A battery thermal runaway early warning method and system

Abstract

The application discloses a battery thermal runaway early warning method and system, relates to the technical field of battery safety, and uses a current step drop process as a diagnostic excitation. Instead of directly predicting a thermal runaway risk, the application first establishes a normal thermal relaxation baseline and a prediction interval, so as to describe how the temperature should change after the current drops under normal circumstances. Deviation of the actual temperature from the baseline is used as an abnormality judgment basis, so that normal thermal responses and abnormal residuals can be more clearly distinguished. Moreover, the prediction is only performed after the window is valid and decoupling is established, so that the influence of invalid sampling points, cooling disturbances and normal load changes on the prediction result of the model can be reduced. The application further discloses a battery thermal runaway early warning system for implementing the battery thermal runaway early warning method.

Description

Technical Field

[0001] This application relates to the field of battery safety technology, and in particular to a method and system for early warning of battery thermal runaway. Background Technology

[0002] Lithium-ion batteries are an important component of existing energy storage systems, such as ternary lithium batteries and lithium iron phosphate batteries. Current solutions to battery thermal runaway mainly involve the following approaches: The first type is a technology that uses multi-source sensor signals such as temperature, voltage, pressure, gas concentration, and casing deformation to detect or provide graded early warning of thermal runaway. This type of technology typically collects data such as temperature, voltage, pressure, smoke, and characteristic gases of individual battery cells or battery modules in real time, and judges whether the battery has a risk of thermal runaway based on preset thresholds, temperature difference changes, deformation changes, or the results of multi-indicator fusion.

[0003] The second type is a technology based on machine learning or deep learning models for battery temperature prediction and thermal runaway warning. This type of technology usually inputs time series data such as current, voltage, temperature, and SOC into CNN, LSTM, attention mechanism or other neural network models to predict the future temperature or internal temperature of the battery, and then outputs the warning level based on the predicted temperature, temperature rise rate or alarm threshold.

[0004] As can be seen from the above, the problems with the existing technology are as follows: 1. The first approach relies heavily on temperature thresholds, temperature rise rates, voltage anomalies, gas concentrations, or multi-sensor fusion signals for judgment. Although this approach can issue an alarm when the battery state is obviously abnormal, it is difficult to distinguish whether the temperature rise is caused by normal load heat generation, cooling disturbances, or abnormal spontaneous heat generation inside the battery when there are high-rate charging and discharging, changes in ambient temperature, or changes in the state of the cooling system. This can easily lead to false alarms or missed alarms. 2. The second approach usually directly inputs time-series data such as current, voltage, temperature, and SOC into the model, which then directly predicts the future temperature or the probability of thermal runaway. The problem with this approach is that although the model may predict that the temperature will rise, it cannot fully explain the reason for the temperature rise. Especially under dynamic load and dynamic cooling conditions, the model may mistake normal high-rate heat generation, changes in cooling state, or step transients for abnormal trends of thermal runaway. Summary of the Invention

[0005] This application provides a battery thermal runaway early warning method and system to solve the above-mentioned technical problems.

[0006] In a first aspect, this application provides a battery thermal runaway early warning method, including: Step S100: Acquire battery operating data and preprocess the data, including current; Step S200: Obtain the current step drop diagnostic window; Step S300: Verify the validity of the current step drop diagnostic window. If the validity requirement is met, proceed to step S400; otherwise, return to step S100. Step S400: Based on historical normal operation data, establish a normal thermal relaxation baseline that matches the current operating condition of the target battery and obtain the prediction range; Step S500: Calculate the current-temperature decoupling characteristic parameters based on the normal thermal relaxation baseline, and determine whether the current and temperature decoupling is valid based on the current-temperature decoupling characteristic parameters. If it is valid, proceed to step S600; otherwise, return to step S100. Step S600: Predict the future state of the battery based on operating data and current-temperature decoupling characteristic parameters; Step S700: Output the thermal runaway warning level based on the future state of the battery.

[0007] Preferably, step S100 includes: Step S110: Obtain battery operating data, including current, and also including at least one of voltage, surface temperature, ambient temperature, and cooling parameters; Step S120: Preprocess the running data; Step S130: Determine whether the running data meets the anomaly judgment conditions and critical danger conditions. If the anomaly determination criteria are met but the critical hazard criteria are not met, proceed to step S200. If the critical hazard conditions are met, a hazard warning will be triggered directly. If neither of them is met, return to step S110.

[0008] Preferably, step S200 includes passively generating a current step drop diagnostic window under the premise of satisfying the first diagnostic window generation conditions. The first diagnostic window generation conditions include that the target battery current is at a first preset level before the current step and drops to a second preset level after the step, and the stability requirements are met before and after the step. Alternatively, the current step drop diagnostic window can be actively generated under the premise of satisfying the second diagnostic window generation conditions. The second diagnostic window generation conditions include that the battery meets preset safety constraints and power availability conditions.

[0009] Preferably, the actively generated current step drop diagnostic window includes: The first step is to generate candidate step actions by at least one of the following methods: short-time current limiting, short-time rate reduction, short-time current interruption, reduction of charging power, or reduction of discharging power, and to generate a set of multiple candidate step actions in this manner. The second step is to calculate the diagnostic information gain and safety cost of each candidate step action based on the operating data and battery condition parameters. The third step is to comprehensively evaluate the optimal candidate step action based on the diagnostic information gain and safety cost, and use it as the diagnostic window for current step drop.

[0010] Preferably, diagnostic information gain As shown below: ; , , , , All indicate ; Security Cost As shown below: ; , , , All indicate ; As shown below: ; Indicates the first m The overall evaluation value of each candidate step jump action. λ This represents the weighting coefficient for security costs.

[0011] Preferably, step S300 includes: if the following conditions are met simultaneously, the system is considered to be valid and proceeds to step S400; otherwise, it returns to step S100. Condition 1: The current drop reaches the preset step amplitude; Condition 2: The current fluctuation within the reference window before the step and the diagnostic window after the step is less than the preset stability threshold. Condition 3: Cooling parameters are within the preset normal sample range; Condition 4: The proportion of valid sampling points meets the requirements.

[0012] Preferably, step S400 includes: Step S410: Select normal samples that match the current operating conditions of the target battery from historical normal operating data. The operating condition matching factors include SOC, SOH, ambient temperature, cooling parameters and pre-step load parameters. Step S420: The ARIMAX model uses the temperature series of historical normal operation data as autoregressive input, and at least two of the following as exogenous variables: current, ambient temperature, SOC, SOH, internal resistance, and cooling parameters. The output is the normal thermal relaxation baseline and prediction range of the target battery within the current step drop diagnostic window. .

[0013] Preferably, step S500 includes: Step S510: Calculate the current-temperature decoupling characteristic parameters, which include abnormal temperature residual, abnormal heat generation power, and current step decoupling index. ; Indicates abnormal temperature residuals. This indicates the actual surface temperature of the target battery. This indicates the temperature corresponding to the normal thermal relaxation baseline generated by the ARIMAX model; ; Indicates abnormal heat generation capacity. This indicates the target battery's equivalent heat capacity. This represents the rate of change of the actual temperature of the target battery over time. This indicates the heat power lost by the target battery to the environment or cooling system. This indicates normal electrochemical heat generation power; ; This represents the current step decoupling index. and Both represent weighting coefficients, and , This represents the integral of abnormal heat production within the diagnostic window after the step jump. This represents the normal heat production integral within the reference window before the step jump. To prevent extremely small positive numbers with a denominator of zero, This indicates that the actual temperature is higher than the upper limit of the prediction range. U(t) The cumulative duration, Indicates the duration of the diagnostic window after the step jump; Step S520, the rules for determining whether current and temperature decoupling is valid are as follows: Within the stable range after a step drop in current, if the abnormal temperature residual... The value remains positive and exceeds the upper bound of the prediction interval. U(t) And the rate of increase of abnormal temperature residuals Acceleration of abnormal temperature residual growth Abnormal heat generation power or current step decoupling index If at least one of the criteria exceeds the corresponding threshold at multiple consecutive valid sampling times, the target battery is determined to have experienced current-temperature decoupling.

[0014] Preferably, step S600 includes: Step S610: Construct a gated Transformer residual evolution prediction model. The input features of the prediction model include core diagnostic features and auxiliary operating condition features. Step S620: The gated Transformer residual evolution prediction model outputs the result parameters of the future state of the battery and the prediction confidence. Step S630: Determine whether the prediction confidence level is lower than the set confidence level threshold and whether the result parameter is within the preset critical range. If either of the two conditions is met, proceed to step S640 for verification; otherwise, proceed directly to step S700. Step S640: Generate a new current step drop diagnostic window and repeat steps S300 to S600 to obtain secondary result parameters. Compare the result parameters obtained from the two diagnoses. If the verification requirements are met, proceed to step S700; otherwise, enter the observation state.

[0015] Preferably, in step S610, the core diagnostic features include abnormal temperature residual, abnormal temperature residual growth rate, abnormal temperature residual growth acceleration, abnormal heat generation power, current step decoupling index, and temperature difference residual between adjacent battery cells, and the auxiliary operating condition features include at least two of the following: step relative time features, current, voltage, surface temperature, ambient temperature, SOC, SOH, internal resistance, and cooling parameters. In step S620, the result parameters include at least one of the following: abnormal temperature residual trajectory, future abnormal heat generation power trajectory, probability of thermal runaway stage, and remaining safe time. The prediction confidence level is as follows: ; in, Indicates the prediction confidence level. This represents the confidence term for the predicted dispersion. This represents the confidence term for the predicted quantile interval width. This indicates the arrangement of input feature terms. This represents the confidence term representing the proportion of valid sampling points. This represents the online calibration error confidence term. , , , and All represent weighting coefficients; In step S630, if the prediction confidence level is lower than the confidence level threshold, or if at least one of the abnormal temperature residual, abnormal heat generation power and remaining safety time is in the preset critical range, the process proceeds to step S640 for verification; otherwise, the process proceeds directly to step S700. In step S640, the review requirement is deemed met if the following conditions are met simultaneously: Condition one: , This represents the confidence level of a prediction. Indicates the confidence level of the second-order prediction; Condition 2: The abnormal temperature residual trajectories obtained from both diagnostic predictions are continuously within a range of no less than [a certain value]. .

