EVTOL power battery thermal runaway risk early warning method and system fused with flight condition

Abstract

The invention discloses an eVTOL power battery thermal runaway risk early warning method and system fused with flight conditions. The eVTOL power battery thermal runaway risk early warning method comprises the following steps of A, multivariate data processing and flight condition fragment division, B, flight condition and battery thermal runaway feature extraction, and C, thermal runaway risk self-adaptive early warning fused with the flight conditions. According to the method, different flight working conditions are fused into the existing eVTOL power battery thermal runaway risk early warning method, and accurate evaluation and early warning of the power battery thermal runaway risk of the eVTOL under different flight working conditions can be realized. According to the invention, the eVTOL airborne safety system can deeply fuse the real-time flight condition so as to realize the early, accurate and adaptive early warning of the thermal runaway risk of the eVTOL power battery.

Description

Technical Field

[0001] This invention belongs to the field of electric aircraft flight safety technology, and more specifically, relates to a dynamic early warning system for thermal runaway of power battery of electric vertical take-off and landing aircraft that integrates real-time flight conditions. Background Technology

[0002] The high-energy-density power batteries and multi-electric drive systems unique to eVTOL aircraft expose them to severe safety risks during flight, such as battery thermal runaway and power loss. If an eVTOL experiences sudden battery thermal runaway or power loss during flight, it could lead to the aircraft itself becoming uncontrollable, seriously threatening the operational safety of other aircraft in the airspace, and even causing catastrophic accidents such as crashes. Therefore, under the complex operating conditions of low-altitude airspace and flight environment, the flight risk modes of eVTOLs exhibit high dynamism and complexity. Currently, eVTOL operational risk early warning methods based on existing aviation surveillance and health management technologies generally fail to fully consider the profound impact of this dynamic flight process on the battery. Therefore, there is an urgent need to establish an adaptive early warning system for eVTOL power battery thermal runaway risk that can integrate real-time flight conditions. Summary of the Invention

[0003] In response to the above-mentioned deficiencies or improvement needs of existing technologies, this invention proposes an eVTOL power battery thermal runaway risk early warning method and system that integrates flight conditions. The aim is to enable the eVTOL airborne safety system to deeply integrate real-time flight conditions to achieve early, accurate and adaptive early warning of eVTOL power battery thermal runaway risk.

[0004] To achieve the above objectives, this invention provides a method for early warning of thermal runaway risk of eVTOL power batteries based on flight operating conditions, comprising the following steps:

[0005] Step A: Multivariate data processing and flight condition segmentation: Flight data is collected and preprocessed through a sensor system; based on multi-parameter fusion logic of flight status, automatic and accurate identification of core flight phases such as takeoff, climb, cruise, hovering, descent and landing is achieved; the sliding time window method is applied to further extract time-series segments representing transient or micro-dynamic features from each macro-condition, providing a data foundation for subsequent refined analysis and modeling;

[0006] Step B: Flight Condition and Battery Thermal Runaway Feature Extraction: Extract feature parameters that characterize the flight state and thermal runaway sensitive features that characterize the consistency of battery voltage from micro-flight segments; then, screen out the key thermal runaway features with the strongest discriminative power from battery fault data using the KS test; and select the subset with the strongest correlation with the key features from the flight condition features using the maximum information coefficient to form the optimal feature set for building a high-precision early warning model.

[0007] Step C: Adaptive Early Warning of Thermal Runaway Risk Based on Fusion Flight Conditions: Based on normal historical data, CNN voltage consistency estimation models are constructed for different flight conditions to predict the normal characteristic baseline value under the current condition; then, the residual sequence between the predicted value and the actual value is calculated, and the abnormal judgment threshold under each condition is adaptively determined using the BOX-COX transform and the 3σ rule; by real-time monitoring and comparison of the residual with the abnormal threshold of the corresponding condition, the adaptive and accurate early warning of thermal runaway risk based on fusion flight conditions is achieved.

[0008] In some alternative implementations, step A specifically includes:

[0009] Step A1: Data Acquisition and Preprocessing: Raw flight time-series data is acquired through the eVTOL airborne multi-sensor fusion system. Subsequently, the raw data undergoes preprocessing including anomaly handling, deduplication, frequency unification, and smoothing filtering.

