Man-machine collaborative battery thermal runaway decision-making method and system based on dynamic trust

Abstract

The invention discloses a man-machine collaborative battery thermal runaway decision-making method and system based on dynamic trust, and the method achieves the efficient response and safety optimization in the risk disposal of battery thermal runaway through the five links of multi-source sensor data fusion, dynamic trust evaluation, hierarchical decision-making mode, linkage control execution, and closed-loop learning feedback. The rapid calculation advantage of AI is ensured, and artificial scene cognition and responsibility judgment are fully utilized, so that an explainable, executable and optimizable fire control system is constructed. According to the method, efficient response, flexible cooperation, reliable execution and continuous optimization can be realized in new energy automobile battery thermal runaway risk disposal, the disposal efficiency and safety are superior to those of the prior art, and the method has remarkable technical progress and practical application value.

Description

Technical Field

[0001] This invention relates to a human-machine collaborative battery thermal runaway decision-making method and system based on dynamic trust, belonging to the field of human-machine collaborative fire protection decision-making and control technology. Background Technology

[0002] Thermal runaway in new energy vehicle batteries is a core cause of major safety accidents such as fires and explosions, and its early warning and handling are directly related to the operational safety of the entire vehicle and the safety of passengers. However, existing battery thermal runaway risk management technologies still have significant limitations.

[0003] In the early warning stage, mainstream methods mostly rely on threshold monitoring of a single sensor, such as monitoring the temperature rise on the surface of the battery cell using a temperature sensor, or using abnormal fluctuations in electrical parameters such as voltage and current as trigger signals. Some studies have attempted to introduce infrared imaging, gas sensors, and acoustic detection, but most still rely on the core logic of "alarm upon exceeding the limit." While these methods achieve the basic function of early warning, they struggle to depict the complex thermo-electro-chemical coupling evolution process inside the battery, often leading to false alarms, missed alarms, and delayed warnings.

[0004] In the emergency response phase, the industry generally adopts fixed plans or relies on human experience for emergency operations. Fixed plans typically pre-set operational steps based on standardized scenarios, such as "detect temperature rise—initiate cooling—power off—isolate," but their rigidity leads to a lack of flexibility in complex and ever-changing accident scenes. Purely manual decision-making relies on operator experience and intuitive judgment, which can easily lead to excessive cognitive load, decision-making delays, or operational errors when rapid integration of multi-source information is required. Some studies have introduced intelligent algorithms to generate response strategies, but the algorithm results often remain at the theoretical optimization level and fail to be deeply integrated with the physical structure of the battery pack and on-site operability, resulting in poor execution performance.

[0005] Therefore, existing technologies suffer from three main defects: ① Disconnect between human and machine decision-making: The rigid strategies generated by automated systems cannot match the dynamic changes on site, while human judgment lacks systematic integration of multi-source data, which can easily lead to delays or misoperations in handling; ② Poor executability of strategies: The fire extinguishing or cooling instructions generated by the algorithm fail to be deeply coupled with the physical scene, resulting in waste of handling resources or even increased risks; ③ Rigid allocation of authority and responsibility: Existing systems cannot dynamically assess the confidence level of AI (artificial intelligence) and the operator's status. Once data is missing or there is a sudden anomaly, the system is prone to losing its adaptive ability. Summary of the Invention

[0006] The technical problem to be solved by this invention is to provide a human-machine collaborative battery thermal runaway decision-making method and system based on dynamic trust, so as to realize the physical precision mapping of risk disposal strategies and the real-time transfer of decision weights, thereby significantly improving the safety and timeliness of disposal in complex scenarios.