[0016] Preferably, step S700 includes: Step S710: Based on the gated Transformer residual evolution prediction model, if the comprehensive prediction confidence level meets the standard, proceed to step S720; otherwise, return to step S100. Step S720: Generate a preliminary warning level according to predetermined rules. Step S730: Calculate the confidence level of abnormal spontaneous heat generation inside the target battery:

[0017] This indicates the confidence level of spontaneous heat generation due to internal anomalies. Indicates the first k The strength of evidence for abnormal heat production in the secondary diagnosis The score representing the consistency between two consecutive diagnostic results. This indicates the overall prediction confidence level. 、 、 and All of these represent weight coefficients, and the sum of all weights is 1; Step S740, if satisfied and If the warning level determined in step S720 is not met, the warning level will be raised by one level before outputting. If not, the warning level determined in step S720 will be output directly.

[0018] Preferably, the predetermined rule in step S720 is as follows: The alert level rules include satisfying any one of criteria one through three. Criterion 1: Satisfies And the duration is greater than 0.2 hours. This indicates the trajectory of future abnormal temperature residuals. H represents the upper bound of the normal forecast interval for the future, and H represents the length of the forecast window. Criterion 2: The condition is satisfied for multiple consecutive moments. , Indicates the trajectory of future abnormal heat generation power; Criterion 3: Furthermore, there is an upward trend in abnormal residuals or abnormal heat generation power. Indicates the predicted remaining safe time; The restriction-level warning rules include satisfying any one of criteria four through six. Criterion 4: Satisfies And the duration is greater than 0.5 hours; Criterion 5: And it lasted for multiple moments, among which Indicates the first heat generation power threshold; Criterion Six: ; Danger level warning rules include satisfying any one of criteria seven through nine. Criterion 7: Satisfied and , This indicates the growth slope of the residual trajectory of future abnormal temperatures. This indicates the threshold for the slope of the residual growth of abnormal temperatures at hazardous levels. Criterion 8: , Indicates the second heat generation power threshold and Greater than ; Criterion Nine: ; Emergency warning rules include satisfying any one of criteria ten to twelfth. Criterion 10: ; Criterion 11: , This indicates a prediction of future temperatures. This indicates a preset dangerous temperature threshold; Criterion 12: , This indicates the third heat production power threshold.

[0019] Preferably, the strength of evidence for anomalous heat production and the consistency score are as follows:

[0020] This indicates the persistence score of abnormal temperature residuals exceeding the upper bound of the prediction interval. This indicates the score for abnormal heat generation power. This represents the score of the current step decoupling index. The score indicates the abnormal heat production trend. , , and All represent weighting coefficients;

[0021] This represents the difference in the strength of evidence for abnormal heat production between two diagnoses. Indicates the maximum permissible difference in evidence;

[0022] This represents the confidence level of a prediction. This indicates the confidence level of the second-order prediction. If it has not been verified, then... =0.

[0023] Secondly, this application provides a battery thermal runaway early warning system for implementing an early warning method. The early warning system includes a data input layer, a diagnostic analysis layer, a prediction and decision layer, and a result output layer.

[0024] The early warning method and system of this application have at least the following beneficial effects: (1) This application uses the current step drop process as a diagnostic excitation. Its technical significance is that after the current drops, the normal current heat generation should decrease. If the temperature continues to deviate from the normal thermal relaxation baseline, it is easier to identify abnormal spontaneous heat generation that cannot be explained by normal load heat generation. Using the thermal relaxation response after the current step drop, a normal thermal relaxation baseline and prediction range that match the current operating condition are established. Based on the current-temperature decoupling characteristic parameters, it is determined whether the target battery has current-temperature decoupling, thereby identifying the risk of abnormal spontaneous heat generation that cannot be explained by normal load heat generation and cooling disturbance. At the same time, after the diagnostic window is effective and decoupling is established, the future state of the battery is predicted, providing a basis for safe disposal. Compared with the prior art, this application does not directly predict the risk of thermal runaway. Instead, it first establishes a normal thermal relaxation baseline and prediction range to describe how the temperature should change after the current drops under normal conditions. The deviation of the actual temperature from the baseline is used as the basis for abnormal judgment. It can more clearly distinguish between normal thermal response and abnormal residual. And prediction is only performed after the window is effective and decoupling is established. This can reduce the impact of invalid sampling points, cooling disturbances and normal load changes on the model prediction results.

[0025] (2) When the target battery meets the initial abnormality but has not reached the critical danger condition, this application uses the thermal relaxation response after the current step drop to make a diagnosis, and verifies the step amplitude, current fluctuation, cooling parameters and effective proportion of sampling points in the diagnostic window. Since the normal current heat generation should decrease after the current drops, it is helpful to judge whether the temperature still deviates abnormally from the normal response. This helps to distinguish between normal load heat generation, cooling disturbance and internal abnormal spontaneous heat generation, and reduce false alarms caused by high-rate operation or cooling changes. Moreover, this application does not wait until the battery temperature reaches the danger threshold before alarming, but uses the thermal relaxation process after the current drop to judge whether the abnormal temperature rise still exists when the battery meets the initial abnormality but has not reached the critical danger condition. This helps to discover abnormal spontaneous heat generation trends that cannot be explained by normal operating conditions earlier.

[0026] (3) When constructing the current step diagnostic window, this application selects candidate step actions based on diagnostic information gain and safety cost, so that the current step becomes a diagnostic action that meets safety constraints, rather than ordinary power adjustment or fixed current limiting.

[0027] (4) This application first establishes a normal thermal relaxation baseline and prediction range that matches the current SOC, SOH, cooling parameters and pre-step load parameters through the ARIMAX model. Then, the actual temperature is compared with the baseline to obtain the abnormal temperature residual, abnormal heat generation power and current step decoupling index. The current step decoupling index is determined based on the normal heat generation integral in the reference window before the step, the abnormal heat generation integral in the diagnostic window after the step, and the duration of the abnormal temperature residual exceeding the upper limit of the prediction range. It can characterize the abnormal thermal response intensity of the target battery that cannot be explained by normal load heat generation and cooling state after the current step decrease.

[0028] (5) This application realizes the division of labor and cooperation between the ARIMAX model and the gated Transformer. The ARIMAX model first establishes a normal thermal relaxation baseline, and the Transformer only predicts the abnormal residual evolution after deviating from the normal baseline, thereby avoiding mistaking normal load changes or cooling disturbances as abnormal thermal runaway trends. Attached Figure Description

[0029] Various other advantages and benefits will become apparent to those skilled in the art upon reading the following detailed description of preferred embodiments. The accompanying drawings are for illustrative purposes only and are not intended to limit the scope of this application. Furthermore, the same reference numerals denote the same parts throughout the drawings. In the drawings: Figure 1 This is a flowchart of Example 1; Figure 2 This is a flowchart of Example 2; Figure 3 This is a flowchart of Example 3; Figure 4 This is a schematic diagram of the diagnostic window for current step drop; Figure 5 This is a flowchart of Example 4; Figure 6 This is a flowchart of Example 5; Figure 7 This is a schematic diagram of the normal thermal relaxation baseline and decoupling determination. Figure 8 This is a flowchart of Example 6; Figure 9 This is a flowchart of Example 7; Figure 10 This is a flowchart of Example 8; Figure 11 This is a flowchart of Example 9; Figure 12 This is a structural block diagram of the early warning system in Implementation Example 10. Detailed Implementation

[0030] The features and exemplary embodiments of various aspects of this application will be described in detail below. To make the objectives, technical solutions, and advantages of this application clearer, the application will be further described in detail below with reference to the accompanying drawings and specific embodiments. It should be understood that the specific embodiments described herein are only intended to explain this application and not to limit it. For those skilled in the art, this application can be implemented without some of these specific details. The following description of the embodiments is merely to provide a better understanding of this application by illustrating examples.

[0031] It should be noted that, in this document, relational terms such as "first" and "second" are used merely to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising..." does not exclude the presence of additional identical elements in the process, method, article, or apparatus that includes the element.

[0032] Example 1: This embodiment discloses a battery thermal runaway early warning method. The main technical idea is: when the battery has shown initial abnormality but has not yet reached a critical dangerous state, instead of directly relying on the temperature threshold or directly using the model to predict the future temperature for alarm, the method first uses the thermal relaxation process after the current step drop for diagnosis. Under normal circumstances, after the current drops, the normal current heat generation of the battery should decrease, the temperature rise rate should decrease, and the temperature change trend should gradually deviate from the original thermal response trajectory. If the current has dropped, but the actual temperature still continues to deviate from the normal thermal relaxation baseline, it indicates that the temperature rise may not be explained by normal load heat generation or cooling disturbance, and there may be a risk of internal abnormal spontaneous heat generation.

[0033] like Figure 1 As shown, the battery thermal runaway early warning method in this embodiment includes the following steps: Step S100: Acquire battery operating data and preprocess the data, including current; Step S200: Obtain the current step drop diagnostic window; Step S300: Verify the validity of the current step drop diagnostic window. If the validity requirement is met, proceed to the next step; otherwise, return to step S100. Step S400: Based on historical normal operation data, establish a normal thermal relaxation baseline that matches the current operating condition of the target battery and obtain the prediction range; Step S500: Calculate the current-temperature decoupling characteristic parameters based on the normal thermal relaxation baseline, and determine whether the current and temperature decoupling is valid based on the current-temperature decoupling characteristic parameters. If it is valid, proceed to the next step; otherwise, return to step S100. Here, current-temperature decoupling means that after the current decreases, the temperature response cannot be explained by the normal current heat generation change. Step S600: Predict the future state of the battery based on operating data and current-temperature decoupling characteristic parameters; Step S700: Output the thermal runaway warning level based on the future state of the battery.

[0034] Example 2: The difference between this embodiment two and embodiment one is that step S100 in this embodiment includes the following steps, such as... Figure 2 As shown: Step S110: Obtain battery operating data Acquire synchronous operating data of each cell in the target battery or battery pack. The operating data includes current, and more preferably, the operating data also includes at least one of voltage, surface temperature, ambient temperature, and cooling parameters. The cooling parameters include at least one of coolant inlet temperature, coolant outlet temperature, coolant flow rate, fan speed, and water pump power. If the corresponding sensor conditions are available, the operating data may also include at least one of tab temperature, adjacent cell gap temperature, gas concentration, casing pressure, and cell expansion pressure.