[0010] Step A2: Flight condition classification: Based on the fusion of multiple parameters such as speed, pitch angle, roll angle, yaw angle, altitude and its rate of change, accurate identification of core flight phases such as takeoff, climb, cruise, hovering, descent and landing is achieved;

[0011] Step A3: Divide micro-operational condition segments based on the sliding time window method: Use the sliding time window method to finely divide the continuous flight data stream into micro-operational condition segments, and accurately capture the features that occur within a short period of time.

[0012] In some alternative implementations, step B specifically includes:

[0013] Step B1: Flight condition feature extraction based on micro-segments. Based on the micro-segments in Step A, calculate and extract characterization parameters from the original data within each segment to comprehensively and quantitatively describe the flight state of that segment.

[0014] Step B2: Thermal runaway feature extraction based on micro-segments: Within each micro-operating condition segment corresponding to the same time, extract thermal runaway sensitive features that characterize the consistency of battery voltage. The relevant features include voltage variation coefficient, voltage sample entropy value, and voltage Mahalanobis distance.

[0015] Step B3: Feature optimization based on KS test and maximum information coefficient: For the extracted thermal runaway features, KS test is performed on normal and faulty battery data respectively, and the features with the most significant statistical distribution difference are selected as the optimal thermal runaway features; then, the maximum information coefficient between each flight condition feature and the optimal thermal runaway feature is calculated, and the subset of conditions features with the strongest correlation is selected to build the early warning model.

[0016] In some alternative implementations, step C specifically includes:

[0017] Step C1: Establish thermal runaway characteristic state estimation models for different operating conditions: Based on normal battery data, a CNN algorithm is used to construct voltage consistency estimation models for six operating conditions: takeoff, climb, cruise, hover, descent and landing, to output normal voltage consistency characteristic estimates under real-time operating conditions.

[0018] Step C2: Calculation of voltage thermal runaway characteristic anomaly judgment threshold based on BOX-COX transformation: Obtain the residual sequence between the estimated value and the actual value of voltage consistency characteristics to form a residual sample set; perform BOX-COX transformation on the sample set to improve the normality of the distribution, and then apply the 3σ criterion to calculate the anomaly threshold of the operating condition adaptive.

[0019] Step C3: Thermal runaway feature anomaly detection and early warning: Calculate the residual between the normal feature value output by the CNN model and the actual thermal runaway feature value. If the residual exceeds the preset anomaly threshold, it indicates that the actual feature value is abnormal, thereby providing an early warning of thermal runaway risk.

[0020] In some alternative implementations, step A1, in which the flight data is acquired through multiple sensor systems, provides the aircraft's pitch, roll, yaw angle (or angular rate), velocity, and three-dimensional acceleration; precise latitude, longitude, altitude, ground speed, and heading information; airspeed, barometric altitude, ambient temperature, and angle of attack; total voltage, total current, individual cell voltage, temperature (at multiple monitoring points), remaining charge, and health status, etc.

[0021] Step A2 uses various data to accurately determine each flight phase:

[0022] Takeoff phase: Vertical velocity increases abruptly from zero, acceleration is significantly higher than cruise value, pitch angle increases rapidly (forward tilt attitude), rotor speed reaches peak; angular velocity rate of change increases sharply, displacement changes from stationary to rising;

[0023] Climbing phase: Maintains positive vertical speed and high acceleration, pitch angle remains large positive value (forward tilt), horizontal speed gradually increases, rotor thrust decreases slightly to takeoff but is still higher than cruise, and energy consumption increases significantly.

[0024] Cruise phase: The speed stabilizes at the design value (e.g., 200–230 km / h), the vertical speed approaches zero, the pitch angle is small (close to horizontal), the rate of change of angular velocity is extremely low, the rotor speed is stable, the energy consumption is low, and the attitude fluctuation is small.

[0025] Hovering phase: All speeds and accelerations approach zero, attitude angles (pitch, roll, yaw) fluctuate slightly, angular velocity change rate is low, rotor speed is stable to counteract disturbances, energy consumption is moderate, and position remains basically unchanged.

[0026] Descent phase: negative vertical velocity and acceleration, pitch angle gradually decreases (backward attitude), horizontal velocity slightly decreases; rotor speed gradually decreases, energy consumption slightly increases, and the rate of change of angular velocity is mainly based on the pitch axis;

[0027] Landing phase: Vertical velocity gradually decreases to zero, acceleration fluctuates slightly, attitude angle returns to positive (horizontal), angular velocity rate of change increases; rotor speed gradually decreases, position stops changing.

