Power battery thermal safety control method, vehicle and storage medium

Abstract

Embodiments of the present application provide a power battery thermal safety control method, a vehicle and a storage medium. The method comprises: collecting multi-source state data and preprocessing to obtain a standardized state vector; performing level prediction according to the standardized state vector to obtain a risk level prediction result; when the risk level prediction result represents that there is a thermal safety risk, performing abnormality identification and mechanism consistency verification according to the standardized state vector to obtain an abnormality identification result and a mechanism consistency result; performing correlation deduction processing according to the abnormality identification result, the mechanism consistency result and the risk level prediction result to obtain a correlation deduction result; generating at least one candidate disposal strategy according to the abnormality identification result, the risk level prediction result and the correlation deduction result; performing safety arbitration processing on the candidate disposal strategy to determine a target disposal strategy; and issuing the target disposal strategy to an execution end. The method is used to improve the pertinence and accuracy of thermal safety control.

Description

Technical Field

[0001] This application relates to the field of battery safety technology, and in particular to a method for thermal safety control of a power battery, a vehicle, and a storage medium. Background Technology

[0002] With the promotion and application of new energy vehicles, the power battery, as the core component of vehicle energy supply, is directly related to vehicle driving safety, charging safety, and passenger safety. Under conditions such as high-rate charging and discharging, high-temperature environment, long-term use and aging, or accumulation of local defects, power batteries may experience local temperature rise, internal short circuit, insulation abnormality, and cooling abnormality. When these conditions continue to develop, they may further cause thermal runaway or even thermal runaway.

[0003] In existing technologies, power battery safety management typically involves the battery management system or vehicle control system monitoring signals such as voltage, current, temperature, insulation status, and cooling status, and comparing the monitoring results with preset thresholds. When a monitoring signal exceeds the threshold, an alarm is triggered, or control actions such as current limiting, power reduction, charging stop, enhanced cooling, and high-voltage disconnection are executed according to preset rules.

[0004] However, in power battery thermal safety incidents, the same abnormal temperature rise may correspond to different sources of failure, and the risk development process and control requirements for different sources of failure are also different. Existing methods usually determine control actions based on a single monitoring data, a single abnormality judgment result, or a preset control rule, which makes it difficult to accurately distinguish the source of the abnormality and its development trend, resulting in insufficient targeting and accuracy of thermal safety control. Summary of the Invention

[0005] This application provides a power battery thermal safety control method, a vehicle, and a storage medium to improve the targeting and accuracy of thermal safety control.

[0006] In a first aspect, embodiments of this application provide a method for thermal safety control of a power battery, including:

[0007] Multi-source state data for characterizing the thermal safety state of the power battery is collected, and the multi-source state data is preprocessed to obtain a standardized state vector.

[0008] Based on the standardized state vector, the thermal safety risk level of the power battery is predicted, and the risk level prediction result is obtained.

[0009] When the risk level prediction result indicates the existence of thermal safety risk, anomaly identification and mechanism consistency verification are performed according to the standardized state vector to obtain anomaly identification results and mechanism consistency results.

[0010] Based on the anomaly identification results, the mechanism consistency results, and the risk level prediction results, correlation inference processing is performed to obtain the correlation inference results;

[0011] Based on the anomaly identification results, the risk level prediction results, and the correlation inference results, at least one candidate disposal strategy is generated;

[0012] Obtain the available status of the executor corresponding to each of the candidate disposal strategies, and perform security arbitration processing on the candidate disposal strategies according to the preset security constraint rules and the available status of each of the executors to determine the target disposal strategy;

[0013] The target handling strategy is sent to the execution end so that the execution end executes the linkage control instructions in the target handling strategy that correspond to the risk level prediction result.

[0014] Secondly, embodiments of this application provide a power battery thermal safety control device, comprising:

[0015] The data processing module is used to collect multi-source state data to characterize the thermal safety state of the power battery, and to preprocess the multi-source state data to obtain a standardized state vector.

[0016] The risk prediction module is used to predict the level of thermal safety risk of the power battery based on the standardized state vector, and obtain the risk level prediction result.

[0017] The identification and verification module is used to perform anomaly identification and mechanism consistency verification according to the standardized state vector when the risk level prediction result indicates the existence of thermal safety risk, so as to obtain the anomaly identification result and the mechanism consistency result.

[0018] The correlation inference module is used to perform correlation inference processing based on the anomaly identification results, the mechanism consistency results, and the risk level prediction results to obtain the correlation inference results;

[0019] The strategy acquisition module is used to generate at least one candidate disposal strategy based on the anomaly identification result, the risk level prediction result, and the correlation inference result.

[0020] The strategy acquisition module is also used to acquire the available status of the executor corresponding to each of the candidate disposal strategies, and to perform security arbitration processing on the candidate disposal strategies according to the preset security constraint rules and the available status of each of the executors to determine the target disposal strategy;

[0021] The result output module is used to send the target disposal strategy to the execution end, so that the execution end executes the linkage control instructions in the target disposal strategy that correspond to the risk level prediction result.

[0022] Thirdly, embodiments of this application provide a vehicle, including: a memory and a processor;

[0023] The memory stores computer-executed instructions;

[0024] The processor executes computer execution instructions stored in the memory, causing the processor to perform the method described above.

[0025] Fourthly, embodiments of this application provide a computer-readable storage medium storing computer-executable instructions, which, when executed by a processor, are used to implement the above-described method.

[0026] Fifthly, embodiments of this application provide a computer program product, including a computer program that, when executed by a processor, implements the above-described method.

[0027] This application provides a power battery thermal safety control method, vehicle, and storage medium. By preprocessing multi-source state data characterizing the thermal safety state of the power battery to obtain standardized state vectors, state data from different sources can form a unified data representation, providing a consistent data foundation for subsequent risk assessment and control decisions. By predicting the risk level based on the standardized state vectors, the degree of current thermal safety risk of the power battery can be determined first. When the risk level prediction indicates the presence of thermal safety risk, anomaly identification and mechanism consistency verification are then performed, focusing subsequent processing on states with thermal safety risks and improving the effectiveness and targeting of the control process. Next, by correlating the anomaly identification results, mechanism consistency results, and risk level prediction results, a correlation deduction result is obtained. This allows thermal safety control to determine subsequent control based not only on the anomaly itself but also on the risk level, fault cause, and propagation trend. By generating at least one candidate disposal strategy based on the anomaly identification results, risk level prediction results, and correlation deduction results, the candidate disposal strategy can achieve a higher degree of matching with the current thermal safety risk state, avoiding the need to determine control actions solely based on fixed rules. Furthermore, by acquiring the available status of the actuators corresponding to each candidate disposal strategy and combining it with preset safety constraint rules, the candidate disposal strategies are subjected to safety arbitration processing, ensuring that the determined target disposal strategy simultaneously meets thermal safety control requirements, safety constraint requirements, and actual execution conditions. By distributing the target disposal strategy to the execution end and having the execution end execute the linkage control instructions corresponding to the risk level prediction results in the target disposal strategy, the final executed control action can correspond to the current risk level, improving the executability, pertinence, and accuracy of power battery thermal safety control. Therefore, this application can solve the technical problem in existing power battery thermal safety control where the control action is not accurately matched with the actual thermal safety risk state, resulting in insufficient pertinence and accuracy of thermal safety control. Attached Figure Description

[0028] The accompanying drawings, which are incorporated in and form part of this specification, illustrate embodiments consistent with this application and, together with the description, serve to explain the principles of this application.

[0029] Figure 1 A schematic flowchart of the power battery thermal safety control method provided in the embodiments of this application;

[0030] Figure 2 This is a schematic diagram of the processing method after strategy execution provided in the embodiments of this application;

[0031] Figure 3 This is a schematic diagram of the structure of the power battery thermal safety control device provided in the embodiments of this application;

[0032] Figure 4 This is a structural schematic diagram of the vehicle provided in an embodiment of this application.

[0033] The accompanying drawings have illustrated specific embodiments of this application, which will be described in more detail below. These drawings and descriptions are not intended to limit the scope of the concept in any way, but rather to illustrate the concept of this application to those skilled in the art through reference to specific embodiments. Detailed Implementation

[0034] Exemplary embodiments will now be described in detail, examples of which are illustrated in the accompanying drawings. When the following description relates to the drawings, unless otherwise indicated, the same numbers in different drawings denote the same or similar elements. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with this application. Rather, they are merely examples of apparatuses and methods consistent with some aspects of this application as detailed in the appended claims.

[0035] The technical solution of this application and how the technical solution of this application solves the above-mentioned technical problems are described in detail below with specific embodiments. These specific embodiments can be combined with each other, and the same or similar concepts or processes may not be described again in some embodiments. The embodiments of this application will be described below with reference to the accompanying drawings.

[0036] Figure 1 This is a schematic flowchart of the power battery thermal safety control method provided in the embodiments of this application, as shown below. Figure 1 As shown, the method includes:

[0037] S11: Collect multi-source state data to characterize the thermal safety state of the power battery, preprocess the multi-source state data, and obtain a standardized state vector.

[0038] In this embodiment, multi-source state data can originate from different detection and control objects during the operation of the power battery, and is used to characterize the thermal safety status of the power battery under current operating conditions. By collecting multi-source state data, it is possible to avoid relying solely on a single state quantity to determine the thermal safety status of the power battery, thus providing a more complete data foundation for subsequent risk level prediction. Preprocessing of multi-source state data refers to unifying the state data from different sources, in different data formats, or with different dimensions after acquisition, converting it into a data expression form that can be used by subsequent processing steps. Preprocessing is at least used to eliminate interfering factors in the original state data that are detrimental to subsequent analysis, and to ensure a consistent data organization among different state data.

[0039] This embodiment transforms multi-source state data from its scattered, raw form into a unified, meaningful, and computationally efficient state representation. The preprocessed multi-source state data is then organized according to a predefined data structure to form a unified input data set for characterizing the thermal safety state of the power battery—a standardized state vector. This standardized state vector serves as the common data foundation for subsequent risk level prediction, anomaly identification, and mechanism consistency verification, enabling each subsequent processing step to analyze the same state representation. Therefore, through the collection and standardization of multi-source state data, a unified characterization of the power battery's thermal safety state is achieved, providing data support for determining the degree of thermal safety risk and generating corresponding control schemes.

[0040] S12, based on the standardized state vector, predicts the level of thermal safety risk of the power battery and obtains the risk level prediction result.

[0041] In this embodiment, the standardized state vector contains multi-source state information that has undergone unified processing. Therefore, it can provide a unified data input for risk level prediction, enabling risk judgment to no longer rely solely on a single state quantity or a single threshold, but to comprehensively consider the changing relationship between the current operating state of the power battery and thermal safety-related states. During level prediction, the characterization result of the power battery's thermal safety risk can be determined based on the standardized state vector, and this characterization result can be converted into a corresponding risk level prediction result. The risk level prediction result represents the degree of current thermal safety risk of the power battery and can serve as a basis for subsequent judgments on whether to trigger anomaly identification, mechanism consistency verification, correlation deduction processing, and control scheme generation. By obtaining the risk level prediction result first, unnecessary subsequent processing can be reduced when no thermal safety risk occurs, and further analysis and control processes can be initiated promptly when a thermal safety risk exists. The risk level prediction result can include level information used to characterize the degree of risk, or it can include the risk characterization value or risk judgment result corresponding to that level information. By converting the standardized state vector into a risk level prediction result, a graded judgment of the power battery's thermal safety state is achieved, enabling subsequent control processes to adopt matching processing methods according to different risk levels, providing a foundation for improving the targeting and accuracy of thermal safety control.

[0042] S13, when the risk level prediction results indicate the presence of thermal safety risks, anomaly identification and mechanism consistency verification are performed according to the standardized state vector to obtain anomaly identification results and mechanism consistency results.