[0035] The data mentioned above needs to be aligned using a unified timestamp to ensure that current, voltage, temperature, and cooling parameters at the same time can correspond, avoiding misjudgments due to asynchronous sampling times. For example, the current may have decreased but the temperature data may still correspond to the previous time, or the changes in the cooling system state and the temperature changes may not be accurately matched.

[0036] Step S120: Preprocess the running data After acquiring synchronized running data, the data is preprocessed, including handling missing values, time synchronization correction, and necessary normalization.

[0037] Preferably, the preprocessing also includes verifying the sampling points to determine whether there are missing, abnormal jumps, timestamp offsets, or sensor saturation in the current, temperature, voltage, and cooling parameters.

[0038] Sampling points that pass verification are marked as valid sampling points, while those that fail verification are marked as invalid. These valid sampling point markers are subsequently used for current-temperature decoupling determination and also serve as valid sampling masks in subsequent prediction models. Verification of valid sampling points can refer to existing technologies; for example, the paper "Data cleaning and restoring method for vehicle battery big data platform" and patent CN118962455A disclose relevant data verification methods. This embodiment preferably verifies the data within the diagnostic window... i For each sampling point, first determine whether the sampling data itself is reliable. If any of the following conditions exist, mark the sampling point as an invalid sampling point: 1. Essential data such as current, voltage, temperature, coolant flow rate, or coolant temperature are missing; 2. Timestamps are discontinuous, repetitive, misaligned, or the time interval between adjacent sampling exceeds the preset allowable range; 3. The sensor data reaches the upper or lower limit of the range and continues to display a saturation state; 4. A single isolated jump occurs in temperature, current, or voltage, and the temperature immediately returns to the level near the level before the jump at the next sampling point; 5. A sensor shows a significant abnormal change, but there is no corresponding change in related signals such as those from adjacent sensors, cooling parameters, voltage, or current, which indicates a single channel abnormality.

[0039] If the temperature, abnormal temperature residual, or abnormal heat generation power shows a continuous increase or remains positive at more than three consecutive sampling points, and the timestamps are normal, the data is not missing, and the sensors are not saturated, then the change is considered to be consistent with the continuity of the battery thermal process and should be retained as a true thermal anomaly signal.

[0040] Step S130: Preliminary anomaly screening Determine whether the running data meets the anomaly judgment conditions and critical danger conditions. If it meets the anomaly judgment conditions but does not meet the critical danger conditions, proceed to the next step. If it meets the critical danger conditions, directly trigger the danger level warning. If neither of them is met, return to step S110.

[0041] Anomaly detection conditions: at least one of the following: target battery temperature exceeds the first temperature threshold, temperature rise rate exceeds the first temperature rise rate threshold, temperature rise acceleration exceeds the first temperature rise acceleration threshold, voltage drop exceeds the first voltage anomaly threshold, temperature difference between adjacent cells exceeds the first temperature difference threshold, and gas concentration change rate exceeds the first gas threshold.

[0042] Critical hazard conditions: These are set according to actual needs. In this embodiment, it is preferable to satisfy any one of the following requirements: Requirement 1: The temperature at any temperature monitoring point in the target battery or battery pack is greater than or equal to 60°C and remains above 60°C for more than 3 seconds; Requirement 2: The temperature difference between the highest and lowest temperatures in the target battery or battery pack must be greater than or equal to 20°C and last for more than 3 seconds. Requirement 3: The highest temperature in the target battery or battery pack rises by more than or equal to 2°C within 5 seconds, or the highest temperature rises by more than or equal to 5°C within 1 second. Requirement 4: The temperature rise rate at any monitoring point in the target battery or battery pack is greater than or equal to 1℃ / s and lasts for more than 3s; Requirement 5: The voltage of the target battery cell experiences a rapid and abnormal drop (the rate of drop is set according to actual needs), and the drop exceeds 25% of the initial voltage or the reference voltage before diagnosis of the cell; Requirement 6: When a gas or smoke sensor is configured, the smoke or carbon monoxide concentration is greater than or equal to 100 ppm, or the hydrogen concentration is greater than or equal to 2% by volume, or the VOCs concentration is greater than or equal to 200 ppm. Requirement 7: When a pressure or expansion detection device is configured, the internal pressure of the target battery reaches its maximum design pressure, or the battery size expansion reaches more than 10% of the design size, or the casing shows obvious bulging, cracking or pressure leakage. Requirement 8: The battery management system has triggered alarms for over-temperature, over-current, short circuit, insulation abnormality, or thermal events.

[0043] When the target battery only meets the preliminary abnormal conditions but does not meet the above critical danger conditions, the current step thermal relaxation diagnostic process is entered. When the target battery meets any of the above critical danger conditions, the system no longer waits for the natural current step drop diagnostic window, nor does it actively construct the current step drop diagnostic window. Instead, it directly enters the current limiting, enhanced cooling, branch isolation, disconnection of charging and discharging circuit or other safety handling processes.

[0044] Example 3: The difference between this embodiment three and embodiment one or embodiment two is that: in this embodiment three, step S200 takes the thermal relaxation process after the current step drop as the diagnostic object, and there are two ways to obtain the current step drop diagnostic window, such as... Figure 3 and Figure 4 As shown: The first method: Under the premise of meeting the first diagnostic window generation conditions, a current step drop diagnostic window is passively generated. The first diagnostic window generation conditions include that the target battery current is at a first preset level before the step, and the current drops to a second preset level after the step, and the stability requirements are met before and after the step. The first preset level is higher than the second preset level.

[0045] Specifically, during the process of vehicle braking energy recovery ending, charging power decreasing, discharging power decreasing, energy storage system power scheduling switching, or protection current limiting, the battery current may naturally experience a step drop. During this process, the conditions for the passively generated current step drop diagnostic window may be met.

[0046] The aforementioned stability requirement refers to selecting a period of time before and after the step jump. If the current fluctuates slightly around a horizontal level during this period without continuous rise, fall, or sudden change, then the interval is considered to meet the stability requirement. The specific rules for determining the stability requirement are as follows: Convert the target battery current to rate current:

[0047] in, For the target battery in t Current at any moment For the target battery rated capacity, for t The current multiplier corresponding to the given moment.

[0048] Taking the time of current step occurrence as Benchmark, in Previously selected reference window before step jump ,exist Then, after avoiding the preset transition time τ, select the post-step diagnostic window. , Used to characterize the high current state before a step jump. Used to characterize low current, current-limiting, or current-interrupting states after a step jump. The length ranges from 5 seconds to 30 seconds. The length is 10s to 60s, and the preset transition time τ is set to 5s.

[0049]

[0050] in, Indicates the reference window before the step jump. This represents the post-step diagnostic window. This indicates the number of sampling points within the reference window before the step jump. This indicates the number of sampling points within the diagnostic window after a step jump.

[0051] Reference window before step Inside, calculate the average rate current before the step jump: ; in, Indicates the first step within the reference window before the step. The current multiplier value at each valid sampling time. ， Indicates the first The actual current at each valid sampling time; Indicates the battery's rated capacity; This represents the absolute value of the actual current.

[0052] In the post-step diagnostic window Inside, calculate the average rate current after the step jump:

[0053] The reference window before the step should satisfy:

[0054] The post-step diagnostic window should meet the following requirements:

[0055] in, Indicates the high current threshold. This indicates the low current threshold.

[0056] When the above conditions are met, it indicates that the target battery is at a higher current level before the step jump, and at a lower current level, current-limited state, or current-interrupted state after the step jump.

[0057] Current step drop amplitude ΔC Calculate according to the following formula:

[0058] The current step drop amplitude should meet the following requirements:

[0059] in, This indicates the preset minimum step amplitude. Set to 0.3C.

[0060] Should meet

[0061]

[0062] when η If the current drop is not less than a preset proportional threshold, the diagnostic requirements are considered met. The preset proportional threshold is 40%.

[0063] Current fluctuation range within the pre-jump reference window Calculate according to the following formula:

[0064] Current fluctuation range within the post-step diagnostic window Calculate according to the following formula:

[0065] The pre-step reference window and the post-step diagnostic window should respectively satisfy the following:

[0066] in, and These represent the current fluctuation threshold before and after the step jump, respectively. Take 0.05C, 4

[0067] To exclude ramp conditions where the current rises or falls slowly and continuously, the current change trend within the pre-step reference window and the post-step diagnostic window is calculated. Taking the pre-step reference window as an example, the slope of its current linear fitting is... It can be represented as:

[0068] in, Indicates the reference window before the step jump Inner The sampling time corresponding to each valid sampling point This represents the average value of the sampling times within the reference window before the step jump. This represents the slope of the linear fit of the rate current change over time within the reference window before the step jump.

[0069] The change in current caused by the slope before the step jump Represented as:

[0070] in, This indicates the length of the reference window before the step jump.

[0071] The reference window before the step should satisfy:

[0072] Similarly, the slope change within the diagnostic window after a step jump can be calculated. And require it to satisfy:

[0073] in, and These represent the slope change thresholds before and after the step jump, respectively, and require... and The current does not exceed 0.05C. When this condition is met, it indicates that the current within the window before and after the step jump does not show a significant and continuous upward or downward trend.

[0074] In summary, when the following conditions are met simultaneously, it can be determined that the natural current step-down process has a stable high-current range before the step and a low-current range after the step, which can be used as a diagnostic window for current step thermal relaxation:

[0075] The second approach is to proactively construct a current step drop diagnostic window if the natural current step conditions are insufficient, provided that the conditions for generating the second diagnostic window are met. The conditions for generating the second diagnostic window include that the battery meets preset safety constraints and power availability conditions.

[0076] The actively generated current step drop diagnostic window includes: The first step involves generating candidate step actions through at least one of the following methods: short-time current limiting, short-time rate reduction, short-time current interruption, reduction of charging power, or reduction of discharging power. This generates a set of multiple candidate step actions, each... ,in, Indicates the first m One candidate step action, Indicates the magnitude of the current drop. This indicates the duration of the low-current state after a step jump. Indicates the execution time. This indicates the current recovery strategy after a step jump; When actively constructing a current step, it is not simply a matter of using a fixed-amplitude current limiting action. Instead, a set of candidate step actions can be formed first. Each candidate step action can include the step amplitude, duration, execution time, and recovery strategy.