[0028] In step B3, KS tests are performed on different thermal runaway characteristics, such as Mahal distance, VVCC, and voltage sample entropy, using normal battery data and fault data. The characteristic with the most significant difference is taken as the optimal thermal runaway characteristic. The KS test steps are as follows:

[0029] (1) Establish the test problem. If the empirical distribution functions of the two groups of samples are respectively expressed as

[0030]

[0031] like If the condition is met, it means that the two sets of samples come from the same distribution, indicating that the parameter has failed to effectively distinguish between normal voltage samples and fault voltage samples.

[0032] (2) Calculate the empirical cumulative distribution function. Calculate the empirical distribution function for each of the two groups of samples. and .

[0033] (3) Calculate the KS statistic for the two-sample test. For each data point, calculate the vertical difference between the cumulative distribution functions of the two samples. The maximum value of these differences is the KS statistic, which can be expressed by the following formula:

[0034]

[0035] In the formula, The KS test statistic is used. and These represent the number of normal voltage samples and fault voltage samples, respectively. In actual calculations, the following formula can be used:

[0036]

[0037] In the formula, i and j represent the order of the two sets of sample data after being sorted in ascending order, respectively. and These are the ordinal numbers of the two groups of samples in the overall sample set, arranged in ascending order.

[0038] (4) Calculate the test p-value: In the test, the test statistic needs to be calculated. The null hypothesis is compared to a critical value, which is related to the chosen significance level (usually 0.05 or 0.01) and the sample size. In practice, the p-value is generally compared with the significance level to determine whether to reject the null hypothesis. Therefore, it is necessary to refer to the statistical test. To calculate the corresponding P value The specific calculation formula is as follows:

[0039]

[0040] (5) Compare the P-value with the significance level α. The smaller the P-value, the greater the difference in this characteristic parameter between the normal sample group and the abnormal sample group. According to hypothesis testing theory, if the P-value is less than the significance level α (the chosen significance level α is 0.05), there is sufficient reason to reject the null hypothesis, and this characteristic parameter can distinguish the actual distribution of normal and unsafe battery voltages at this significance level.

[0041] In some optional implementations, the Box-Cox transform in step C2 can automatically determine the form and parameters of the data transformation function based on the distribution of the original data. Data that does not conform to a normal distribution will conform to or approximate a normal distribution after transformation. The normality, additivity, and homoscedasticity of the transformed data can generally be significantly improved. The formula for calculating the Box-Cox transform is as follows:

[0042]

[0043] In the formula, Represents the original variable data. Indicates the transformation parameters. This represents the transformed data. The above transformation requires the original variables to be positive; if the variable value is negative, a constant must first be added to all the original data. make If the value is positive, the above transformations can be performed after this data shift operation.

[0044] The parameters in the Box-Cox transform are generally calculated using the maximum likelihood estimation method. :

[0045]

[0046] In the formula, The transformed data The arithmetic mean can be calculated using methods such as gradient descent and quasi-Newton's method. The maximum value corresponding to .

[0047] A thermal runaway risk warning system for eVTOL power batteries that integrates flight operating conditions, applicable to the aforementioned method and system for thermal runaway risk warning of eVTOL power batteries that integrates flight operating conditions, includes the following modules:

[0048] The multivariate data processing and flight condition segmentation module is used for multivariate data processing and flight condition segmentation. The multivariate data processing and flight condition segmentation module includes a data acquisition and preprocessing unit, a flight condition type segmentation unit, and a micro-condition segmentation unit based on the sliding time window method.

[0049] The flight operating condition and battery thermal runaway feature extraction module is used for extracting features of flight operating conditions and battery thermal runaway. The flight operating condition and battery thermal runaway feature extraction module includes a flight operating condition feature extraction unit based on micro-segments, a thermal runaway feature extraction unit based on micro-segments, and a feature optimization unit based on KS test and maximum information coefficient.

