An unmanned aerial vehicle health prediction method based on multi-source time sequence signal collaborative analysis driving

Abstract

The application discloses a kind of unmanned plane health prediction methods based on multi-source time series signal collaborative analysis driving, belong to unmanned plane health management technical field, the unmanned plane health prediction method includes the following steps: S1: for the unmanned plane health prediction method, individualized frequency domain health baseline migration is carried out, S2: for the unmanned plane health prediction method, dynamic weighted cross-dimension correlation matrix will be constructed, S3: for the unmanned plane health prediction method, between hybridization of physical model and data driving will be carried out, S4: for the unmanned plane health prediction method, example data and output will be carried out.The unmanned plane health prediction method based on multi-source time series signal collaborative analysis driving realizes unmanned plane health prediction by collecting the multi-source heterogeneous time series data generated in the process of unmanned plane flight.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of unmanned aerial vehicle health management, in particular to an unmanned aerial vehicle health prediction method based on multi-source time sequence signal collaborative analysis driving. BACKGROUND

[0002] The current unmanned aerial vehicle health monitoring method fuses unmanned aerial vehicle sensor data or flight log parameters, cooperates with multi-dimensional time sequence information to construct a deep space-time correlation model, so as to realize accurate prediction of the degradation trend of key components. However, in use, the existing unmanned aerial vehicle health monitoring method is not convenient for multi-source data acquisition and preprocessing, affecting the accuracy of the data.

[0003] In order to overcome the above defects, the prior art (application number CN202311369886.2, application date 20231023 Chinese patent application) a kind of electric aeroengine health management system and method, by perception unit, data transmission unit, cloud data storage and processing unit, state monitoring unit, state prediction unit, health examination unit, decision unit, precision control unit is made of, and it is combined to utilize the deep network perception model, solve the problem that current unmanned aerial vehicle electric aeroengine lacks health management and precision control, reach the purpose of unmanned aerial vehicle electric aeroengine fault detection and diagnosis, with the beneficial effects of comprehensive health management, high intelligence, high control precision;There is prior art (application number CN202411285072.5, application date 20240913 Chinese patent application) a kind of unmanned aerial vehicle battery monitoring method and system for intelligent hangar, input the real-time charge-discharge times of battery to the preset experience degradation model to obtain the initial value of battery health assessment;The real-time working condition information of unmanned aerial vehicle battery is input into the trained error compensation model to obtain the battery health error evaluation value;The battery health error evaluation value and the initial value of battery health assessment are superimposed to obtain the battery health evaluation value;Error compensation model can effectively supplement the uncertainty in the battery degradation process, enhance the adaptability and prediction accuracy of the model;And prior art (application number CN202111109826.8, application date 20210918 Chinese patent application) a kind of unmanned aerial vehicle health operation evaluation method based on digital twinning, adopts digital twinning technology, effectively realizes the cooperation of unmanned aerial vehicle safe operation and cost reduction.Combined with 5G technology, effectively improve the communication problem of unmanned aerial vehicle.Using finite element technology, or unmanned aerial vehicle digital twinning model constructed in Simulink environment, the unmanned aerial vehicle in physical environment is displayed in virtual environment.The constructed unmanned aerial vehicle visualization management and control platform can quickly and intuitively display the health status of unmanned aerial vehicle, so as to make health assessment in time and reduce the corresponding loss.Through the interaction feedback of unmanned aerial vehicle data in virtual environment and real environment, the real condition of unmanned aerial vehicle is accurately and efficiently displayed in virtual environment, which provides support for the development, operation and maintenance of unmanned aerial vehicle;Although the prior art can predict the health of unmanned aerial vehicle, in the working process, current unmanned aerial vehicle health monitoring depends on time domain threshold alarm or data-driven black box model, and is limited to single parameter threshold alarm, lacks systematic and predictive comprehensive evaluation, and it is difficult to capture early performance degradation, the latter has "illusion" risk and is difficult to explain.

[0004] In view of the above problems, it is urgent to make innovative design on the basis of the original unmanned aerial vehicle health prediction method. SUMMARY

[0005] The unmanned aerial vehicle health prediction method based on multi-source time sequence signal collaborative analysis driving aims to solve the problem that the current unmanned aerial vehicle health monitoring is mainly dependent on time domain threshold alarm or data-driven black box model, and is limited to single parameter threshold alarm, lacks systematic and predictive comprehensive evaluation, and is difficult to capture early performance degradation, and the latter has the risk of "illusion" and is difficult to explain.

