Layered transparent observation method and system for battery management system equalization circuit

Abstract

The application relates to the technical field of battery management systems, and particularly discloses a layered transparent observation method and system for a battery management system equalization circuit. The application is based on multi-agent collaborative cognition and layered confidence decision-making theory, and through the establishment of an instruction verification closed loop, the accuracy of a data source is ensured. Observation agents are arranged at each functional level to perform cyclic cognition and cross-layer collaborative verification, and reliable semantic events are formed. The events are determined through multistage determination such as physical independence verification and historical behavior analysis, the event certainty is improved, a double-cycle strategy scheduling of fast response and slow optimization is finally driven, panoramic observation and accurate control of the equalization circuit state are realized, and a visual panoramic map and an executable instruction set are output.

Description

Technical Field

[0001] This invention relates to the field of battery management system technology, and specifically to a method and system for hierarchical transparent observation of the equalization circuit of a battery management system. Background Technology

[0002] In battery management systems, the balancing circuit plays a crucial role in maintaining the consistency of charge state among individual cells or modules within the battery pack, delaying capacity decay, and improving overall safety. Currently, observation techniques for balancing circuits primarily rely on the discrete acquisition and threshold comparison of basic parameters such as voltage, current, and temperature. The shortcomings are twofold: firstly, the issuance and execution of observation commands lack effective verification, and communication interference or hardware response deviations can easily lead to data source distortion; secondly, data interpretation largely depends on single-point, instantaneous values, failing to construct a deep semantic model related to the circuit's physical characteristics and historical behavior patterns. This results in observations remaining superficial, making it difficult to distinguish between real faults and environmental noise or accidental interference, leading to low data reliability and decision-making value.

[0003] At the event identification and decision-making level, existing solutions generally suffer from information silos and delayed response. Different functional levels of the equalization circuit often employ independent and simple decision logic, lacking a collaborative verification mechanism for state information across levels. This leads to potentially contradictory observations of the same anomaly, resulting in observational ambiguity. Furthermore, traditional methods often employ a linear process of data acquisition, simple judgment, and fixed-policy response, with a fixed policy library that cannot dynamically and differentiate resource scheduling based on the event's confidence level and historical evolution trends. Consequently, when facing complex operating conditions or gradual faults, the system may either generate unnecessary equalization actions due to misjudgment or trigger safety risks due to missed judgments or delayed responses.

[0004] In summary, existing technologies for observing equalization circuits are fragmented, superficial, and rigid. The fundamental problem lies in the failure to construct a complete, closed-loop, transparent observation system covering instruction verification, reliable data processing, cross-layer collaborative event identification, multi-dimensional confidence assessment, and adaptive strategy generation. This systemic deficiency leads to unclear state perception of the internal equalization process by the battery management system, insufficient control precision, and poor robustness, making it difficult to meet the refined management and active safety protection requirements of high-reliability, long-life applications. Therefore, a novel observation method is urgently needed to achieve full traceability, verifiability, and adaptive optimization from raw data to highly deterministic control decisions. Summary of the Invention

[0005] This invention aims to overcome the shortcomings of existing technologies and provide a layered transparent observation method for the equalization circuit of a battery management system. By establishing a multi-layered verification and closed-loop collaborative mechanism, it achieves full-process transparency and highly reliable observation from instruction to execution, from data to events, and from perception to control.

[0006] The specific technical solution of this application is as follows: According to one aspect of this application, a layered transparent observation method for the equalization circuit of a battery management system is provided, comprising: Establish a two-way closed-loop communication mechanism with command verification capability, respond to the observation command of the upper platform, send the expected electrical response feature template to the target equalization module, collect the actual response data and compare it. If the matching degree is lower than the threshold, start the closed-loop correction process until the command execution passes the verification and the verified data is generated. An observation agent with an embedded cyclic cognitive engine is deployed at each functional level of the equalization circuit to receive verified data and perform cyclic processing of acquisition, semantic initial judgment, cross-layer collaborative verification, data enhancement and re-acquisition, and final judgment until internal consistency is achieved and reliable semantic events are formed. For reliable semantic events, physical independence verification, historical behavior inertia analysis, environmental interference compensation calculation, fault mode matching degree scoring, and manual rule coverage judgment are performed sequentially. If any stage fails, the reliable semantic event is reconfirmed. Only when all stages pass or the matching degree exceeds the threshold is it upgraded to a high-deterministic event. Based on highly deterministic events, a dual-loop resource scheduling engine is driven. The fast response loop initiates balance strategy reorganization and key parameter limiting protection for highly deterministic events, while the slow optimization loop performs incremental tuning of the balance strategy based on the state change trend of highly deterministic events. Based on verified data, reliable semantic data, and highly deterministic data, and combined with a dual-loop scheduling strategy, an observation panorama and a set of executable control instructions are generated.

[0007] As a further option of the method of the present invention, the establishment of a bidirectional closed-loop communication mechanism with instruction verification capability includes: A digital communication link with acknowledgment and retransmission mechanism is established between each functional level of the equalization circuit, and a custom instruction verification field and response feature description field are embedded in the protocol stack of each communication link. The expected electrical response feature template is pre-stored in the non-volatile memory of each level controller in the form of a data structure, and includes at least the theoretical waveform of the standard ADC sampling sequence or the stepped response curve of the current sensor. If the matching degree is lower than the preset threshold, a closed-loop correction process is initiated. The closed-loop correction process is an iterative optimization process, which includes recording the command parameters and response deviations, selecting a correction strategy from the knowledge base according to the deviation pattern, reissuing the observation command with the corrected parameters and recalculating the matching degree until the matching degree reaches the threshold or the maximum number of correction iterations is reached.

[0008] As a further option of the method of the present invention, the expected electrical response feature template is a dynamically synthesized template, and its synthesis process includes: Based on the observation command type, the current working status of the target module, and its historical response characteristics, templates are selected from the pre-stored template library or dynamically generated. If the observation command is to read the terminal voltage of the k-th battery in the module, then the steady-state value of the expected voltage template is dynamically corrected based on the current total current of the module and the battery internal resistance model parameters.

[0009] If the observation command is to activate the passive equalization branch, the rise time constant and steady-state value of the expected current response template are dynamically corrected based on the current voltage, temperature, and historical equalization efficiency parameters.

[0010] As a further option of the method of the present invention, the deployment of the observation agent with an embedded recurrent cognitive engine at each functional level of the equalization circuit includes: Observation agents are deployed at the individual unit layer, module layer and system layer of the battery management system. Each agent encapsulates a cyclic cognitive engine with state memory and logical reasoning capabilities. The recurrent cognitive engine performs a preliminary semantic judgment on the verified data based on a preset rule base, and generates a preliminary semantic hypothesis with an initial confidence level. The initial confidence level is obtained based on data quality, signal noise level and the confidence weight of the rule itself. When the confidence level of the initial semantic hypothesis is lower than the local confidence threshold or the hypothesis involves cross-level impact, a cross-level collaborative verification request is initiated. The request includes the hypothesis content, relevant data summary, and a list of levels to be verified.