[0077] The second step involves calculating the diagnostic information gain and safety cost for each candidate step action based on operational data and battery operating parameters. Operational data includes temperature, temperature rise rate, etc., while battery operating parameters include SOC, SOH, power demand, cooling system margin, and estimated temperature rise, etc. The diagnostic information gain indicates whether the step action helps to observe the thermal relaxation response after the current decreases and whether it helps to distinguish between normal heat generation and abnormal spontaneous heat generation. The safety cost indicates the impact of the step action on power demand, power output, temperature rise risk, and system operational stability.

[0078] The third step is to select the candidate step action that satisfies the safety constraints and has the best comprehensive evaluation of diagnostic information gain and safety cost as the target diagnostic action, which is also the current step drop diagnostic window.

[0079] one,

[0080]

[0081] in, Indicates execution of the first m The highest temperature predicted during each candidate step action. Indicates the upper limit of the allowable temperature. This indicates the actual power the system can provide after executing the candidate step action. Indicates the system's minimum power requirements. Indicates the currently available cooling capacity. If a candidate step action does not meet the above safety constraints, it will not be considered as an optional diagnostic action.

[0082] two, Diagnostic information gain can be determined by the magnitude of current decrease, the decrease in normal heat production, the coverage of normal samples, and the proportion of expected effective sampling points. As shown below: ; , , , , All indicate The values ​​are 0.30, 0.30, 0.25, and 0.15, respectively.

[0083] The normal decrease in heat production can be expressed as:

[0084] in, This represents the normal electrochemical heat generation power before the step transition. Indicates execution of the first m In a simplified implementation, the normal electrochemical heat generation power after each candidate step action can be approximated using the Ohmic heat generation method:

[0085] in, I Indicates battery current. R This indicates the battery's internal resistance.

[0086] Normal sample coverage can be expressed as:

[0087] in, This represents the number of normal samples that match the operating conditions corresponding to the candidate step action. This represents the minimum number of samples required to establish a normal thermal relaxation baseline.

[0088] The expected proportion of effective sampling points can be expressed as:

[0089] in, This indicates the expected number of sampling points to pass the verification. This indicates the total number of sampling points within the candidate diagnostic window.

[0090] three, The safety cost can be determined jointly by the power disturbance cost, temperature risk cost, cooling margin cost, diagnostic duration cost, and recovery impact cost:

[0091] , , , , , , and All indicate The values ​​are 0.25, 0.35, 0.20, 0.10, and 0.10, respectively.

[0092] The various safety costs can be calculated using the following formula:

[0093] in, This indicates the actual power the system can provide after executing the candidate step action. ; Indicates execution of the first m The highest temperature predicted during each candidate step action. , Indicates the upper limit of the allowable temperature; , Indicates the currently available cooling capacity; This indicates the duration of the low-current state after a step jump. ; This indicates the magnitude of the rate change during the current recovery process after the diagnosis is completed. .

[0094] Four,

[0095]

[0096] in, Indicates the first m The overall evaluation value of each candidate step jump action. λThis represents the safety cost weighting coefficient. The system selects from the set of candidate step actions that satisfy safety constraints. The candidate step action with the highest comprehensive evaluation value is selected as the target diagnostic action.

[0097]

[0098] in, This indicates the final selected target current step diagnostic action, also known as the current step drop diagnostic window. If there is no candidate step action that meets the safety constraints, the system will not actively construct a current step drop diagnostic window, but will wait for the next natural current step drop window, or directly enter the corresponding safety handling process.

[0099] In this embodiment, the current step is not just treated as a normal load change, but becomes a thermal relaxation diagnostic excitation that satisfies safety constraints.

[0100] Example 4: The difference between this embodiment and embodiments one through three is that step S300 in this embodiment includes: if all of the following conditions are met, the current step drop diagnostic window is considered valid, and the process proceeds to step S400; otherwise, the process returns to step S110. Figure 5 As shown: Condition 1: The current drop reaches the preset step size. This determines whether the current drop reaches the preset step size to avoid insufficient thermal relaxation response due to small current changes.

[0101] Convert the target battery current into rate current.

[0102] in, I For the target battery current, The target battery rated capacity.

[0103] Let the average rate current within the reference window before the step be The average rate current within the diagnostic window after the step jump is The current step drop amplitude is:

[0104] The preset step amplitude can be set to: In this embodiment, the preset step amplitude is preferably not less than .

[0105] Furthermore, the relative decrease percentage:

[0106] when When the current decrease reaches %, it is considered to meet the diagnostic requirements. and If the change in current is too small, the reduction in heat generation from normal current is not significant, and it is not suitable as a diagnostic window for effective thermal relaxation.

[0107] Condition 2: Stability interval verification before and after step jump. Determine whether the current fluctuation within the reference window before the step jump and the diagnostic window after the step jump is less than the corresponding stability threshold, so as to avoid the inability to judge the temperature response due to continuous and violent current fluctuations.

[0108] Current fluctuation range within the reference window before step and the range of current fluctuations within the diagnostic window after a step jump. As already described above, it will not be repeated here. Preferably, when the current after the step jump is not close to zero, the relative fluctuation range can also be used for judgment: .

[0109] Condition 3: Determine whether cooling parameters such as coolant flow rate, coolant inlet temperature, coolant outlet temperature, fan speed, or water pump power remain stable within the diagnostic window. If the cooling status changes little, the cooling status is considered to be met. If the cooling status changes but remains within the normal sample coverage range, the cooling status will be used as an exogenous variable for compensation when establishing a normal thermal relaxation baseline. If the cooling disturbance is too large or exceeds the normal sample coverage range, the window will be marked as an invalid window.

[0110] Cooling state maintenance or compensation verification is used to determine whether the heat dissipation boundary conditions within the diagnostic window have changed significantly. For liquid-cooled battery systems, coolant flow rate, coolant inlet temperature, and water pump power or water pump speed are prioritized as the primary judgment objects, with coolant outlet temperature as a secondary judgment object, as detailed below: Construct the cooling state vector:

[0111] in, Indicates coolant flow rate. Indicates the coolant inlet temperature. Indicates the coolant outlet temperature. This indicates the power or speed of the water pump.

[0112] At the moment of current step Using this as a baseline, calculate the pre-step reference window separately. and post-step diagnostic window Average values ​​of internal cooling parameters.

[0113] For control parameters such as coolant flow rate, water pump power, or fan speed, their relative changes can be expressed as:

[0114] in, x This indicates one of the following: coolant flow rate, water pump power, or fan speed. This parameter indicates the reference window before the step jump. x The average value, This parameter is displayed in the post-step diagnostic window. x The average value, To prevent extremely small positive numbers with a denominator of zero.

[0115] For temperature parameters such as coolant inlet temperature and coolant outlet temperature, their absolute changes can be expressed as:

[0116] in, This indicates the change in coolant inlet temperature. This indicates the change in coolant outlet temperature; This represents the average coolant inlet temperature within the reference window before the step change. This represents the average coolant inlet temperature within the diagnostic window after the step change. This represents the average coolant outlet temperature within the reference window before the step change. This represents the average coolant outlet temperature within the diagnostic window after the step change.

[0117] When the following conditions are met simultaneously within the diagnostic window, the cooling state change can be considered small, and the cooling state maintenance condition is satisfied:

[0118] in, This indicates the relative change in coolant flow rate. This indicates the relative change in water pump power or water pump speed. The above conditions suggest that the coolant flow rate, water pump operating status, and coolant inlet temperature within the diagnostic window remain relatively stable, and the heat dissipation boundary conditions have not changed significantly. However, since the coolant outlet temperature is affected by the heat generated by the battery itself, therefore... It is mainly used as an auxiliary verification parameter and should not be used as the sole basis for judging sudden changes in cooling status.

[0119] When the change in cooling state exceeds the aforementioned holding threshold, but is still within the modelable compensation range, the cooling parameters can be used as exogenous variables for compensation.

[0120] Preferably, the cooling state change is considered compensable when the following conditions are met simultaneously:

[0121] At the same time, the number of normally operating samples corresponding to the current cooling state should meet the following requirements.： ， in, This represents the number of normal samples that match the current SOC, SOH, cooling status, pre-step load parameters, and current step amplitude. This represents the minimum number of samples required to establish a normal thermal relaxation baseline.

[0122] When the above conditions are met, it indicates that although the cooling state has changed to some extent, the change is still within the normal sample coverage range and can be used as an exogenous variable for compensation when establishing a normal thermal relaxation baseline in the future. In step S400, a normal thermal relaxation baseline is established using the ARIMAX model.

[0123] When the cooling status changes excessively, the diagnostic window is marked as invalid. Excessive cooling disturbance can be considered to occur if any of the following conditions are met:

[0124] In addition, if the diagnostic window displays the following: water pump stops, coolant flow suddenly drops, cooling control mode switches, cooling system fault alarm, or the current cooling status exceeds the normal sample coverage range, i.e.: If the cooling condition does not meet the conditions for maintaining or compensating, the diagnostic window will not be used as a diagnostic window for effective current step drop.

[0125] In the above, a normal sample refers to a historical operating segment or test segment of the target battery or the same model battery that is operating normally, cooling normally, without triggering a safety alarm and without subsequently developing into a thermal event. A normal sample should have undergone data quality verification, and the temperature, temperature rise rate, voltage, temperature difference between adjacent cells, gas concentration, and cooling system status within the sample window should not have reached the initial abnormal conditions.

[0126] Condition 4: The proportion of valid sampling points meets the requirements. At the diagnostic window level, the proportion of effective sampling points is calculated:

[0127] in, Indicates the number of valid sampling points Indicates the total number of sampling points within the diagnostic window; when When the percentage is %, the data integrity of the diagnostic window is considered to meet the requirements; when... If the data quality of the diagnostic window is deemed insufficient, the window can be marked as invalid.

[0128] In this embodiment, if the diagnostic window fails the validity check, the system will not perform the current-temperature decoupling judgment, nor will it trigger the prediction step. Instead, it will continue to monitor and wait for the next valid diagnostic window, or perform safety measures based on the current risk status.

[0129] Example 5: The difference between this fifth embodiment and embodiments one through four is that step S400 in this embodiment includes the following steps, such as... Figure 6 and Figure 7 As shown: After a step drop in current, the heat generated by the current in a normal battery should decrease, and the battery temperature should generally tend to stabilize or gradually decrease. However, the normal thermal relaxation baseline of different batteries under different SOC, SOH, ambient temperature, cooling conditions and pre-step load parameters are not the same. Therefore, this embodiment selects a normal sample that matches the current operating condition of the target battery from historical normal operation data.