[0006] To achieve the above object, the present application provides the following technical scheme: an unmanned aerial vehicle health prediction method based on multi-source time sequence signal collaborative analysis driving, the unmanned aerial vehicle health prediction method comprising the following steps: S1: for the unmanned aerial vehicle health prediction method, individualized frequency domain health baseline migration is performed, and the structure of the individualized frequency domain health baseline migration is health evaluated, and baseline establishment is controlled, and real-time monitoring and migration calculation are performed, and the unmanned aerial vehicle is health scored; S2: for the unmanned aerial vehicle health prediction method, a dynamic weighted cross-dimension correlation matrix is constructed for fault prediction, and whether the system overall state is stable is detected through correlation matrix construction, dynamic learning and adjustment and comprehensive health entropy calculation in the dynamic weighted cross-dimension correlation matrix; S3: for the unmanned aerial vehicle health prediction method, mixing between a physical model and data driving is performed, the remaining useful life is predicted, and the trend is anchored according to the physical model, and the data driving is corrected for deviation; S4: for the unmanned aerial vehicle health prediction method, example data and output are performed, and different dimension states are displayed by cooperating with a radar chart.

[0007] Preferably, the baseline is established after a new machine or overhaul, and calibration flight is performed under a standard environment to collect high-frequency vibration signals (sampling rate >=1kHz) of the structure (IMU); high-resolution Fourier transform (FFT) is performed on the signals to generate a standard frequency energy distribution map (F0) of the unmanned aerial vehicle in a healthy state, and the map is unique.

[0008] Preferably, the real-time monitoring and migration calculation also generates a frequency energy distribution map (Ft) of the current flight after each flight; the migration degree (D) of the current map and the health baseline is calculated, and the spectral correlation density (SCD) or Wasserstein distance is used to quantify the difference between the two distributions, wherein the index is extremely sensitive to small frequency band energy changes, and the calculation formula is Dt=SCD(F0,Ft).

[0009] Preferably, the smaller the health score migration degree (D) is, the higher the health degree is, and the structure health score Sstructure is calculated according to the migration degree: Sstructure=5*exp(-lambda*Dt) (wherein lambda is an attenuation coefficient calibrated according to historical data), and this method can find early imbalance or small structure cracks that cannot be found by traditional time domain indicators (such as total vibration).

[0010] Preferably, the correlation matrix is defined as a 6x6 health state correlation matrix M, with elements M ij representing the weight of the i-th dimension health state on the j-th dimension (e.g. the weight of power system anomaly on structure health); the initial weight is determined by domain expert knowledge and historical failure data.

[0011] Preferably, the dynamically learned and adjusted matrix weight is not fixed, the system dynamically updates the weight matrix M through continuous monitoring using Granger causality test or transfer entropy analysis, for example, if historical data shows that "battery voltage drop" (i) is frequently followed by "navigation module restart" (j) multiple times, the system will automatically increase the weight of M ij .

[0012] Preferably, the comprehensive health entropy value (H) calculation introduces the concept of "health entropy" to evaluate the overall chaos degree of the system, first calculates the independent score Si of each dimension, and then modifies it through the correlation matrix M to obtain the comprehensive health entropy value H, the calculation formula is: H = Σ (Pi * log (Pi)), where Pi = (ΣjMji * Sj) / ΣSj; the lower the entropy value H, the more stable and healthy the overall state of the system; an increase in entropy value indicates that the system is in chaos and the risk of failure increases.

[0013] Preferably, for components such as batteries that have a clear degradation model, instead of simply using long short-term memory networks for prediction, a physical model is used to anchor the trend: an empirical degradation model based on electrochemical model (such as double exponential model) is used to predict the macroscopic degradation curve, and data-driven correction is used to correct the deviation: a Kalman filter is used to take the real-time monitored health indicators (such as internal resistance growth) as observations to real-time correct the prediction results of the physical model; this method not only utilizes the stability of the physical model, but also absorbs the feedback of real-time data, making the prediction results more reliable and interpretable.