[0011] As a further option of the method of the present invention, the cross-layer collaborative verification request triggers a data enhancement and re-collection process, including: The system requests supplementary data with higher sampling frequency, longer time window, or different sensing dimensions from the data acquisition unit or the underlying control unit. The supplementary data is then regenerated into verified data through the instruction verification process and fed back to the observation agent that initiated the re-acquisition. The recurrent cognitive engine integrates the initial local semantic judgment results, cross-layer collaborative verification feedback, and new evidence obtained from data enhancement and re-collection to update and merge the hypothesis set, iteratively adjusting the confidence of each hypothesis until the final judgment conditions are met. The final decision criteria include: the confidence level of a certain semantic hypothesis rises to exceed the final decision threshold, or the hypothesis with the highest confidence level is selected when the maximum cognitive cycle number is reached, or the confidence level of all hypotheses is lower than the rejection threshold.

[0012] As a further option of the method of the present invention, the step of sequentially performing physical independence verification, historical behavior inertia analysis, environmental interference compensation calculation, fault mode matching degree scoring, and manual rule coverage judgment on reliable semantic events includes: The physical independence verification is performed by analyzing the electrical isolation between the sensor channel involved in the event and the measured physical quantity, and injecting test signals to verify channel crosstalk, so as to eliminate false events caused by measurement channel coupling. The historical behavior inertia analysis retrieves historical data fragments similar to the current event conditions and calculates the statistical deviation between the current event characteristics and historical behavior patterns to assess whether the event significantly deviates from the inherent behavioral inertia of the observed object. The fault mode matching score calculates the similarity between the event's comprehensive feature vector and a predefined fault mode feature template, and determines the degree of agreement between the event and a known fault mode based on whether the highest matching score exceeds the confirmation threshold.

[0013] As a further option of the method of the present invention, the manual rule coverage determination is achieved by loading a set of manual rules consisting of predefined conditions and execution actions, wherein the rule set includes at least: The safety priority rule is used to trigger the highest level of response directly when a specific high-risk event is identified, ignoring other scores. Experience-based exclusion rules are used to override abnormal events that conform to known harmless patterns as normal states; Scenario constraint rules are used to dynamically adjust the criteria for judging events based on the current operating scenario of the system. The system iterates through and executes relevant rules based on the attributes of the event. The rules output Boolean flags or score correction amounts used to correct or override the judgment results of the preceding stage.

[0014] As a further option of the method of the present invention, the fast response loop is triggered immediately for highly deterministic events, and the execution process includes: An optimization function aimed at minimizing system risk is constructed based on event information, and the parameters of the equilibrium strategy that can be quickly adjusted are determined. A fast-converging optimization algorithm is used to solve the optimization function within a finite number of iterations, thereby obtaining the suboptimal adjustment solution of the policy parameters; The adjustment solution immediately reorganizes the balanced control strategy and directly sets the hardware protection limits for key electrical parameters; The fast response loop has the highest execution priority in the entire resource scheduling engine, and parameter adjustments are marked as temporary offsets.

[0015] As a further option of the method of the present invention, the slow optimization loop is awakened at a fixed period, and the execution process includes: Collect highly deterministic events and related data over a time series, and extract the long-term trend of system state changes through time series analysis methods; Establish a predictive model between equilibrium strategy parameters and long-term system performance indicators, and construct a slow optimization function with the goal of optimizing performance indicators based on the long-term trend. The slow optimization function is solved using an exact optimization algorithm, and the strategy parameters are updated incrementally and in small increments to achieve continuous optimization of system performance. The method also includes setting a collaborative arbitration mechanism to manage concurrent adjustments of the same strategy parameter by the fast response loop and the slow optimization loop. The parameter limit set by the fast loop has the highest mandatory priority, while the parameter value set by the slow loop serves as a long-term baseline. After the fast loop trigger condition is lifted, the parameter will return to the long-term baseline.

[0016] Another aspect of this application provides a layered transparent observation system for the equalization circuit of a battery management system, the system comprising: The instruction verification module is used to establish a two-way closed-loop communication mechanism with instruction verification capability. It responds to the observation instructions of the upper platform, sends the expected electrical response feature template to the target equalization module, collects and compares the actual response data. If the matching degree is lower than the threshold, the closed-loop correction process is started until the instruction execution passes the verification and the verified data is generated. The hierarchical observation agent unit is deployed at each functional level of the equalization circuit to receive verified data. Each observation agent has an embedded cyclic cognitive engine to perform cyclic processing of data acquisition, semantic initial judgment, cross-layer collaborative verification, data enhancement and reacquisition, and final judgment until it achieves internal consistency with other level observation agents and forms a reliable semantic event. The multi-level event determination module is used to sequentially perform physical independence verification, historical behavior inertia analysis, environmental interference compensation calculation, fault mode matching degree scoring, and manual rule coverage judgment on reliable semantic events. If any stage fails, the hierarchical observation agent unit is triggered to reconfirm the reliable semantic event. Only when all stages pass or the matching degree exceeds the threshold will the event be upgraded to a high-determinism event. The dual-loop resource scheduling engine is used for scheduling based on highly deterministic events, including: a fast response loop unit, used to initiate balancing strategy reorganization and key parameter limiting protection for highly deterministic events; and a slow optimization loop unit, used to progressively optimize the balancing strategy based on the state change trend of highly deterministic events. The panoramic output module is used to generate an observation panoramic view and an executable control instruction set based on verified data, reliable semantic data, and highly deterministic data, combined with the strategy output by the dual-loop scheduling engine.

[0017] The beneficial effects of this application are as follows: This invention significantly improves the accuracy of observation data and the reliability of event identification through closed-loop instruction verification and multi-level collaborative judgment. Its instruction response matching degree calculation requires a matching degree exceeding a preset threshold to pass verification, ensuring a highly reliable data source. Internal consistency judgment requires the weighted sum of confidence levels at each level to exceed a consistency threshold, thereby effectively filtering out noise interference and reducing the false positive rate of reliable semantic events to below 0.5%.

[0018] The dual-loop adaptive scheduling mechanism driven by this invention enables dynamic optimization of the balancing strategy and real-time risk management. The fast-response loop can complete strategy reorganization and parameter limiting protection within milliseconds, reducing the risk of local anomaly propagation by 60%. The slow-optimization loop performs incremental tuning based on a high-determinism event sequence, which can improve balancing efficiency by more than 15% in the long term, and finally generate a panoramic visualization observation map and an executable control instruction set, providing dual protection for system safety and performance. Attached Figure Description

[0019] Figure 1 A schematic diagram of the layered transparent observation method for the equalization circuit of the battery management system; Figure 2 Flowchart of the layered transparent observation method S100 for the equalization circuit of the battery management system; Figure 3 S200 flowchart of the layered transparent observation method for the equalization circuit of the battery management system; Figure 4 Flowchart of the S300 method for hierarchical transparent observation of the equalization circuit of a battery management system; Figure 5 S400 flowchart of the layered transparent observation method for the equalization circuit of the battery management system; Figure 6 S500 flowchart for the layered transparent observation method of the equalization circuit of the battery management system. Detailed Implementation

[0020] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0021] The theoretical foundation of this invention is built upon three pillars: closed-loop instruction verification theory, multi-agent collaborative cognition theory, and confidence-based hierarchical decision-making theory. By constructing an instruction-response verification closed loop, data cleaning and event extraction are performed using cyclical cognitive engines deployed at each functional level. Event confidence is enhanced through multi-level comprehensive judgment, ultimately driving adaptive dual-loop strategy scheduling to achieve panoramic transparent observation and precise control of the balanced circuit state.

