interface: The main and auxiliary data structure management unit is used to maintain the data required for computing tasks. The main data is the entity content of the data to be processed, and the auxiliary data is metadata describing the partitions, status and dependencies of the main data. The main and auxiliary data structure management unit ensures that the CPU and GPU threads can efficiently locate and load the main data based on the auxiliary data when processing in parallel, thereby achieving load balancing. The user-defined function execution engine is used to load and execute the computational kernel logic issued by the main control scheduling module. The user-defined function execution engine provides a unified programming interface, which allows the core computational logic of graph neural network aggregation and different shortest path algorithms to be encapsulated into user-defined functions for replacement and execution. The hardware abstraction layer and task offloading interface are used to shield the differences in underlying heterogeneous hardware and support the offloading of computing tasks to a dedicated coprocessor via remote procedure call protocol. The hardware abstraction layer and task offloading interface manage shared memory, hardware FIFO and doorbell interrupt mechanism to achieve efficient parameter transfer and result return between the main controller and the coprocessor. The dynamic partitioning and load balancer's critical path dynamic awareness and reconstruction algorithm has the following formula: ,in This is the estimated critical path time for the task graph, in seconds. Representative task diagram The set of all paths from the entrance to the exit. For the task vertices on the path, As vertices The estimated time for the target resources As vertices The data communication time between it and its predecessor vertex; The algorithm periodically monitors the actual execution time of each vertex. ,like This triggers a local repartition, migrating lagging vertices or their subsequent dependent vertices to a less loaded collaborative computing proxy module. This is a dynamic threshold.

[0012] As a preferred embodiment of the present invention, the main control and collaborative computing module further includes a resource collaborative arbitration module, which executes a distributed resource state synchronization algorithm based on a consistency protocol, the specific steps of which are as follows: Each collaborative computing agent module maintains a local resource state vector. ,in To calculate utilization rate, For memory usage, The average network latency to the main controller; In each synchronization cycle, the module Its state vector Send to its neighbor module and receive the status of neighbors. ; According to the formula Update its own state estimate, where The convergence coefficient is . For connection weights; After a limited number of iterations, all modules reach an approximate consensus on the global resource load status. At the same time, all modules need to know each other's CPU utilization. The main control scheduling module makes task allocation decisions based on this consensus status. The master control scheduling module periodically instructs each collaborative computing agent module to persist the graph computing state it maintains to reliable storage. When a node failure is detected, the task and its state are rolled back to the most recent consistent checkpoint and execution is resumed on other healthy nodes. The graph computing state includes vertex state, edge state, global progress state, and persistent checkpoint. The rollback mechanism provides the BMS with an intelligent path to safely exit from an anomaly and attempt to recover to a known health baseline. It is a key strategy to prevent the escalation of failures, ensure the continuous availability of the system, and achieve safe recovery. When the collaborative computing agent module predicts that the resources of this node are about to be overloaded or fail, it migrates the delegated computing tasks and context states it carries to other edge computing platforms or cloud instances through the delegated computing application management interface and continues to predict, so as to achieve uninterrupted service.

[0013] As a preferred embodiment of the present invention, the collaborative computing method between the main control module and the collaborative computing module includes the following steps: The main control scheduling module receives computing tasks and transforms them into a directed acyclic graph of tasks through a task graph compiler. The resource status monitor works in conjunction with the resource coordination arbitration module to obtain a globally consistent view of heterogeneous resource status. The dynamic partitioning and load balancer divides the task graph and distributes it to each collaborative computing agent module based on the critical path dynamic perception and reconstruction algorithm. Each collaborative computing agent module executes its assigned subtasks in parallel through its user-defined function execution engine and hardware abstraction layer, while performing efficient data access through the main and auxiliary data structure management unit. The superstep iterative controller coordinates each round of computation for the iterative task until the task termination condition is met. The cross-layer awareness elastic fault tolerance and recovery module runs in the background to ensure the reliability of the computing process.

[0014] Compared with the prior art, the advantages of this invention are: This invention utilizes distributed intelligent acquisition modules, each of which is a miniature autonomous unit with signal conditioning, analog-to-digital conversion, edge computing, and intelligent communication capabilities. This completely eliminates long-distance analog signal transmission; voltage and temperature signals are converted into digital signals near the point of generation, fundamentally preventing transmission interference and ensuring high precision at the data source. Secondly, it greatly simplifies physical connections. Modules and cells are directly connected or connected over short distances, and modules are networked via simple digital communication cables, replacing traditional, complex analog wiring harnesses. This significantly improves system reliability, reduces weight, and creates conditions for automated assembly. It gives the system unprecedented flexibility and scalability. Adding or removing battery modules only requires adding or removing intelligent acquisition modules accordingly. The system automatically networks itself through adaptive communication algorithms, and the main controller automatically integrates new data through a collaborative computing framework. This allows the system to easily adapt to battery packs of different sizes and models, achieving true universal design and greatly facilitating subsequent maintenance, module replacement, and reuse.