[0014] Preferably, the example data and output data show that after this flight, the power system health score Spower dropped from 4.5 to 3.0 (due to increased motor torque fluctuation), while the structure system health score Sstructure remained temporarily at 4.8 (time domain indicators are normal), the system analysis: querying the dynamic correlation matrix, it is found that Mpower->structure (the weight of power on structure) is high, at 0.8, the system immediately calculates that the abnormal fluctuation of the power system, even if it does not immediately cause the structure vibration to exceed the standard, will continue to cause hidden fatigue damage to the arm.

[0015] Preferably, the example data and the output in the output pass through the early warning information: "detecting abnormal fluctuation of power system, it is expected that in the future 15 take-off and landing cycles, resulting in structural health degree to drop to early warning level (<2.5 points)"; then make maintenance suggestions: "it is suggested to check the motor and propeller balance immediately, and carry out nondestructive testing on the key connecting structure".

[0016] Compared with the prior art, the present application has the beneficial effects that:

[0017] 1. The unmanned aerial vehicle health prediction method based on multi-source time sequence signal collaborative analysis driving acquires multi-source heterogeneous time sequence data (such as collecting structure, battery voltage and current, signal strength, temperature and the like) generated in the flight process of the unmanned aerial vehicle, designs a set of refined health degree evaluation model fusing time domain and frequency domain analysis for six core dimensions of battery, structure, communication, navigation, hardware and power, calculates five-point system scores for each dimension through segmented scoring or weighted formula method, and finally integrates into an intuitive multi-dimensional health score radar chart;

[0018] 2. The unmanned aerial vehicle health prediction method based on multi-source time sequence signal collaborative analysis driving determines the scoring parameters by using historical big data analysis, realizes the change from single fault alarm to system health trend prediction, can comprehensively, quantitatively and in advance evaluate the overall health state of the unmanned aerial vehicle, provides core decision basis for predictive maintenance, significantly improves the operation safety and reliability of the unmanned aerial vehicle, and initiates two core mechanisms of "individualized frequency domain health baseline" and "cross-dimension fault propagation graph";

[0019] 3. The unmanned aerial vehicle health prediction method based on multi-source time sequence signal collaborative analysis driving, the system realizes the micro change perception of the structural health degree by establishing the high-frequency vibration frequency domain energy distribution baseline unique to each unmanned aerial vehicle and monitoring the migration thereof; further, by constructing a dynamic weighted cross-dimension correlation matrix, the health state coupling relationship between different systems (such as power and structure, battery and temperature) is quantitatively analyzed, so as to predict the chain reaction failure caused by single component failure; through multi-dimensional health entropy comprehensive evaluation, the output is not only accurate but also has strong explainability, which realizes the leap from "post-alarm" to "pre-prediction", from "single-point monitoring" to "system-level insight". BRIEF DESCRIPTION OF DRAWINGS

[0020] Figure 1 The figure is a structural schematic diagram of the unmanned aerial vehicle health prediction method based on multi-source time sequence signal collaborative analysis driving of the present application.

[0021] Figure 2 The figure is a structural schematic diagram of the detailed steps of the unmanned aerial vehicle health prediction method based on multi-source time sequence signal collaborative analysis driving of the present application.

[0022] Figure 3 For the battery health dimension score rule data structure diagram of the application;

[0023] Figure 4 For the structure health dimension score rule data structure diagram of the application;

[0024] Figure 5 For the multi-dimensional fusion and health radar chart structure diagram of the application;

[0025] Figure 6 For the example data structure diagram of the application. DETAILED DESCRIPTION

[0026] The technical solutions in the embodiments of the application will be clearly and completely described below with reference to the drawings in the embodiments of the application. Obviously, the described embodiments are only part of the embodiments of the application, rather than all the embodiments of the application. Based on the embodiments in the application, all other embodiments obtained by a person of ordinary skill in the art without creative work fall within the protection scope of the application.

[0027] Please refer to Figures 1-6 , the application provides a technical solution: a UAV health prediction method based on multi-source time series signal collaborative analysis driving, the UAV health prediction method comprises the following steps:

[0028] Step 1: For the UAV health prediction method, individual frequency domain health baseline migration is performed, and the structure of the individual frequency domain health baseline migration is health evaluated, and the baseline is established and controlled, and real-time monitoring and migration calculation are performed, and the UAV is health scored;

[0029] Step 2: For the UAV health prediction method, a dynamic weighted cross-dimension correlation matrix is constructed for fault prediction, and through correlation matrix construction, dynamic learning and adjustment and comprehensive health entropy calculation in the dynamic weighted cross-dimension correlation matrix, it is detected whether the overall state of the system is stable;

[0030] Step 3: For the UAV health prediction method, mixing between a physical model and data driving is performed to predict the remaining useful life, and the trend is anchored according to the physical model, and the data driving is corrected for deviation;

[0031] Step 4: For the UAV health prediction method, example data and output are performed, and a radar chart is used to display different dimension states (such as Figure 1 and Figure 2 as shown).