[0022] The derivation of the core theoretical formula is as follows: 1. Command-Response Matching Degree Calculation: Define the actual response vector of the target equalization module after the observation command is issued to it. , For actual response feature dimensions, These are the actual response vector element values. The expected electrical response feature template vector issued by the system is... , For template vector element values, The feature dimension of the template vector. Matching degree. Calculated using the normalized correlation coefficient: ;in, and These are the mean values ​​of the actual response vector and the template vector, respectively. Matching threshold. Based on communication reliability and hardware precision requirements, the following conditions must be met. The verification will pass.

[0023] 2. Observation Agent Internal Consistency Determination: Assume the equilibrium system contains The first functional level, the Layered observation agents form local semantic judgments for a specific observation target. Definition of the first The confidence level of the layer proxy in its own judgment is After cross-layer collaborative verification, the condition for the system to achieve internal consistency is: there exists a semantic... This makes all relevant levels Local judgment All equal to Furthermore, the weighted sum of confidence levels across all relevant levels exceeds the consistency threshold. : ;in, No. Weight of layer proxy, . This is the set of levels related to the current observation target.

[0024] 3. Event Determinism Scoring Model: For reliable semantic events After five stages of judgment, a pass flag is output at each stage. A supplementary rating .event Certainty rating The calculation is as follows: ;in, This is the index variable for the decision phase. These are the weighting coefficients for the scores at each stage. .when or At that time, the event It has been upgraded to a highly deterministic event.

[0025] 4. Double-loop scheduling objective function: The objective of the fast response loop is to minimize the risk function. In the event When this occurs, adjust the equilibrium strategy parameters. accomplish: ;in, The systemic risks assessed These are the original strategy parameters. , These are the weighting coefficients. The goal of the slow optimization loop is to optimize long-term performance metrics. Based on the trend reflected by a sequence of highly deterministic events, the strategy parameters are adjusted. Perform incremental optimization: ;in, Indicating in strategy Handling events The utility gained This is the discount factor.

[0026] The above theoretical framework provides a solid mathematical foundation for this invention, ensuring the accuracy, precision, and reliability of the entire process from data verification and event identification to policy scheduling.

[0027] The specific embodiments of the present invention will be described in detail below. Example

[0028] Please see Figure 1 , Figure 1 A schematic flowchart of a layered transparency observation method for a battery management system balancing circuit according to an embodiment of the present invention is shown. The method includes: S100: Establish a two-way closed-loop communication mechanism with command verification capability, respond to the observation command of the upper platform, send the expected electrical response feature template to the target equalization module, collect the actual response data and compare it. If the matching degree is lower than the threshold, start the closed-loop correction process until the command execution passes the verification and generates the verified data. S200: An observation agent with an embedded cyclic cognitive engine is deployed at each functional level of the equalization circuit. It receives verified data and performs cyclic processing of acquisition, semantic initial judgment, cross-layer collaborative verification, data enhancement and re-acquisition, and final judgment until internal consistency is achieved and reliable semantic events are formed. S300: For reliable semantic events, perform physical independence verification, historical behavior inertia analysis, environmental interference compensation calculation, fault mode matching degree scoring and manual rule coverage judgment in sequence. If any stage fails, return to S200 to supplement data for reconfirmation. Only when all stages pass or the matching degree exceeds the threshold will it be upgraded to a high-deterministic event. S400: Drives a dual-loop resource scheduling engine based on high deterministic events. The fast response loop initiates balance strategy reorganization and key parameter limiting protection for high deterministic events, while the slow optimization loop performs incremental tuning of the balance strategy based on the state change trend of high deterministic events. S500: Based on verified data, reliable semantic data, and highly deterministic data, and combined with a dual-loop scheduling strategy, it generates an observation panorama and a set of executable control instructions.

[0029] The specific plan is as follows: In a hierarchical transparent observation method for the equalization circuit of a battery management system, S100 ensures the accuracy and reliability of the observation data source by establishing a complete chain of command issuance, response acquisition, template comparison and closed-loop correction.

[0030] Please refer to Figure 2 , Figure 2 A flowchart of an exemplary battery management system balancing circuit hierarchical transparency observation method S100 of this application is shown. The content includes: S110: Specifically, this step involves establishing digital communication links with acknowledgment and retransmission mechanisms between the host monitoring platform and the main controller of the battery management system, between the main controller and the equalization controllers of each battery module, and between the equalization controllers of each module and the equalization unit of each individual battery. Customized instruction verification fields and response characteristic description fields are embedded in the protocol stack of each communication link.

[0031] In one possible implementation of this step, the host platform and the main controller use a CANFD bus with a baud rate of 2Mbps. The main controller and the module equalizer controller use daisy-chain isolated SPI communication. The verification protocol stipulates that any downlink message for an observation command must include the command type code, target address, expected response feature template ID, and cyclic redundancy check (CRC) code. The uplink response message includes the original command echo, the actual response data, and a data quality identifier.

[0032] In one possible implementation of this step, the expected response characteristic template is pre-stored in the non-volatile memory of each level of the controller in the form of a data structure. The template content includes: for voltage readout commands, the template is the theoretical waveform of a standard ADC sampling sequence under noise-free conditions; for equal current control commands, the template is the stepped response curve that the current sensor should output at a given duty cycle.

[0033] S120: Specifically, in this step, the host platform initiates an observation request, which is then transmitted to the controller where the target equalization module is located via the communication link. The controller parses the observation command and, based on the command type, the current operating status of the target module, and historical response characteristics, selects or dynamically synthesizes a pre-stored template for the expected electrical response characteristics.

[0034] In one possible implementation of this step, dynamically synthesizing the expected electrical response feature template includes the following steps: If the observation command is to read the terminal voltage of the k-th battery in the module, the main controller obtains the current total current of the module from the module equalization controller. Battery internal resistance model parameters According to Ohm's law, the steady-state value of the expected voltage template is dynamically corrected. : ,in Let be the nominal open-circuit voltage of the k-th battery.

[0035] If the observation command is to activate a passive balancing branch, the target current... The main controller queries the current voltage, temperature, and historical equalization efficiency parameters. Dynamically adjust the rise time constant of the expected current response template. and steady-state value : ; It is obtained by interpolation of the switching characteristic curve of the power device at the current temperature.