[0050] The thermal runaway risk adaptive early warning module for integrated flight conditions is used for adaptive early warning of thermal runaway risk under integrated flight conditions. The thermal runaway risk adaptive early warning module for integrated flight conditions includes a unit for establishing thermal runaway characteristic state estimation model for each flight condition, a unit for calculating the threshold for judging voltage thermal runaway characteristic anomalies based on BOX-COX transformation, and a unit for detecting and warning of thermal runaway characteristic anomalies.

[0051] This invention integrates real-time flight conditions into an adaptive early warning system for the thermal runaway risk of eVTOL power batteries. It considers all factors and can solve the key problem of thermal runaway risk of eVTOL power batteries under complex flight conditions. Compared with existing technologies, this invention has the following advantages: (1) First, the KS test is used to obtain the feature with the most significant difference as the optimal thermal runaway feature. Then, the maximum information coefficient between each operating condition feature and the optimal thermal runaway feature is calculated to select the operating condition feature for early warning modeling. This achieves a preliminary screening followed by a fine screening, ensuring the quality and reliability of the final feature set. (2) By using the collected data to classify flight conditions and then deeply integrating real-time flight conditions, the system can intelligently determine the cause of battery state changes and understand the normal physical impact of different flight stages on the battery. Thus, it only alarms when abnormal conditions occur, achieving accurate early warning. Attached Figure Description

[0052] Figure 1 This is a schematic diagram of an eVTOL power battery thermal runaway risk adaptive early warning system that integrates flight conditions, provided by an embodiment of the present invention.

[0053] Figure 2 This is a flowchart of the multi-data processing and flight condition segmentation module implementation method provided in this embodiment of the invention;

[0054] Figure 3This is a flowchart of the flight condition and battery thermal runaway feature extraction module implementation method provided in this embodiment of the invention;

[0055] Figure 4 This is a flowchart of the implementation method of the thermal runaway risk adaptive early warning module based on fusion flight conditions provided in this embodiment of the invention. Detailed Implementation

[0056] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention. Furthermore, the technical features involved in the various embodiments of this invention described below can be combined with each other as long as they do not conflict with each other.

[0057] like Figure 1 The diagram shown is a schematic diagram of an eVTOL power battery thermal runaway risk adaptive early warning system that integrates flight conditions according to an embodiment of the present invention. The system includes: a multi-data processing and flight condition segmentation module A, a flight condition and battery thermal runaway feature extraction module B, and a thermal runaway risk adaptive early warning module that integrates flight conditions.

[0058] Among them, the above-mentioned multi-data processing and flight condition segmentation module collects flight data through various sensor systems and preprocesses it, and then uses these data to divide the flight condition types into takeoff, climb, cruise, hover, descent and landing conditions, and then divides the micro condition segments based on the sliding time window method.

[0059] The aforementioned flight condition and battery thermal runaway feature extraction module extracts flight condition features and battery thermal runaway features from each micro-segment, and then combines the KS test and the maximum information coefficient to perform feature optimization.

[0060] The aforementioned thermal runaway risk adaptive early warning module, which integrates flight conditions, establishes a thermal runaway characteristic state estimation model based on normal battery data. Then, it imports the optimized features of each condition into the model, calculates the residual between the output value and the actual feature value, and uses the BOX-COX transformation and the 3σ rule to calculate the abnormal threshold. If the residual exceeds the threshold, it indicates that there is an abnormality in the actual feature value, and thus a thermal runaway risk early warning is issued.

[0061] The specific process of the above-mentioned multi-source data processing and flight condition segmentation module is as follows: Figure 2 As shown, it is mainly used to perform the following operations:

[0062] Step A1: Data Acquisition and Preprocessing: Raw flight time-series data is acquired through the eVTOL airborne multi-sensor fusion system. Subsequently, the raw data undergoes preprocessing including anomaly handling, deduplication, frequency unification, and smoothing filtering.

[0063] In this embodiment of the invention, the flight data is collected through multiple sensor systems, providing the aircraft's pitch, roll, yaw angle (or angular rate), speed, and three-dimensional acceleration; precise latitude, longitude, altitude, ground speed, and heading information; airspeed, barometric altitude, ambient temperature, and angle of attack; total voltage, total current, individual cell voltage, temperature (at multiple monitoring points), remaining charge, and health status, etc.

[0064] Step A2: Flight Condition Classification: Based on the fusion of multiple parameters such as speed, pitch angle, roll angle, yaw angle, altitude and its rate of change, accurate identification of core flight phases such as takeoff, climb, cruise, hovering, descent and landing is achieved.