[0007] To solve the above-mentioned technical problems, the present invention adopts the following technical solution: A human-machine collaborative decision-making method for thermal runaway of new energy vehicle batteries based on dynamic trust includes the following steps: Step 1: Real-time acquisition of multi-dimensional status information of the new energy vehicle battery and the surrounding environment, including battery cell surface temperature, concentration of harmful and combustible gases in the battery compartment, battery voltage and current, smoke and flame conditions in the battery compartment, abnormal sounds in the battery compartment, and the vehicle's location information. Step 2: After preprocessing the multi-dimensional state information, a multi-source fusion algorithm based on Bayesian inference is used to fuse the preprocessed multi-dimensional state information to obtain the risk situation index. ; The risk of thermal runaway in new energy vehicle batteries is classified into three categories: low risk, medium risk, and high risk, based on risk status indicators. If the value is greater than the lower limit of the low-risk range, proceed to step 3; otherwise, store the preprocessed multi-dimensional status information in the case library and return to step 1 to obtain the multi-dimensional status information for the next moment. Step 3: A dual-channel trust review mechanism is adopted to calculate the credibility of AI handling and human handling under the current multi-dimensional state information, and to calculate the comprehensive credibility through dynamic weight allocation. ; Step 4, based on overall credibility and risk situation indicators Under the combined effect of multiple thresholds, the corresponding battery disposal mode is selected based on the current multi-dimensional state information, and the selected battery disposal mode is converted into physical actions and executed. Step 5: After the physical action is performed, key indicators that can reflect the treatment effect are collected in real time to determine whether the risk of thermal runaway has been eliminated. If not, return to step 1 and repeat steps 1-4 until the risk of thermal runaway is eliminated. Step 6: After the thermal runaway risk is eliminated, record the multi-dimensional status information, battery disposal mode, physical execution actions and disposal effects throughout the entire thermal runaway risk elimination process, form a complete disposal report, store it in the case library, and realize the human-machine collaborative decision-making closed loop.

[0008] Compared with the prior art, the present invention, employing the above technical solution, has the following technical effects: 1. This invention achieves real-time fusion between human and AI decision-making through dual-channel trust review and dynamic weight allocation, avoiding the delay caused by human judgment alone and overcoming the uncontrollable risks brought about by isolated AI calculations. This enables the system to make a reasonable response quickly in the early stages of a crisis, thereby significantly improving the timeliness of the response.

[0009] 2. This invention uses a multi-threshold hierarchical decision-making mechanism to flexibly switch to automatic, semi-automatic, or manually-led modes based on the dynamic evolution of the risk situation. This avoids the rigidity and limitations caused by fixed permission allocation in existing technologies, enabling the handling strategy to be dynamically adjusted according to the level of risk, making it more suitable for complex and ever-changing actual environments and enhancing the flexibility of human-machine collaboration.

[0010] 3. This invention introduces a physical feasibility verification mechanism at the algorithm output end to ensure that the generated disposal plan can directly correspond to the execution devices such as on-site spraying, heat insulation, smoke exhaust, and power cut-off. This solves the problem that the theoretically optimal solution in the existing system is not feasible in reality, thus enabling the disposal instructions to be successfully implemented.

[0011] 4. This invention introduces dynamic trust assessment into the human-machine collaboration mechanism. When the AI's credibility is insufficient, it can be taken over by humans. When the human's condition is not good, the AI ​​provides assistance, thereby avoiding the risk of failure of a single entity and making the entire process more redundant and reliable, ensuring the safety and reliability of the system.

[0012] 5. This invention employs a closed-loop learning and feedback mechanism to continuously feed the handling effects and manual operation records back into the system model, forming an iterative optimization process. This mechanism not only accumulates scenario experience but also continuously improves the accuracy and interpretability of the model, enabling the system to have the ability to evolve and self-improve over the long term.

[0013] 6. In semi-automatic mode, the present invention achieves human-machine consensus through a manual confirmation process, which fully leverages the rapid computing advantages of AI while retaining the human's right to judge responsibility and scenario. This effectively solves the distrust problem caused by machine black box decision-making in existing technologies and improves the acceptability and scalability of the system. Attached Figure Description

[0014] Figure 1 This is a flowchart of the human-machine collaborative decision-making method for thermal runaway of new energy vehicle batteries based on dynamic trust, as described in this invention. Detailed Implementation

[0015] Embodiments of the present invention are described in detail below, examples of which are illustrated in the accompanying drawings. The embodiments described below with reference to the accompanying drawings are exemplary and are only used to explain the present invention, and should not be construed as limiting the present invention.

[0016] New energy vehicles face the potential risk of battery thermal runaway during charging, driving, and parking. Thermal runaway refers to the irreversible, self-heating chain reaction that occurs when a battery cell or module is subjected to internal defects or external stimuli, leading to a rapid rise in temperature, the release of large amounts of gas, and the ejection of flames. Because thermal runaway is sudden, conductive, and difficult to control, improper handling can easily trigger serious fires or even explosions.

[0017] Traditional handling methods typically rely on manual judgment by firefighters or automated response strategies based on fixed algorithms. However, relying on a single entity has significant limitations: human response is often delayed, making it difficult to take immediate action; while automated algorithms, although responding quickly, often lack coupling between their strategies and actual physical devices, resulting in "optimal on paper, impractical in reality." Based on this, this invention proposes a human-machine collaborative battery thermal runaway decision-making method based on dynamic trust, such as... Figure 1 As shown, its overall structure consists of five core components: multi-source sensor data fusion, dynamic trust assessment, hierarchical decision-making model, physical execution mapping, and closed-loop learning optimization. The following will provide a detailed explanation of each of the five steps.