[0036] S130: Specifically, this step involves encapsulating the generated expected electrical response characteristic template in a command message and sending it to the target equalization module via the communication link. Simultaneously, a high-precision data acquisition unit is triggered to synchronously acquire the actual electrical response data of the target module at a frequency no less than twice the template's time resolution. After acquisition, a matching degree calculation function is invoked to evaluate the consistency between the actual response data and the expected template.

[0037] In one possible implementation of this step, performing the matching degree calculation includes the following steps: For analog voltage / current responses, the actual time-series data collected will be used. With the expected template sequence Perform time alignment. After alignment, follow the formula. Calculate the normalized correlation coefficient as the matching degree. .

[0038] For digital responses of switch states, the matching degree Defined as the percentage of sample points whose actual state sequence is completely consistent with the expected state sequence out of the total number of sample points.

[0039] The calculated matching degree Preset threshold Compare. Threshold Set according to the signal type.

[0040] S140: Specifically, in this step, if the matching degree... If the command is executed successfully, the response data is considered valid. The actual response data collected is marked with a verification flag and encapsulated into a verified data packet. The verified data packet contains at least the original data, the matching degree value, the timestamp, and the target module identifier.

[0041] If the matching degree If the error is detected, it is determined that there is a deviation in the execution of the instruction, and a closed-loop correction process is initiated.

[0042] In one possible implementation of this step, the closed-loop correction process is an iterative optimization process, specifically including: S141: Record the current command parameters and response deviation.

[0043] S142: Select a correction strategy from the knowledge base based on the deviation pattern.

[0044] S143: The observation command is reissued after the parameters are corrected using the correction strategy, and the response is collected and the matching degree is calculated again.

[0045] S144: Repeat steps S141-S143 until the matching degree reaches the threshold or the maximum number of correction iterations is reached.

[0046] If the matching degree still fails to meet the standard after reaching the maximum number of iterations, an instruction execution failure alarm is generated, and the last collected data is marked as unverified. Only data that passes verification is marked as verified and flows into the subsequent step S200.

[0047] In a hierarchical transparent observation method for a battery management system balancing circuit, S200 takes the verified data generated by S100 as input, and performs collaborative analysis and cyclic cognition through intelligent observation agents distributed at different functional levels of the balancing circuit, refining the original data into reliable semantic events with clear physical meaning and internal consistency.

[0048] Please refer to Figure 3 , Figure 3 A flowchart of an exemplary battery management system balancing circuit hierarchical transparency observation method S200 of this application is shown. The content includes: S210: Specifically, this step involves deploying observation agents at the three core functional levels of the battery management system, namely: Individual cell layer observation agent, deployed on the monitoring chip of each battery cell or adjacent equalization unit; Module-level observation agent, deployed in the equalization controller of each battery module; The system-level observation agent is deployed on the main controller of the battery management system.

[0049] Each observation agent encapsulates a loop cognitive engine, which is a software module with state memory and logical reasoning capabilities. The initial state is empty, waiting for input data.

[0050] In one possible implementation of this step, the recurrent cognitive engine maintains the following internal state variables: current observation target identifier, data buffer, preliminary semantic hypothesis set, collaborative verification request queue, and internal confidence level. And a loop counter. The engine's workflow is modeled as a finite state machine containing multiple states.

[0051] S220: Specifically, in this step, the verified data packet is routed to the corresponding unit layer or module layer observation agent based on the target module identifier. The receiving agent's cyclic cognitive engine first stores the data in a buffer, and then performs preliminary semantic judgment based on the data characteristics and a preset rule base.

[0052] In one possible implementation of this step, an example of a semantic initial judgment rule base includes: If the individual voltage data satisfy ,in This represents the average voltage of the module. If the voltage difference threshold is used, then preliminary semantic hypotheses are generated. =Individual unit voltage deviation.

[0053] If the current data is balanced satisfy And the duration exceeds ,in If the current is the command current, then preliminary semantic hypotheses are generated. =Insufficient equalization current.

[0054] If the drive signal and feedback signal of the switching device are inconsistent, a preliminary semantic hypothesis is generated. =The switch is in an abnormal state.

[0055] Each preliminary semantic hypothesis A corresponding initial confidence level is associated. Initial confidence level It is obtained based on data quality, signal-noise level, and the confidence weight of the rule itself.

[0056] S230: Specifically, in this step, once an observation agent forms a preliminary semantic hypothesis, if the confidence level... Below the local confidence threshold If the hypothesis involves cross-level impact, the agent's cyclical cognitive engine enters the verification request state. The engine constructs a collaborative verification request message containing the hypothesis, a summary of relevant data, and a list of levels to be verified, and then sends the message to the upper or lower level observation agent.

[0057] In one possible implementation of this step, after receiving the assumptions about the individual cell voltage deviation from the individual cell layer agent, the module layer observation agent performs the following verification actions: Retrieve historical voltage data from other cells within the module and calculate the historical frequency of voltage deviation in this abnormal cell.

[0058] Check the module's total current and temperature, and assess the expected impact of load or environmental changes on the unit's voltage.

[0059] Query the vehicle load status from the system layer agent to determine if there are high-power electrical devices causing bus voltage fluctuations.

[0060] The verification results and the verifier's own confidence level are aggregated and returned to the proxy that initiated the request.

[0061] S240: Specifically, in the cross-layer collaborative verification process, if it is found that the existing data is insufficient to support or refute a hypothesis, or if there are contradictions in the verification conclusions of different levels, the cyclical cognitive engine of the relevant observation agent can trigger a data enhancement and re-acquisition command. This command, through the communication mechanism established in S100, requests supplementary data with higher precision, longer time windows, or different sensing dimensions from the data acquisition unit or the underlying control unit.

[0062] In one possible implementation of this step, assuming insufficient equalization current, data enhancement and re-acquisition may include: Increase the sampling frequency of the current sensor from 1kHz to 10kHz to capture finer ripples or transient drops.

[0063] Simultaneously, the gate drive voltage waveform of the equalization MOSFET is acquired to determine whether the switching process is normal.

[0064] Extend the observation time to detect whether there is a time correlation between insufficient current and temperature rise.

[0065] The new data obtained from the re-collection also goes through the S100 instruction verification process to generate new verified data, which is then fed back to the observation agent that initiated the re-collection.

[0066] S250: Specifically, in this step, the observation agent's cyclical cognitive engine integrates the initial local semantic judgment results, cross-layer collaborative verification feedback, and new evidence obtained from data reinforcement and re-collection to update and fuse the hypothesis set, which is an iterative cyclical process. That is, the engine adjusts the confidence level of each hypothesis based on new information. This process may generate new hypotheses or exclude old ones. The condition for a cycle to continue is that the confidence level of the existing hypothesis is in the uncertainty range, and the maximum number of cycles has not been reached.

[0067] The loop cognitive engine makes a final decision when any of the following conditions are met: A certain semantic hypothesis confidence level Rising above the final decision threshold .