[0043] In this embodiment, anomaly identification is used to determine the abnormal manifestation and location corresponding to the current thermal safety risk, while mechanism consistency verification is used to determine whether the current state change conforms to the mechanistic characteristics of the development of power battery thermal safety risks. By triggering these processes only when the risk level prediction results indicate the existence of thermal safety risks, subsequent analysis can focus on operating states with actual risks, reducing invalid calculations in risk-free states. Anomaly identification based on standardized state vectors refers to using the power battery thermal safety state change information represented in the standardized state vectors to judge the possible anomaly types, anomaly regions, or anomaly degrees of the power battery, thus obtaining anomaly identification results. The anomaly identification results are used to characterize the specific manifestation of power battery thermal safety risks at the state level, enabling subsequent processing to clearly identify the abnormal objects and characteristics corresponding to the thermal safety risks, rather than simply using general control based on risk level prediction results.

[0044] Mechanism consistency verification based on standardized state vectors refers to matching and verifying the state changes reflected by the standardized state vectors with the mechanistic laws of thermal safety risks in power batteries to obtain mechanism consistency results. Mechanism consistency results characterize the degree of conformity between the current state change and the corresponding fault mechanism, enabling subsequent processing to determine whether the current thermal safety risk has a corresponding mechanistic basis. By simultaneously obtaining anomaly identification results and mechanism consistency results, thermal safety risks can be analyzed from both anomaly manifestation and mechanistic rationality perspectives, providing a more reliable basis for subsequent correlation and deduction processing. By performing anomaly identification and mechanism consistency verification separately when risks exist, power battery thermal safety control no longer relies solely on risk level judgment but can further clarify the corresponding abnormal states and mechanistic basis of the risks. This achieves further decomposition and confirmation of thermal safety risks, improving the targeting of subsequent processing.

[0045] S14. Based on the anomaly identification results, mechanism consistency results, and risk level prediction results, correlation inference processing is performed to obtain the correlation inference results.

[0046] In this embodiment, the abnormal object or abnormal region corresponding to the current thermal safety risk is determined based on the anomaly identification results. The correspondence between the abnormal object or abnormal region and the fault mechanism is determined based on the mechanism consistency results. Finally, the development degree of the current thermal safety risk is determined by combining the risk level prediction results. Thus, relevant information about the current thermal safety risk can be inferred from three aspects: anomaly manifestation, mechanism basis, and risk degree, resulting in a correlation inference.

[0047] The correlation simulation results provide input for the generation of subsequent candidate disposal strategies. These results characterize the potential sources of failure, possible directions or ranges of development corresponding to the current thermal safety risk, as well as the objects or areas requiring priority attention during subsequent control processes. By obtaining the correlation simulation results, the generation of subsequent candidate disposal strategies can move beyond solely relying on risk level predictions or anomaly identification results, and instead incorporate mechanistic consistency results for a comprehensive assessment of the current thermal safety risk. Furthermore, the correlation simulation results provide a more complete description of the risk scenario for the generation of subsequent candidate disposal strategies, enabling subsequent control schemes to more accurately match the current thermal safety risk state of the power battery.

[0048] S15. Based on the anomaly identification results, risk level prediction results, and correlation inference results, generate at least one candidate disposal strategy.

[0049] In this embodiment, the intensity of the response is determined based on the risk level prediction results, the target or scope of the response is determined based on the anomaly identification results, and the direction of the response is determined based on the correlation deduction results. The resulting candidate response strategies may include one or more response schemes available for subsequent selection, so that the target response strategy can be further determined based on safety constraints and execution conditions. By generating at least one candidate response strategy based on the anomaly identification results, risk level prediction results, and correlation deduction results, the conversion from thermal safety risk analysis results to selectable control schemes is realized, enabling the subsequent control process to form targeted candidate response strategies based on the current risk status.

[0050] S16: Obtain the available status of the executors corresponding to each candidate disposal strategy, and perform security arbitration processing on the candidate disposal strategies according to the preset security constraint rules and the available status of each executor to determine the target disposal strategy.

[0051] In this embodiment, obtaining the actuator availability status corresponding to each candidate disposal strategy refers to determining, after generating at least one candidate disposal strategy, whether the actuator required to execute each candidate disposal strategy has the conditions to perform the corresponding control action at the current moment. The actuator availability status characterizes whether the actuator is in a callable, responsive, or executable state, reflecting whether the candidate disposal strategy has a practical execution basis in the current power battery operating environment. By obtaining the actuator availability status corresponding to each candidate disposal strategy before determining the target disposal strategy, it is possible to avoid directly using candidate disposal strategies that are currently unexecutable or whose execution conditions are not met as the final control scheme.

[0052] Based on preset safety constraints and the availability status of each actuator, safety arbitration processing of candidate disposal strategies refers to judging the safety and feasibility among multiple candidate disposal strategies and determining the disposal scheme that meets the current thermal safety control requirements. Preset safety constraints define the safety boundaries and control conditions that candidate disposal strategies must meet during execution, while actuator availability status determines whether the control actions in the candidate disposal strategies can be actually executed. By simultaneously considering both preset safety constraints and actuator availability status, the process of determining candidate disposal strategies focuses not only on the control effect but also on the safety and feasibility of the control actions under the current state.

[0053] During safety arbitration, candidate disposal strategies that do not meet preset safety constraints or lack execution conditions can be excluded from the selection pool. From the remaining candidate strategies, a target disposal strategy matching the current thermal safety risk state is determined. The target disposal strategy refers to the disposal strategy determined after safety arbitration and used to be issued to the execution end. This process reduces the possibility of control failure due to insufficient execution conditions or unmet safety boundaries of candidate disposal strategies, improving the reliability of the target disposal strategy. By introducing the actuator availability state and preset safety constraints after the candidate disposal strategies are generated, and performing safety arbitration on the candidate disposal strategies, the screening and determination from candidate disposal strategies to the target disposal strategy is achieved. This ensures that the final determined target disposal strategy simultaneously meets the current thermal safety risk control needs, safety constraint requirements, and execution condition requirements, improving the targeting, accuracy, and executability of power battery thermal safety control.

[0054] S17, the target handling strategy is sent to the execution end so that the execution end can execute the linkage control instructions in the target handling strategy that correspond to the risk level prediction results.

[0055] In this embodiment, the execution terminal receives the target handling strategy and triggers corresponding control actions based on the control content contained in the target handling strategy. By distributing the target handling strategy to the execution terminal, the results obtained from the aforementioned risk analysis, correlation deduction, and safety arbitration processing can be transformed into actual control actions, thereby completing the transition from thermal safety status judgment to thermal safety control execution. The target handling strategy includes linkage control instructions corresponding to the risk level prediction results. Linkage control instructions refer to control instructions used to control one or more execution objects to execute collaboratively, and their control strength, control range, or control timing corresponds to the degree of thermal safety risk represented by the risk level prediction results. By having the execution terminal execute the linkage control instructions corresponding to the risk level prediction results, the same control actions can be avoided under different risk levels, allowing the control process to adapt to the current thermal safety risk level.

[0056] During execution, the execution terminal determines the linkage control command that matches the risk level prediction result based on the received target handling strategy, and executes the corresponding action according to the control content defined by the target handling strategy. This realizes the implementation of thermal safety risk analysis results into actual control actions. The power battery thermal safety control process in this embodiment can determine the control direction based on the current risk level, candidate fault root cause information, and thermal diffusion path information, and complete the corresponding control through the execution terminal, improving the pertinence, accuracy, and executability of power battery thermal safety control.

[0057] In one embodiment, the multi-source state data collected in step S11 above is used to obtain a standardized state vector, which will be further explained here. Based on the above embodiment, it includes:

[0058] S111, collect at least one of the following: battery electrical state data, battery thermal state data, battery health state data, environmental state data, execution state data, and vehicle operating condition state data, to obtain multi-source state data.

[0059] Specifically, battery electrical state data refers to data that reflects the electrical operating status of the power battery, such as individual cell voltage, module voltage, total battery pack voltage, charging and discharging current, insulation resistance, contactor status, and charging connection status. Battery thermal state data refers to data that reflects the thermal distribution and temperature changes of the power battery, such as individual cell temperature, module temperature, battery pack temperature, maximum temperature, minimum temperature, temperature difference, temperature rise rate, cooling inlet temperature, and cooling outlet temperature. Battery health state data refers to data that reflects the aging degree and usability of the power battery, such as state of charge (SOC), state of health (SOH), internal resistance change, capacity decay, and cycle count. Environmental state data refers to data that reflects the external operating environment of the power battery, such as ambient temperature, ambient humidity, combustible gas signals, smoke signals, external air pressure, and the external temperature conditions of the area where the vehicle is located.

[0060] Execution status data refers to data that reflects the current operating status of actuators related to thermal safety control, such as the operating status of the cooling system, cooling pump, fan, valve body, heating components, contactor on / off status, and actuator fault status. Vehicle operating condition data refers to data that reflects the current operating scenario of the vehicle, such as parking status, driving status, charging status, discharging status, energy recovery status, and power request status. By collecting at least one of the above data sources to obtain multi-source status data, the thermal safety status of the power battery can be characterized from multiple aspects, providing a complete data source for the subsequent formation of standardized state vectors.

[0061] S112, Based on the preset sliding time window length and preset state feature dimensions, organize the multi-source state data into a sliding time window to construct an initial state matrix; wherein, the rows of the initial state matrix correspond to time series sampling points, and the columns correspond to state features.

[0062] Specifically, let the time window length be... The state feature dimension is Then the first Each sample is constructed into a standardized state matrix:

[0063]

[0064] Wherein, the state vector at a single time moment is x t It can be represented as:

[0065]

[0066] Among them, U cell U represents the voltage of a single unit. mod I represents the module voltage, and T represents the current. cell T represents the temperature of the monomer. mod Indicates the module temperature. Indicates the rate of temperature rise. The temperature difference is indicated by SOC (State of Charge), which is the remaining charge of the battery. SOH (State of Health) indicates the battery's aging or health status. R in Z represents internal resistance, P represents pressure signal, G represents combustible gas concentration, Smoke represents smoke characteristics, and R represents internal resistance. ins Indicates insulation resistance, I leak S represents the leakage current. con S indicates the contactor status. cool S indicates the status of the cooling system. chg Indicates charging status, v represents vehicle speed, and S represents charging speed. park Indicates the parking status, T amb Indicates ambient temperature.

[0067] S113, based on the preset time reference, synchronize the data corresponding to different sampling frequencies in the initial state matrix to obtain the aligned state matrix.

[0068] Specifically, the preset time reference can be a unified sampling period, a unified timestamp sequence, or a synchronization clock in the power battery control system. Since battery electrical state data, battery thermal state data, environmental state data, execution state data, and vehicle operating condition data may originate from different sensors, controllers, or communication channels, their sampling frequencies and upload periods may differ. Directly using these data for subsequent processing can easily result in the state characteristics corresponding to the same row of time series sampling points not being at the same time or within the same time interval. By synchronizing the data corresponding to different sampling frequencies in the initial state matrix, the accuracy of subsequent processing is improved.

[0069] Furthermore, during time synchronization, for data with a sampling frequency higher than the preset time base, resampling, aggregation, or neighboring sample values ​​can be performed according to the preset time base; for data with a sampling frequency lower than the preset time base, forward hold, adjacent time point interpolation, or time proximity matching can be used to fill in the gaps. Through the above processing, different state features in the initial state matrix can be adjusted to the same time base, so that the rows of the aligned state matrix still correspond to time series sampling points, the columns still correspond to state features, and the data in the same row can jointly characterize the thermal safety state of the power battery at the corresponding time position.

[0070] S114, perform anomaly removal processing on out-of-bounds values, mutation values ​​and communication error values ​​in the alignment state matrix to obtain the removed state matrix.