[0018] First, the system acquires multi-dimensional state information of the new energy vehicle battery and its environment in real time through a multi-source sensor acquisition module, including temperature, gas concentration, visual, acoustic, electrical parameters, and location information. This heterogeneous data is input into the fusion module after time synchronization. Temperature and acoustic state information need to be compared with the actual temperature and ambient noise before fusion. For example, in summer, when the battery is about to experience thermal runaway, its temperature will definitely rise, but the rate of temperature change is generally not very large (not as noticeable as in winter). Therefore, the actual ambient temperature must be considered, prioritizing the battery temperature over its rate of temperature change. Similarly, the acoustic state information fusion also requires comparison with ambient noise to ensure accuracy. To guarantee information consistency and reliability, this invention employs a Bayesian inference-based multi-source fusion method, the basic form of which is: , in, This indicates the risk status of the system, where D represents multi-source sensor data. For posterior risk probability, Let be the likelihood function of the sensor data. This represents the prior probability. This method can map heterogeneous data to a unified risk posture indicator and ultimately output a risk index. The formula is: , in, Indicates the first Sensor-like data, This represents the corresponding feature extraction function. For weighting parameters. Risk index. They are divided into three categories: low risk, medium risk, and high risk. This classification result will serve as input for subsequent trust assessment and decision-making model selection.

[0019] Secondly, based on risk assessment, this invention addresses the disconnect in human-machine collaboration through a "dual-channel trust review" mechanism. This mechanism simultaneously calculates the credibility of AI actions and the credibility of human actions, and forms a comprehensive credibility score through dynamic weight allocation.

[0020] 1) Credibility of AI processing It is obtained by weighting the data integrity coefficient, model consistency coefficient, and historical accuracy factor. It can be generally expressed as: , in, Indicates data integrity. Indicates model consistency. Indicates historical accuracy. and All are weighted coefficients.

[0021] Data integrity is defined as whether the sensor data received by the AI ​​in the current situation is comprehensive, continuous, and without any abnormal omissions. In thermal runaway response, data such as temperature, gas concentration, voltage, current, and smoke images must be available simultaneously. If some key data is missing, the AI's generated response plan may be biased. If the AI's decision is based on comprehensive information, the data integrity is high; if data is lost, delayed, or interfered with, the data integrity is low.

[0022] Model consistency is defined as whether an AI model's internal inference results remain stable and consistent when faced with heterogeneous data from multiple sources, and whether they align with existing physical models. For example, if a temperature sensor shows a rapid rise, but a visual camera does not capture significant smoke, and the AI ​​model still outputs a "high-risk" response, it indicates that its decision is inconsistent with multi-source evidence, resulting in low model consistency. Conversely, if the trends from different input sources logically corroborate each other, and the AI's output is consistent with the physical model (such as a thermal diffusion model), it indicates high model consistency. A model with high consistency means that the AI ​​is not merely making "black box judgments," but rather can match multi-source data and physical models, making it more credible.

[0023] Historical accuracy is defined as the degree of agreement between the intervention plans predicted by the AI ​​model in similar past scenarios and the actual intervention results. Historical accuracy is obtained through cumulative statistics from the closed-loop learning module. For example, it measures whether the intervention plans predicted early by the AI ​​in several past battery anomaly events matched the final outcomes. If a model version can correctly identify early signs of thermal runaway and provide effective intervention suggestions in 90% of cases, it is considered to have high historical accuracy; conversely, if it cannot, its historical accuracy is considered low.

[0024] 2) Credibility of manual handling This is obtained by weighting the operator status factor, experience level factor, and subjective confirmation factor. The formula can be expressed as: , in, Indicates the operator status. Indicates experience level. Indicates subjective confirmation level. and All are weighted coefficients.

[0025] Operator Status Definition: In this invention, operator status refers to the real-time physiological and cognitive load level of a human operator during the handling process, including fatigue level, concentration level, and task load. In the scenario of thermal runaway of new energy vehicle batteries, if the operator is on duty for a long time or performing continuous tasks, fatigue or decreased attention may occur, leading to an increased risk of judgment errors. The system can indirectly assess the operator's status through indicators such as operator interaction frequency, response latency, and heart rate monitoring (if wearable devices are available). Operators in good condition have a higher operator status; those in poor condition have a lower operator status.