[0032] Baseline is established after new machine or overhaul, calibration flight is carried out in standard environment, high-frequency vibration signal (sampling rate ≥ 1 kHz) of structure (IMU) is collected; high-resolution Fourier transform (FFT) is carried out on the signal, and the standard frequency domain energy distribution map (F0) of the unmanned aerial vehicle in the healthy state is generated, which is the unique "health fingerprint" of the unmanned aerial vehicle.

[0033] Real-time monitoring and migration calculation generates the frequency domain energy distribution map (Ft) of the flight after each flight; the migration degree (D) of the current map and the health baseline is calculated, and the spectral correlation density (SCD) or Wasserstein distance is used to quantify the difference between the two distributions, wherein the index is extremely sensitive to the small frequency band energy change, and the calculation formula is: Dt=SCD(F0,Ft).

[0034] The smaller the migration degree (D) is, the higher the health degree is, and the structure health score Sstructure is calculated according to the migration degree: Sstructure=5*exp(-λ*Dt) (wherein λ is an attenuation coefficient, which is calibrated according to historical data), and this method can find early imbalance or small structure cracks that cannot be found by traditional time domain indicators (such as total vibration).

[0035] Correlation matrix construction defines a 6x6 health state correlation matrix M, and the element M ij represents the influence weight of the health state of the i-th dimension on the j-th dimension (for example, the influence weight of the power system anomaly on the structure health); the initial weight is determined by the domain expert knowledge and historical fault data.

[0036] Dynamic learning and adjustment of matrix weight is not fixed, the system dynamically updates the weight matrix M through continuous monitoring, using Granger causality test or transfer entropy analysis, for example, if historical data shows that "battery voltage drop" (i) frequently occurs after "navigation module restarts" (j) many times, the system will automatically increase the weight of M ij .

[0037] Comprehensive health entropy (H) calculation introduces the concept of "health entropy" to evaluate the overall chaos degree of the system, first calculates the independent score Si of each dimension, and then modifies it through the correlation matrix M to obtain the comprehensive health entropy H, the calculation formula is: H=Σ(Pi*log(Pi)), wherein Pi=(ΣjMji*Sj) / ΣSj; the lower the entropy value H is, the more stable and healthy the overall state of the system is; the increase of the entropy value indicates that the system is in chaos, and the fault risk increases.

[0038] For components with clear degradation models, such as batteries, instead of using long short-term memory networks for prediction, a physical model is used to anchor the trend: an empirical degradation model based on electrochemical models (such as the double exponential model) is used to predict the macroscopic degradation curve, and data-driven correction is used to correct deviations: using a Kalman filter, real-time monitoring of health indicators (such as internal resistance growth) as observations, real-time correction of the prediction results of the physical model; This method not only takes advantage of the stability of the physical model, but also absorbs the feedback of real-time data, making the prediction results more reliable and interpretable.

[0039] Example data and data in output, after this flight, the power system health score Spower dropped from 4.5 to 3.0 (due to increased motor torque fluctuation), and the structure system health score Sstructure remained at 4.8 (time domain indicators were normal), system analysis: query dynamic correlation matrix, found Mpower->structure (the impact weight of power on structure) is high, 0.8, the system immediately calculates: abnormal fluctuation of the power system, even if it does not immediately cause structural vibration to exceed the standard, it will continue to cause hidden fatigue damage to the arm.

[0040] Example data and output in output through early warning information: "detected abnormal fluctuation of power system, predicted to cause structure health to drop to early warning level (<2.5 points) within the next 15 take-off and landing cycles"; then make maintenance recommendations: "suggest checking motor and propeller balance immediately, and perform non-destructive testing on key connecting structures".

[0041] Working method

[0042] Multi-source data acquisition and preprocessing

[0043] Data sources:

[0044] Battery: voltage, current, temperature, internal resistance, charge and discharge cycle number.