[0015] This invention integrates a series of intelligent algorithms into a cohesive whole. At the sensing level, the intelligent processing unit dynamically adjusts the sampling frequency based on time series prediction, achieving on-demand sampling. This ensures monitoring effectiveness while minimizing the energy consumption of the acquisition module. At the transmission level, the adaptive communication unit not only selects protocols based on network conditions but also dynamically optimizes multi-hop transmission paths based on asynchronous distributed algorithms, minimizing the total energy consumption of the entire sensor network. This is significant for large battery systems consisting of hundreds of nodes. At the computing level, the main control and collaborative computing modules parallelize and efficiently distribute complex battery management algorithms across all available computing resources through heterogeneous resource abstraction, task graph compilation, and dynamic scheduling of critical paths. This enables the real-time execution of advanced AI algorithms in embedded environments, significantly improving the accuracy and speed of estimating core states such as SOC and SOH. At the protection level, rapid self-testing and protection circuits are linked with the acquisition and communication networks, enabling rapid and accurate fault location and isolation. Finally, all these components are coordinated and managed by the collaborative management unit and configuration center, forming an intelligent closed-loop system capable of self-optimization, self-adaptation, and self-protection. This not only significantly improves the performance limits of battery management, but also extends the range and lifespan of the system itself or the battery pack through comprehensive energy efficiency optimization, representing a clear development direction for the next generation of battery management systems. Attached Figure Description

[0016] Figure 1 This is a schematic diagram of a high-precision data acquisition and processing system for a battery module according to the present invention. Figure 2 This is a schematic diagram of a distributed intelligent acquisition module in a high-precision data acquisition and processing system for a battery module 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. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are within the scope of protection of the present invention.

[0018] Example: Please see Figure 1-2 A high-precision data acquisition and processing system for a battery module includes: multiple distributed intelligent acquisition modules, digital communication connection lines, a main control and collaborative computing module, and a fast self-test and protection circuit; Multiple distributed intelligent acquisition modules are respectively set on each cell or cell module of the battery module. The distributed intelligent acquisition is used to connect and collect analog signals of electrical parameters and environmental parameters, and convert them into raw digital data. Then, the raw data is subjected to online adaptive weighted data fusion, sparse sampling compression based on compressed sensing, and feature extraction to generate a concise situation data packet. The data acquisition cycle is dynamically adjusted according to the time series prediction model to optimize energy consumption. At the same time, the communication protocol and transmission path are dynamically selected according to the network status and data priority. The multi-hop transmission energy efficiency is optimized based on the asynchronous distributed aggregation layout construction algorithm. The situation data packet is uploaded, and the policy update event of the configuration center is monitored. The new acquisition strategy is dynamically loaded and immediately takes effect to coordinate the working sequence and energy consumption. The digital communication connection base supports fully digital, bidirectional, and multi-station communication, enabling real-time information interaction and command transmission between the distributed intelligent acquisition module, the main control and collaborative computing module, and the rapid self-test and protection circuit. It also employs a mechanism for rapid retransmission of change information and timed transmission of no change information to ensure real-time communication. Furthermore, it actively queries and alarms the status of adjacent nodes through a mirror reflection echo location method to ensure network reliability. The main control and collaborative computing are used to receive global computing tasks, compile them into a directed acyclic graph of tasks through the task graph compiler, obtain the globally consistent heterogeneous computing resource status, divide and allocate the task graph according to the critical path dynamic perception and reconstruction algorithm, execute sub-tasks after allocation, achieve efficient data access, and coordinate the iterative computing process by the superstep iterative controller. The rapid self-test and protection circuit is used to continuously perform automatic checks on its own hardware, software, input / output channels, communication system and key components during the device's power-on and operation. When the self-test detects a fault or a line fault is sensed through the distributed intelligent acquisition and communication network, it can quickly identify the fault type and scope, delay trip to cut off the corresponding fault area, realize rapid isolation of the fault and restoration of power supply to the non-faulty area, and generate a fault report containing complete information such as the fault time, action behavior and sampling data after the fault.

[0019] In a specific embodiment of this invention, the architecture of battery data acquisition and processing is fundamentally reconstructed through the deep collaboration and closed-loop design of four core functions. Compared with traditional centralized, wired BMS, this system achieves high-precision digitization of the data source through distributed acquisition modules, eliminating interference and accuracy loss caused by long-distance analog signal transmission; it ensures real-time and reliable interaction of massive amounts of data between nodes through intelligent communication mechanisms; it efficiently handles complex battery state estimation tasks through a collaborative computing framework; and finally, it ensures the safe operation of the entire system through rapid self-testing and protection circuits. This integrated design of acquisition, transmission, computing, and protection enables the system to achieve a qualitative leap in accuracy, reliability, real-time performance, and scalability, meeting the stringent requirements of next-generation high-energy-density, long-life battery modules for refined management.