[0065] After data preprocessing, various data points are used to accurately determine each flight phase. Takeoff phase: Vertical velocity increases abruptly from zero, acceleration is significantly higher than cruise value, pitch angle increases rapidly (forward tilt), rotor speed reaches peak, angular velocity rate of change increases sharply, and displacement changes from stationary to ascending; Climb phase: Continuous positive vertical velocity and high acceleration, pitch angle maintains a large positive value (forward tilt), horizontal velocity gradually increases, rotor thrust decreases slightly compared to takeoff but remains higher than cruise, and energy consumption increases significantly; Cruise phase: Speed ​​stabilizes at design value (e.g., 200–230 km / h). During the hovering phase: all speeds and accelerations approach zero, the attitude angles (pitch, roll, yaw) fluctuate slightly, the rate of change of angular velocity is low, the rotor speed is stable to counteract disturbances, energy consumption is moderate, and the position remains basically unchanged; during the descent phase: negative vertical speed and acceleration, pitch angle gradually decreases (backward attitude), horizontal speed decreases slightly; rotor speed decreases slowly, energy consumption increases slightly, and the rate of change of angular velocity is dominated by the pitch axis; during the landing phase: vertical speed gradually decreases to zero, acceleration fluctuates slightly, attitude angles return to positive (horizontal), and the rate of change of angular velocity increases again; rotor speed gradually decreases, and the position stops changing.

[0066] Step A3: Divide micro-operational condition segments based on the sliding time window method: Use the sliding time window method to finely divide the continuous flight data stream into micro-operational condition segments, and accurately capture the features that occur within a short period of time.

[0067] The specific process of the above-mentioned flight condition and battery thermal runaway feature extraction module is as follows: Figure 3 As shown,

[0068] This can be achieved in the following ways:

[0069] Step B1: Flight condition feature extraction based on micro-segments. Based on the micro-segments in Step A, calculate and extract characterization parameters from the original data within each segment to comprehensively and quantitatively describe the flight state of that segment.

[0070] Among the characterization parameters that can describe flight operating conditions are velocity, acceleration, ambient temperature, and air resistance. Step B2: Extraction of thermal runaway features based on micro-segments: Within each micro-operating condition segment corresponding to the same time, extract thermal runaway sensitive features that characterize the consistency of battery voltage. Relevant features include voltage variation coefficient, voltage sample entropy value, and voltage Mahalanobis distance.

[0071] Step B3: Feature optimization based on KS test and maximum information coefficient: For the extracted thermal runaway features, KS test is performed on normal and faulty battery data respectively, and the features with the most significant statistical distribution difference are selected as the optimal thermal runaway features; then, the maximum information coefficient between each flight condition feature and the optimal thermal runaway feature is calculated, and the subset of conditions features with the strongest correlation is selected to build the early warning model.

[0072] The steps of the KS test are as follows:

[0073] (1) Establish the test problem. If the empirical distribution functions of the two groups of samples are respectively expressed as

[0074]

[0075] like If the condition is met, it means that the two sets of samples come from the same distribution, indicating that the parameter has failed to effectively distinguish between normal voltage samples and fault voltage samples.

[0076] (2) Calculate the empirical cumulative distribution function. Calculate the empirical distribution function for each of the two groups of samples. and .

[0077] (3) Calculate the KS statistic for the two-sample test. For each data point, calculate the vertical difference between the cumulative distribution functions of the two samples. The maximum value of these differences is the KS statistic, which can be expressed by the following formula:

[0078]

[0079] In the formula, The KS test statistic is used. , These represent the number of normal voltage samples and fault voltage samples, respectively. In actual calculations, the following formula can be used:

[0080]

[0081] In the formula, i and j represent the order of the two sets of sample data after being sorted in ascending order, respectively. and These are the ordinal numbers of the two groups of samples in the overall sample set, arranged in ascending order.