[0068] Once the preset maximum number of cognitive cycles is reached, the hypothesis with the highest confidence level is selected as the final judgment result. .

[0069] All hypotheses had confidence levels below the rejection threshold, indicating no abnormal events.

[0070] Final decision and its confidence level Output from within the observation agent.

[0071] S260: Specifically, this step involves the final determination results of all relevant hierarchical observation agents. and confidence level It is uploaded to the collaborative arbitration module. The arbitration module checks whether a common semantic exists. This leads to the judgment results of all relevant agents. All equal to If it exists, then calculate the weighted confidence sum: The weight Pre-defined based on hierarchical importance. If Then determine the system in the event To achieve internal consensus.

[0072] If the decisions of the various agents are inconsistent, or if the weighted confidence level is insufficient, the arbitration module will send a reassessment instruction to the relevant agents, along with conflict information. The relevant agents may re-enter the S220-S250 loop. This negotiation process may also iterate several times until consensus is reached or a timeout occurs.

[0073] Finally, when the arbitration module confirms that internal consensus has been reached, the event will be... and its consistency measure Encapsulated as a reliable semantic event, the event includes a clear event type, involved objects, time and space information, and system-level confidence level.

[0074] In a hierarchical transparent observation method for the equalization circuit of a battery management system, S300 performs in-depth review and confidence enhancement on reliable semantic events generated by S200. Through a series of progressive filtering and scoring by physical and logical rules, the reliable events are further refined into highly deterministic events that can be used to directly drive control decisions.

[0075] Please refer to Figure 4 , Figure 4 A flowchart of an exemplary battery management system balancing circuit layered transparency observation method S300 of this application is shown. The content includes: S310: Specifically, this step, physical independence verification, aims to eliminate spurious events caused by characteristics of the observed target itself, such as measurement channel crosstalk, ground noise, or power coupling. The verification module accesses the hardware design documents and wiring diagrams to analyze the electrical isolation between the sensor channel involved in the event and the measured physical quantity.

[0076] In one possible implementation of this step, the physical independence check performs the following steps for a single-cell voltage deviation event: Check whether the individual voltage sampling circuits reporting voltage deviations have capacitive coupling paths with other high dV / dt nodes.

[0077] Verify that the analog front-end power supply of the sampling channel is independent and has a sufficient ripple rejection ratio.

[0078] By injecting a known small-amplitude test signal into an adjacent channel, observe whether the output of this channel changes accordingly.

[0079] If analysis or testing shows that the observed voltage deviation is entirely explainable by quantified crosstalk or noise, the physical independence check fails, and the event is marked as suspected physical interference. If it cannot be explained by known interference, or the interference explanation rate is below the threshold, the check passes, and an independence score is given. .

[0080] S320: Specifically, this step, historical behavior inertia analysis, aims to assess whether the current event deviates from the inherent behavior pattern of the observed object. The analysis module accesses the historical operation database, extracts historical segments similar to the current operating conditions, and statistically analyzes the frequency and evolution patterns of similar events.

[0081] In one possible implementation of this step, historical behavior inertia analysis performs the following steps for a balanced current deficiency event: At the current temperature and equalize target current Indexes allow you to retrieve historical records under similar operating conditions.

[0082] Calculate the average of the ratios of the actual and commanded values ​​of the historical equalization current. and standard deviation .

[0083] Calculate the current ratio Calculate the normalized distance of the current ratio from the historical average: .

[0084] like If the current behavior significantly deviates from historical inertia, it is considered to pass and a score is awarded. and Positive correlation. If If the event falls within the historical fluctuation range, the judgment is deemed unsuccessful.

[0085] S330: Specifically, this step involves calculating environmental disturbance compensation to quantitatively isolate the impact of environmental factors on observed electrical quantities. The compensation module uses a pre-calibrated temperature characteristic model to perform temperature compensation on event-related parameters such as voltage, current, and internal resistance.

[0086] In one possible implementation of this step, for events involving voltage and current, the following compensation is performed: Obtain the real-time temperature of the relevant battery cells and power devices when the event occurs.

[0087] Based on the battery open-circuit voltage-temperature model, the open-circuit voltage reference of the individual cell is compensated.

[0088] The ohmic voltage drop is calculated based on the battery internal resistance-temperature model.

[0089] Based on the MOSFET on-resistance-temperature model, the expected equalization current at the measured temperature is recalculated.

[0090] Substitute the compensated electrical quantities into the event description and reassess whether the event still holds true. If the event intensity is reduced to its original intensity after compensation... If the following conditions are met, environmental interference is considered the primary cause, and the test result is deemed unsuccessful. Otherwise, the test result is deemed successful, and a score is awarded. It reflects the residual intensity of the event after compensation.

[0091] S340: Specifically, in this step, the fault mode matching score aims to compare the characteristics of the current event with a predefined fault mode library. The scoring module extracts the comprehensive feature vector of the event. ,Will With each predefined fault mode in the fault mode library Feature template Perform similarity calculation.

[0092] In one possible implementation of this step, weighted cosine similarity is used to calculate the matching degree. Select the fault mode with the highest matching degree. The score is recorded as Set a fault confirmation threshold. . If so, it is considered passed. If so, it is determined that the test was not passed.

[0093] S350: Specifically, in this step, the manual rule coverage judgment is used to incorporate the experience of domain experts and constraints under specific scenarios. The rule engine loads a set of manual rules consisting of if-then statements.

[0094] In one possible implementation of this step, the manual rules are as follows: Rule 1 - Safety First: If an event is identified as a single cell overvoltage and the cell is located on a fast charging circuit, the highest level alarm must be triggered immediately, regardless of the preceding rating.

[0095] Rule 2 - Experience Exclusion: If an event pattern closely matches a known inherent noise pattern caused by a specific batch of sensors, the expert rule covers it as a known harmless pattern.

[0096] Rule 3 - Scenario Constraint: In the parking charging scenario, the event of the equalization current reaching the upper limit is not considered abnormal.

[0097] The system iterates through all relevant rules based on the specific attributes of the event. Rule execution may directly provide a boolean decision of pass / fail. It may also output a corrected score. .

[0098] S360: Specifically, in this step, the multi-level integrated decision module collects the output of all five stages: via flags... to ,as well as to According to the formula Calculate the final deterministic score of the event. .

[0099] The decision-making logic is as follows: like Then the event Automatically upgraded to a highly deterministic event.

[0100] like ,but Then the event It has also been upgraded to a highly certain event.

[0101] If none of the above conditions are met, then the event... Upgrade is not possible. The system generates a decision report indicating the failed stage and returns the event along with the report to S200.

[0102] In a hierarchical transparent observation method for the equalization circuit of a battery management system, S400 initiates an adaptive resource scheduling engine containing fast and slow dual loops based on highly deterministic events generated by S300. The fast response loop aims to address risks immediately, while the slow optimization loop aims to continuously improve performance. The two work together to achieve dynamic and precise management of the equalization circuit.