[0020] Specifically, the distributed intelligent acquisition module includes a multimodal sensing interface unit, an intelligent processing unit, an adaptive communication unit, a collaborative management unit, a hub, a host computer, and a configuration center; The multimodal sensing interface unit is used to connect to and acquire analog signals from various types of sensors, including electrical parameter sensors and environmental parameter sensors, through different interface protocols, and converts the analog signals into raw digital data. The intelligent processing unit is used to perform online data fusion, compression, and feature extraction on the raw digital data to generate a concise situational data package; The adaptive communication unit is used to dynamically select the communication protocol and transmission path according to the network status and data priority, and upload the situation data packet to the hub or host computer. The collaborative management unit is connected to the multimodal sensing interface unit, the intelligent processing unit, and the adaptive communication unit, respectively, to coordinate the working sequence of each unit, manage the energy consumption of the module, and respond to configuration and query commands from the host computer. The hub connects multiple adaptive communication units through a star topology and polls to receive data uploaded by each module. The host computer connects to the hub to configure the acquisition parameters of each module, receive and process global data, and realize data visualization and analysis. The configuration center connects to the host computer and stores global policy configuration parameters; The collaborative management unit of the distributed intelligent acquisition module listens for update events in the configuration center and dynamically loads new acquisition strategies to make them take effect immediately. The intelligent processing unit executes an adaptive weighted fusion algorithm for data fusion, where the formula is: , The fused output value has the same units as the original physical quantities from the sensor. For the raw observation value of the i-th sensor, Let be the adaptive weighting coefficient of the i-th sensor, satisfying The number of sensors participating in the fusion, and the weighting coefficient. Based on the sensor's real-time confidence level and historical error Dynamic calculation: .

[0021] In a specific embodiment of the present invention, multimodal sensing, edge intelligent processing, and dynamic strategy loading capabilities are integrated into a miniaturized module close to the battery cell. In traditional solutions, signals from different sensors such as voltage and temperature need to be collected to the central processing unit through complex wiring harnesses, resulting in complex circuit design and susceptibility to interference. The multimodal interface unit supports multiple protocols, achieving standardization and flexible configuration of sensor access. The intelligent processing unit performs fusion, compression, and feature extraction at the data source to generate situational data packets, which greatly reduces the communication bandwidth pressure and the computational load of the main control unit. The collaborative management unit monitors the configuration center and dynamically loads strategies, enabling the system to remotely optimize parameters such as acquisition frequency and fusion algorithm without interrupting operation. This achieves software definition and continuous optimization of the battery management system, which is impossible with traditional fixed-function acquisition circuits.

[0022] Specifically, the multimodal sensing interface unit includes multiple analog-to-digital conversion modules and protocol adaptation subunits; Multiple analog-to-digital converter modules form a multi-input channel, used to convert analog signals from different sensors into digital signals; The protocol adapter subunit supports SPI, IIC, RS485 and analog voltage / current interfaces to accommodate the differences in data format and sampling rate of different modal sensors; The channel selection switch is controlled by the collaborative management unit and is used for time-division multiplexing of the analog-to-digital conversion module to sequentially acquire sensor data connected to different interfaces.

[0023] In a specific embodiment of the present invention, efficient and unified access to heterogeneous sensors is achieved with a simple hardware structure through protocol adaptation and channel time-division multiplexing mechanism. In the prior art, to collect multiple temperature signals, each sensor usually needs to occupy an independent ADC channel or add a complex multiplexer, resulting in high resource consumption or reduced circuit reliability. This solution uses a protocol adaptation subunit to be compatible with digital and analog interfaces, and then uses a collaborative management unit to intelligently schedule the channel selection switch, so that a single or a few ADC modules can collect all sensor data in sequence. This method optimizes hardware resources to the maximum extent while ensuring real-time sampling, reduces module complexity and cost, and improves system integration and reliability. It is particularly suitable for application scenarios with limited internal space in battery modules and numerous sensors.

[0024] Specifically, the data compression performed by the intelligent processing unit adopts a sparse sampling algorithm based on compression sensing, as follows: The intelligent processing unit has a built-in one OK Random perceptual matrix of columns ,in <6, in discrete time right The original data vector of each channel Compress to obtain Dimensional measurement The perception matrix Each row has only one non-zero element, and each column has at most one non-zero element, enabling distributed random data collection where only a portion of the node data is transmitted. Measured values The unit is a linear combination of the original data units, and its dimension is... It is determined by the preset compression ratio and reconstruction accuracy requirements.