[0082] (4) Calculate the test p-value: In the test, the test statistic needs to be calculated. The null hypothesis is compared to a critical value, which is related to the chosen significance level (usually 0.05 or 0.01) and the sample size. In practice, the p-value is generally compared with the significance level to determine whether to reject the null hypothesis. Therefore, it is necessary to refer to the statistical test. To calculate the corresponding P value The specific calculation formula is as follows:

[0083]

[0084] (5) Compare the P-value with the significance level α. The smaller the P-value, the greater the difference in this characteristic parameter between the normal sample group and the abnormal sample group. According to hypothesis testing theory, if the P-value is less than the significance level α (the chosen significance level α is 0.05), there is sufficient reason to reject the null hypothesis, and this characteristic parameter can distinguish the actual distribution of normal and unsafe battery voltages at this significance level.

[0085] The specific process of the aforementioned adaptive early warning module for thermal runaway risk under integrated flight conditions is as follows: Figure 4 As shown, this can be achieved in the following way:

[0086] Step C1: Establish thermal runaway characteristic state estimation models for different operating conditions: Based on normal battery data, a CNN algorithm is used to construct voltage consistency estimation models for six operating conditions: takeoff, climb, cruise, hovering, descent and landing, to output normal voltage consistency characteristic estimates under real-time operating conditions.

[0087] Step C2: Calculation of voltage thermal runaway characteristic anomaly judgment threshold based on BOX-COX transformation: Obtain the residual sequence between the estimated and actual voltage consistency characteristics to form a residual sample set; perform BOX-COX transformation on the sample set to improve the normality of the distribution, and then apply the 3σ criterion to calculate the anomaly threshold of the operating condition adaptive.

[0088] The formula for calculating the BOX-COX transformation is as follows:

[0089]

[0090] In the formula, Represents the original variable data. Indicates the transformation parameters. This represents the transformed data. The above transformation requires the original variables to be positive; if the variable value is negative, a constant must first be added to all the original data. make If the value is positive, the above transformations can be performed after this data shift operation.

[0091] The parameters in the Box-Cox transform are generally calculated using the maximum likelihood estimation method. :

[0092]

[0093] In the formula, The transformed data The arithmetic mean can be calculated using methods such as gradient descent and quasi-Newton's method. The maximum value corresponding to .

[0094] Step C3: Thermal runaway feature anomaly detection and early warning: Calculate the residual between the normal feature value output by the CNN model and the actual thermal runaway feature value. If the residual exceeds the preset anomaly threshold, it indicates that the actual feature value is abnormal, thereby providing an early warning of thermal runaway risk.

[0095] The eVTOL power battery thermal runaway risk warning model and warning strategy are the key to achieving accurate warning of eVTOL operation risks. The eVTOL power battery thermal runaway risk warning model and warning strategy in this invention need to consider different flight conditions. Under different flight conditions, different eVTOL power battery thermal runaway risk warning models and warning strategies should be adopted to achieve accurate warning.

[0096] It should be noted that, depending on the implementation needs, the various steps / components described in this application can be broken down into more steps / components, or two or more steps / components or parts of the operation of steps / components can be combined into new steps / components to achieve the purpose of this invention.