[0071] Specifically, out-of-bounds values ​​refer to data that exceed the physical value range or preset valid range of the corresponding state characteristic. Examples include obviously unreasonable values ​​for state characteristics such as temperature, voltage, current, insulation resistance, pressure signal, combustible gas concentration, smoke characteristics, vehicle speed, or ambient temperature. Abrupt values ​​refer to data whose variation exceeds a preset range between adjacent time-series sampling points and cannot be explained by the actual operating state of the power battery. Communication error values ​​refer to invalid data caused by packet loss, abnormal message parsing, verification failure, or abnormal acquisition channels. By removing these anomalies, the impact of acquisition noise, sensor anomalies, and communication anomalies on the subsequent standardized state vector can be reduced.

[0072] For example, during anomaly removal, out-of-bounds values, abrupt changes, and communication errors can be identified based on the physical boundaries corresponding to each state feature, the variation amplitude of adjacent sampling points, the consistency between similar state features, and communication status flags. For identified anomalous data, they can be deleted or marked as null values, while retaining their temporal and feature positions in the aligned state matrix, so that subsequent missing data completion processing can correct for the corresponding positions. The resulting state matrix after removal, while maintaining the original time series structure and state feature structure, reduces the interference of anomalous data on subsequent calculations.

[0073] S115, perform missing data completion processing on the short-term missing data in the removed state matrix to obtain the corrected state matrix; wherein, the missing data completion processing includes using Kalman filtering to complete the short-term missing data.

[0074] Specifically, short-term missing data refers to data whose missing duration does not exceed a preset time range and can be estimated based on valid data before and after the missing position or the change pattern of adjacent states. Since the thermal safety state of a power battery usually has a certain degree of temporal continuity, in the case of short-term missing data, the state change relationship of adjacent time positions can be used to fill in the missing data, avoiding the interruption of subsequent feature extraction due to local missing data.

[0075] Missing data completion processing includes using Kalman filtering to complete short-term missing data. Kalman filtering is a data processing method that recursively estimates data based on historical state estimates and current observation information. It can estimate the current state value even in the presence of measurement noise and short-term missing data. In this embodiment, Kalman filtering can be used to complete the state matrix after removal, based on the state change relationships between adjacent time series sampling points, for temperature, voltage, current, temperature rise rate, temperature difference, insulation resistance, pressure signal, combustible gas concentration, smoke characteristics, or other continuous state features. This preserves the continuity of state changes while reducing noise influence, resulting in a corrected state matrix. This allows subsequent feature extraction to be based on data with higher integrity and lower anomaly interference, improving the window feature matrix's ability to characterize changes in the thermal safety state of the power battery.

[0076] S116, extract features from the modified state matrix to obtain a window feature matrix; wherein, the window feature matrix includes at least one of the following: mean feature, variance feature, extreme value feature, rate of change feature, slope feature, and trend feature.

[0077] Specifically, the window feature matrix is ​​used to transform the original time-series state data in the modified state matrix into a feature representation that reflects the changing patterns of thermal safety status. The window feature matrix includes at least one of the following: mean feature, variance feature, extreme value feature, rate of change feature, slope feature, and trend feature. The mean feature characterizes the average level of the corresponding state feature within the sliding time window; the variance feature characterizes the degree of fluctuation of the corresponding state feature within the sliding time window; the extreme value feature characterizes the maximum or minimum value of the corresponding state feature within the sliding time window; the rate of change feature characterizes the rate of change of the corresponding state feature between adjacent time-series sampling points; the slope feature corresponds to the overall rise or fall of the state feature within the time window; and the trend feature characterizes whether the corresponding state feature exhibits fluctuating changes or tends to stabilize within the sliding time window. By extracting these features, the ability to express the impact of continuously rising temperature, abnormal fluctuations in electrical state, abnormal cooling state, or changes in vehicle operating conditions on thermal safety status can be enhanced.

[0078] S117, normalize the window feature matrix to obtain the standardized state matrix.

[0079] Specifically, since the features in the window feature matrix are data of different types, their dimensions and value ranges differ. Through normalization, various window features can be brought to a relatively consistent data scale, improving the stability of subsequent risk level prediction, anomaly identification, and mechanism consistency verification.

[0080] In one implementation, normalization can be achieved using a minimum-maximum normalization method. Minimum-maximum normalization maps a feature to a preset numerical range based on its minimum and maximum values. For example, the difference between a window feature value and its corresponding minimum value can be divided by the difference between the corresponding maximum and minimum values ​​to obtain the normalized feature value. That is, the formula is:

[0081]

[0082] Obtain the standardized state matrix This refers to the standardized state vector. Where x represents the current feature, x... min Denotes the minimum eigenvalue, x max This represents the maximum value of the characteristic. This embodiment achieves a unified representation of the thermal safety state data of power batteries from different sources, with different sampling frequencies, and different dimensions. It can reduce the impact of asynchronous sampling, outliers, short-term missing values, and dimensional differences on subsequent analysis, and improve the accuracy and computability of the standardized state vector in representing the thermal safety state of power batteries.

[0083] In one embodiment, based on the above embodiments, the thermal safety risk level of the power battery is predicted according to the standardized state vector to obtain the risk level prediction result, including:

[0084] S121, Obtain the battery topology graph corresponding to the power battery system; wherein, the battery topology graph includes a set of nodes, a set of edges, and an adjacency matrix used to represent the connection relationship between nodes;

[0085] S122, Based on the nodes in the battery topology diagram, the standardized state vector is organized into nodes to obtain the node feature matrix;

[0086] S123, input the node feature matrix and adjacency matrix into the graph convolutional network to extract topological features containing node state association information and topological coupling information;

[0087] S124: Input the topological features corresponding to the continuous time window into the temporal recursive network to obtain the risk value at the current time.

[0088] S125, based on the preset risk level mapping relationship, map the risk value to the risk level to obtain the risk level prediction result including the risk value and the risk level.

[0089] In this embodiment, the battery topology graph refers to a data structure constructed based on the structural relationships between cells, modules, battery packs, sensors, and actuators in the power battery system, capable of representing the objects within the power battery system and the relationships between them. The battery topology graph includes a set of nodes, a set of edges, and an adjacency matrix representing the connection relationships between nodes. The node set represents the entity objects in the power battery system, the edge set represents the relationships between different entity objects, and the adjacency matrix records, in matrix form, whether there are connections between nodes and the strength of those connections.

[0090] An example battery topology diagram:

[0091]

[0092] Where V is the set of nodes, E is the set of edges, and A is the adjacency matrix. Nodes can include cell nodes, module nodes, battery pack nodes, temperature sensor nodes, pressure sensor nodes, contactor nodes, and cooling pump nodes, etc.; edges can include series-parallel connection edges, thermal coupling edges, propagation edges, and control-dependent edges. The adjacency matrix is ​​a matrix representation corresponding to the set of nodes. The rows and columns in the adjacency matrix correspond to the nodes in the set, and the matrix elements are used to indicate whether there is an edge between two nodes and the corresponding connection strength or association weight. When there is an electrical connection, thermal coupling, fault propagation, or control dependency relationship between two nodes, a corresponding non-zero element can be set in the adjacency matrix; when there is no association relationship between two nodes, a corresponding zero element can be set in the adjacency matrix. Different types of relationships can be expressed through different weights, different channels, or different matrix layers. By constructing a battery topology graph, the power battery system can be transformed from dispersed physical components and detection objects into a computable topological expression, enabling subsequent risk level prediction to utilize the internal structural relationships of the power battery system.

[0093] Next, the state features associated with each node in the standardized state vector are assigned to the corresponding nodes, resulting in a node feature matrix. Each row in the node feature matrix corresponds to a node in the node set, and each column corresponds to a state feature associated with that node. This node-based organization process establishes a correspondence between the state information in the standardized state vector and the nodes in the battery topology diagram. When a state feature in the standardized state vector is associated with multiple nodes, it can be assigned to multiple related nodes based on its physical meaning or acquisition location, or node features can be formed for each related node according to a preset allocation rule. When a node lacks directly acquired data, corresponding node features can be supplemented for that node based on the states of other nodes connected to it, the node's hierarchy, or historical state features. The node feature matrix obtained in this way not only contains the state information of each node itself but also maintains the correspondence between the state information and the structural location of the power battery system, providing an input basis for subsequent extraction of node state association information and topology coupling information.

[0094] Subsequently, the node feature matrix and adjacency matrix are input into the graph convolutional network to extract topological features. A graph convolutional network (GCN) is a neural network model capable of feature aggregation and propagation based on graph-structured data. It can determine the information transmission range between nodes based on the adjacency matrix and determine the initial state features of each node based on the node feature matrix. In a graph convolutional network, each node can update its features based on the state features of its neighboring nodes and their corresponding connections, so that the node's output features simultaneously reflect its own state and the states of its neighboring nodes. The graph convolution process is represented as follows:

[0095]

[0096] in, Indicates the first The node feature matrix of a layered graph convolutional network; Indicates the first The node feature matrix output by the layer graph convolutional network; Indicates the layer index of the graph convolutional network; This represents the adjacency matrix after adding self-connections, used to characterize the connection relationships between a node and its neighboring nodes; Indicates and The corresponding degree matrix is ​​used to normalize the adjacency matrix to avoid the difference in node degree having too much of an impact on feature propagation. Indicates the first The trainable weight matrix of a layered graph convolutional network; This represents a nonlinear activation function. By extracting topological features through graph convolutional networks, the risk level prediction process can consider the internal structural relationships and state transit relationships of the power battery system, rather than relying solely on the independent values ​​of each state data.

[0097] Inputting the topological features corresponding to continuous time windows into a temporal recursive network (TRRN) means inputting the topological features extracted from multiple continuous time windows into the TRRN network in chronological order. The TRRN network then models the process of topological feature changes over time. TRRN networks are used to process feature sequences with temporal relationships. They can retain state change information from previous time windows or multiple historical time windows and fuse historical change information with the topological features of the current time window. Through this processing, the gradual development of thermal safety risks in power batteries from slight state changes can be identified, such as a continuous increase in temperature, a gradual increase in temperature difference, the spread of local node anomalies to adjacent nodes, or the persistent existence of abnormal execution states.

[0098] Continuous time windows can be formed by the aforementioned sliding time windows, with each window corresponding to topological features within a specific time range. The temporal recursive network recursively calculates the topological features corresponding to the continuous time windows and outputs the risk value at the current time. For example, by modeling the risk evolution process using temporal recursion, the risk value R at the current time is obtained. t :

[0099]

[0100] in, Indicates the first Topological features corresponding to each time window; This indicates the hidden state at the previous moment, used to store risk evolution information within the historical time window; It represents the hidden state at the current moment, used to characterize the temporal risk features after fusing the current topological structure features and historical states; Represents the recursive computation function for a gated recurrent unit network; This represents the weight matrix of the risk value output layer; This represents the bias term of the risk value output layer; This represents the activation function used to map the output to a preset risk value range; This represents the risk value at the current moment. By inputting topological features into a temporal recursive network to obtain the risk value at the current moment, risk assessment can simultaneously consider spatial topological relationships and temporal evolution relationships, improving the ability to identify the development process of thermal safety risks.

[0101] Next, according to pre-defined risk value ranges, risk thresholds, or classification rules, the continuous numerical risk values ​​are converted into discrete risk levels. The pre-defined risk level mapping relationship can be set based on the power battery thermal safety control requirements, historical thermal safety event data, test calibration results, or safety management rules. For example, a risk value can be mapped to a risk level L. t :

[0102]

[0103] Where L1 represents minor anomaly; L2 represents moderate risk; L3 represents high risk; L4 represents critical thermal runaway; and L5 represents thermal runaway propagation. r1, r2, r3, and r4 represent progressively increasing risk thresholds; R t This represents the risk value at the current moment. Different risk levels correspond to different degrees of thermal safety risk and control requirements. By mapping risk values ​​to risk levels, subsequent processing can trigger corresponding anomaly identification, mechanism consistency verification, candidate disposal strategy generation, and linkage control command execution based on a clear risk level.