[0025] In a specific embodiment of this invention, a sparse sampling algorithm based on compressed sensing is introduced for data compression. Its innovation lies in overturning the traditional model of first collecting all data and then compressing and transmitting it. In battery module monitoring, voltage and temperature data often exhibit spatiotemporal correlation. Traditional methods collect and transmit all raw data, resulting in significant redundancy and energy consumption. This solution utilizes a specially structured random sensing matrix built into the intelligent processing unit. Only a far fewer linear measurements need to be collected than the number of sensors. This allows the original data trend to be restored using reconstruction algorithms in the cloud or on a host computer. This method of collecting and compressing data simultaneously reduces the amount of data in wireless or wired communication by an order of magnitude without losing any key information. This significantly reduces the communication power consumption of distributed nodes, extends system battery life, and removes the data transmission bottleneck for the large-scale deployment of wireless BMS.

[0026] Specifically, the intelligent processing unit also performs time series forecasting and dynamic scheduling, using the P-order autoregressive model AR(P) formula: ,in For the current moment The sensor prediction value, in the same unit as the sampled value, These are the regression coefficients of the model, dimensionless. For the past Historical observations at each moment, With a mean of 0 and a variance of A white noise sequence; The collaborative management unit compares the current actual sampled values. Compared with the predicted value ,like If the sampling period is not exceeded, the next sampling period for that channel will be automatically extended; otherwise, the sampling period will be shortened or data will be reported immediately. This is the preset error threshold.

[0027] In a specific embodiment of the present invention, the combination of time series prediction model and dynamic sampling scheduling enables on-demand data acquisition, representing a key breakthrough in energy efficiency optimization. Traditional battery data acquisition systems typically employ fixed-frequency sampling, consuming the same sampling power regardless of whether the battery is in a quiescent, slow-charging, or fast-charging state, resulting in unnecessary energy waste. This solution uses an AR model to predict sensor readings for the next moment based on historical data, only detecting data when the deviation between the actual and predicted values ​​exceeds a threshold. Only when the data changes significantly is the sampling period considered to have changed, thus triggering high-frequency sampling or immediate reporting. Otherwise, the sampling period is automatically extended. This event-triggered adaptive sampling mechanism allows the system to enter a low-power listening mode when the battery status is stable, and to respond quickly when the status changes drastically. It dynamically balances the contradiction between data accuracy and system power consumption, which is especially beneficial for the long-term operation of power-sensitive wireless sensor nodes.

[0028] Specifically, the transmission path selection performed by the adaptive communication unit employs an asynchronous distributed aggregation layout construction algorithm, including: The distributed intelligent acquisition module acts as a network node, constructing an initial acquisition tree with neighboring nodes based on the received signal strength. Calculate the local topology overhead with the distributed intelligent acquisition module as the aggregation node. ,in It is a collection of child nodes with the distributed intelligent acquisition module as the parent node. For nodes The transmission energy consumption, measured in joules. Processing data from nodes for aggregation nodes The energy consumption of the data; By asynchronously exchanging cost information with neighboring nodes, the aggregation node and multi-hop transmission path that minimizes global cost are determined iteratively.

[0029] In a specific embodiment of the present invention, an asynchronous distributed aggregation layout construction algorithm is proposed for multi-hop transmission networks. This algorithm achieves dynamic self-organization and optimal energy efficiency of the communication topology. In a network composed of dozens or hundreds of distributed acquisition modules, efficiently aggregating data to the gateway is a challenge. Traditional static routing or centralized scheduling algorithms are difficult to adapt to changes in network topology and fluctuations in channel conditions. This algorithm allows each node to asynchronously construct an initial acquisition tree based on local information and to negotiate the aggregation path with the minimum global transmission energy consumption by iteratively exchanging local topology overhead information. This decentralized approach not only avoids single points of failure but also enables the network to adapt to the addition, removal, or movement of nodes, always maintaining optimal transmission energy efficiency. This greatly improves the data collection reliability and overall energy efficiency of large-scale battery networks.

[0030] Specifically, the main control and collaborative computing module includes a main control scheduling module and a collaborative computing agent module; The main control scheduling module is used to receive global computing tasks, parse, divide and schedule the tasks, and monitor the status of all collaborative computing resources. The collaborative computing agent module is used to receive segmented subtasks, drive local or remote heterogeneous computing resources to perform computations, and return the results; The main control scheduling module has a built-in heterogeneous computing resource abstract model, which is used to uniformly describe the computing power, memory and power consumption constraints of CPU, GPU, FPGA, DSP and edge computing nodes. The collaborative computing agent module supports task dependency parsing based on directed acyclic graphs to realize automatic data flow synchronization between subtasks.