[0026] Experience Level Definition: In this invention, experience level refers to a tiered description of an operator's professional knowledge and practical experience accumulated in areas such as new energy vehicle battery fires and thermal runaway response. Experienced operators are often able to identify complex risk signals more quickly and provide emergency judgments that surpass algorithms in special circumstances. Experience level can be quantified and graded based on factors such as training and certification level, the number of past handling cases, and the effectiveness of historical decisions. The higher the training and certification level, the more past handling cases, and the better the historical decision-making effectiveness, the higher the experience level, and vice versa.

[0027] Definition of Subjective Confirmation: In this invention, subjective confirmation refers to the degree of trust an operator has in the correctness and feasibility of a system-recommended solution. In semi-automatic mode, the AI ​​will provide a solution, but whether to execute it often requires operator confirmation. If the operator believes the system solution is reasonable and feasible and confirms it quickly, the subjective confirmation degree is high; conversely, if the operator hesitates, repeatedly modifies, or delays confirmation, the subjective confirmation degree is low. Subjective confirmation degree can be quantified through user interface interaction (confirmation button click speed, number of modifications). The faster the confirmation button click speed and the fewer the modifications, the higher the subjective confirmation degree, and vice versa.

[0028] 3) Overall credibility It can be represented as: , in, These are dynamic weighting parameters that can be adaptively adjusted according to the needs of the scenario. Through this mechanism, the system can dynamically determine when AI should lead the decision-making, when human intervention is needed, or when both should be combined.

[0029] In terms of overall credibility Risk Index Under the combined effect of these factors, the hierarchical decision-making module adopts multi-threshold judgment to automatically select the appropriate handling mode: ① When ≥ or When the risk level is high, it will directly enter automatic mode, where AI will autonomously trigger and execute the response; ② When < < or When the risk level is medium, the system enters semi-automatic mode, where AI provides a response plan, which is then confirmed and executed by the operator; ③ At that time or When the risk level is low, the system switches to manual mode, allowing operators to make independent decisions while the system provides auxiliary references. This tiered mechanism breaks the rigid pattern of traditional systems that are either "fully manual" or "fully automatic," enabling flexible transfer of responsibilities between humans and machines.

[0030] Following the determination of the decision-making model, the human-machine collaborative system of this invention maps the generated disposal strategy to the on-site physical execution chain, ensuring the practical operability of the solution. Typical execution actions include: activating sprinkler or aerosol devices to cool the battery; activating heat insulation curtains or isolation devices to prevent the fire from spreading; activating smoke extraction devices to reduce the concentration of toxic gases; executing emergency power-off protection to avoid the risk of electric arcs or short circuits; and activating the evacuation guidance system to ensure the safe evacuation of personnel. This process ensures that the "theoretical strategy" output by AI can be mapped to directly drivable physical actions, solving the problem in existing technologies where "the theory is optimal but not executable on-site."

[0031] After the response is implemented, the effect monitoring module collects key indicators in real time, such as the rate of temperature change, the trend of gas concentration decline, and the flame / smoke situation. These are compared with expected targets, such as the rate of temperature change becoming zero or negative, the concentration of harmful gases decreasing to a safe range, the absence of open flames, and low smoke concentration, to determine whether the response has met requirements. If the effect is insufficient, the system will prompt manual intervention or a mode switch to ensure the situation remains under control. Simultaneously, the closed-loop learning module records the inputs, outputs, and feedback during the execution process as training samples, iteratively optimizing the risk model and trust mechanism. Through continuous learning, the system can continuously improve its judgment accuracy and response efficiency in similar scenarios, achieving a self-evolutionary closed loop of "experience accumulation - model evolution - capability enhancement."

[0032] To facilitate the explanation of the specific application of the human-machine collaborative decision-making mechanism based on dynamic trust assessment of the present invention, this embodiment provides a detailed simulation of a sudden battery thermal runaway scenario at a new energy vehicle charging station. The scenario is set as follows: ① Scenario location: A public fast-charging station in an urban area, where multiple new energy vehicles are simultaneously charging, equipped with fire sprinkler systems, ventilation systems, automatic power-off switches, and emergency spray explosion suppression equipment. ② Monitoring object: A pure electric sedan in fast-charging mode, whose battery pack may experience thermal runaway and spread due to abnormal temperature rise in individual cells. ③ Participating entities: Including the artificial intelligence decision engine (hereinafter referred to as "AI engine"), the human-machine collaborative decision-making terminal (i.e., the operator interaction interface), and the on-site execution device in the system of this invention. ④ Risk characteristics: Rapid rise in battery temperature, abnormal gas concentration inside the chamber, alarm of harmful gas sensors, and risks such as flame spray, toxic gas leakage, and chain explosions. Against this backdrop, the present invention, through dynamic trust assessment and a multi-round human-machine interaction mechanism, achieves rapid, accurate, and executable generation and implementation of handling strategies.