[0104] The risk level prediction results include a risk value and a risk level. By retaining both the risk value and the risk level simultaneously, subsequent steps can determine whether a thermal safety risk exists based on the risk level, and also make a precise judgment on the degree of risk based on the magnitude and changes in the risk value. This embodiment achieves structured and time-series prediction of the thermal safety risks of power batteries, improving the accuracy of the risk level prediction results in representing the actual thermal safety state of the power battery.

[0105] In one embodiment, anomaly identification is performed in step S3 above, and an implementation method is provided herein. Based on the above embodiment, it includes:

[0106] S1311 When the risk level prediction result indicates the presence of thermal safety risk, the standardized state vector is input into the perception and diagnosis agent, and local temporal features are extracted through the one-dimensional convolutional network in the perception and diagnosis agent.

[0107] Specifically, the perception and diagnostic agent is used to identify abnormal behaviors in the thermal safety state of power batteries. After receiving a standardized state vector, it extracts features and identifies anomalies from the time-series state change information contained within it. Local temporal features are extracted using a one-dimensional convolutional network (1D-CNN) within the agent. A 1D-CNN is a network structure that convolves the input data along the time dimension to extract state change features within a local time range from the standardized state vector. Local temporal features refer to features that reflect the local change patterns of state data within adjacent sampling times or short time windows, such as local temperature increases, local voltage fluctuations, abrupt changes, or continuous shifts in state quantities within a short period. Extracting local temporal features through a 1D-CNN enhances the ability to perceive early local changes in thermal safety anomalies, enabling subsequent processing to focus on short-term state changes related to the occurrence of anomalies.

[0108] For example, a one-dimensional convolutional network extracts local temporal features F. conv Its expression is:

[0109]

[0110] Where X represents the input data to the one-dimensional convolutional network, corresponding to the standardized state vector, and W... conv b represents the convolution kernel parameters, i.e., the convolution weight parameters. conv For bias terms, It is a non-linear activation function.

[0111] S1312, based on local temporal domain features, extracts long-term temporal dependency features through a bidirectional gated recurrent unit network in the perception and diagnostic agent.

[0112] Specifically, the Bidirectional Gated Recurrent Unit (BiGRU) is a network structure that recursively extracts features in both the forward and reverse directions of a time series, used to characterize the dependencies between state data over a longer time span. Long-term time-series dependencies refer to state evolution features formed by local time-domain features from multiple consecutive time points, reflecting the continuous change in the thermal safety state of a power battery from normal fluctuations to abnormal development. By processing local time-domain features through the BiGRU, the temporal correlation information before and after an anomaly can be utilized simultaneously, improving the ability to identify phenomena such as continuous temperature rise, continuous temperature difference expansion, continuous electrical anomalies, or continuous abnormal execution states.

[0113] For example, the convolutional output features are input into a bidirectional gated recurrent unit network to extract long-term temporal dependencies and obtain the forward hidden state. and backward hidden state And combine the two into a fusion feature. The specific process is as follows:

[0114]

[0115] Where t represents the current time step or the current sampling time, F conv,t This represents the local temporal features of the output of a one-dimensional convolutional network at time step t, i.e., the convolutional output features corresponding to the current time step. This represents the forward hidden state at time step t-1, i.e., the historical state information before the current time step. This represents the backward hidden state at time step t+1, i.e., the state information after the current time step. GRU() represents the recursive computation function of the Gated Recurrent Unit (GRU).

[0116] S1313 employs an attention mechanism to perform weighted enhancement processing on anomalous key segments in long-term time-dependent features, thereby obtaining anomalous representation data.

[0117] Specifically, the attention mechanism is a processing method that assigns different weights to feature segments based on their contribution to the anomaly identification results. Key anomaly segments refer to time segments within long-term time-dependent features that are highly correlated with anomaly type, anomaly location, or anomaly initiation judgments. Examples include time segments corresponding to a sudden increase in the rate of temperature change, a continuous widening of local temperature differences, asynchronous changes in electrical and thermal states, or execution state anomalies. By weighting and enhancing key anomaly segments, the proportion of anomaly-related features in the anomaly representation data can be increased, while the impact of irrelevant fluctuations or noise segments on the anomaly identification results can be reduced.

[0118] For example, attention mechanisms are used to enhance key anomalous fragments, resulting in an aggregated anomalous representation Z:

[0119]

[0120] in, Indicates the first Long-term temporal dependency features corresponding to each time location; Indicates the first Long-term temporal dependency features corresponding to each time location; Indicates the total length of a time series segment or the total number of time steps; This represents the trainable weight matrix in the attention mechanism; This represents the bias term in the attention mechanism; This represents the context vector or attention parameter vector used in attention weight calculation. Represents the hyperbolic tangent activation function; Indicates exponentiation; Indicates the first The attention weights corresponding to each time segment are used to characterize the degree of contribution of that time segment to the anomaly identification results; This represents the anomaly representation data after attention-weighted aggregation. The anomaly representation data is a feature expression jointly formed by local temporal features, long-term temporal dependency features, and weighted enhanced key anomaly fragments, used to characterize the anomaly patterns corresponding to the current thermal safety risks of the power battery. This formula can increase the weight of key anomaly fragments in the final anomaly representation, reducing the impact of irrelevant fluctuations or noise fragments on the anomaly identification results.

[0121] S1314, perform anomaly classification and anomaly localization processing based on the anomaly characterization data to obtain anomaly identification results; wherein, the anomaly identification results include at least one of anomaly type, anomaly location, anomaly confidence level, and anomaly start time.

[0122] Specifically, anomaly classification is used to determine the type of the current anomaly, while anomaly localization is used to determine the location or region corresponding to the current anomaly. By simultaneously performing anomaly classification and localization, the anomaly identification results can characterize not only whether an anomaly exists in the power battery, but also the form in which the anomaly occurs and where it occurs. That is, the sensing and diagnostic agent outputs a unified message:

[0123]

[0124] Among them, e type Indicates the exception type, e loc Indicates an abnormal location, c anom Indicates the confidence level of the anomaly, t start The anomaly initiation time is indicated. The anomaly type indicates the category of the current thermal safety risk, including but not limited to: localized overheating, internal short circuit, external heat intrusion, poor local contact, cooling failure, insulation anomaly, and signs of heat diffusion. The anomaly location indicates the cell, module, or battery pack area corresponding to the anomaly, or the location corresponding to the object being collected. The anomaly confidence level indicates the credibility of the anomaly identification results by the sensing and diagnostic agent. The anomaly initiation time indicates the time when the anomaly state begins to appear or when the anomaly identification conditions are first met. By outputting the above anomaly identification results, clear evidence of anomaly behavior can be provided for subsequent mechanism consistency verification, correlation deduction processing, and candidate disposal strategy generation.

[0125] In one embodiment, a mechanism consistency check is performed in step S13 above, and an implementation method is provided herein. Based on the above embodiment, it includes:

[0126] S1321, when the risk level prediction result indicates the existence of thermal safety risk, the standardized state vector is input into the mechanism verification agent, so that the mechanism verification agent can call the mechanism rule base and determine at least one fault mechanism to be verified from the mechanism rule base according to the standardized state vector.

[0127] S1322, based on each fault mechanism to be verified, the thermal safety state of the power battery corresponding to the standardized state vector is predicted by the lumped parameter thermal network model in the mechanism verification agent, and the predicted state trajectory is obtained. The lumped parameter thermal network model is parameterized using the thermal model parameters obtained by parameter identification.

[0128] S1323, determine the measured state trajectory based on the standardized state vector, calculate the trajectory deviation between the predicted state trajectory and the measured state trajectory, and quantify the trajectory deviation to obtain the mechanism difference quantification value;

[0129] S1324, Based on the quantification value of mechanism differences, determine the consistency probability corresponding to each fault mechanism to be verified, and determine the mechanism consistency score based on the consistency probability;

[0130] S1325, when the consistency probability is greater than the preset probability threshold, the corresponding fault mechanism to be verified is determined as a candidate fault mechanism type, and the mechanism prediction diffusion path is determined according to the mechanism rule corresponding to the candidate fault mechanism type.

[0131] S1326 outputs mechanism consistency results including candidate fault mechanism type, mechanism consistency score, and mechanism prediction diffusion path.

[0132] In this embodiment, the mechanism verification agent determines whether the current state change conforms to the development law of the preset fault mechanism based on the thermal state change, electrical state change, execution state change, and operating condition change represented by the standardized state vector. By triggering the mechanism verification agent when the risk level prediction result indicates the existence of thermal safety risk, it is possible to further determine whether the abnormal state has a corresponding physical mechanism or control mechanism basis beyond anomaly identification, thereby improving the reliability of the determination of candidate fault root cause information.

[0133] The mechanism verification agent invokes a pre-established mechanism rule base and, based on the state change characteristics reflected in the standardized state vector, determines at least one fault mechanism to be verified from the mechanism rule base. The mechanism rule base describes the thermal safety fault mechanisms of the power battery and their state manifestation relationships. It can include discrimination rules corresponding to fault mechanisms such as internal short circuits, external heat intrusion, localized poor contact, cooling failure, abnormal heating, insulation abnormalities, and thermal effects from adjacent components. Different fault mechanisms to be verified correspond to different state change constraints. For example, internal short circuits typically manifest as localized abnormal heating accompanied by abnormal changes in electrical state; external heat intrusion typically manifests as an abnormal temperature change in the area affected by the external heat source; and cooling failure typically manifests as an overall increase in the temperature of the object being cooled or insufficient cooling effect.

[0134] For each fault mechanism to be verified, the mechanism verification agent predicts the thermal safety state of the power battery corresponding to the standardized state vector using a lumped parameter thermal network model, obtaining the predicted state trajectory. The lumped parameter thermal network model describes the thermal balance relationship of a cell or module. It can set corresponding heat generation terms, heat transfer terms, heat dissipation terms, or boundary conditions according to different fault mechanisms to be verified, so that the predicted state trajectory can reflect the theoretical state evolution process under the corresponding fault mechanism. For example, the thermal balance equation of a single cell or module can be expressed as:

[0135]

[0136] Among them, C th Q represents the heat capacity parameter. gen,i Indicates fever item, This represents the heat exchange term between the device and the environment. This represents the heat conduction term between adjacent nodes. The heat generation term Q. gem It can be composed of ohmic heat, polarization heat, etc., and can be represented as:

[0137]

[0138] Where I represents current, U oc U represents the open-circuit voltage. t ∂T and ∂U represent the terminal voltage, T represents the temperature, and ∂T and ∂U are respectively. oc This represents the partial derivative of the open-circuit voltage with respect to temperature.

[0139] The lumped-parameter thermal network model is parameterized using thermal model parameters obtained through parameter identification. Parameter identification refers to determining the parameters in the lumped-parameter thermal network model based on historical, experimental, or current operating data of the power battery. For example, parameter identification can be based on at least one of the following experimental data: hybrid pulse power characteristic test, constant current charge-discharge test, and thermal response test. The thermal model parameters to be identified may include internal resistance parameters, polarization parameters, heat capacity parameters, heat conduction parameters, heat dissipation parameters, and parameters related to cooling effects. The objective function for parameter identification can be expressed as:

[0140]

[0141] in, The parameters to be optimized are used to identify the objective function value; This represents the parameter vector to be identified, used to characterize the thermal model parameters in the lumped parameter thermal network model; This represents the total number of sampling points involved in parameter identification; Indicates the sampling point number; and This represents the weighting coefficient, used to adjust the relative importance of electrical state error and temperature state error in the objective function; Indicates the first Measured electrical state characteristic values ​​of each sampling point; Indicates the first The electrical state characteristic values ​​of each sampling point are obtained from the model prediction; Indicates the first Measured temperature status values ​​at each sampling point; Indicates the first The temperature state values ​​at each sampling point are predicted by the model. By parameterizing the lumped-parameter thermal network model using the thermal model parameters obtained through parameter identification, the model can be made closer to the actual thermal response characteristics of the current power battery system, reducing the verification error caused by the inconsistency between the fixed model parameters and the actual battery state.