[0031] In a specific embodiment of this invention, a heterogeneous computing resource abstraction model and task dependency resolution based on directed acyclic graphs are introduced to achieve intelligent offloading and parallel scheduling of computing tasks. Traditional BMS main control units have limited computing power, making it difficult to run complex algorithms. This solution upgrades the main control module to a collaborative computing center, which can uniformly describe and manage heterogeneous computing power such as CPUs, GPUs, FPGAs, and even edge nodes. By compiling tasks such as SOC estimation, balancing strategies, and fault diagnosis into DAGs, the system can automatically resolve data dependencies between tasks and dynamically schedule subtasks to the most suitable computing resources for execution. This breaks the bottleneck of traditional BMS computing tasks being executed serially on a single MCU, fully utilizing the computing potential inside and outside the system, making it possible to run advanced and complex AI algorithms in real time on the battery management system, and significantly improving the accuracy and speed of state estimation.

[0032] Specifically, the main control scheduling module includes a task graph compiler, a resource status monitor, an overstep iteration controller, and a dynamic partitioning and load balancer. The task graph compiler is used to compile user-submitted serial or parallel programs into a directed acyclic graph of tasks containing vertices, edges, and dependencies. The resource status monitor is used to periodically collect computing resource utilization, memory usage, network latency and power consumption data from each collaborative computing agent module, and update the heterogeneous computing resource abstract model. The superstep iteration controller is used to manage iterative computing tasks. Each iteration is defined as a superstep. The controller broadcasts the computing start instruction to all collaborative computing agent modules and collects completion signals to trigger the next superstep. The dynamic partitioning and load balancer adopts a chemometrics-inspired dynamic partitioning algorithm to divide the task directed acyclic graph into multiple subgraphs and distribute them to each collaborative computing agent module. Its objective function is to minimize the sum of global computing completion time and communication overhead. The collaborative computing agent module includes a primary and secondary data structure management unit, a user-defined function execution engine, and a hardware abstraction layer and task offloading interface. The main and auxiliary data structure management unit is used to maintain the data required for computing tasks. The main data is the entity content of the data to be processed, and the auxiliary data is the metadata describing the partitions, status and dependencies of the main data. The main and auxiliary data structure management unit ensures that the CPU and GPU threads can efficiently locate and load the main data based on the auxiliary data when processing in parallel, so as to achieve load balancing. The user-defined function execution engine is used to load and execute the computational kernel logic issued by the main control scheduling module. The user-defined function execution engine provides a unified programming interface, which allows the core computational logic of graph neural network aggregation and different shortest path algorithms to be encapsulated into user-defined functions for replacement and execution. The hardware abstraction layer and task offloading interface are used to shield the differences in underlying heterogeneous hardware. They support the offloading of computing tasks to a dedicated coprocessor via remote procedure call protocol. The hardware abstraction layer and task offloading interface manage shared memory, hardware FIFO and doorbell interrupt mechanism, and realize efficient parameter transfer and result return between the main controller and the coprocessor. The dynamic partitioning and load balancer's critical path dynamic awareness and reconstruction algorithm has the following formula: ,in This is the estimated critical path time for the task graph, in seconds. Representative task diagram The set of all paths from the entrance to the exit. For the task vertices on the path, As vertices The estimated time for the target resources As vertices The data communication time between it and its predecessor vertex; The algorithm periodically monitors the actual execution time of each vertex. ,like This triggers a local repartition, migrating lagging vertices or their subsequent dependent vertices to a less loaded collaborative computing proxy module. This is a dynamic threshold.

[0033] In a specific embodiment of the present invention, for complex computational task graphs, static task partitioning cannot adapt to the dynamic changes in runtime resource load, which can easily lead to overload of some nodes and extended overall completion time. This solution monitors the actual execution time of task vertices in real time and compares it with the estimated time, dynamically triggering task repartitioning and migration to ensure that optimization always revolves around the critical path and minimizes the total task completion time. At the same time, the main and auxiliary data structure management units efficiently guide computation threads to access actual data through metadata for heterogeneous collaborative computing scenarios such as CPU-GPU, solving the bottleneck problem of data access in heterogeneous computing. The combination of these two makes the collaborative computing module not only intelligently scheduled but also efficiently executed, and can cope with the dual challenges of uneven load and data density in real-time computing of the battery management system.

[0034] Specifically, the main control and collaborative computing module also includes a resource collaborative arbitration module. This module executes a distributed resource state synchronization algorithm based on a consistency protocol. The specific steps are as follows: Each collaborative computing agent module maintains a local resource state vector. ,in To calculate utilization rate, For memory usage, The average network latency to the main controller; In each synchronization cycle, the module Its state vector Send to its neighbor module and receive the status of neighbors. ; According to the formula Update its own state estimate, where The convergence coefficient is . For connection weights; After a limited number of iterations, all modules reach an approximate consensus on the global resource load status, and the main control scheduling module makes task allocation decisions based on this consensus status. The master scheduling module periodically instructs each collaborative computing agent module to persist the graph computing state it maintains to reliable storage. When a node failure is detected, the task and its state are rolled back to the most recent consistent checkpoint and execution is resumed on other healthy nodes. When the collaborative computing agent module predicts that the resources of this node are about to be overloaded or fail, it migrates the delegated computing tasks and context states it carries to other edge computing platforms or cloud instances through the delegated computing application management interface to achieve uninterrupted service.