[0033] When the vehicle enters the charging state, the system first collects data from multiple sensors in real time, including: a cell surface temperature sensor; a voltage and current real-time detection module; a gas sensor; a camera and infrared imaging equipment; and an ambient temperature and humidity monitoring module. The AI ​​engine performs preliminary processing on the above data using a multi-source fusion algorithm and establishes a risk trend curve. When the temperature rise rate of a single cell exceeds a threshold and is accompanied by abnormal voltage fluctuations, the AI ​​engine determines that there are potential signs of thermal runaway. At this time, the system marks the status as "Level 1 Risk," automatically generates an early warning prompt on the operator terminal, and provides the first round of handling suggestions, such as: "It is recommended to slow down the charging rate and closely monitor the gas concentration."

[0034] Within the next 30 seconds, the gas sensor detected an increase in hydrogen fluoride concentration, while the infrared image showed a cluster of bright spots in a localized area of ​​the battery pack. The AI ​​engine upgraded the risk level to "Level 2 Risk". ① AI Engine's suggested solutions: Automatically cut off the fast charging current; activate low-power spray cooling; close the exhaust system to prevent gas diffusion. ② Operator intervention: The operator confirmed the suggestion through the human-machine interaction terminal but offered a correction: Since the site was a semi-open space, immediately closing the exhaust system could lead to the accumulation of toxic gases in a localized area, endangering personnel. Therefore, the operator selected "Maintain partial exhaust and set low speed operation" in the system interface. ③ Trust weight adjustment: Based on this human-machine interaction record, the system adjusted the weight allocation between AI and human decision-making. In the "Ventilation Control" dimension, the weight of human opinion was increased, while in the "Electrical Disposal" and "Spray Control" dimensions, the AI ​​weight remained high.

[0035] After the first round of collaborative decision-making, the system issued the following instructions: disconnect the vehicle's charging power; activate the low-power spray device; maintain low-speed ventilation. Following execution, sensor feedback indicated a limited rate of temperature reduction in the battery cells, with some areas still showing a continuous warming trend. ① Based on this, the AI ​​engine automatically triggered a second calculation, proposing an upgrade plan: increase the spray intensity to medium power; activate the isolation air curtain device to separate high-risk areas; prepare an emergency smoke extraction passage. ② At this point, the operator confirmed through video monitoring that no other vehicles were too close to the vehicle and approved the deployment of the isolation air curtain; however, regarding the smoke extraction passage, the operator, considering wind direction, chose to postpone it. The system then updated the trust weights again, further increasing the weight of human opinion in the "environmental control" dimension.

[0036] During continuous monitoring, if the battery cell temperature begins to slowly decrease after the spray is enhanced, but the gas concentration does not decrease significantly, the third round of coordination is initiated: ① AI suggests: immediately activate high-power smoke extraction to prevent toxic accumulation. ② The operator, concerned that the smoke extraction direction may spread to the personnel evacuation route, proposes an alternative: delay smoke extraction and add wind direction guiding devices. ③ After weighted calculation by the system, the final execution strategy is: first deploy wind direction guiding devices, then implement tiered smoke extraction. This approach ensures the dispersion of toxic gases while avoiding secondary hazards. During this process, the AI ​​engine learns that "in semi-open spaces, smoke extraction control needs to combine wind direction and personnel evacuation paths" and records this in the case library as a basis for subsequent model iterations.

[0037] After approximately 5 minutes of continuous intervention, the cell temperature gradually returned to a controllable range, the gas concentration decreased, and the flame jet disappeared. The system determined that the risk had been eliminated and automatically lowered the risk level to "safe observation state." At this point, the system of this invention archives the entire multi-round human-machine collaboration process: ① Decision process record: each round of AI suggestions, human corrections, and final strategy; ② Trust weight change curve: dynamic adjustment of the AI / human ratio in each dimension; ③ Status and feedback of execution devices: including the real-time action trajectories of spraying, ventilation, and isolation equipment. Finally, a complete incident handling report is generated and entered into the case library for subsequent training and optimization.