[0097] Those skilled in the art will readily understand that the above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A method for early warning of thermal runaway risk of eVTOL power battery in fusion flight operating mode, characterized in that, Comprising the following steps: Step A: Multi-data processing and flight condition segment division: Collecting flight data through the sensor system and preprocessing it; Based on the fusion logic of multiple flight state parameters, automatically and accurately distinguishing the core flight stages such as take-off, climb, cruise, hovering, descent and landing; Applying the sliding time window method to further extract time series segments representing transient or micro-dynamic characteristics from each macro-condition, providing a data basis for subsequent refined analysis and modeling; Specifically including: Step A1: Data acquisition and preprocessing: Collecting original flight time series data through the multi-sensor fusion system on board the eVTOL; Then, the original data is preprocessed by anomaly processing, de-duplication, frequency unification and smoothing filtering; Step A2: Flight condition type division: Based on the fusion of multiple parameters such as speed, pitch angle, roll angle, yaw angle, height and its rate of change, accurately distinguish the core flight stages such as take-off, climb, cruise, hover, descent and landing; Step A3: Micro-condition segment division based on sliding time window method: Using the sliding time window method to finely segment the continuous flight data stream to divide micro-condition segments and accurately capture the characteristics occurring in a short period of time; Step B: Flight condition and battery thermal runaway feature extraction: Extracting feature parameters that characterize flight conditions and thermal runaway sensitive features that represent battery voltage consistency from micro flight segments; Then, through K-S test, select the key features with the strongest discrimination degree from the battery fault data; Use the maximum information coefficient to select the strongest subset of flight condition features associated with the key features to form the optimal feature set for building a high-precision early warning model; Specifically including: Step B1: Flight condition feature extraction based on micro-segment: Based on the micro-condition segments in step A, calculate and extract the representative parameters that can comprehensively and quantitatively characterize the flight state of each segment from the original data within the segment; Step B2: Thermal runaway feature extraction based on micro-segment: Within each micro-condition segment corresponding to the same time, extract thermal runaway sensitive features that represent battery voltage consistency, including voltage coefficient of variation, voltage sample entropy value, and voltage Mahalanobis distance; Step B3: Feature optimization based on K-S test and maximum information coefficient: For each type of thermal runaway feature extracted, use K-S test on normal and fault battery data to select the feature with the most significant statistical distribution difference as the optimal thermal runaway feature; Then, calculate the maximum information coefficient between each flight condition feature and the optimal thermal runaway feature, and select the strongest subset of condition features for building a warning model; Step C: Fusion flight condition thermal runaway risk adaptive early warning: Based on normal historical data, a CNN voltage consistency estimation model is constructed for different flight conditions to predict the normal characteristic reference value under the current condition. Then, the residual sequence of the predicted value and the actual value is calculated, and the BOX-COX transformation and 3σ rule are used to adaptively determine the abnormal judgment threshold under each condition. By real-time monitoring and comparing the residual with the abnormal threshold of the corresponding condition, the adaptive and accurate early warning of thermal runaway risk under the fusion flight state is realized. Specifically, it includes: Step C1: Establish thermal runaway feature state estimation model under different conditions: Based on normal battery data, a CNN algorithm is used to construct a voltage consistency estimation model for take-off, climb, cruise, hover, descent and landing six conditions, which is used to output the normal voltage consistency feature estimation value under real-time condition. Step C2: Abnormal judgment threshold calculation of voltage thermal runaway feature based on BOX-COX transformation: Obtain the residual sequence between the voltage consistency feature estimation value and the actual value to form a residual sample set. The sample set is subjected to BOX-COX transformation to improve the normality of the distribution, and then the 3σ rule is applied to calculate the adaptive abnormal threshold of the condition. Step C3: Thermal runaway feature anomaly detection and early warning: Calculate the residual of the normal feature value output by the CNN model and the actual thermal runaway feature value. If the residual exceeds the preset abnormal threshold, it indicates that the actual feature value is abnormal, and the thermal runaway risk early warning is performed.

2. The eVTOL power cell thermal runaway risk pre-alarm method of fusing flight operating conditions according to claim 1, characterized in that, In step A1, the flight data is collected by multiple sensor systems, and the flight data includes the pitch, roll, yaw angle, speed and three-dimensional acceleration of the aircraft; The flight data also includes accurate latitude, longitude, altitude, ground speed and heading information; The flight data also includes airspeed, barometric altitude, ambient temperature and attack angle; Total voltage, total current, single cell voltage, temperature, remaining capacity, health status.

3. The eVTOL power cell thermal runaway risk pre-alarming method of fusing flight operating conditions according to claim 1, characterized in that, In step A2, the flight phases are accurately distinguished by various data: Take-off phase: The vertical speed increases from zero, the acceleration is significantly higher than the cruising value, the pitch angle increases rapidly, the rotor speed reaches the peak value, the angular velocity changes rapidly, and the displacement changes from static to ascending; Climbing phase: Continuous positive vertical speed and higher acceleration, pitch angle maintains a large positive value, horizontal speed gradually increases, rotor thrust slightly decreases from take-off, but is still higher than cruising, energy consumption increases significantly; Cruise phase: The speed is stable at the design value, the vertical speed tends to zero, the pitch angle is small, the angular velocity change rate is very low, the rotor speed is stable, the energy consumption is low, and the attitude fluctuation is small; Hover phase: All speeds and accelerations tend to zero, attitude angles fluctuate slightly, angular velocity change rate is low, rotor speed is stable to offset disturbances, energy consumption is moderate, and position is basically unchanged; Descent phase: Negative vertical speed and acceleration, pitch angle gradually decreases, horizontal speed slightly decreases; Rotor speed decreases slowly, energy consumption increases slightly, and angular velocity change rate is mainly in the pitch axis; Landing phase: The vertical speed gradually decreases to zero, the acceleration fluctuates slightly, the attitude angle returns to positive, and the angular velocity change rate rises; The rotor speed gradually decreases, and the position stops changing.