[0035] In a specific embodiment of the present invention, two major guarantee mechanisms are added to the collaborative computing system: distributed resource state consistency synchronization and cross-layer elastic fault tolerance. In a distributed computing environment, the accuracy and timeliness of the master control module's acquisition of the global resource view are the basis for correct scheduling. This solution, through a distributed synchronization algorithm based on a consistency protocol, enables all computing agent nodes to quickly achieve a consistent understanding of the global load, avoiding the delay and single point of failure caused by the central node collecting information. More importantly, it designs two fault tolerance mechanisms: checkpoint persistence and task migration. The checkpoint mechanism deals with sudden node failures, ensuring that the computing state is recoverable, while the task migration mechanism addresses node overload or predictive maintenance, achieving seamless service switching. This builds comprehensive resilience from hardware failure to performance degradation, ensuring the continuity and high availability of the battery management core computing service, and meeting the stringent requirements for functional safety and reliability in fields such as automobiles and energy storage.

[0036] Specifically, the collaborative computing method between the main control module and the collaborative computing module includes the following steps: The main control scheduling module receives computing tasks and transforms them into a directed acyclic graph of tasks using a task graph compiler; The resource status monitor works in conjunction with the resource coordination arbitration module to obtain a globally consistent view of heterogeneous resource status. Dynamic partitioning and load balancer use a critical path dynamic perception and reconstruction algorithm to segment and distribute the task graph to each collaborative computing agent module; Each collaborative computing agent module executes its assigned subtasks in parallel through its user-defined function execution engine and hardware abstraction layer, while performing efficient data access through the main and auxiliary data structure management unit. The superstep iterative controller coordinates each round of computation for the iterative task until the task termination condition is met. The cross-layer awareness elastic fault tolerance and recovery module runs in the background to ensure the reliability of the computing process.

[0037] In specific embodiments of the present invention, a complete collaborative computing process is summarized, from task compilation, resource synchronization, dynamic scheduling, parallel execution to iterative coordination and fault tolerance assurance. It connects the innovative points in the aforementioned device claims into an automated, adaptive, and highly reliable complete computing workflow. This workflow clearly demonstrates how to automatically and efficiently decompose and implement a high-level battery management algorithm requirement into a heterogeneous computing resource cluster for execution. It embodies system-level intelligence, knowing not only how to allocate tasks but also how to ensure efficient task execution and guaranteeing system robustness even when unexpected events occur during task execution. This method-level protection ensures that this efficient and reliable task processing paradigm has lasting value and wide applicability regardless of how the underlying hardware iterates.

[0038] The above description is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and its improved concept, should be covered within the scope of protection of the present invention.

Claims

1. A high-precision data acquisition and processing system for a battery module, characterized in that, include: Distributed intelligent acquisition module, digital communication connection line, main control and collaborative computing module, and fast self-test and protection circuit; The distributed intelligent acquisition module is used to connect and acquire analog signals of electrical and environmental parameters of each battery cell or battery cell module, and convert them into raw digital data. Then, the raw data is subjected to online adaptive weighted data fusion, sparse sampling compression and feature extraction to generate situation data packets. The data acquisition cycle is dynamically adjusted according to the time series prediction model to optimize energy consumption. At the same time, the communication protocol and transmission path are dynamically selected according to the network status and data priority. The multi-hop transmission energy efficiency is optimized based on the asynchronous distributed aggregation layout construction algorithm. The situation data packets are uploaded, and the policy update event of the configuration center is monitored. The new acquisition strategy is dynamically loaded and immediately takes effect to coordinate the working sequence and energy consumption. The digital communication connection line supports fully digital, bidirectional, and multi-station communication, enabling real-time information interaction and command transmission within the system. It also employs a mechanism for rapid retransmission of change information and timed transmission of no change information to ensure real-time communication. Furthermore, it actively queries and alerts the status of adjacent nodes through a mirror reflection echolocation method to ensure network reliability. The main control and collaborative computing are used to receive global computing tasks, compile them into a task directed acyclic graph, obtain a globally consistent heterogeneous computing resource status, divide and allocate the task graph according to the critical path dynamic perception and reconstruction algorithm, execute sub-tasks after allocation, and coordinate the iterative computing process by the superstep iterative controller. The rapid self-test and protection circuit is used to continuously perform automatic checks on its own hardware, software, input / output channels, communication system and key components during the device's power-on and operation. When the self-test detects a fault or a line fault is sensed through the distributed intelligent acquisition and communication network, it can quickly identify the fault type and scope, delay trip to cut off the corresponding fault area, realize rapid isolation of the fault and restoration of power supply to the non-faulty area, and generate a fault report after the fault occurs.