[0038] As can be seen from the above description of specific implementation methods, this invention not only provides a complete human-machine collaborative fire-fighting decision-making method, but also establishes a close coupling relationship at the algorithm and execution levels. Its core value lies in: achieving complementary advantages between humans and machines through dynamic trust assessment; achieving flexible response through a hierarchical decision-making model; achieving on-the-ground handling through physical executability verification; and achieving system self-evolution through a closed-loop learning mechanism.

[0039] This invention also proposes a human-machine collaborative decision-making system for thermal runaway of new energy vehicle batteries based on dynamic trust, including a multi-dimensional state information acquisition module, a fusion module, a dual-channel trust review module, a hierarchical decision-making module, an effect monitoring module, and a disposal report recording module; wherein, The multi-dimensional status information acquisition module is used to acquire multi-dimensional status information of new energy vehicle batteries and the surrounding environment in real time, including battery cell surface temperature, concentration of harmful gases and combustible gases in the battery compartment, battery voltage and current, smoke and flame conditions in the battery compartment, abnormal sounds in the battery compartment, and vehicle location information. The fusion module is used to preprocess multi-dimensional state information and then employs a Bayesian inference-based multi-source fusion algorithm to fuse the preprocessed multi-dimensional state information to obtain risk status indicators. ; The risk of thermal runaway in new energy vehicle batteries is classified into three categories: low risk, medium risk, and high risk, based on risk status indicators. If the value is greater than the lower limit of the low-risk range, the system enters the dual-channel trust review module; otherwise, the pre-processed multi-dimensional status information is stored in the case library and returned to the multi-dimensional status information acquisition module to acquire the multi-dimensional status information for the next moment. The dual-channel trust review module employs a dual-channel trust review mechanism to calculate the credibility of AI handling and human handling under the current multi-dimensional state information, and calculates the overall credibility through dynamic weight allocation. ; The hierarchical decision-making module is used to assess overall credibility. and risk situation indicators Under the combined effect of multiple thresholds, the corresponding battery disposal mode is selected based on the current multi-dimensional state information, and the selected battery disposal mode is converted into physical actions and executed. The effect monitoring module is used to collect key indicators that reflect the treatment effect in real time after the physical action is performed, to determine whether the risk of thermal runaway has been eliminated. If it has not been eliminated, it returns to the multi-dimensional status information acquisition module to reacquire multi-dimensional status information until the risk of thermal runaway is eliminated. The disposal report recording module is used to record multi-dimensional status information, battery disposal mode, physical execution actions and disposal effects throughout the entire process of thermal runaway risk disposal after the thermal runaway risk has been eliminated, forming a complete disposal report, which is stored in the case library to realize a closed loop of human-machine collaborative decision-making.

[0040] In existing technologies, human-machine collaboration mostly relies on a single dominant entity (either human-first or AI-first), lacking quantification and review of the trustworthiness of both parties. This invention proposes for the first time a dual-channel trust review, which quantifies the trustworthiness of both AI and human handling separately and introduces a comprehensive trust score as the basis for decision-making. This mechanism achieves dynamic complementarity between humans and machines, avoiding delays or misjudgments caused by the failure of a single entity, enabling the system to automatically select the optimal decision path in complex environments and improving overall reliability.

[0041] Existing technologies typically employ static permission allocation models, with the responsibility fixed to either humans or AI. This invention, based on a comprehensive trust score, sets a multi-threshold hierarchical decision-making model that can flexibly switch between automatic, semi-automatic, and manual modes. This significantly enhances the flexibility of human-machine collaboration, enabling response strategies to adapt to the dynamic evolution of the risk situation. It avoids the rigidity caused by fixed permission allocations and ensures a rapid and appropriate response at different stages of the crisis.

[0042] Existing AI algorithms largely remain at the theoretical level, and the generated solutions may not correspond to actual on-site devices, resulting in a problem where the solutions are calculable but not executable. This invention establishes a mapping relationship between the algorithm's output and on-site devices such as sprinklers, heat insulation, smoke extraction, and power cutoff, and ensures the solutions are implemented through executability verification. This guarantees that all decision results match the physical execution chain, fundamentally solving the problem of strategies being divorced from reality. It ensures that the solutions are not only calculable but also feasible, significantly enhancing the system's practical value.

[0043] Existing technologies mostly involve single-step decision-making and lack continuous optimization capabilities. This invention introduces a feedback collection and learning mechanism into the handling process, feeding the handling effects and manual operation records back into the risk model to achieve continuous iterative optimization. This enables the system to have self-evolution capabilities, gradually improving the accuracy of risk prediction and the rationality of decision-making with the increase in the number of applications, forming a long-term accumulated experience base, and thus exhibiting higher efficiency and stability in future scenarios.