[0130] Step S410: Select a normal operating sample that matches the current operating condition of the target battery from the historical normal operating data. The normal operating sample is preferably a historical operating segment that has not experienced any initial abnormalities. The operating condition matching factors include SOC, SOH, cooling status and pre-step load parameters.

[0131] Operating condition matching refers to key operating condition parameters such as SOC (State of Charge), SOH (State of Health), cooling status, and pre-step load parameters being sufficiently close to the current diagnostic window, and the overall similarity meeting the requirements. Since actual operating data is continuously changing, it is impossible to require parameters such as SOC, SOH, flow rate, and current to be completely equal; they should be within the allowable deviation range. Therefore, the operating condition matching conditions between the normal operating sample and the current diagnostic window include:

[0132] In the formula for the above working condition matching conditions, subscripts are used. s The parameters represent the corresponding parameters in the normally operating sample, for example, , The corresponding normal operation sample , , , , , and These represent the allowable matching deviations, To prevent extremely small positive numbers with a denominator of zero, Take 5%, Take 3%, Take 2℃ Take 2℃, Take 10%, Take 0.1C.

[0133] When a normally operating sample meets all of the above operating condition matching conditions, it is considered that the normally operating sample matches the current diagnostic window operating condition.

[0134] Step S420: The ARIMAX model (dynamic regression model) uses historical temperature series as autoregressive input and at least two of the following as exogenous variables: current, ambient temperature, SOC, SOH, internal resistance, and cooling parameters. It outputs the normal thermal relaxation baseline and prediction range of the target battery within the diagnostic window. The normal thermal relaxation baseline represents how the target battery temperature should change after a step drop in current under normal conditions, and the prediction range is... This represents the allowable fluctuation range of the normal thermal relaxation response. This prediction range can be individually calibrated based on the ARIMAX multi-step prediction uncertainty and the residual distribution of the operating condition matching normal operation sample.

[0135] The ARIMAX model outputs the predicted normal temperature at each prediction time after the current step drop, and uses this predicted normal temperature as the normal thermal relaxation baseline, as follows:

[0136] in, This indicates the temperature corresponding to the normal thermal relaxation baseline. This represents the normal temperature prediction value output by the ARIMAX model.

[0137] Forecast range It can be obtained in the following ways: First, based on the multi-step prediction standard deviation of the ARIMAX model, the prediction interval can be expressed as follows, at a confidence level of 95%:

[0138] in, L(t) Indicates the lower bound of the prediction interval. U(t) Indicates the upper bound of the prediction interval. This indicates that the ARIMAX model is at the prediction time. t Multi-step prediction standard deviation.

[0139] II. Alternatively, it can be obtained by matching the residual distribution of the normal operating sample under the operating conditions. Let the residual of the normal operating sample be:

[0140] in, This indicates the actual temperature in a normally operating sample. This represents the normal thermal relaxation baseline obtained from the ARIMAX model for the corresponding normal sample. This represents the residual of a normally operating sample.

[0141] Based on the residual distribution of normal samples matched to multiple operating conditions, the lower and upper quantiles of the residuals are taken to obtain the calibrated prediction interval:

[0142] in, and These represent the lower and upper quantiles of the residual distribution of the normal sample, respectively.

[0143] In this embodiment, the initial prediction interval is first obtained based on the standard deviation of ARIMAX multi-step prediction. Then, the prediction interval is calibrated using the residual distribution of the normal operating sample matched with the operating conditions. If the actual temperature is consistently higher than the upper limit of the prediction interval, it indicates that the temperature change has exceeded the normal thermal relaxation response range, which can be used for subsequent abnormal residual calculation and current-temperature decoupling determination.

[0144] In addition, in this embodiment, ARIMAX is used to establish an interpretable baseline for normal thermal response, which can clearly represent the influence of factors such as current, ambient temperature, SOC, SOH, internal resistance and cooling state on normal temperature changes; Transformer is used for subsequent nonlinear evolution prediction of abnormal residuals. The two have different functions: ARIMAX is responsible for describing how it should be under normal conditions, while gated Transformer is responsible for predicting what will happen next in the abnormal part.

[0145] Example 6: The difference between this embodiment six and embodiments one through five is that step SS500 in this embodiment includes the following steps, such as... Figure 8 As shown: Step S510: Calculate the current-temperature decoupling characteristic parameters. In this embodiment, the current-temperature decoupling characteristic parameters include at least one of abnormal temperature residual, abnormal heat generation power, and current step decoupling index, and more preferably all three. The specific calculation method is as follows: ; Indicates abnormal temperature residuals. This indicates the actual surface temperature of the target battery. This indicates the temperature corresponding to the normal thermal relaxation baseline generated by the ARIMAX model; Furthermore, this embodiment also calculates the growth rate of the abnormal temperature residual. and the acceleration of residual growth at abnormal temperatures :

[0146] in, Indicates the sampling time interval.

[0147] Estimate the power of abnormal heat generation based on the law of conservation of energy. As shown below:

[0148] This indicates the target battery's equivalent heat capacity. For the target battery at time t The actual measured temperature This represents the rate of change of the actual temperature of the target battery over time, i.e., the temperature rise rate. This represents the heat power lost by the target battery to the environment or cooling system (which can be obtained by fitting historical normal data). The normal electrochemical heat generation power can be calculated from at least one of the following: current, voltage, internal resistance, entropy-heat term, and polarization term; furthermore, This represents the heat storage power corresponding to changes in the target battery temperature. Specifically, it calculates the abnormal heat generation power when the cooling condition meets stable or compensated conditions. The abnormal heat generation power is considered diagnostically valid only when the cooling state meets the stable or compensation conditions, and is then used for subsequent internal abnormal spontaneous heat generation judgment.

[0149] Normal electrochemical heat generation power Represented as:

[0150] in, I(t) Indicates the target battery at time t The current, R(t) This indicates the internal resistance of the target battery under the current conditions.

[0151] If the current decreases If the value remains positive or continues to increase, it indicates that the target battery may have abnormal spontaneous heat generation that cannot be explained by normal load heat generation. To further enhance the specificity of current step thermal relaxation diagnosis, this embodiment can also calculate the current step decoupling index. This index is determined based on the normal heat generation integral in the reference window before the step, the abnormal heat generation integral in the diagnosis window after the step, and the duration for which the abnormal temperature residual exceeds the upper limit of the prediction interval. Its function is to characterize the intensity of the abnormal thermal response of the target battery that cannot be explained by normal load heat generation and cooling state after the current step decrease. Furthermore, this embodiment defines the current step decoupling index as a normalized comprehensive index:

[0152] This represents the current step decoupling index. and Both represent weighting coefficients, and , This represents the integral of abnormal heat production within the diagnostic window after the step jump. This represents the normal heat production integral within the reference window before the step jump. To prevent extremely small positive numbers with a denominator of zero, This indicates that the actual temperature is higher than the upper limit of the prediction range. U(t) The cumulative duration, This indicates the duration of the diagnostic window after the step jump.

[0153] Calculate the normal heat production integral within the reference window before the step jump:

[0154] Calculate the integral of abnormal heat production within the diagnostic window after the step jump:

[0155] Among them, adopt This is to only count the portion of abnormal heat generation power that is positive.

[0156] Calculate the duration for which the actual temperature exceeds the upper limit of the normal prediction range:

[0157] in, This indicates an indicator function that takes the value 1 if the condition is true, and 0 otherwise.

[0158] Step S520: Determine current-temperature decoupling. The determination rules are as follows: Within the stable region after the current step drop, the abnormal temperature residual The value remains positive and exceeds the upper bound of the prediction interval. U(t) Under the premise that any of the following conditions are met during multiple consecutive valid sampling times, it is determined that current-temperature decoupling has occurred; otherwise, it is not decoupling.

[0159] I. Growth rate of abnormal temperature residuals Exceeding the first corresponding threshold; II. Acceleration of Abnormal Temperature Residual Growth Exceeding the second corresponding threshold; III. Abnormal heat generation power Exceeding the third corresponding threshold; IV. Current Step Decoupling Index Exceeding the fourth corresponding threshold.

[0160] The aforementioned stable interval after a current step drop refers to the continuous sampling interval during which the current remains relatively stable after a brief transition period following the current drop from the high current level before the step to the low current level after the step. This stable interval can be obtained as follows: At the moment of current step Based on this, first avoid the transition time at the instant of the step jump. , and then from Start selecting candidate stable intervals after step jump ,in, Select 2s. This indicates the duration of the stable interval, which can be set to 30 seconds.

[0161] exist If the average rate current after the step jump satisfies the low current condition:

[0162] And the current fluctuation range meets the following requirements:

[0163] in, This represents the average rate current within the stable region after a step jump. Indicates the low current threshold. This indicates the range of current fluctuations within the stable region after a step jump. This represents the threshold of current fluctuation after a step jump.

[0164] If the above conditions are not met, it is considered that the current after the step has not yet entered a stable low current platform, and this interval is not used for current-temperature decoupling determination. More preferably, when selecting the above stable interval, it is necessary to ensure that there is no second large current step, frequent charge-discharge switching, or continuous slope change in the interval, and the proportion of effective sampling points is not less than 80%.

[0165] Example 7: The difference between this embodiment seven and embodiments one through six is ​​that step S600 in this embodiment includes the following steps, such as... Figure 9 As shown: First, the Transformer residual evolution prediction model in this embodiment does not directly predict the temperature sequence under all operating conditions. Instead, it is triggered only after both conditions are met: the diagnostic window is valid and the current-temperature decoupling is established.

[0166] Step S610: Construct a gated Transformer residual evolution prediction model. The input features of the prediction model include core diagnostic features and auxiliary operating condition features. The Transformer residual evolution prediction model includes the following modules: I. Input Feature Module The core diagnostic features are used to characterize whether the target battery has an abnormal thermal response. These features include at least the abnormal temperature residual, and further include the abnormal temperature residual growth rate, the abnormal temperature residual growth acceleration, the abnormal heat generation power, the current step decoupling index, and the temperature difference residual between adjacent cells (i.e., the difference between the abnormal temperature residual of the target battery and the representative value of the abnormal temperature residual of the adjacent cells).