[0142] The predicted state trajectory is a prediction of the change in the thermal safety state of the power battery over time, calculated by a lumped-parameter thermal network model. It can include at least one of the following: temperature prediction trajectory, voltage prediction trajectory, pressure prediction trajectory, temperature rise rate prediction trajectory, and local temperature difference prediction trajectory. Subsequently, time-series data related to the actual changes in the thermal safety state of the power battery are extracted from the standardized state vector to form a measured state trajectory that can be compared with the predicted state trajectory. The measured state trajectory can include at least one of the following: temperature measured trajectory, voltage measured trajectory, and pressure measured trajectory. To ensure comparability, the predicted and measured state trajectories can be aligned using the same time reference, the same state object, and the same feature dimensions.

[0143] Next, the trajectory deviation between the predicted and measured state trajectories is calculated. This refers to the difference calculation between predicted and measured values ​​for the same state characteristic and at the same time location, resulting in a trajectory deviation used to characterize the degree of inconsistency between the predicted and actual results. Trajectory deviation can include at least one of temperature trajectory deviation, voltage trajectory deviation, and pressure trajectory deviation. For example, the difference between the measured and predicted temperature trajectory, the difference between the measured and predicted voltage trajectory, and the difference between the measured and predicted pressure trajectory can be calculated separately. This process allows for evaluation from multiple state dimensions to determine whether the corresponding fault mechanism can explain the current thermal safety risk state, rather than relying solely on a single temperature change.

[0144] Quantifying trajectory deviations to obtain mechanistic difference quantification values ​​refers to converting the trajectory deviations between different state trajectories into a unified numerical evaluation result. The mechanistic difference quantification value represents the overall degree of difference between the predicted state trajectory and the measured state trajectory under the corresponding fault mechanism to be verified. For example, the mechanistic difference quantification value can be determined based on the weighted sum of temperature trajectory deviation, voltage trajectory deviation, and pressure trajectory deviation, as expressed as follows:

[0145]

[0146] Among them, C mech This represents the quantified value of the mechanism difference, where λ1, λ2, and λ3 represent the weighting coefficients, and T obs Represents the measured temperature trajectory, T pred U represents the temperature prediction trajectory. obs Represents the measured voltage trajectory, U pred P represents the voltage prediction trajectory. obs P represents the measured pressure trajectory. pred This represents the pressure prediction trajectory. Through the above quantification process, the multidimensional trajectory deviation can be unified into a comparable quantified value of mechanism difference, which facilitates the consistency evaluation of different fault mechanisms to be verified.

[0147] Furthermore, the quantified value of the mechanism difference is converted into a probability value characterizing the degree of consistency between the corresponding fault mechanism to be verified and the current measured state trajectory. The smaller the quantified value of the mechanism difference, the closer the predicted state trajectory obtained under the corresponding fault mechanism to be verified is to the measured state trajectory, and the more consistent the fault mechanism to be verified is with the current abnormal state; conversely, the larger the quantified value of the mechanism difference, the greater the difference between the predicted state trajectory under the corresponding fault mechanism to be verified and the measured state trajectory, and the lower the consistency between the fault mechanism to be verified and the current abnormal state. For example, an exponential mapping method can be used to determine the consistency probability, for example, expressed as: .

[0148] The consistency probability can be used as the scoring criterion for the corresponding fault mechanism to be verified, or the consistency probability can be converted into a corresponding score according to a preset scoring rule. The higher the consistency probability, the higher the corresponding mechanism consistency score; the lower the consistency probability, the lower the corresponding mechanism consistency score. By determining the mechanism consistency score based on the consistency probability, the mechanism consistency verification results can be continuously and quantitatively expressed, which facilitates subsequent correlation and inference processing to make a comprehensive judgment based on the credibility of different candidate fault mechanism types.

[0149] When P mech If the consistency probability exceeds a preset probability threshold, the corresponding fault mechanism to be verified is identified as a candidate fault mechanism type. When the consistency probability of multiple fault mechanisms to be verified is greater than the preset probability threshold, multiple candidate fault mechanism types can be retained, and their corresponding mechanism consistency scores can be retained respectively, to avoid prematurely eliminating possible fault sources during the mechanism verification stage. Otherwise, the credibility of the mechanism is reduced. After determining the candidate fault mechanism type, the direction or range of possible thermal safety risk development is predicted according to the fault development law and propagation constraints corresponding to the candidate fault mechanism type. Different candidate fault mechanism types correspond to different diffusion laws. For example, the mechanism rule corresponding to an internal short circuit can point to a path of expansion from the abnormal cell to adjacent cells or the module to which it belongs; the mechanism rule corresponding to cooling failure can point to the same cooling circuit or the area affected by the decrease in cooling capacity. By determining the mechanism prediction diffusion path according to the mechanism rule corresponding to the candidate fault mechanism type, the mechanism consistency result can not only characterize the fault mechanism type and credibility, but also provide an initial path basis for the subsequent deduction of thermal diffusion path information.

[0150] The mechanism type judgment result, quantization score result, and diffusion path prediction result obtained during the mechanism consistency verification process are all used as output. That is, the mechanism verification agent outputs a unified message:

[0151]

[0152] Where, m type Indicates the mechanism type, m score The consistency score represents the mechanism, m. path This indicates that the possible diffusion path is predicted based on the mechanism rules.

[0153] In one embodiment, correlation and deduction processing is performed based on anomaly identification results, mechanism consistency results, and risk level prediction results to obtain candidate fault root cause information and heat propagation path information, including:

[0154] S141, Obtain a battery thermal safety cognitive model constructed based on a prior knowledge base of mechanisms; wherein, the battery thermal safety cognitive model uses a knowledge graph to represent at least one of the following relationships among battery components, sensors and execution objects: connection relationship, thermal coupling relationship, fault propagation relationship, control dependency relationship and operating condition constraint relationship.

[0155] Specifically, in this embodiment, the mechanism prior knowledge base is used to store structural topology knowledge, thermal diffusion law knowledge, fault propagation law knowledge, typical thermal safety event knowledge, historical handling experience, and constraint relationships under different operating conditions related to the thermal safety of power batteries. The battery thermal safety cognitive model is used to provide mechanism constraints for the correlation deduction between anomaly identification results, mechanism consistency results, and risk level prediction results, so that the subsequently obtained candidate fault root cause information and thermal diffusion path information have data basis and mechanism basis.

[0156] The battery thermal safety cognitive model uses a knowledge graph format to represent at least one of the following relationships among battery components, sensors, and actuators: connectivity, thermal coupling, fault propagation, control dependence, and operating condition constraints. Battery components may include cells, modules, and battery packs; sensors may include temperature sensors, pressure sensors, gas sensors, smoke sensors, voltage acquisition units, current acquisition units, and insulation detection units; actuators may include contactors, cooling pumps, fans, valve bodies, thermal management control objects, charging control objects, and alarm control objects, etc. Through the expression of these nodes and relationships, the battery thermal safety cognitive model can simultaneously describe the structural relationships, heat transfer relationships, fault development relationships, and control action relationships of the power battery system.

[0157] S142, Based on the abnormal location in the anomaly identification results, determine the corresponding candidate fault initiation node in the battery thermal safety cognitive model.

[0158] Specifically, the abnormal location can point to a cell, module, battery pack area, sensor acquisition location, or other locatable object; the candidate fault initiation node is a node in the battery thermal safety cognitive model that corresponds to or is associated with the abnormal location. For example, when the abnormal location points to a certain cell, the corresponding cell node in the knowledge graph can be identified as the candidate fault initiation node; when the abnormal location points to a certain module area, the corresponding module node and its related cell nodes can be identified as candidate fault initiation nodes.

[0159] S143. Based on the candidate fault mechanism type and mechanism consistency score in the mechanism consistency results, candidate propagation directions associated with the candidate fault initiation node are screened in the battery thermal safety cognitive model, and candidate propagation directions that do not conform to the fault propagation relationship and mechanism law are eliminated to obtain the retained candidate propagation directions.

[0160] Specifically, different candidate fault mechanisms correspond to different propagation patterns. For example, the propagation direction corresponding to an internal short circuit can focus on the direction of expansion from the abnormal cell to adjacent cells or the module to which it belongs; the propagation direction corresponding to external heat intrusion can focus on the direction of expansion from the externally heated area to adjacent battery modules; and the propagation direction corresponding to cooling failure can focus on the range of battery modules affected by the same cooling circuit. Through screening and elimination, candidate propagation directions that do not match the current fault mechanism can be reduced.

[0161] S144. Based on the risk level prediction results and the thermal coupling relationship in the battery thermal safety cognitive model, the thermal diffusion range of the retained candidate propagation directions is extrapolated to obtain candidate thermal diffusion paths.

[0162] Specifically, the risk level prediction results are used to characterize the severity of the current thermal safety risk, while the thermal coupling relationship is used to characterize the degree of correlation between thermal effects between different battery components. By combining the risk level and the thermal coupling relationship for extrapolation, the node range, propagation sequence, and expansion path that the thermal safety risk may affect can be determined, so that the candidate thermal diffusion path is both compatible with the current risk level and conforms to the heat transfer conditions of the power battery system.

[0163] S145, through the consistency verification of multi-source evidence and the interpretability verification of mechanism in the battery thermal safety cognitive model, the candidate fault initiation node and candidate thermal diffusion path are verified, and the verified candidate fault initiation node is determined as candidate fault root cause information, and the verified candidate thermal diffusion path is determined as thermal diffusion path information.

[0164] Specifically, candidate fault initiation nodes and candidate thermal diffusion paths are verified through multi-source evidence consistency checks and mechanism interpretability checks in the battery thermal safety cognitive model. Multi-source evidence consistency checks determine whether anomaly identification results, mechanism consistency results, risk level prediction results, and state data are mutually supportive. For example, whether the anomaly location corresponds to the temperature change region, whether the candidate fault mechanism type corresponds to changes in electrical and thermal states, and whether the candidate thermal diffusion path corresponds to the temperature rise trend of adjacent nodes. Mechanism interpretability checks determine whether candidate fault initiation nodes and candidate thermal diffusion paths conform to the fault mechanism, thermal diffusion law, and operating condition constraints.

[0165] Verified candidate fault initiation nodes are identified as candidate fault root cause information, and verified candidate heat propagation paths are identified as heat propagation path information. Candidate fault root cause information represents the source or initiating object of a fault that may cause the current thermal safety risk, while heat propagation path information represents the direction, scope, or node path from which the current thermal safety risk may expand. Finally, after the above deduction, the output is:

[0166]

[0167] Msg cog The system uses the following parameters: root_cause (representing the correlation and deduction results), path (representing the thermal propagation path information), and priority (representing the handling priority or risk priority, used to characterize the objects, paths, or control directions that should be given priority when generating subsequent candidate handling strategies or conducting safety arbitration). By correlating and deducing the anomaly identification results, mechanism consistency results, and risk level prediction results through the battery thermal safety cognitive model, the system determines the candidate fault root cause information and thermal propagation path information, improving the matching degree between subsequent candidate handling strategies and the current sources and development trends of power battery thermal safety risks.

[0168] In one embodiment, step S15 is further described below. Based on the above embodiment, at least one candidate disposal strategy is generated, including:

[0169] S151 organizes the risk level prediction results, candidate fault root cause information, heat diffusion path information, anomaly identification results, and the current available set of actuators into structured prompt information.

[0170] Specifically, structured prompts are used to describe the current thermal safety risk scenario in a unified format. These prompts may include risk description fields, anomaly description fields, root cause description fields, propagation path description fields, and execution condition description fields. Through this field-based processing, the current risk level, anomaly manifestations, possible fault sources, possible propagation directions, and callable execution conditions can be uniformly expressed, providing structured input for subsequent retrieval and candidate treatment strategy generation. This process improves the completeness and parsability of the structured prompts.