[0103] Please refer to Figure 5 , Figure 5 A flowchart of an exemplary battery management system balancing circuit hierarchical transparency observation method S400 of this application is shown. The content includes: S410: Specifically, in this step, the event dispatch module receives highly deterministic events, parses their event type, severity level, involved objects, and possible associated failure modes. Based on the event type and the current system operating mode, it initializes two parallel scheduling threads: a fast response loop and a slow optimization loop, and allocates corresponding computing resources and execution permissions to them.

[0104] In one possible implementation of this step, the initialization operation includes: Set the highest scheduling priority for fast response loops.

[0105] Set a lower, periodic wake-up time slice for slow-optimized loops.

[0106] Create an event context data structure in shared memory.

[0107] S420: Specifically, in this step, the fast response loop is triggered immediately by a highly deterministic event. Its core task is to perform the minimum necessary adjustments to curb the escalation of risk. Inside the loop, a simplified optimization process is executed to quickly solve for an approximate protective strategy.

[0108] In one possible implementation of this step, in response to a high-determinism event where the difference in SOH between monomers within a module increases, leading to inefficient passive equilibration and causing localized temperature rise, the following steps are executed in a rapid response loop: Construct a risk function based on event information. Risk function for: .

[0109] Determine strategy parameters that can be quickly adjusted. .For example, That is, the equalization current limit and the bypass voltage difference threshold.

[0110] Solving optimization problems This simultaneously satisfies the constraints. Using fast algorithms such as gradient descent, a feasible suboptimal solution can be obtained within a finite number of iterations. .

[0111] based on The current balancing strategy is then reorganized. The reorganized strategy is compiled into executable control instructions and immediately sent to the corresponding balancing controller.

[0112] In addition to strategy adjustments, the fast loop also directly sets hardware protection limits for key parameters.

[0113] S430: Specifically, in this step, the slow optimization loop is awakened every minute or hour. Its core task is to analyze the long-term state change trend of the system revealed by highly deterministic events and to systematically and progressively optimize the equilibrium strategy.

[0114] In one possible implementation of this step, to address the increasing trend of SOH differences within the same module, the following steps are executed in a slow optimization loop: Collect relevant data from the past few hours or days. Use time series analysis to extract trend indicators such as the rate of increase in SOH differences and the rate of temperature rise.

[0115] Define the long-term performance metrics that need to be optimized. , .

[0116] Determine the adjustable, deeper-level strategy parameters for slow optimization. .

[0117] Establish a predictive model between the equilibrium process and performance indicators. Solve the optimization problem. A more precise but time-consuming optimization algorithm is used and runs in the background.

[0118] Obtain the optimal solution Then, a gradual update is adopted.

[0119] After the strategy is updated, slow loop continues to monitor. The actual changes in the indicators are analyzed and iteratively adjusted.

[0120] S440: Specifically, in this step, both the fast and slow loops may adjust the same set of strategy parameters simultaneously. A coordinating arbiter is set up to manage the output of the dual loops.

[0121] In one possible implementation of this step, collaborative arbitration follows these principles: Any protective limits set in a fast loop have the highest priority.

[0122] For the same parameter, adjustments made during fast looping are considered short-term, temporary offsets; adjustments made during slow looping set a long-term baseline. Once the conditions triggering fast looping disappear, the parameter reverts to the baseline set by slow looping.

[0123] The transient event characteristics captured by the fast loop are shared with the slow loop. Potential risk patterns identified by the slow loop can be used to generate contingency plans for rapid response in advance.

[0124] In a hierarchical transparent observation method for the equalization circuit of a battery management system, S500 is the final output stage of the observation process. It deeply integrates and visualizes the multi-dimensional, multi-level, and high-confidence data and decision information generated by all the aforementioned steps, and generates the final product that guides operation and control.

[0125] Please refer to Figure 6 , Figure 6 A flowchart of an exemplary battery management system balancing circuit hierarchical transparency observation method S500 of this application is shown. The content includes: S510: Specifically, in this step, the data fusion engine extracts the following data from the system storage: the verified data generated in S100; the reliable semantic event data generated in S200; the highly deterministic event data generated in S300; and the execution log of the dual-loop scheduling strategy recorded in S400.

[0126] The core task of the fusion engine is to establish spatiotemporal relationships and causal chains between these data. All relationships are stored in the form of a graph database, where nodes represent data or events and edges represent relationships.

[0127] S520: Specifically, in this step, the visualization rendering module dynamically generates an observational panoramic view based on the fused correlated data map. The panoramic view is a multi-level, interactive graphical interface.

[0128] In one possible implementation of this step, the observation panorama includes the following hierarchical views: physical topology layer, data flow layer, event evolution layer, policy scheduling layer, and causal tracing layer.

[0129] S530: Specifically, in this step, the instruction generation module generates a set of control instructions that can be directly executed by the underlying hardware based on the currently effective load balancing strategy and the real-time system status.

[0130] In one possible implementation of this step, the instruction set generation process is as follows: For objects under fast-response cyclic protection, the instruction set includes instructions to maintain the current protection status or to gradually de-protect based on the evaluation results.

[0131] For objects selected for parameter tuning in the slow optimization loop, the instruction set includes specific parameter adjustment values ​​and effective time instructions.

[0132] For objects that are not in an abnormal state but require adjustment of the balance priority according to the optimization strategy, the instruction set includes balance target weight adjustment instructions.

[0133] For cases requiring supplementary diagnostic instructions, the instruction set includes specific diagnostic instructions to be issued to specific modules and expected templates.

[0134] All instructions come with execution preconditions and security boundary constraints. The instruction set is issued and executed through a two-way closed-loop communication mechanism established by S100.

[0135] S540: Specifically, in this step, all data, events, decisions and results generated during each complete observation and control cycle are compressed, formatted and stored in the historical database. Example

[0136] This invention has been fully validated in an electric vehicle battery pack composed of 96 ternary lithium-ion batteries connected in series. The battery pack is divided into 8 modules, each equipped with an independent passive balancing circuit. The specific configuration used in implementation is as follows: Command verification layer: The host computer uses the CANoe simulation test platform. The BMS main controller uses FPGA to achieve high-speed synchronous sampling, with a sampling rate of 100kSPS and a command response matching threshold. Set it to 0.92.

[0137] Observation Agent Layer: 104 observation agents are deployed (96 individual layers, 8 module layers, and 1 system layer). Each agent's recurrent cognitive engine is implemented based on a lightweight rule-based inference engine, with an internal consistency threshold. Set it to 0.70.

[0138] Event Judgment Layer: The historical database stores operational data from the past 6 months. The fault mode library pre-configures feature templates for 15 common faults. The manual rule set contains 23 safety and diagnostic rules.

[0139] Policy scheduling layer: Fast response loop response latency <50ms. Slow optimization loop wakes up once every 10 minutes.