[0167] The auxiliary operating condition characteristics are used to characterize the current operating state and heat dissipation boundary conditions of the target battery, including at least the step relative time characteristics, and further including at least one of the following: current, voltage, surface temperature, ambient temperature, SOC, SOH, internal resistance, and cooling parameters.

[0168] II. Diagnostic Window Validity Gating Module This module is used to determine whether the diagnostic window passes the checks for current step amplitude, stable range before and after the step, cooling status, data integrity, and effective sampling point ratio. If the diagnostic window is invalid, Transformer prediction will not be performed.

[0169] III. Current-Temperature Decoupling Gating Module This module is used to determine whether abnormal temperature residual, abnormal heat generation power, or current step decoupling index meets the decoupling conditions. If decoupling does not occur, Transformer prediction is not performed.

[0170] IV. Step Relative Time Encoding Module This module is based on the timing of current step occurrence. Based on this, relative time features are generated for each sampling point, enabling the model to distinguish the thermal relaxation response in the early, middle, and late stages of a step jump.

[0171] V. Effective Sampling Mask Module This module is used to mask invalid sampling points or invalid diagnostic windows, reducing the impact of missing data, sensor anomalies, cooling disturbances, or timestamp offsets on self-attention calculations.

[0172] VI. Self-Attention Feature Extraction Module This module is used to learn the temporal correlation and evolution trend of abnormal temperature residuals, abnormal heat generation power, current step decoupling index, and other input features.

[0173] VII. Multi-task output module This module is used to output at least two of the following: future abnormal temperature residual trajectory, future abnormal heat generation power trajectory, probability of thermal runaway stage, remaining safe time, and prediction confidence level.

[0174] The above modules enable ARIMAX and gated Transformer to work together: the ARIMAX model first establishes a normal thermal relaxation baseline, and the Transformer only predicts the abnormal residual evolution after deviating from the normal thermal relaxation baseline, thereby avoiding the direct temperature prediction model from mistaking normal load changes or cooling disturbances as abnormal thermal runaway trends.

[0175] Step S620: The gated Transformer residual evolution prediction model outputs the result parameters of the future state of the battery and the prediction confidence. The result parameters of the future state of the battery include at least one of the following: abnormal temperature residual trajectory, future abnormal heat generation power trajectory, probability of thermal runaway stage and remaining safe time. In this embodiment, all four are preferred.

[0176] Furthermore, the probability of thermal runaway stages can be obtained through the classification output head. The classification output head maps the temporal features extracted by the model to the probabilities of different thermal runaway stages, such as normal stage, early abnormal stage, heat accumulation stage, dangerous stage, and thermal runaway stage. Its output can be expressed as: m

[0177] in, These represent the probabilities of each stage of thermal runaway. This represents the temporal features extracted by the gated Transformer residual evolution prediction model. and This represents the model parameters of the classification output head.

[0178] Furthermore, the remaining safety time can be obtained in at least one of the following ways: The first method outputs directly from the remaining safety time regression output header:

[0179] in, Indicates the predicted remaining safe time. This represents the temporal features extracted by the gated Transformer residual evolution prediction model. Indicated based on time series features The regression output function.

[0180] The second method is based on calculations derived from future abnormal temperature residual trajectories, future abnormal heat generation power trajectories, or future temperature trajectories. If the predicted time is the first time the value of the predicted result reaches a preset danger threshold... The current time is t The remaining safe time is:

[0181] in, This indicates the future moment when the predicted temperature, abnormal temperature residual, or abnormal heat generation power first reaches the corresponding danger threshold.

[0182] After the gated Transformer residual evolution prediction model outputs the prediction results, the prediction confidence is calculated. The prediction confidence can be determined based on the prediction dispersion, prediction quantile interval width, distance of input features relative to training samples, effective sampling point ratio, diagnostic window validity score, or online calibration error of prediction residuals of multiple gated Transformer residual evolution prediction models. The prediction confidence is expressed as a value between 0 and 1, and the larger the value, the more reliable the prediction result.

[0183] Furthermore, the prediction confidence level can be obtained by weighting multiple confidence sub-items, where each confidence sub-item is obtained by dimensionless normalization mapping of the corresponding evaluation index:

[0184] Indicates the prediction confidence level. This represents the confidence term for the predicted dispersion. This represents the confidence term for the predicted quantile interval width. This indicates the arrangement of input feature terms. This represents the confidence term representing the proportion of valid sampling points. This represents the online calibration error confidence term. , , , and All represent weighting coefficients, preferred .

[0185] When using a gated Transformer residual evolution prediction model or multiple random inactivation predictions, the dispersion between multiple prediction results is calculated. The more dispersed the prediction results, the higher the uncertainty and the lower the confidence level.

[0186] in, To represent the dispersion among multiple prediction results, one can take the standard deviation of the predicted values ​​of future abnormal temperature residuals, abnormal heat generation power, or remaining safe time. This indicates the maximum allowable prediction dispersion.

[0187] When the model outputs the predicted quantile interval, the width of the quantile interval is calculated. The wider the interval, the more uncertain the prediction, and the lower the confidence level.

[0188] in, Indicates the width of the predicted quantile interval. This indicates the maximum allowed quantile interval width.

[0189] The distance between the input features and the training samples is used to determine whether the current working condition is within the data range familiar to the model. The larger the distance, the more the current working condition deviates from the training sample distribution, and the lower the confidence level.

[0190] in, This represents the distance between the current input feature and the feature distribution of the training samples, and can be calculated using Euclidean distance, Mahalanobis distance, or nearest neighbor distance. Indicates the maximum allowable distribution distance.

[0191] The confidence term for the proportion of valid sampling points can be expressed as:

[0192] in, Indicates the number of valid sampling points. This indicates the total number of sampling points within the diagnostic window.

[0193] The online calibration error confidence term can be determined based on the error between recent model predictions and actual observations:

[0194] in, This indicates the predicted residual online calibration error. This indicates the maximum allowable calibration error.

[0195] Step S630: When the prediction confidence level is lower than the set confidence level threshold, or when at least one of the abnormal temperature residual, abnormal heat generation power, or remaining safe time is in the preset critical range, a high-level warning is not directly output based on the single prediction result. Instead, step S640 is entered to perform a secondary current step drop diagnosis and verification, or a verification is performed in combination with gas, pressure, and voltage signals. Otherwise, step S700 is entered directly.

[0196] Step S640: Perform secondary current step drop diagnosis, that is, actively or passively generate the next current step drop window, and after the above steps S300 to S620, generate secondary diagnosis result parameters. Compare the result parameters obtained from the current diagnosis with the result parameters obtained from the secondary current step drop diagnosis. The consistency comparison includes: whether the duration of the abnormal temperature residual exceeding the prediction range, the trend of abnormal heat generation power change, the current step decoupling index and the prediction results of the remaining safe time all meet the preset consistency requirements. If the requirements are met, proceed to step S700; otherwise, enter the observation state.

[0197] The consistency requirement for verification must meet the following conditions simultaneously: Condition 1: Both diagnoses must be valid.

[0198] This represents the confidence level of a prediction. Indicates the confidence level of the second-order prediction; Condition 2: The abnormal temperature residuals in the two diagnoses show consistent trends. This means that continuous prediction points are used to determine the abnormal trend. If, in the first diagnosis, the future abnormal temperature residuals increase for at least three consecutive prediction time points, and in the second diagnosis, the future abnormal temperature residuals also increase for at least three consecutive prediction time points... As shown below: The residual of future abnormal temperatures increases over at least three consecutive prediction time points:

[0199] In the second diagnosis, the residual of future abnormal temperatures also increased for at least three consecutive prediction time points:

[0200] in, Indicates a future moment in the first diagnosis. The predicted abnormal temperature residual, Indicates a future moment in the second diagnosis. The predicted abnormal temperature residual; , , Let represent three consecutive prediction steps arranged chronologically, satisfying the following conditions: < < .

[0201] Example 8: The difference between this embodiment eight and embodiments one through seven is that step S700 in this embodiment eight includes: first, determining whether the prediction confidence level meets the standard, which determines whether the result is reliable; then, determining the corresponding thermal runaway warning level based on the duration of the abnormal temperature residual exceeding the prediction range, the trend of abnormal heat generation power change, the current step decoupling index, and the prediction result of the remaining safe time. The thermal runaway warning levels include attention level warning, restriction level warning, danger level warning, and emergency level warning. The attention level warning is used to prompt for enhanced sampling, continuous monitoring, or waiting for the next effective diagnostic window; the restriction level warning is used to limit the charging and discharging power and may appropriately enhance cooling; the danger level warning is used to execute maximum cooling, branch isolation, or limit the continued operation of the battery pack; and the emergency level warning is used to cut off the charging and discharging circuit and activate fire protection, pressure relief, or other safety protection measures, such as... Figure 10As shown, the details are as follows: Step S710: Determine whether the comprehensive prediction confidence level meets the standard. This is discussed in two cases: those where the review step has not been performed (i.e., step S640) and those where the review step has been performed.

[0202] In the first scenario, the review process was not performed: ， =0 This indicates the overall prediction confidence level. This represents the confidence level of a single prediction, that is, the confidence level of the predicted outcome parameter at the first diagnosis. This represents the confidence level of the secondary prediction. Since no verification step was performed, the confidence level of the secondary prediction is configured to 0.

[0203] In the second scenario, a review process has already been performed: ; This represents the confidence level of the second prediction, that is, the confidence level of the predicted result parameter at the second diagnosis; if the combined prediction confidence level of the above... If the value is greater than 0.6 or greater than 0.7, the prediction result can be used to determine the warning level, and proceed to step S720; otherwise, the result is unreliable, and return to step S110 to continue collecting data for observation.

[0204] Step S720: Generate a preliminary warning level according to the following rules: First, let the prediction window length be H, and the future abnormal temperature residual trajectory be... The future trajectory of abnormal heat generation power is as follows The remaining safe time is predicted to be The rules for determining each warning level are as follows: I. Attention Level Warning: An attention level warning is generated when the future abnormal temperature residual begins to exceed the normal range, or the future abnormal heat production power begins to reach the minimum effective abnormal heat production level, or the remaining safe time is long and an abnormal trend has already appeared, i.e., when any of the following criteria are met: Criterion 1: Satisfies And the duration is greater than 0.2 hours. This indicates the trajectory of future abnormal temperature residuals. This indicates the upper bound of the normal forecast range for the future; Criterion 2: The condition is satisfied for multiple consecutive moments. , Indicates the trajectory of future abnormal heat generation power; Criterion 3: Furthermore, there is an upward trend in abnormal residuals or abnormal heat generation power.