[0171] S152, using structured prompts as query conditions, access a pre-configured external knowledge base through enhanced retrieval generation to obtain at least one of historical handling cases, control rules, and safety constraint information that match the current thermal safety risk scenario.

[0172] Specifically, the retrieval-augmented generation (RAG) method utilizes a process of first retrieving and then generating data, using relevant knowledge from an external knowledge base as the basis for generating candidate disposal strategies. This approach combines the current risk scenario with existing knowledge during the candidate strategy generation process, reducing the inadequacy of scenario adaptation caused by relying solely on internal model generation. The pre-configured external knowledge base is a pre-established set of knowledge related to power battery thermal safety control, which may include at least one of the following: a thermal runaway case library, a cell model parameter library, a fault rule library, a control strategy library, a regulatory constraint library, and a historical execution result library.

[0173] The database includes several key components: a thermal runaway case library (which stores the conditions for occurrence, abnormal manifestations, fault sources, propagation processes, and handling results of different thermal safety events); a cell model parameter library (which stores the thermal characteristic parameters, electrical characteristic parameters, temperature thresholds, and safety limits of different cell models); a fault rule library (which stores the state manifestations and handling principles corresponding to faults such as internal short circuits, external thermal intrusion, poor local contact, cooling failure, and insulation abnormalities); a control strategy library (which stores control actions, execution sequences, triggering conditions, and termination conditions under different risk scenarios); a regulatory constraint library (which stores constraint requirements related to the safe operation of power batteries, charging safety, vehicle safety, and alarm prompts); and a historical execution result library (which stores the risk changes, execution effects, and side effects after the execution of historical handling strategies).

[0174] During the search, the risk scope is determined based on the risk level prediction results, the fault type is determined based on the candidate fault root cause information, the diffusion scenario is determined based on the thermal diffusion path information, the abnormal behavior is determined based on the anomaly identification results, and the execution conditions that need to be prioritized or excluded are determined based on the current available actuator set. Thus, historical handling cases, control rules, and safety constraints matching the current thermal safety risk scenario are obtained from a pre-configured external knowledge base, providing a basis for subsequently generating candidate handling strategies.

[0175] S153, generate a large model from at least one of the input strategies, including structured prompt information, historical handling cases, control rules, and safety constraint information, and generate at least one candidate handling strategy through the strategy generation large model.

[0176] Specifically, the strategy generation model is used to generate candidate disposal strategies based on the current thermal safety risk scenario and retrieved external knowledge. By inputting structured prompts and retrieval results into the strategy generation model, the generated candidate disposal strategies can simultaneously possess scenario-specificity and knowledge constraints. The strategy generation model can determine the control strength of the candidate disposal strategy based on the risk level prediction results, the location of action of the candidate disposal strategy based on the anomaly identification results, the control direction of the candidate disposal strategy based on the candidate fault root cause information, the control range of the candidate disposal strategy based on the thermal diffusion path information, and the executable control actions in the candidate disposal strategy based on the current available actuator set. The candidate disposal strategy can include at least one of the following: action set, target, execution order, triggering condition, execution duration or termination condition, expected risk reduction effect, possible side effects, and generation confidence level. For example, a candidate disposal strategy could be to implement a 30% current limit on the abnormal module, shut down the charging request, activate high-level cooling, prohibit high-power discharge, and reassess the risk after continuous monitoring for 60 seconds. By generating at least one candidate disposal strategy, a comparable and screenable set of strategies can be provided for subsequent safety arbitration processing.

[0177] In one embodiment, the available status of the executors corresponding to each candidate disposal strategy is obtained, and a security arbitration process is performed on the candidate disposal strategies according to preset security constraint rules and the available status of each executor to determine the target disposal strategy, including:

[0178] S161, based on each candidate disposal strategy, the security arbitration agent performs strategy validity testing on each candidate disposal strategy, and selects a set of valid candidate strategies based on the strategy validity testing results; wherein, strategy validity testing includes at least one of strategy content integrity testing, generation timeliness testing, and generation confidence testing.

[0179] Specifically, the security arbitration agent is used to determine the effectiveness, security, execution conditions, conflict resolution, and risk-benefit evaluation of candidate disposal strategies. Strategy effectiveness detection includes at least one of the following: strategy content completeness detection, generation timeliness detection, and generation confidence detection. Strategy content completeness detection determines whether the candidate disposal strategy contains necessary information such as control actions, target objects, execution order, triggering conditions, execution duration, or termination conditions. Generation timeliness detection determines whether the candidate disposal strategy meets the response time requirements of the current hot security risk scenario. Generation confidence detection determines whether the generation confidence of the candidate disposal strategy meets the preset confidence requirements. Through these processes, candidate disposal strategies that are incomplete, timed out, or lack sufficient credibility can be eliminated.

[0180] S162, obtain the available status of the executor corresponding to each candidate disposal strategy in the set of valid candidate strategies.

[0181] Specifically, the actuator availability status characterizes whether the actuators required to execute the candidate disposal strategy are currently in a callable, responsive, and executable state. This can include whether the actuator is online, whether it is faulty, whether it can receive control commands, whether it can complete the corresponding action, and whether it is limited by the current vehicle operating conditions. By obtaining the actuator availability status, it can be determined whether the candidate disposal strategy has a practical basis for execution.

[0182] S163, based on the preset security constraint rules and the available status of the actuator, the candidate disposal strategies in the effective candidate strategy set are subject to security constraint review, and the first candidate strategy set is obtained by screening based on the security constraint review results.

[0183] Specifically, the safety constraint review is used to determine whether the control actions in the candidate disposal strategies meet the safety boundary requirements of the current thermal safety state, vehicle operating condition, and actuator state. For example, cooling actions need to meet the availability conditions of the cooling system, disconnection or isolation actions need to meet the corresponding electrical safety conditions, and control actions in charging scenarios need to meet charging safety requirements. The first set of candidate strategies is obtained based on the results of the safety constraint review.

[0184] For example, the security constraints that are allowed to be executed are defined as follows:

[0185]

[0186] Among them, T lim This is the upper limit of temperature. R is the upper limit of the rate of temperature rise. lim S is the insulation threshold. con S indicates that the contactor is in an available state. cool This indicates the cooling system is in an available state. Safety constraints include at least the upper limit of temperature, the upper limit of temperature rise rate, the insulation threshold, the contactor availability constraint, the cooling system availability constraint, the parking / driving condition constraint, and the charging state constraint.

[0187] S164, based on the available status of the executor, perform executability verification on the candidate disposal strategies in the first candidate strategy set, and filter and obtain the second candidate strategy set based on the executability verification results.

[0188] Specifically, executability verification is used to determine whether the control actions, target objects, and execution conditions in candidate disposal strategies match the capabilities of currently available actuators. For example, when a candidate disposal strategy includes a cooling enhancement action, it is necessary to determine whether the corresponding cooling actuator is available; when a candidate disposal strategy includes a disconnection action, it is necessary to determine whether the corresponding contactor has the necessary operating conditions; when a candidate disposal strategy includes an alarm action, it is necessary to determine whether the corresponding alarm channel can be invoked. Through executability verification, candidate disposal strategies that meet safety constraints but cannot be actually executed at present can be eliminated, resulting in a second set of candidate strategies.

[0189] S165, perform action conflict identification on the candidate disposal strategies in the second candidate strategy set, and perform conflict resolution processing on the candidate disposal strategies in the second candidate strategy set based on the action conflict identification results to obtain the third candidate strategy set.

[0190] Specifically, action conflict identification is used to determine whether there are situations such as mutually exclusive control actions, conflicting execution orders, conflicting target objects, or conflicting control objectives within or between candidate disposal strategies. Conflict resolution processing can, based on preset priorities, safety constraints, risk level prediction results, dependencies between control actions, and actuator availability, perform actions such as deletion, replacement, execution order adjustment, target object adjustment, or merging of conflicting candidate disposal strategies. This processing ensures that the candidate disposal strategies in the third candidate strategy set possess action consistency and implementability.

[0191] S166. Based on the preset risk-return evaluation indicators, evaluate the risk-return of the candidate disposal strategies in the third candidate strategy set to obtain the arbitration score of each candidate disposal strategy.

[0192] Specifically, the pre-defined risk-return evaluation indicators may include at least one of the following: expected risk reduction benefit, potential side effect risk, execution cost, execution feasibility, and strategy generation confidence level. Expected risk reduction benefit characterizes the reduction effect of the candidate disposal strategy on thermal safety risk after execution; potential side effect risk characterizes the potential impact of the candidate disposal strategy on the operation of the power battery, vehicle operation, or charging process; execution cost characterizes the resource consumption or functional limitations required to execute the candidate disposal strategy; and execution feasibility characterizes the feasibility of implementing the candidate disposal strategy under current execution conditions. Through risk-return evaluation, an arbitration score can be obtained to compare the overall suitability of each candidate disposal strategy.

[0193] For example, each candidate disposal strategy a i Calculate its score:

[0194]

[0195] Among them, B i R indicates that expected risk reduces returns. i Indicating potential side effects risk, C i F represents the execution cost. i Indicates feasibility of implementation, Q i The confidence level for strategy generation is represented by w1, w2, w3, w4, and w5, which are weighting coefficients. For example, candidate disposal strategies can carry fields such as expected risk reduction effect, possible side effects, execution cost, and generation confidence level. The security arbitration agent determines the expected risk reduction benefit, potential side effect risk, execution cost, and strategy generation confidence level based on the fields carried by the candidate disposal strategy.

[0196] S167. Based on the arbitration score, prioritize the candidate disposal strategies in the third candidate strategy set and determine the highest-ranked candidate disposal strategy as the target disposal strategy.

[0197] Specifically, the highest-ranked candidate disposal strategy represents a comprehensive and suitable solution that, under the current thermal safety risk scenario, simultaneously satisfies the requirements of strategy effectiveness, safety constraints, actuator availability, action consistency, and risk-benefit evaluation. This process ensures that the final determined target disposal strategy balances safety, executability, and risk control effectiveness. The security arbitration agent outputs a unified message:

[0198]

[0199] Where a* represents the target disposal strategy, score represents the arbitration score, and reason represents the arbitration basis.

[0200] In some implementations, when the effective candidate strategy set, the first candidate strategy set, the second candidate strategy set, or the third candidate strategy set is empty, or when the target handling strategy cannot be determined based on the arbitration score, a preset safety rollback rule can be invoked to determine the preset safety strategy from the local conservative strategy set as the target handling strategy. The preset safety rollback rule ensures that a handling strategy meeting basic safety protection requirements can still be output when candidate handling strategies are invalid, execution conditions are insufficient, action conflicts cannot be resolved, or the arbitration result cannot be determined. This safety rollback process improves the continuity and robustness of the power battery thermal safety control process under abnormal conditions.

[0201] Furthermore, the target handling strategy includes linkage control instructions corresponding to the risk level prediction results. After receiving the target handling strategy, the execution terminal can determine the current level of thermal safety risk of the power battery based on the risk level prediction results and execute linkage control instructions matching that risk level. The execution terminal includes one or more of the following: battery management system, thermal management controller, vehicle controller, and charging control unit; the linkage instructions include one or more of the following: current limiting, power reduction, cooling, isolation, disconnection, alarm, and evacuation guidance actions.

[0202] Specifically, when the risk level prediction result indicates a slight anomaly, the linkage control command may include at least one of increasing the sampling frequency, triggering a warning, and limiting unnecessary power requests, in order to enhance condition monitoring and reduce unnecessary loads without significantly affecting the normal operation of the vehicle. When the risk level prediction result indicates a moderate risk, the linkage control command may include at least one of implementing current limiting, power reduction, enhanced cooling, and limiting charging current, in order to reduce the possibility of the power battery continuing to heat up or the risk escalating. When the risk level prediction result indicates a high risk, the linkage control command may include at least one of stopping charging, limiting drive output, forcibly activating high-level cooling, and locking some high-risk functions, in order to suppress the continued development of thermal safety risks.