2. The high-precision data acquisition and processing system for a battery module according to claim 1, characterized in that, The distributed intelligent acquisition module includes a multimodal sensing interface unit, an intelligent processing unit, an adaptive communication unit, a collaborative management unit, a hub, a host computer, and a configuration center. The multimodal sensing interface unit is used to connect to and acquire analog signals from various types of sensors through different interface protocols, including electrical parameter sensors and environmental parameter sensors, and converts the analog signals into raw digital data. The intelligent processing unit is used to perform online data fusion, compression, and feature extraction on the raw digital data to generate a concise situational data package; The adaptive communication unit is used to dynamically select the communication protocol and transmission path according to the network status and data priority, and upload the situation data packet to the hub or host computer. The collaborative management unit is connected to the multimodal sensing interface unit, the intelligent processing unit, and the adaptive communication unit respectively, and is used to coordinate the working sequence of each unit, manage the energy consumption of the module, and respond to configuration and query commands from the host computer. The hub connects multiple adaptive communication units through a star topology and polls to receive data uploaded by each module. The host computer is connected to the hub and is used to configure the acquisition parameters of each module, receive and process global data, and realize data visualization and analysis. The configuration center is connected to the host computer and stores global policy configuration parameters; The collaborative management unit of the distributed intelligent acquisition module listens to the update events of the configuration center and dynamically loads new acquisition strategies to make them take effect immediately. The data fusion adaptive weighted fusion algorithm executed by the intelligent processing unit is given by the formula: , The fused output value has the same units as the original physical quantities from the sensor. For the raw observation value of the i-th sensor, Let be the adaptive weighting coefficient of the i-th sensor, satisfying The weighting coefficient represents the number of sensors participating in the fusion. Based on the sensor's real-time confidence level and historical error Dynamic calculation: .

3. The high-precision data acquisition and processing system for a battery module according to claim 2, characterized in that, The multimodal sensing interface unit specifically includes multiple analog-to-digital conversion modules and protocol adaptation subunits; Multiple analog-to-digital conversion modules constitute a multi-input channel for converting analog signals from different sensors into digital signals; The protocol adapter subunit supports SPI, IIC, RS485 and analog voltage / current interfaces to accommodate the differences in data format and sampling rate of different modal sensors.

4. The high-precision data acquisition and processing system for a battery module according to claim 2, characterized in that, The data compression performed by the intelligent processing unit employs a sparse sampling algorithm based on compression sensing, specifically: The intelligent processing unit has a built-in one OK Random perceptual matrix of columns ,in <6, in discrete time right The original data vector of each channel Compress to obtain Dimensional measurement The perception matrix Each row has only one non-zero element, and each column has at most one non-zero element, enabling distributed random data collection where only a portion of the node data is transmitted. The measured value The unit is a linear combination of the original data units, and its dimension is... It is determined by the preset compression ratio and reconstruction accuracy requirements.

5. The high-precision data acquisition and processing system for a battery module according to claim 2, characterized in that, The intelligent processing unit also performs time series forecasting and dynamic scheduling, using the P-order autoregressive model AR(P) formula: ,in For the current moment The sensor prediction value, in the same unit as the sampled value, These are the regression coefficients of the model, dimensionless. For the past Historical observations at each moment, With a mean of 0 and a variance of A white noise sequence; The collaborative management unit compares the current actual sampled value. Compared with the predicted value ,like If the sampling period is not exceeded, the next sampling period for that channel will be automatically extended; otherwise, the sampling period will be shortened or data will be reported immediately. This is the preset error threshold.

6. The high-precision data acquisition and processing system for a battery module according to claim 1, characterized in that, The main control and collaborative computing module includes a main control scheduling module and a collaborative computing agent module; The main control scheduling module is used to receive global computing tasks, parse, segment and schedule the tasks, and monitor the status of all collaborative computing resources. The collaborative computing agent module is used to receive the segmented subtasks, drive local or remote heterogeneous computing resources to perform calculations, and return the results; The main control scheduling module has a built-in heterogeneous computing resource abstract model, which is used to uniformly describe the computing power, memory and power consumption constraints of CPU, GPU, FPGA, DSP and edge computing nodes. The collaborative computing agent module supports task dependency parsing based on directed acyclic graphs to realize automatic data flow synchronization between subtasks.