[0044] In the handling of thermal runaway risks in new energy vehicle batteries, this invention can simultaneously improve response efficiency and handling safety, which is significantly better than existing technologies and has broad application prospects and promotional value.

[0045] Based on the same inventive concept, this application provides a computer device, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it implements the steps of the aforementioned human-machine collaborative decision-making method for thermal runaway of new energy vehicle batteries based on dynamic trust.

[0046] Based on the same inventive concept, embodiments of this application provide a computer-readable storage medium storing a computer program that, when executed by a processor, implements the steps of the aforementioned human-machine collaborative decision-making method for thermal runaway of new energy vehicle batteries based on dynamic trust.

[0047] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0048] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0049] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0050] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0051] The above embodiments are merely illustrative of the technical concept of the present invention and should not be construed as limiting the scope of protection of the present invention. Any modifications made to the technical solutions based on the technical concept proposed in this invention shall fall within the scope of protection of this invention.

Claims

1. A human-machine collaborative decision-making method for thermal runaway of new energy vehicle batteries based on dynamic trust, characterized in that, Includes the following steps: Step 1: Real-time acquisition of multi-dimensional status information of the new energy vehicle battery and the surrounding environment, including battery cell surface temperature, concentration of harmful and combustible gases in the battery compartment, battery voltage and current, smoke and flame conditions in the battery compartment, abnormal sounds in the battery compartment, and the vehicle's location information. Step 2: After preprocessing the multi-dimensional state information, a multi-source fusion algorithm based on Bayesian inference is used to fuse the preprocessed multi-dimensional state information to obtain the risk situation index. ; The risk of thermal runaway in new energy vehicle batteries is classified into three categories: low risk, medium risk, and high risk, based on risk status indicators. If the value is greater than the lower limit of the low-risk range, proceed to step 3; Otherwise, store the preprocessed multi-dimensional state information into the case library and return to step 1 to obtain the multi-dimensional state information for the next moment; Step 3: A dual-channel trust review mechanism is adopted to calculate the credibility of AI handling and human handling under the current multi-dimensional state information, and to calculate the comprehensive credibility through dynamic weight allocation. ; Step 4, based on overall credibility and risk situation indicators Under the combined effect of multiple thresholds, the corresponding battery disposal mode is selected based on the current multi-dimensional state information, and the selected battery disposal mode is converted into physical actions and executed. Step 5: After the physical action is performed, key indicators that can reflect the treatment effect are collected in real time to determine whether the risk of thermal runaway has been eliminated. If not, return to step 1 and repeat steps 1-4 until the risk of thermal runaway is eliminated. Step 6: After the thermal runaway risk is eliminated, record the multi-dimensional status information, battery disposal mode, physical execution actions and disposal effects throughout the entire thermal runaway risk elimination process, form a complete disposal report, store it in the case library, and realize the human-machine collaborative decision-making closed loop.

2. The human-machine collaborative decision-making method for thermal runaway of new energy vehicle batteries based on dynamic trust as described in claim 1, characterized in that, In step 2, preprocessing the multi-dimensional state information includes synchronizing the multi-dimensional state information over time. The formula for the multi-source fusion algorithm based on Bayesian inference is as follows: ， in, This indicates the risk status of the system, where D represents the preprocessed multi-dimensional status information data. For posterior risk probability, The likelihood function is the result of preprocessing the multi-dimensional state information data. This is the prior probability; The preprocessed multi-dimensional state information is fused to obtain risk status indicators. The formula is: ， in, Indicates the obtained first Class data, Indicates the first Feature extraction function corresponding to class data, Here, n is the weight parameter, and n is the number of categories of data to be acquired. The risk of thermal runaway in new energy vehicle batteries is classified into three categories: low risk, medium risk, and high risk, based on risk status indicators. If the value is greater than the lower limit of the low-risk range, proceed to step 3.