4. The eVTOL power cell thermal runaway risk warning method fusing flight operating conditions according to claim 1, characterized in that, The k-s test step in step B3 is as follows: (1) Establish the test problem: if the empirical distribution functions of two groups of samples are represented as If holds, it indicates that the two groups of samples come from the same distribution, suggesting that the parameter fails to effectively distinguish normal voltage samples from fault voltage samples; (2) Calculate the empirical cumulative distribution function: Calculate the empirical distribution function of the two groups of samples respectively and ; (3) Calculate the K-S statistic of the two-sample K-S test: for each data point, calculate the vertical gap between the cumulative distribution functions of the two samples, and the maximum of these gaps is the K-S statistic, which is expressed as follows: In the formula, is the K-S test statistic, , are the number of normal voltage samples and the number of fault voltage samples, respectively; for , the following formula can be used in actual calculation: In the formula, i and j are the order of two groups of sample data arranged in ascending order, respectively, and are the serial numbers of the order statistics of two groups of samples arranged in ascending order under the total sample set, respectively. (4) Calculate the test P-value: in the test, the statistic is compared with a critical value, which is related to the selected significance level and sample size. In practical applications, the P-value is generally compared with the significance level to determine whether to reject the null hypothesis. Therefore, the corresponding P-value needs to be calculated according to the statistic , and the specific calculation formula is as follows: ​ (5) Compare the P-value with the significance level a: the smaller the P-value, the greater the difference between the normal sample group and the abnormal sample group in this characteristic parameter; if the P-value is less than the significance level a, there is enough reason to reject the null hypothesis, and this characteristic parameter can distinguish the actual distribution of normal and unsafe battery voltages at this significance level.

5. The eVTOL power cell thermal runaway risk pre-alarming method fusing flight operating conditions according to claim 1, characterized in that, The Box-Cox transformation in step C2 can automatically determine the form and parameters of the data transformation function according to the distribution of the original data. The data that does not conform to the normal distribution will conform or approximately conform to the normal distribution after transformation. The normality, additivity and homoscedasticity of the transformed data are generally significantly improved. The calculation formula of the Box-Cox transformation is as follows: In the formula, X represents the original variable data, Y represents the transformation parameter, Z represents the transformed data, the above transformation requires the original variable to be positive, and when the variable value is negative, a constant needs to be added to all original data X is positive, Z is positive, after this data translation operation, the above transformation can be performed; The parameters in the Box-Cox transformation are calculated using the maximum likelihood estimation method : In the formula, is the arithmetic mean of the transformed data , which can be calculated by gradient descent, quasi-Newton method, etc. corresponding to the maximum value .

6. A fusion flight operating mode eVTOL power battery thermal runaway risk early warning system, characterized in that, The eVTOL power battery thermal runaway risk early warning method suitable for the fusion flight working condition of any one of claims 1-8 comprises the following modules: A multi-element data processing and flight working condition segment division module for multi-element data processing and flight working condition segment division; A flight working condition and battery thermal runaway feature extraction module for flight working condition and battery thermal runaway feature extraction; A fusion flight working condition thermal runaway risk adaptive early warning module for fusion flight working condition thermal runaway risk adaptive early warning.

7. The eVTOL power cell thermal runaway risk pre-alarm system fusing flight operating conditions according to claim 6, characterized in that, The multi-element data processing and flight working condition segment division module includes a data acquisition and preprocessing unit, a flight working condition type division unit, and a micro working condition segment division unit based on a sliding time window method. The flight working condition and battery thermal runaway feature extraction module includes a flight working condition feature extraction unit based on micro segments, a thermal runaway feature extraction unit based on micro segments, and a feature optimization unit based on K-S test and maximum information coefficient. The fusion flight working condition thermal runaway risk adaptive early warning module includes a thermal runaway feature state estimation model establishment unit based on working condition division, a voltage thermal runaway feature abnormality judgment threshold calculation unit based on BOX-COX transformation, and a thermal runaway feature abnormality detection and early warning unit.

Images