[0203] When the risk level prediction indicates critical thermal runaway, the coordinated control commands may include at least one of the following: disconnecting the contactor, implementing fault isolation, issuing a high-level alarm, and triggering personnel evacuation guidance, to reduce the risk of further fault escalation and prompt relevant personnel to take safety measures. When the risk level prediction indicates thermal runaway propagation, the coordinated control commands may include at least one of the following: performing full-enclosure isolation, emergency power cut-off, continuous alarm, and coordinating with the vehicle safety management module to execute emergency response logic, to provide a higher level of safety protection for the power battery system.

[0204] Figure 2 This is a schematic flowchart illustrating the post-strategy processing method provided in an embodiment of this application. Figure 2 As shown, based on the above embodiments, it also includes:

[0205] S21, Collect feedback data after the target handling strategy is executed; the feedback data includes multi-source state data used to characterize the thermal safety status of the power battery;

[0206] S22, Preprocess the feedback data to obtain the updated standardized state vector;

[0207] S23, Based on the updated standardized state vector, the thermal safety risk level of the power battery is predicted to obtain the updated risk level prediction result;

[0208] S24, when the updated risk level prediction results indicate the presence of thermal safety risks, anomaly identification and mechanism consistency verification are performed again based on the updated standardized state vector to obtain updated anomaly identification results and updated mechanism consistency results.

[0209] S25. Based on the updated anomaly identification results, updated mechanism consistency results, and updated risk level prediction results, the correlation and deduction process is re-performed to obtain updated candidate fault root cause information and updated heat diffusion path information.

[0210] S26. Based on the updated anomaly identification results, updated risk level prediction results, updated candidate fault root cause information and updated heat spread path information, candidate disposal strategies are regenerated, and a new target disposal strategy is determined through safety arbitration.

[0211] S27, the linkage control command corresponding to the new target disposal strategy is sent to the execution end for execution until the risk level prediction result obtained again indicates that there is no thermal safety risk.

[0212] In this embodiment, feedback data is used to reflect the execution effect of the target treatment strategy and the changes in the thermal safety status of the power battery. It includes multi-source state data characterizing the thermal safety status of the power battery, and may also include execution feedback information related to the execution process of the target treatment strategy. Execution feedback information may include whether the execution end received the target treatment strategy, whether the linkage control command was successfully executed, whether the actuator completed the corresponding action, the actuator action response time, and whether fault feedback occurred during execution. By collecting feedback data, the actual state changes after the execution of the target treatment strategy can be obtained, providing a basis for subsequent judgment on whether the thermal safety risk has decreased, persisted, or continued to expand.

[0213] The feedback data is preprocessed to obtain an updated standardized state vector. Preprocessing can employ methods corresponding to the multi-source state data preprocessing described above, including time synchronization, outlier removal, missing data completion, feature extraction, normalization, and matrix construction. Based on the updated standardized state vector, the thermal safety risk level of the power battery is predicted, resulting in an updated risk level prediction. This updated risk level prediction includes an updated risk value and an updated risk level, representing the change in thermal safety risk of the power battery after the implementation of the target mitigation strategy. When the updated risk level prediction indicates no thermal safety risk, the current risk is determined to be eliminated, and subsequent control procedures based on the current risk event are stopped. When the updated risk level prediction indicates a thermal safety risk, the aforementioned power battery thermal safety control method continues until a newly obtained risk level prediction indicates no thermal safety risk.

[0214] Figure 3 This is a schematic diagram of the structure of the power battery thermal safety control device provided in the embodiments of this application, as shown below. Figure 3 As shown, the power battery thermal safety control device 30 provided in this embodiment includes:

[0215] Data processing module 301 is used to collect multi-source state data for characterizing the thermal safety state of power batteries, preprocess the multi-source state data, and obtain a standardized state vector.

[0216] The risk prediction module 302 is used to predict the level of thermal safety risk of the power battery based on the standardized state vector and obtain the risk level prediction result.

[0217] The identification and verification module 303 is used to perform anomaly identification and mechanism consistency verification according to the standardized state vector when the risk level prediction result indicates the presence of thermal safety risk, and to obtain the anomaly identification result and mechanism consistency result.

[0218] The correlation inference module 304 is used to perform correlation inference processing based on the anomaly identification results, mechanism consistency results, and risk level prediction results to obtain the correlation inference results;

[0219] The strategy acquisition module 305 is used to generate at least one candidate disposal strategy based on the anomaly identification results, risk level prediction results, and correlation inference results.

[0220] The strategy acquisition module 305 is also used to acquire the available status of the executor corresponding to each candidate disposal strategy, and to perform security arbitration processing on the candidate disposal strategies according to the preset security constraint rules and the available status of each executor to determine the target disposal strategy;

[0221] The result output module 306 is used to send the target disposal strategy to the execution end, so that the execution end executes the linkage control instructions in the target disposal strategy that correspond to the risk level prediction results.

[0222] In one possible implementation, the risk prediction module 302 is specifically used for: obtaining a battery topology graph corresponding to the power battery system; wherein the battery topology graph includes a set of nodes, a set of edges, and an adjacency matrix for characterizing node connection relationships; organizing the standardized state vectors into nodes according to the nodes in the battery topology graph to obtain a node feature matrix; inputting the node feature matrix and the adjacency matrix into a graph convolutional network to extract topological features containing node state association information and topological coupling information; inputting the topological features corresponding to continuous time windows into a temporal recursive network to obtain the risk value at the current moment; mapping the risk value to a risk level according to a preset risk level mapping relationship to obtain a risk level prediction result including the risk value and the risk level.

[0223] In one possible implementation, the identification and verification module 303 is specifically used for: when the risk level prediction result indicates the existence of thermal safety risk, inputting a standardized state vector into the perception and diagnosis agent, and extracting local temporal features through a one-dimensional convolutional network in the perception and diagnosis agent; based on the local temporal features, extracting long-term time-dependent features through a bidirectional gated recurrent unit network in the perception and diagnosis agent; using an attention mechanism to perform weighted enhancement processing on the abnormal key segments in the long-term time-dependent features to obtain abnormal representation data; and performing abnormal classification and abnormal localization processing based on the abnormal representation data to obtain an abnormal identification result; wherein the abnormal identification result includes at least one of abnormal type, abnormal location, abnormal confidence level, and abnormal start time.

[0224] In one possible implementation, the identification and verification module 303 is further specifically used for: when the risk level prediction result indicates the existence of a thermal safety risk, inputting a standardized state vector into the mechanism verification agent, so that the mechanism verification agent can call the mechanism rule base, and determine at least one fault mechanism to be verified from the mechanism rule base according to the standardized state vector; according to each fault mechanism to be verified, predicting the thermal safety state of the power battery corresponding to the standardized state vector through the lumped parameter thermal network model in the mechanism verification agent, and obtaining the predicted state trajectory, wherein the lumped parameter thermal network model is parameterized using thermal model parameters obtained through parameter identification; The measured state trajectory is determined based on the standardized state vector, and the trajectory deviation between the predicted state trajectory and the measured state trajectory is calculated. The trajectory deviation is then quantified to obtain the mechanism difference quantification value. Based on the mechanism difference quantification value, the consistency probability corresponding to each fault mechanism to be verified is determined, and the mechanism consistency score is determined based on the consistency probability. When the consistency probability is greater than a preset probability threshold, the corresponding fault mechanism to be verified is determined as a candidate fault mechanism type, and the mechanism prediction diffusion path is determined according to the mechanism rules corresponding to the candidate fault mechanism type. The output includes the mechanism consistency result, which includes the candidate fault mechanism type, the mechanism consistency score, and the mechanism prediction diffusion path.

[0225] In one possible implementation, the correlation inference results include candidate fault root cause information and thermal diffusion path information.

[0226] In one possible implementation, the correlation inference module 304 is specifically used for: acquiring a battery thermal safety cognitive model constructed based on a mechanism prior knowledge base; wherein the battery thermal safety cognitive model uses a knowledge graph to represent at least one of the following relationships among battery components, sensors, and actuators: connection relationships, thermal coupling relationships, fault propagation relationships, control dependency relationships, and operating condition constraint relationships; determining the corresponding candidate fault initiation node in the battery thermal safety cognitive model based on the anomaly location in the anomaly identification results; and filtering the candidate fault initiation node in the battery thermal safety cognitive model based on the candidate fault mechanism type and mechanism consistency score in the mechanism consistency results. Candidate propagation directions are identified based on point association, and those that do not conform to the fault propagation relationship and mechanism are eliminated to obtain the retained candidate propagation directions. Based on the risk level prediction results and the thermal coupling relationship in the battery thermal safety cognitive model, the thermal diffusion range of the retained candidate propagation directions is extrapolated to obtain candidate thermal diffusion paths. Through the multi-source evidence consistency verification and mechanism interpretability verification in the battery thermal safety cognitive model, the candidate fault initiation nodes and candidate thermal diffusion paths are verified, and the verified candidate fault initiation nodes are determined as candidate fault root cause information, and the verified candidate thermal diffusion paths are determined as thermal diffusion path information.

[0227] In one possible implementation, the strategy acquisition module 305 is specifically used to: organize the risk level prediction results, candidate fault root cause information, heat spread path information, anomaly identification results, and the current available set of actuators into structured prompt information; use the structured prompt information as query conditions to access a pre-configured external knowledge base through a retrieval-enhanced generation method to obtain at least one of historical handling cases, control rules, and safety constraint information that match the current thermal safety risk scenario; input the structured prompt information and at least one of the historical handling cases, control rules, and safety constraint information into a large model to generate at least one candidate handling strategy through the strategy generation large model.

[0228] In one possible implementation, the strategy acquisition module 305 is further specifically configured to: perform strategy validity testing on each candidate disposal strategy through a security arbitration agent, and obtain a set of valid candidate strategies based on the strategy validity testing results; wherein, the strategy validity testing includes at least one of strategy content integrity testing, generation timeliness testing, and generation confidence testing; acquire the executor availability status corresponding to each candidate disposal strategy in the set of valid candidate strategies; perform security constraint review on the candidate disposal strategies in the set of valid candidate strategies according to preset security constraint rules and executor availability status, and obtain a first set of candidate strategies based on the security constraint review results; and, based on the executor... In the available state, the feasibility of candidate disposal strategies in the first candidate strategy set is verified, and a second candidate strategy set is obtained based on the feasibility verification results. Action conflicts are identified in the candidate disposal strategies in the second candidate strategy set, and conflict resolution is performed based on the action conflict identification results to obtain a third candidate strategy set. Risk and return evaluation is performed on the candidate disposal strategies in the third candidate strategy set according to preset risk and return evaluation indicators to obtain an arbitration score for each candidate disposal strategy. Based on the arbitration scores, the candidate disposal strategies in the third candidate strategy set are prioritized, and the candidate disposal strategy with the highest ranking is determined as the target disposal strategy.

[0229] In one possible implementation, a feedback monitoring module is further included. This module is configured to: collect feedback data after the execution of the target handling strategy; wherein the feedback data includes multi-source state data characterizing the thermal safety state of the power battery; preprocess the feedback data to obtain an updated standardized state vector; predict the level of thermal safety risk of the power battery based on the updated standardized state vector to obtain an updated risk level prediction result; when the updated risk level prediction result indicates the existence of a thermal safety risk, re-identify the anomaly and verify the mechanism consistency based on the updated standardized state vector to obtain an updated anomaly identification result and an updated mechanism consistency result; re-perform correlation deduction processing based on the updated anomaly identification result, the updated mechanism consistency result, and the updated risk level prediction result to obtain updated candidate fault root cause information and updated thermal diffusion path information; regenerate a candidate handling strategy based on the updated anomaly identification result, the updated risk level prediction result, the updated candidate fault root cause information, and the updated thermal diffusion path information, and determine a new target handling strategy through safety arbitration processing; and issue the linkage control command corresponding to the new target handling strategy to the execution end for execution until the re-obtained risk level prediction result indicates that there is no thermal safety risk.