3. The human-machine collaborative decision-making method for thermal runaway of new energy vehicle batteries based on dynamic trust as described in claim 1, characterized in that, In step 3, the credibility of AI processing The formula is obtained by weighting data integrity, model consistency, and historical accuracy, and is expressed as follows: ， in, Indicates data integrity. Indicates model consistency. Indicates historical accuracy. and All are weighted coefficients; data integrity indicates whether the acquired multi-dimensional state information is comprehensive, continuous, and without abnormal omissions; model consistency indicates the consistency between the AI's reasoning results for the acquired multi-dimensional state information and the physical model results corresponding to the multi-dimensional state information; historical accuracy indicates the degree of consistency between the AI's handling decisions and the actual handling decisions at historical moments similar to the currently acquired multi-dimensional state information. Credibility of manual handling It is obtained by weighting operator status, experience level, and subjective confirmation, and the formula is expressed as follows: ， in, Indicates the operator status. Indicates experience level. Indicates subjective confirmation level. and All are weighted coefficients; Operator status represents the operator's real-time physiological and cognitive load level, including fatigue level, concentration level, and task load; Experience level represents a graded description of the professional knowledge and practical experience accumulated by the operator in the process of handling thermal runaway of new energy vehicle batteries; Subjective confirmation level represents the operator's subjective confidence in the correctness and feasibility of the handling decisions provided by AI. Overall credibility Represented as: ， in, These are dynamic weight parameters.

4. The human-machine collaborative decision-making method for thermal runaway of new energy vehicle batteries based on dynamic trust as described in claim 1, characterized in that, The specific process of step 4 is as follows: Preset first threshold Second threshold , ,when or When the risk level is high, select automatic mode, where AI provides a battery disposal solution and executes it automatically; when... or When the risk level is medium, select the semi-automatic mode, where AI provides a battery disposal plan, which is then executed after manual confirmation; when... or When the risk level is low, select manual mode, which allows the operator to independently decide on and execute the battery disposal plan. Physical actions include: activating sprinkler or aerosol devices to cool the battery; activating heat shields or isolation devices to prevent the fire from spreading; activating smoke extraction devices to reduce the concentration of toxic gases; implementing emergency power cut-off protection to avoid the risk of electric arcs or short circuits; and activating the evacuation guidance system to ensure the safe evacuation of personnel.

5. A human-machine collaborative decision-making system for thermal runaway of new energy vehicle batteries based on dynamic trust, characterized in that, The system includes a multi-dimensional status information acquisition module, a fusion module, a dual-channel trust review module, a hierarchical decision-making module, an effect monitoring module, and a disposal report recording module; among which... The multi-dimensional status information acquisition module is used to acquire multi-dimensional status information of the new energy vehicle battery and the surrounding environment in real time, including the surface temperature of the battery cell, the concentration of harmful gases and combustible gases in the battery compartment, the battery voltage and current, the smoke and flame situation in the battery compartment, abnormal sounds in the battery compartment, and the vehicle's location information. The fusion module is used to preprocess multi-dimensional state information and then fuse the preprocessed multi-dimensional state information using a Bayesian inference-based multi-source fusion algorithm to obtain a risk status index. ; The risk of thermal runaway in new energy vehicle batteries is classified into three categories: low risk, medium risk, and high risk, based on risk status indicators. If the value is greater than the lower limit of the low-risk range, the system enters the dual-channel trust review module; otherwise, the pre-processed multi-dimensional status information is stored in the case library and returned to the multi-dimensional status information acquisition module to acquire the multi-dimensional status information for the next moment. The dual-channel trust review module employs a dual-channel trust review mechanism to calculate the credibility of AI handling and human handling under the current multi-dimensional state information, and calculates the comprehensive credibility through dynamic weight allocation. ; The hierarchical decision-making module is used to assess overall credibility. and risk situation indicators Under the combined effect of multiple thresholds, the corresponding battery disposal mode is selected based on the current multi-dimensional state information, and the selected battery disposal mode is converted into physical actions and executed. The effect monitoring module is used to collect key indicators that reflect the treatment effect in real time after the physical action is performed, and to determine whether the risk of thermal runaway has been eliminated. If it has not been eliminated, it returns to the multi-dimensional status information acquisition module to reacquire multi-dimensional status information until the risk of thermal runaway is eliminated. The disposal report recording module is used to record multi-dimensional status information, battery disposal mode, physical execution actions and disposal effects throughout the entire process of thermal runaway risk disposal after the thermal runaway risk is eliminated, forming a complete disposal report, which is stored in the case library to realize a closed loop of human-machine collaborative decision-making.

6. A computer device comprising a memory, a processor, and a computer program stored in the memory and capable of running on the processor, characterized in that, When the processor executes the computer program, it implements the steps of the human-machine collaborative decision-making method for thermal runaway of new energy vehicle batteries based on dynamic trust as described in any one of claims 1 to 4.

7. A computer-readable storage medium storing a computer program, characterized in that, When the computer program is executed by the processor, it implements the steps of the human-machine collaborative decision-making method for thermal runaway of new energy vehicle batteries based on dynamic trust as described in any one of claims 1 to 4.

Images