[0205] II. Limitation Level Warning: A limitation level warning is generated when the residual temperature difference continues to exceed the normal range, the abnormal heat generation power reaches the threshold for concern-level abnormal heat generation, or the remaining safe time is shortened to the limitation level range, i.e., when any of the following criteria are met: Criterion 4: Satisfies And the duration is greater than 0.5 hours; Criterion 5: And it lasted for multiple moments, among which This represents the first heat generation power threshold that is greater than zero; Criterion Six: .

[0206] III. Danger Level Warning: A danger level warning is generated when the future abnormal temperature residual not only continues to exceed the normal range, but also the growth trend reaches the danger level threshold, or the future abnormal heat generation power reaches the danger level threshold, or the remaining safe time shortens to the danger level range, i.e., when any of the following criteria are met: Criterion 7: Satisfied and , This indicates the growth slope of the residual trajectory of future abnormal temperatures. The threshold for the slope of the residual growth of abnormal temperature at hazardous levels is: The growth slope of the future abnormal temperature residual trajectory It can be obtained by linear fitting of the future abnormal temperature residual sequence within the prediction window:

[0207] in, Indicates the first i One prediction step size, This represents the average of the prediction step size. Indicates future time The predicted abnormal temperature residual, This represents the average value of the residuals of future abnormal temperatures within the prediction window. n This indicates the number of predicted points involved in the fitting, and n Not less than 3; Criterion 8: , Indicates the second heat generation power threshold and Greater than ; Criterion Nine: .

[0208] IV. Emergency Level Warning: An emergency level warning is generated when the prediction results indicate that the target battery is about to reach a dangerous temperature, or the future abnormal heat generation power reaches the emergency level threshold, or the remaining safe time is extremely short, i.e., when any of the following criteria are met: Criterion 10: ; Criterion 11: , This indicates a prediction of future temperatures. Indicates the preset dangerous temperature threshold, preferred For battery systems with low temperature resistance or high safety requirements, it is also possible to... Set to 55℃ to 60℃, which can be adjusted according to the battery model and safety policy; Criterion 12: , This indicates the third heat production power threshold.

[0209] in, , and It can be based on the target battery's equivalent heat capacity. Perform normalization settings:

[0210] If all of the above different warning levels are met simultaneously, the highest warning level will be used.

[0211] Step S730: Calculate the confidence level of abnormal spontaneous heat generation inside the target battery.

[0212] Confidence level of spontaneous heat generation due to abnormality inside the battery Unlike the prediction confidence of gated Transformers The prediction confidence score mainly indicates whether the model prediction results are reliable, while the confidence score for internal anomalous spontaneous heat generation is used to indicate whether the target battery actually has internal anomalous heat generation that cannot be explained by normal load heat generation and cooling state. First, regarding the first k Secondary diagnosis, calculating the strength of evidence for abnormal heat production. :

[0213] In the above formula, the subscript k Indicates the first k Secondary diagnosis. Indicates the first k The strength of evidence for abnormal heat production in the secondary diagnosis This indicates the persistence score of abnormal temperature residuals exceeding the upper bound of the prediction interval. This indicates the score for abnormal heat generation power. This represents the score of the current step decoupling index. The score indicates the abnormal heat production trend. to All represent weighting coefficients, and each weighting coefficient is set as follows: ; The persistence score of abnormal temperature residuals can be expressed as:

[0214] The abnormal heat generation power score can be expressed as:

[0215] The current step decoupling index score can be expressed as:

[0216] Abnormal heat production trend score According to the k The continuous trend of abnormal temperature residuals and abnormal heat generation power within the secondary diagnostic window is determined according to the following rules: If the following conditions are met within at least 3 consecutive valid sampling points: and

[0217] It is assumed that both the abnormal temperature residual and the abnormal heat generation power show a continuous upward trend, and we take:

[0218] If either the abnormal temperature residual or the abnormal heat generation power shows a continuous upward trend, and the other remains positive but the upward trend is not obvious, then take: The condition of "not obvious upward trend" means that within at least three consecutive valid sampling points, the slope of the corresponding parameter change is not less than a preset significant upward slope threshold. The significant upward slope threshold for abnormal temperature residuals is set as follows: The threshold for the significant increase in abnormal heat generation power is set as follows: ; If only one of the abnormal temperature residual or abnormal heat generation power shows a continuous upward trend, while the other does not show a consistent change, then take: ; If the abnormal temperature residual or abnormal heat generation power is still positive, but has not formed a continuous upward trend, take: ; If the abnormal temperature residual falls back to the normal prediction range or the abnormal heat generation power drops below the threshold, then: ; In the above formula, Indicates the first k The cumulative duration for which abnormal temperature residuals exceed the upper bound of the prediction interval in the second diagnosis. Indicates the first k Duration of the next diagnostic window This represents the integral of abnormal heat production after a step jump. This represents the integral of normal heat production before the step jump. Indicates the first k Current step decoupling index for secondary diagnosis. This indicates the preset decoupling index threshold. To prevent extremely small positive numbers with a denominator of zero; Indicates the first i Abnormal temperature residuals at each valid sampling point Indicates the first i The abnormal heat generation power of each valid sampling point can be set according to the sampling frequency. In this embodiment, 5 valid sampling points are set.

[0219] Secondly, the consistency score between two adjacent diagnostic results is calculated. It should be noted that because the diagnostic window is continuously acquired, the consistency score between two diagnostic results can be determined even if the review step (i.e., step S640) has not been performed.

[0220]

[0221] in, The score representing the consistency between two consecutive diagnostic results. The difference in the strength of evidence for abnormal heat production between two diagnoses, i.e., the first... k+ 1st and 2nd k The difference in the strength of evidence for abnormal heat production in the secondary diagnoses. This indicates the maximum permissible difference in evidence.

[0222] Finally, the confidence level of the anomalous spontaneous heat generation inside the target battery can be expressed as:

[0223] in, This indicates the confidence level of spontaneous heat generation due to internal anomalies. This indicates the overall prediction confidence level. to Each represents a weight coefficient, and the sum of all weights is 1. This is set as follows: .

[0224] Step S740, if satisfied and If the warning level determined in step S720 is raised by one level, the warning level will be output again (if the warning level initially determined in step S720 is the highest warning level, the warning level will be output directly). If the condition is not met, the warning level determined in step S720 will be output.

[0225] In this preferred embodiment, the cause of anomalies can be distinguished by combining the thermal relaxation responses of adjacent battery cells. Specifically, the differential residual between the abnormal temperature residual of the target battery and the median of the abnormal temperature residuals of adjacent batteries is calculated. When the abnormal residuals of multiple adjacent batteries increase synchronously and are related to changes in the cooling system state, it can be determined that the cooling system is abnormal. When the abnormal residual of the target battery continues to increase while that of adjacent batteries does not increase synchronously, and the differential residual of the target battery continues to exceed the neighborhood differential threshold, the confidence level of abnormal spontaneous heat generation inside the target battery can be increased. When the abnormal residual gradually decreases with decreasing current and returns to the prediction range, it can be determined that it is normal charging and discharging heat generation. Furthermore, whether abnormal residuals are related to changes in the cooling system state can be determined by considering cooling parameter disturbances, synchronicity of adjacent units, and time correlation. For liquid cooling systems, cooling parameter disturbances can be determined by coolant flow rate, coolant inlet temperature, coolant outlet temperature, pump power, or pump speed. If at least one of the following conditions occurs within the diagnostic window or within a preset time range before or after it, the cooling system state is considered to have changed:

[0226] in, This indicates the relative change in coolant flow rate. This indicates the relative change in pump power or pump speed. This indicates the change in coolant inlet temperature. This indicates the change in coolant outlet temperature; Furthermore, the abnormal temperature residuals of the target cell and its adjacent cells are calculated. If the abnormal temperature residuals of multiple adjacent cells increase synchronously within a similar time period, and more than half of the adjacent cells satisfy the following condition: or

[0227] It is then assumed that anomalous residuals exhibit neighborhood synchronicity, where, Indicates the first j At time 1, adjacent battery cells t Abnormal temperature residual, This indicates the upper bound of the prediction interval corresponding to the adjacent battery.

[0228] Example 9: like Figure 11 As shown, the difference between this embodiment nine and embodiment one is that: This embodiment adopts step S100 of embodiment two; This embodiment adopts step S200 of embodiment three; This embodiment adopts step S300 of embodiment four; This embodiment adopts step S400 of embodiment five; This embodiment adopts step S500 of embodiment six; This embodiment adopts step S600 of embodiment seven; This embodiment adopts step S700 of embodiment eight.

[0229] Example 10: This embodiment discloses a battery thermal runaway early warning system, used to implement the early warning method in Embodiment Nine, such as... Figure 12 As shown, the early warning system includes: The data input layer includes a data acquisition module, a time synchronization module, and a data preprocessing module connected in sequence. The diagnostic analysis layer includes a preliminary anomaly screening module, a current step diagnosis module, a window validity verification module, an ARIMAX thermal relaxation baseline module, a decoupling feature calculation module, a decoupling judgment module, and an adjacent single-unit residual difference diagnosis module, which are connected in sequence. The preliminary anomaly screening module is connected to the data preprocessing module. The prediction decision layer includes a dual-gated trigger module, a gated Transformer residual prediction module, a verification module, and an early warning decision module connected in sequence. The decoupling judgment module is connected to the dual-gated trigger module, and the adjacent single-unit differential diagnosis module is connected to the verification module. The results output layer is connected to the early warning decision module and is used to output the early warning level, remaining safe time, confidence level, and safety handling strategy.

[0230] The above description is merely a specific implementation of this application. Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working processes of the systems, modules, and units described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here. It should be understood that the protection scope of this application is not limited thereto. Any person skilled in the art can easily conceive of various equivalent modifications or substitutions within the technical scope disclosed in this application, and these modifications or substitutions should all be covered within the protection scope of this application.