[0230] The power battery thermal safety control device 30 provided in this embodiment can execute the method provided in the above method embodiment. Its implementation principle and technical effect are similar, and will not be described in detail here.

[0231] Figure 4 This is a structural schematic diagram of a vehicle provided in an embodiment of this application. Figure 4 As shown, the vehicle 40 provided in this embodiment includes at least one processor 401 and a memory 402. Optionally, the vehicle 40 also includes a communication component 403. The processor 401, memory 402, and communication component 403 are connected via a bus 404.

[0232] In a specific implementation, at least one processor 401 executes computer execution instructions stored in memory 402, causing at least one processor 401 to perform the above-described method.

[0233] The specific implementation process of processor 401 can be found in the above method embodiments, and its implementation principle and technical effect are similar. It will not be repeated here.

[0234] In the above embodiments, it should be understood that the processor can be a Central Processing Unit (CPU), or other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), etc. The general-purpose processor can be a microprocessor or any conventional processor. The steps of the method disclosed in this invention can be directly implemented by a hardware processor, or implemented by a combination of hardware and software modules within the processor.

[0235] The memory may include random access memory (RAM) and may also include non-volatile memory (NVM), such as at least one disk storage device.

[0236] The bus can be an Industry Standard Architecture (ISA) bus, a Peripheral Component Interconnect (PCI) bus, or an Extended Industry Standard Architecture (EISA) bus, etc. Buses can be categorized as address buses, data buses, control buses, etc. For ease of illustration, the buses shown in the accompanying drawings are not limited to a single bus or a single type of bus.

[0237] This application also provides a computer program product, including a computer program that, when executed by a processor, implements the above-described method.

[0238] This application also provides a computer-readable storage medium storing computer-executable instructions, which, when executed by a processor, implement the above-described method.

[0239] Finally, it should be noted that other embodiments of the invention will readily occur to those skilled in the art upon consideration of the specification and practice of the invention disclosed herein. This invention is intended to cover any variations, uses, or adaptations of the invention that follow the general principles of the invention and include common knowledge or customary techniques in the art not disclosed herein, and is not limited to the precise structures described above and shown in the accompanying drawings, and various modifications and changes can be made without departing from its scope. The scope of the invention is limited only by the appended claims.

Claims

1. A method for thermal safety control of a power battery, characterized in that, include: Multi-source state data for characterizing the thermal safety state of the power battery is collected, and the multi-source state data is preprocessed to obtain a standardized state vector. Based on the standardized state vector, the thermal safety risk level of the power battery is predicted, and the risk level prediction result is obtained. When the risk level prediction result indicates the existence of thermal safety risk, anomaly identification and mechanism consistency verification are performed according to the standardized state vector to obtain anomaly identification results and mechanism consistency results. Based on the anomaly identification results, the mechanism consistency results, and the risk level prediction results, correlation inference processing is performed to obtain the correlation inference results; Based on the anomaly identification results, the risk level prediction results, and the correlation inference results, at least one candidate disposal strategy is generated; Obtain the available status of the executor corresponding to each candidate disposal strategy, and perform security arbitration processing on the candidate disposal strategies according to the preset security constraint rules and the available status of each executor to determine the target disposal strategy; The target handling strategy is sent to the execution end so that the execution end executes the linkage control instructions in the target handling strategy that correspond to the risk level prediction result.

2. The method according to claim 1, characterized in that, The step of predicting the level of thermal safety risk of the power battery based on the standardized state vector, and obtaining the risk level prediction result, includes: Obtain the battery topology graph corresponding to the power battery system; wherein, the battery topology graph includes a set of nodes, a set of edges, and an adjacency matrix used to characterize the connection relationship between nodes; Based on the nodes in the battery topology diagram, the standardized state vector is organized into nodes to obtain a node feature matrix; The node feature matrix and the adjacency matrix are input into a graph convolutional network to extract topological features containing node state association information and topological coupling information; Input the topological features corresponding to continuous time windows into the temporal recursive network to obtain the risk value at the current time. Based on the preset risk level mapping relationship, the risk value is mapped to the risk level, resulting in a risk level prediction result that includes both the risk value and the risk level.

3. The method according to claim 1, characterized in that, When the risk level prediction result indicates the existence of a thermal safety risk, anomaly identification is performed based on the standardized state vector to obtain anomaly identification results, including: When the risk level prediction result indicates the existence of thermal safety risk, the standardized state vector is input into the perception and diagnosis agent, and local temporal features are extracted through the one-dimensional convolutional network in the perception and diagnosis agent. Based on the local temporal features, long-term temporal dependency features are extracted through the bidirectional gated recurrent unit network in the perceptual diagnostic agent. An attention mechanism is used to perform weighted enhancement processing on the abnormal key segments in the long temporal dependency features to obtain abnormal representation data; Anomaly classification and localization processes are performed on the anomaly characterization data to obtain anomaly identification results; wherein, the anomaly identification results include at least one of anomaly type, anomaly location, anomaly confidence level, and anomaly start time.

4. The method according to claim 1, characterized in that, When the risk level prediction result indicates the existence of a thermal safety risk, a mechanism consistency check is performed based on the standardized state vector to obtain a mechanism consistency result, including: When the risk level prediction result indicates the presence of thermal safety risk, the standardized state vector is input into the mechanism verification agent, so that the mechanism verification agent can call the mechanism rule base and determine at least one fault mechanism to be verified from the mechanism rule base according to the standardized state vector; Based on each of the fault mechanisms to be verified, the thermal safety state of the power battery corresponding to the standardized state vector is predicted by using the lumped parameter thermal network model in the mechanism verification agent, and the predicted state trajectory is obtained. The lumped parameter thermal network model is parameterized using the thermal model parameters obtained through parameter identification. The measured state trajectory is determined based on the standardized state vector, and the trajectory deviation between the predicted state trajectory and the measured state trajectory is calculated. The trajectory deviation is then quantified to obtain the mechanism difference quantification value. Based on the quantified value of the mechanism difference, the consistency probability corresponding to each of the fault mechanisms to be verified is determined, and the mechanism consistency score is determined based on the consistency probability. When the consistency probability is greater than a preset probability threshold, the corresponding fault mechanism to be verified is determined as a candidate fault mechanism type, and the mechanism prediction diffusion path is determined according to the mechanism rules corresponding to the candidate fault mechanism type. The output includes the candidate fault mechanism type, the mechanism consistency score, and the mechanism consistency result of the predicted diffusion path.

5. The method according to claim 1, characterized in that, The correlation simulation results include candidate fault root cause information and thermal diffusion path information; Accordingly, the correlation inference process based on the anomaly identification result, the mechanism consistency result, and the risk level prediction result to obtain the correlation inference result includes: Obtain a battery thermal safety cognitive model based on a prior knowledge base of mechanisms; wherein, the battery thermal safety cognitive model uses a knowledge graph to represent at least one of the following relationships among battery components, sensors, and actuators: connection relationship, thermal coupling relationship, fault propagation relationship, control dependency relationship, and operating condition constraint relationship; Based on the abnormal location in the anomaly identification result, the corresponding candidate fault initiation node is determined in the battery thermal safety cognitive model; Based on the candidate fault mechanism type and mechanism consistency score in the mechanism consistency results, candidate propagation directions associated with the candidate fault initiation node are screened in the battery thermal safety cognitive model, and candidate propagation directions that do not conform to the fault propagation relationship and mechanism law are eliminated to obtain the retained candidate propagation directions. Based on the risk level prediction results and the thermal coupling relationship in the battery thermal safety cognitive model, the thermal diffusion range of the retained candidate propagation directions is extrapolated to obtain candidate thermal diffusion paths; The candidate fault initiation node and the candidate thermal diffusion path are verified through the multi-source evidence consistency verification and mechanism interpretability verification in the battery thermal safety cognitive model. The candidate fault initiation node that passes the verification is determined as the candidate fault root cause information, and the candidate thermal diffusion path that passes the verification is determined as the thermal diffusion path information.

6. The method according to claim 5, characterized in that, The step of generating at least one candidate disposal strategy based on the anomaly identification result, the risk level prediction result, and the correlation inference result includes: The risk level prediction results, the candidate fault root cause information, the heat diffusion path information, the anomaly identification results, and the current available actuator set are organized into structured prompt information; Using the structured prompt information as the query condition, the system accesses a pre-configured external knowledge base through enhanced retrieval generation to obtain at least one of the following: historical handling cases, control rules, and safety constraint information that match the current thermal safety risk scenario. The structured prompt information and at least one of the input strategies from the historical handling cases, control rules and security constraints are used to generate a large model, and at least one candidate handling strategy is generated through the strategy generation large model.

7. The method according to claim 1, characterized in that, The step of obtaining the available status of the executors corresponding to each candidate disposal strategy, and performing security arbitration processing on the candidate disposal strategies according to preset security constraint rules and the available status of each executor to determine the target disposal strategy includes: Based on each of the candidate disposal strategies, a security arbitration agent performs strategy effectiveness detection on each of the candidate disposal strategies, and obtains a set of effective candidate strategies based on the strategy effectiveness detection results; wherein, the strategy effectiveness detection includes at least one of strategy content integrity detection, generation timeliness detection, and generation confidence detection; Obtain the available status of the executor corresponding to each candidate disposal strategy in the set of effective candidate strategies; Based on the preset security constraint rules and the available status of the actuator, the candidate disposal strategies in the effective candidate strategy set are subject to security constraint review, and the first candidate strategy set is obtained by screening based on the security constraint review results. Based on the available status of the executor, the executability of the candidate disposal strategies in the first candidate strategy set is verified, and a second candidate strategy set is obtained by filtering based on the executability verification results. Action conflict identification is performed on the candidate handling strategies in the second candidate strategy set, and conflict resolution processing is performed on the candidate handling strategies in the second candidate strategy set based on the action conflict identification results to obtain a third candidate strategy set. Based on the preset risk-return evaluation index, the candidate disposal strategies in the third candidate strategy set are evaluated for risk and return, and an arbitration score is obtained for each candidate disposal strategy. Based on the arbitration score, the candidate disposal strategies in the third candidate strategy set are prioritized and ranked, and the candidate disposal strategy with the highest ranking is determined as the target disposal strategy.

8. The method according to any one of claims 1-7, characterized in that, Also includes: Collect feedback data after the target handling strategy is executed; wherein, the feedback data includes multi-source state data for characterizing the thermal safety state of the power battery; The feedback data is preprocessed to obtain the updated standardized state vector. Based on the updated standardized state vector, the thermal safety risk level of the power battery is predicted, and the updated risk level prediction result is obtained. When the updated risk level prediction result indicates the presence of thermal safety risk, anomaly identification and mechanism consistency verification are performed again based on the updated standardized state vector to obtain updated anomaly identification results and updated mechanism consistency results. Based on the updated anomaly identification results, the updated mechanism consistency results, and the updated risk level prediction results, the correlation and deduction processes are re-performed to obtain updated candidate fault root cause information and updated heat diffusion path information. Based on the updated anomaly identification results, the updated risk level prediction results, the updated candidate fault root cause information, and the updated heat spread path information, candidate handling strategies are regenerated, and a new target handling strategy is determined through safety arbitration. The linkage control command corresponding to the new target disposal strategy is sent to the execution end for execution until the risk level prediction result obtained again indicates that there is no thermal safety risk.

9. A vehicle, characterized in that, include: Memory, processor; The memory stores computer-executed instructions; The processor executes computer execution instructions stored in the memory, causing the processor to perform the method as described in any one of claims 1 to 8.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores computer-executable instructions, which, when executed by a processor, are used to implement the method as described in any one of claims 1 to 8.

Images