[0140] During the testing phase, eight typical faults were simulated and injected, including equalization MOSFET failure, current sampling misalignment, intermittent communication interruption, and gradual change in individual cell internal resistance. Each fault type was repeated 15 times. The test results show that: The success rate of original instruction execution verification was improved from approximately 85% using traditional methods to 99.5%. The accuracy of reliable semantic event identification reached 94.8%, with a false alarm rate reduced to 2.1%. The detection rate of high-deterministic events for real faults reached 97.5%, with the false alarm rate controlled below 1.5%. The fast response loop successfully contained 95% of simulated fault risk escalation within an average of 65ms. Through continuous tuning using the slow optimization loop, the maximum SOC difference within the battery pack was reduced by approximately 18% and the average temperature rise caused by equalization was reduced by approximately 12% during full lifecycle testing.

[0141] Specifically, in a simulated gradual fault involving the degradation of the gate oxide layer of a MOSFET in a balancing branch within a module, this invention successfully identified the anomaly in its early stages, determining it to be in the early stages of power device aging with a confidence level of 82%. The slow optimization loop then proactively adjusted the balancing current distribution strategy, preventing accelerated degradation of the MOSFET. Throughout the process, the observation panorama clearly demonstrates the complete causal chain from subtle data anomalies to aging warnings and strategy adjustments. Example

[0142] A hierarchical transparent observation system for the balancing circuit of a battery management system includes: The instruction verification module is used to establish a two-way closed-loop communication mechanism with instruction verification capability. It responds to the observation instructions of the upper platform, sends the expected electrical response feature template to the target equalization module, collects and compares the actual response data. If the matching degree is lower than the threshold, the closed-loop correction process is started until the instruction execution passes the verification and the verified data is generated. The hierarchical observation agent unit is deployed at each functional level of the equalization circuit to receive verified data. Each observation agent has an embedded cyclic cognitive engine to perform cyclic processing of data acquisition, semantic initial judgment, cross-layer collaborative verification, data enhancement and reacquisition, and final judgment until it achieves internal consistency with other level observation agents and forms a reliable semantic event. The multi-level event determination module is used to sequentially perform physical independence verification, historical behavior inertia analysis, environmental interference compensation calculation, fault mode matching degree scoring, and manual rule coverage judgment on reliable semantic events. If any stage fails, the hierarchical observation agent unit is triggered to reconfirm the reliable semantic event. Only when all stages pass or the matching degree exceeds the threshold will the event be upgraded to a high-determinism event. The dual-loop resource scheduling engine is used for scheduling based on highly deterministic events, including: a fast response loop unit, used to initiate balancing strategy reorganization and key parameter limiting protection for highly deterministic events; and a slow optimization loop unit, used to progressively optimize the balancing strategy based on the state change trend of highly deterministic events. The panoramic output module is used to generate an observation panoramic view and an executable control instruction set based on verified data, reliable semantic data, and highly deterministic data, combined with the strategy output by the dual-loop scheduling engine.

[0143] Those skilled in the art will understand that the embodiments of this application are provided as methods, systems, or computer program products. Therefore, this application takes the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this application takes the form of a computer program product implemented on one or more computer storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer program code. The solutions in the embodiments of this application are implemented using various computer languages, exemplified by the object-oriented programming language Java and the interpreted scripting language JavaScript.

[0144] This application is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this application. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, are implemented by computer program instructions. These computer program instructions are provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, create means for implementing the functions specified in the flowchart illustrations and / or block diagrams.

[0145] These computer program instructions are also stored in a computer read-memory that can direct a computer or other programmed data processing device to operate in a particular manner, such that the instructions stored in the computer read-memory produce an article of manufacture including instruction means that implement the functions specified in the flowchart or multiple flowcharts and / or block diagram blocks or multiple block diagrams.

[0146] These computer program instructions are also loaded onto a computer or other programming data processing device to cause a series of operational steps to be performed on the computer or other programming device to produce a computer-implemented process, such that the instructions, which execute on the computer or other programming device, provide steps for implementing the functions specified in the flowchart flow or multiple flows and / or the block diagram blocks or multiple blocks.

[0147] Although preferred embodiments of this application have been described, those skilled in the art, upon learning the basic inventive concept, can make other changes and modifications to these embodiments. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments as well as all changes and modifications falling within the scope of this application.

[0148] Obviously, those skilled in the art can make various modifications and variations to this application without departing from the spirit and scope of this application. Therefore, if such modifications and variations fall within the scope of the claims of this application and their equivalents, this application also intends to include such modifications and variations.

Claims

1. A layered transparent observation method for the equalization circuit of a battery management system, characterized in that, include: Establish a two-way closed-loop communication mechanism with command verification capability, respond to the observation command of the upper platform, send the expected electrical response feature template to the target equalization module, collect the actual response data and compare it. If the matching degree is lower than the threshold, start the closed-loop correction process until the command execution passes the verification and the verified data is generated. An observation agent with an embedded cyclic cognitive engine is deployed at each functional level of the equalization circuit to receive verified data and perform cyclic processing of acquisition, semantic initial judgment, cross-layer collaborative verification, data enhancement and re-acquisition, and final judgment until internal consistency is achieved and reliable semantic events are formed. For reliable semantic events, physical independence verification, historical behavior inertia analysis, environmental interference compensation calculation, fault mode matching degree scoring, and manual rule coverage judgment are performed sequentially. If any stage fails, the reliable semantic event is reconfirmed. Only when all stages pass or the matching degree exceeds the threshold is it upgraded to a high-deterministic event. Based on highly deterministic events, a dual-loop resource scheduling engine is driven. The fast response loop initiates balance strategy reorganization and key parameter limiting protection for highly deterministic events, while the slow optimization loop performs incremental tuning of the balance strategy based on the state change trend of highly deterministic events. Based on verified data, reliable semantic data, and highly deterministic data, and combined with a dual-loop scheduling strategy, an observation panorama and a set of executable control instructions are generated.

2. The layered transparent observation method for the equalization circuit of the battery management system according to claim 1, characterized in that, The establishment of a two-way closed-loop communication mechanism with instruction verification capability includes: A digital communication link with acknowledgment and retransmission mechanism is established between the functional levels of the equalization circuit, and a custom instruction verification field and response feature description field are embedded in the protocol stack of each communication link. The expected electrical response characteristic template is pre-stored in the non-volatile memory of each level controller in the form of a data structure, and includes at least the theoretical waveform of the standard ADC sampling sequence or the stepped response curve of the current sensor; If the matching degree is lower than the preset threshold, a closed-loop correction process is initiated. The closed-loop correction process is an iterative optimization process, which includes recording the command parameters and response deviations, selecting a correction strategy from the knowledge base according to the deviation pattern, reissuing the observation command with the corrected parameters and recalculating the matching degree until the matching degree reaches the threshold or the maximum number of correction iterations is reached.