7. The high-precision data acquisition and processing system for a battery module according to claim 6, characterized in that, The main control scheduling module specifically includes a task graph compiler, a resource status monitor, an overstep iteration controller, and a dynamic partitioning and load balancer. The task graph compiler is used to compile user-submitted serial or parallel programs into a directed acyclic graph of tasks containing vertices, edges, and dependencies. The resource status monitor is used to periodically collect computing resource utilization, memory usage, network latency and power consumption data from each collaborative computing agent module, and update the heterogeneous computing resource abstract model. The superstep iteration controller is used to manage iterative computing tasks. Each iteration is defined as a superstep. The controller broadcasts a computing start instruction to all collaborative computing agent modules and collects completion signals to trigger the next superstep. The dynamic partitioning and load balancer adopts a chemometrics-inspired dynamic partitioning algorithm to divide the task directed acyclic graph into multiple subgraphs and distribute them to each collaborative computing agent module. Its objective function is to minimize the sum of global computation completion time and communication overhead.

8. The high-precision data acquisition and processing system for a battery module according to claim 6, characterized in that, The collaborative computing agent module includes a primary and secondary data structure management unit, a user-defined function execution engine, and a hardware abstraction layer and task offloading interface. The main and auxiliary data structure management unit is used to maintain the data required for computing tasks. The main data is the entity content of the data to be processed, and the auxiliary data is metadata describing the partitions, status and dependencies of the main data. The main and auxiliary data structure management unit ensures that the CPU and GPU threads can efficiently locate and load the main data based on the auxiliary data when processing in parallel, thereby achieving load balancing. The user-defined function execution engine is used to load and execute the computational kernel logic issued by the main control scheduling module. The user-defined function execution engine provides a unified programming interface, which allows the core computational logic of graph neural network aggregation and different shortest path algorithms to be encapsulated into user-defined functions for replacement and execution. The hardware abstraction layer and task offloading interface are used to shield the differences in underlying heterogeneous hardware and support the offloading of computing tasks to a dedicated coprocessor via remote procedure call protocol. The hardware abstraction layer and task offloading interface manage shared memory, hardware FIFO and doorbell interrupt mechanism to achieve efficient parameter transfer and result return between the main controller and the coprocessor. The dynamic partitioning and load balancer's critical path dynamic awareness and reconstruction algorithm has the following formula: ,in This is the estimated critical path time for the task graph, in seconds. Representative task diagram The set of all paths from the entrance to the exit. For the task vertices on the path, As vertices The estimated time for the target resources As vertices The data communication time between it and its predecessor vertex; The algorithm periodically monitors the actual execution time of each vertex. ,like This triggers a local repartition, migrating lagging vertices or their subsequent dependent vertices to a less loaded collaborative computing proxy module. This is a dynamic threshold.

9. The high-precision data acquisition and processing system for a battery module according to claim 8, characterized in that, The main control and collaborative computing module also includes a resource collaborative arbitration module, which executes a distributed resource state synchronization algorithm based on a consistency protocol. The specific steps are as follows: Each collaborative computing agent module maintains a local resource state vector. ,in To calculate utilization rate, For memory usage, The average network latency to the main controller; In each synchronization cycle, the module Its state vector Send to its neighbor module and receive the status of neighbors. ; According to the formula Update its own state estimate, where The convergence coefficient is . For connection weights; After a limited number of iterations, all modules reach an approximate consensus on the global resource load status. At the same time, all modules need to know each other's CPU utilization. The main control scheduling module makes task allocation decisions based on this consensus status. The master control scheduling module periodically instructs each collaborative computing agent module to persist the graph computing state it maintains to reliable storage. When a node failure is detected, the task and its state are rolled back to the most recent consistent checkpoint and execution is resumed on other healthy nodes. The graph computing state includes vertex state, edge state, global progress state, and persistent checkpoint. The rollback mechanism provides the BMS with an intelligent path to safely exit from an anomaly and attempt to recover to a known health baseline. It is a key strategy to prevent the escalation of failures, ensure the continuous availability of the system, and achieve safe recovery. When the collaborative computing agent module predicts that the resources of this node are about to be overloaded or fail, it migrates the delegated computing tasks and context states it carries to other edge computing platforms or cloud instances through the delegated computing application management interface and continues to predict, so as to achieve uninterrupted service.

10. A high-precision data acquisition and processing system for a battery module according to claim 9, characterized in that, The collaborative computing method between the main control module and the collaborative computing module includes the following steps: The main control scheduling module receives computing tasks and transforms them into a directed acyclic graph of tasks through a task graph compiler. The resource status monitor works in conjunction with the resource coordination arbitration module to obtain a globally consistent view of heterogeneous resource status. The dynamic partitioning and load balancer divides the task graph and distributes it to each collaborative computing agent module based on the critical path dynamic perception and reconstruction algorithm. Each collaborative computing agent module executes its assigned subtasks in parallel through its user-defined function execution engine and hardware abstraction layer, while performing efficient data access through the main and auxiliary data structure management unit. The superstep iterative controller coordinates each round of computation for the iterative task until the task termination condition is met. The cross-layer awareness elastic fault tolerance and recovery module runs in the background to ensure the reliability of the computing process.

Images