Claims

1. A method for early warning of battery thermal runaway, characterized in that, include: Step S100: Acquire battery operating data and preprocess the data, including current; Step S200: Obtain the current step drop diagnostic window; Step S300: Verify the validity of the current step drop diagnostic window. If the validity requirement is met, proceed to step S400; otherwise, return to step S100. Step S400: Based on historical normal operation data, establish a normal thermal relaxation baseline that matches the current operating condition of the target battery and obtain the prediction range; Step S500: Calculate the current-temperature decoupling characteristic parameters based on the normal thermal relaxation baseline, and determine whether the current and temperature decoupling is valid based on the current-temperature decoupling characteristic parameters. If it is valid, proceed to step S600; otherwise, return to step S100. Step S600: Predict the future state of the battery based on operating data and current-temperature decoupling characteristic parameters; Step S700: Output the thermal runaway warning level based on the future state of the battery.

2. The battery thermal runaway early warning method according to claim 1, characterized in that, Step S100 includes: Step S110: Obtain battery operating data, including current, and also including at least one of voltage, surface temperature, ambient temperature, and cooling parameters; Step S120: Preprocess the running data; Step S130: Determine whether the running data meets the anomaly judgment conditions and critical danger conditions. If the anomaly determination criteria are met but the critical hazard criteria are not met, proceed to step S200. If the critical hazard conditions are met, a hazard warning will be triggered directly. If neither of them is met, return to step S110.

3. The battery thermal runaway early warning method according to claim 1 or 2, characterized in that, Step S200 includes passively generating a current step drop diagnostic window under the premise of satisfying the first diagnostic window generation conditions. The first diagnostic window generation conditions include that the target battery current is at a first preset level before the current step and drops to a second preset level after the step, and the stability requirements are met before and after the step. Alternatively, the current step drop diagnostic window can be actively generated under the premise of satisfying the second diagnostic window generation conditions. The second diagnostic window generation conditions include that the battery meets preset safety constraints and power availability conditions.

4. The battery thermal runaway early warning method according to claim 3, characterized in that, The actively generated current step drop diagnostic window includes: The first step is to generate candidate step actions by at least one of the following methods: short-time current limiting, short-time rate reduction, short-time current interruption, reduction of charging power, or reduction of discharging power, and to generate a set of multiple candidate step actions in this manner. The second step is to calculate the diagnostic information gain and safety cost of each candidate step action based on the operating data and battery condition parameters. The third step is to comprehensively evaluate the optimal candidate step action based on the diagnostic information gain and safety cost, and use it as the diagnostic window for current step drop.

5. The battery thermal runaway early warning method according to claim 4, characterized in that, Diagnostic information gain As shown below: ； , , , , All indicate ; Security Cost As shown below: ； , , , All indicate ; As shown below: ； Indicates the first m The overall evaluation value of each candidate step jump action λ This represents the weighting coefficient for security costs.

6. The battery thermal runaway early warning method according to claim 1, characterized in that, Step S300 includes: if the following conditions are met simultaneously, the system is considered valid and proceeds to step S400; otherwise, it returns to step S100. Condition 1: The current drop reaches the preset step amplitude; Condition 2: The current fluctuation within the reference window before the step and the diagnostic window after the step is less than the preset stability threshold. Condition 3: Cooling parameters are within the preset normal sample range; Condition 4: The proportion of valid sampling points meets the requirements.

7. The battery thermal runaway early warning method according to claim 1, characterized in that, Step S400 includes: Step S410: Select normal samples that match the current operating conditions of the target battery from historical normal operating data. The operating condition matching factors include SOC, SOH, ambient temperature, cooling parameters and pre-step load parameters. Step S420: The ARIMAX model uses the temperature series of historical normal operation data as autoregressive input, and at least two of the following as exogenous variables: current, ambient temperature, SOC, SOH, internal resistance, and cooling parameters. The output is the normal thermal relaxation baseline and prediction range of the target battery within the current step drop diagnostic window. .

8. The battery thermal runaway early warning method according to claim 7, characterized in that, Step S500 includes: Step S510: Calculate the current-temperature decoupling characteristic parameters, which include abnormal temperature residual, abnormal heat generation power, and current step decoupling index. ； Indicates abnormal temperature residuals. This indicates the actual surface temperature of the target battery. This indicates the temperature corresponding to the normal thermal relaxation baseline generated by the ARIMAX model; ； Indicates abnormal heat generation capacity. This indicates the target battery's equivalent heat capacity. This represents the rate of change of the actual temperature of the target battery over time. This indicates the heat power lost by the target battery to the environment or cooling system. This indicates normal electrochemical heat generation power; ； This represents the current step decoupling index. and Both represent weighting coefficients, and , This represents the integral of abnormal heat production within the diagnostic window after the step jump. This represents the normal heat production integral within the reference window before the step jump. To prevent extremely small positive numbers with a denominator of zero, This indicates that the actual temperature is higher than the upper limit of the prediction range. U(t) The cumulative duration, Indicates the duration of the diagnostic window after the step jump; Step S520, the rules for determining whether current and temperature decoupling is valid are as follows: Within the stable range after a step drop in current, if the abnormal temperature residual... The value remains positive and exceeds the upper bound of the prediction interval. U(t) And the rate of increase of abnormal temperature residuals Acceleration of abnormal temperature residual growth Abnormal heat generation power or current step decoupling index If at least one of the criteria exceeds the corresponding threshold at multiple consecutive valid sampling times, the target battery is determined to have experienced current-temperature decoupling.

9. The battery thermal runaway early warning method according to claim 8, characterized in that, Step S600 includes: Step S610: Construct a gated Transformer residual evolution prediction model. The input features of the prediction model include core diagnostic features and auxiliary operating condition features. Step S620: The gated Transformer residual evolution prediction model outputs the result parameters of the future state of the battery and the prediction confidence. Step S630: Determine whether the prediction confidence level is lower than the set confidence level threshold and whether the result parameter is within the preset critical range. If either of the two conditions is met, proceed to step S640 for verification; otherwise, proceed directly to step S700. Step S640: Generate a new current step drop diagnostic window and repeat steps S300 to S600 to obtain secondary result parameters. Compare the result parameters obtained from the two diagnoses. If the verification requirements are met, proceed to step S700; otherwise, enter the observation state.

10. The battery thermal runaway early warning method according to claim 9, characterized in that, In step S610, the core diagnostic features include abnormal temperature residual, abnormal temperature residual growth rate, abnormal temperature residual growth acceleration, abnormal heat generation power, current step decoupling index, and temperature difference residual between adjacent battery cells. The auxiliary operating condition features include at least two of the following: step relative time features, current, voltage, surface temperature, ambient temperature, SOC, SOH, internal resistance, and cooling parameters. In step S620, the result parameters include at least one of the following: abnormal temperature residual trajectory, future abnormal heat generation power trajectory, probability of thermal runaway stage, and remaining safe time. The prediction confidence level is as follows: ； in, Indicates the prediction confidence level. This represents the confidence term for the predicted dispersion. This represents the confidence term for the predicted quantile interval width. This indicates the input feature distribution information term. This represents the confidence term representing the proportion of valid sampling points. This represents the online calibration error confidence term. , , , and All represent weighting coefficients; In step S630, if the prediction confidence level is lower than the confidence level threshold, or if at least one of the abnormal temperature residual, abnormal heat generation power and remaining safety time is in the preset critical range, the process proceeds to step S640 for verification; otherwise, the process proceeds directly to step S700. In step S640, the review requirement is deemed met if the following conditions are met simultaneously: Condition one: , This represents the confidence level of a prediction. Indicates the confidence level of the second-order prediction; Condition 2: The abnormal temperature residual trajectories obtained from both diagnostic predictions are continuously within a range of no less than [a certain value]. .

11. The battery thermal runaway early warning method according to claim 9 or 10, characterized in that, Step S700 includes: Step S710: Based on the gated Transformer residual evolution prediction model, if the comprehensive prediction confidence level meets the standard, proceed to step S720; otherwise, return to step S100. Step S720: Generate a preliminary warning level according to predetermined rules. Step S730: Calculate the confidence level of abnormal spontaneous heat generation inside the target battery: This indicates the confidence level of spontaneous heat generation due to internal anomalies. Indicates the first k The strength of evidence for abnormal heat production in the secondary diagnosis The score representing the consistency between two consecutive diagnostic results. This indicates the overall prediction confidence level. 、 、 and Each of these represents a weighting coefficient, and the sum of all weights is 1. Step S740, if satisfied and If the warning level determined in step S720 is not met, the warning level will be raised by one level before outputting. If not, the warning level determined in step S720 will be output directly.

12. The battery thermal runaway early warning method according to claim 11, characterized in that, The predefined rules in step S720 are as follows: The alert level rules include meeting any one of criteria one through three. Criterion 1: Satisfies And the duration is greater than 0.2 hours. This indicates the trajectory of future abnormal temperature residuals. H represents the upper bound of the normal forecast interval for the future, and H represents the length of the forecast window. Criterion 2: The condition is satisfied for multiple consecutive moments. , Indicates the trajectory of future abnormal heat generation power; Criterion 3: Furthermore, there is an upward trend in abnormal residuals or abnormal heat generation power. Indicates the predicted remaining safe time; The restriction-level warning rules include satisfying any one of criteria four through six. Criterion 4: Satisfies And the duration is greater than 0.5 hours; Criterion 5: And it lasted for multiple moments, among which Indicates the first heat generation power threshold; Criterion Six: ; Danger level warning rules include satisfying any one of criteria seven through nine. Criterion 7: Satisfies and , This indicates the growth slope of the residual trajectory of future abnormal temperatures. This indicates the threshold for the slope of the residual growth of abnormal temperatures at hazardous levels. Criterion 8: , Indicates the second heat generation power threshold and Greater than ; Criterion Nine: ; Emergency warning rules include satisfying any one of criteria ten to twelfth. Criterion 10: ; Criterion 11: , This indicates a prediction of future temperatures. This indicates a preset dangerous temperature threshold; Criterion 12: , This indicates the third heat production power threshold.

13. The battery thermal runaway early warning method according to claim 11, characterized in that, The strength of evidence for anomalous heat production and the consistency scores are shown below: This indicates the persistence score of abnormal temperature residuals exceeding the upper bound of the prediction interval. This indicates the score for abnormal heat generation power. This represents the score of the current step decoupling index. The score indicates the abnormal heat production trend. , , and All represent weighting coefficients; This represents the difference in the strength of evidence for abnormal heat production between two diagnoses. Indicates the maximum permissible difference in evidence; This represents the confidence level of a prediction. This indicates the confidence level of the second-order prediction. If it has not been verified, then... =0.

Images