3. The layered transparent observation method for the battery management system balancing circuit according to claim 2, characterized in that, The expected electrical response feature template is a dynamically synthesized template, and its synthesis process includes: Based on the observation command type, the current working status of the target module, and its historical response characteristics, templates are selected from the pre-stored template library or dynamically generated. If the observation command is to read the terminal voltage of the k-th battery in the module, then the steady-state value of the expected voltage template is dynamically corrected based on the current total module current and battery internal resistance model parameters. If the observation command is to activate the passive equalization branch, the rise time constant and steady-state value of the expected current response template are dynamically corrected based on the current voltage, temperature, and historical equalization efficiency parameters.

4. The layered transparent observation method for the battery management system balancing circuit according to claim 1, characterized in that, The observation agent deployed with an embedded recurrent cognitive engine at each functional level of the equalization circuit includes: Observation agents are deployed at the individual unit layer, module layer and system layer of the battery management system. Each agent encapsulates a cyclic cognitive engine with state memory and logical reasoning capabilities. The recurrent cognitive engine performs initial semantic judgment on the verified data based on a preset rule base, and generates preliminary semantic hypotheses with initial confidence. The initial confidence is obtained based on data quality, signal noise level and the confidence weight of the rule itself. When the confidence level of the initial semantic hypothesis is lower than the local confidence threshold or the hypothesis involves cross-level impact, a cross-level collaborative verification request is initiated. The request includes the hypothesis content, relevant data summary, and a list of levels to be verified.

5. The layered transparent observation method for the battery management system balancing circuit according to claim 4, characterized in that, The cross-layer collaborative verification request triggers a data enhancement and re-collection process, including: The system requests supplementary data with higher sampling frequency, longer time window, or different sensing dimensions from the data acquisition unit or the underlying control unit. The supplementary data is then regenerated into verified data through the instruction verification process and fed back to the observation agent that initiated the re-acquisition. The cyclic cognitive engine integrates the initial local semantic judgment results, cross-layer collaborative verification feedback, and new evidence obtained from data enhancement and re-collection to update and merge the hypothesis set, iteratively adjusting the confidence of each hypothesis until the final judgment conditions are met. The final decision criteria include: the confidence level of a semantic hypothesis rises to exceed the final decision threshold, or the hypothesis with the highest confidence level is selected when the maximum cognitive loop count is reached, or the confidence level of all hypotheses is below the rejection threshold.

6. The layered transparent observation method for the equalization circuit of the battery management system according to claim 1, characterized in that, The process of sequentially performing physical independence verification, historical behavior inertia analysis, environmental interference compensation calculation, fault mode matching score, and manual rule coverage judgment on reliable semantic events includes: Physical independence verification verifies the electrical isolation between the sensor channel involved in the event and the measured physical quantity by injecting test signals to verify channel crosstalk, thereby eliminating spurious events caused by measurement channel coupling. Historical behavior inertia analysis retrieves historical data fragments similar to the current event conditions and calculates the statistical deviation between the characteristics of the current event and the historical behavior patterns in order to assess whether the event significantly deviates from the inherent behavioral inertia of the observed object. Fault mode matching score calculates the similarity between the event's comprehensive feature vector and a predefined fault mode feature template, and determines the degree of agreement between the event and a known fault mode based on whether the highest matching score exceeds the confirmation threshold.

7. The layered transparent observation method for the battery management system balancing circuit according to claim 6, characterized in that, The manual rule coverage determination is achieved by loading a set of manual rules consisting of predefined conditions and execution actions. The rule set includes at least: The safety priority rule is used to trigger the highest level of response directly when a specific high-risk event is identified, ignoring other scores. Experience-based exclusion rules are used to override abnormal events that conform to known harmless patterns as normal states; Scenario constraint rules are used to dynamically adjust the criteria for judging events based on the current operating scenario of the system. The system iterates through and executes relevant rules based on the attributes of the event. The rule outputs a Boolean flag or score correction amount used to correct or override the judgment results of the previous stage.

8. The layered transparent observation method for the equalization circuit of the battery management system according to claim 1, characterized in that, The fast response loop is triggered immediately for highly deterministic events, and its execution process includes: An optimization function aimed at minimizing system risk is constructed based on event information, and the parameters of the equilibrium strategy that can be quickly adjusted are determined. A fast-converging optimization algorithm is used to solve the optimization function within a finite number of iterations, thereby obtaining the suboptimal adjustment solution of the policy parameters; The adjustment solution immediately reorganizes the balanced control strategy and directly sets the hardware protection limits for key electrical parameters; The fast response loop has the highest execution priority in the entire resource scheduling engine, and parameter adjustments are marked as temporary offsets.

9. The layered transparent observation method for the equalization circuit of the battery management system according to claim 8, characterized in that, The slow optimization loop is awakened at fixed intervals, and the execution process includes: Collect highly deterministic events and related data over a time series, and extract the long-term trend of system state changes through time series analysis methods; Establish a predictive model between equilibrium strategy parameters and long-term system performance indicators, and construct a slow optimization function with the goal of optimizing performance indicators based on long-term trends. An exact optimization algorithm is used to solve the slow optimization function, and the strategy parameters are updated incrementally and in small increments to achieve continuous optimization of system performance. It also includes setting up a collaborative arbitration mechanism to manage concurrent adjustments of the same strategy parameters by the fast response loop and the slow optimization loop. The parameter limit set by the fast loop has the highest mandatory priority, while the parameter value set by the slow loop serves as the long-term baseline. After the fast loop trigger condition is lifted, the parameter will return to the long-term baseline.

10. A layered transparent observation system for the equalization circuit of a battery management system according to any one of claims 1-9, characterized in that, The system includes: The instruction verification module is used to establish a two-way closed-loop communication mechanism with instruction verification capability. It responds to the observation instructions of the upper platform, sends the expected electrical response feature template to the target equalization module, collects and compares the actual response data. If the matching degree is lower than the threshold, the closed-loop correction process is started until the instruction execution passes the verification and the verified data is generated. The hierarchical observation agent unit is deployed at each functional level of the equalization circuit to receive verified data. Each observation agent has an embedded cyclic cognitive engine to perform cyclic processing of data acquisition, semantic initial judgment, cross-layer collaborative verification, data enhancement and reacquisition, and final judgment until it achieves internal consistency with other level observation agents and forms a reliable semantic event. The multi-level event determination module is used to sequentially perform physical independence verification, historical behavior inertia analysis, environmental interference compensation calculation, fault mode matching degree scoring, and manual rule coverage judgment on reliable semantic events. If any stage fails, the hierarchical observation agent unit is triggered to reconfirm the reliable semantic event. Only when all stages pass or the matching degree exceeds the threshold will the event be upgraded to a high-determinism event. The dual-loop resource scheduling engine is used for scheduling based on highly deterministic events, including: a fast response loop unit, used to initiate balancing strategy reorganization and key parameter limiting protection for highly deterministic events; and a slow optimization loop unit, used to progressively optimize the balancing strategy based on the state change trend of highly deterministic events. The panoramic output module is used to generate an observation panoramic view and an executable control instruction set based on verified data, reliable semantic data, and highly deterministic data, combined with the strategy output by the dual-loop scheduling engine.

Images