[0045] Structure: three-axis accelerometer (vibration standard deviation, kurtosis), gyroscope (angular velocity standard deviation, kurtosis).

[0046] Communication: signal strength (RSSI), packet loss rate, communication delay.

[0047] Navigation: GPS positioning accuracy (HDOP), number of satellites, positioning drift standard deviation.

[0048] Hardware: CPU temperature, memory usage, storage read / write error rate.

[0049] Power: motor speed, electronic speed controller output current, thrust efficiency.

[0050] Sampling frequency:

[0051] High frequency signal (e.g. vibration): 100 Hz

[0052] Medium frequency signal (e.g. voltage, current): 10 Hz

[0053] Low frequency signal (e.g. temperature, communication quality): 1 Hz

[0054] Preprocessing:

[0055] De-noising: Wavelet transform (Daubechies5 wavelet) is used to de-noise the vibration and current signals.

[0056] Normalization: Standard deviation standardization method (Z-score standardization method) is used to normalize all indicators to the same dimension.

[0057] Missing value processing: Linear interpolation or forward-backward frame padding is used.

[0058] Health indicator calculation and scoring rules

[0059] The health score of each dimension is based on multiple sub-indicators, calculated by a rule engine, and does not rely on machine learning models to ensure interpretability and stability.

[0060] Battery health dimension

[0061] Indicators: voltage smoothness (discharge curve variance), maximum voltage drop, internal resistance growth rate,

[0062] Scoring rules: Each indicator is scored on a five-point scale (e.g. as shown in Figure 3 ), and the final score is taken as the lowest score;

[0063] Structural health dimension

[0064] Time domain indicators: vibration standard deviation, vibration kurtosis, angular velocity standard deviation, angular velocity kurtosis

[0065] Frequency domain indicators: Fourier transform (FFT) is performed on the vibration signal to calculate the energy proportion (Eband) of a specific frequency band (e.g. 100-200 Hz)

[0066] Scoring rules: Each indicator is scored separately, and the final score is taken as the lowest score among all indicators (e.g. as shown in Figure 4 );

[0067] Communication health dimension

[0068] Weighted scoring method is used:

[0069] Scomm=0.5*ScoreRSSI+0.3*(1-ScoreLoss)+0.2*(1-ScoreDelay)

[0070] Each sub-index score rule is similar (e.g. signal strength >-70dBm is 5 points, <-90dBm is 1 point);

[0071] Other dimensions

[0072] Similarly, set the threshold table to score, and finally take the lowest score or weighted score according to the dimension attribute;

[0073] Multi-dimensional fusion and health radar chart generation

[0074] Each dimension score is summarized as a six-dimensional health vector, which can be visualized by a radar chart to intuitively display the health status (e.g. Figure 5 As shown).

[0075] Trend prediction and maintenance recommendations

[0076] Trend analysis: calculate the score slope of each dimension based on historical data (e.g. battery score trend of the past 10 flights).

[0077] Early warning rules:

[0078] If any dimension score ≤2, immediately alarm;

[0079] If the score decreases continuously for 3 times, prompt "performance degradation";

[0080] Combine the score changes of multiple dimensions (e.g. battery score decreases and temperature score decreases) to predict potential failures.

[0081] Example data and output

[0082] Example data (a flight segment) (e.g. Figure 6 As shown):

[0083] Health score output:

[0084] Battery health score = min(3, 3, 4) = 3 (pass)

[0085] Structural health score = min(2, 2) = 2 (warning)

[0086] Communication health score = 0.5*3 + 0.3*(1-0.2) + 0.2*(1-0.1) = 3.2 (rounded to 3)

[0087] Hardware health score = 3 (CPU temperature is high)

[0088] Radar chart and maintenance recommendations:

[0089] Radar chart: shows that the structural dimension score is the lowest (2 points), and the battery and hardware dimension scores are lower (3 points).

[0090] Maintenance recommendations:

[0091] "Structural Health Alert: Recommend checking motor mounting screws and blade balance."

[0092] "Battery Performance Degradation: Recommend deep discharge calibration."

[0093] "CPU Temperature High: Recommend cleaning thermal vents."

[0094] The contents of the present specification not described in detail are the prior art known to those skilled in the art.

[0095] While embodiments of the present application have been shown and described, it is to be understood that the embodiments described are merely divergences of the principles and spirit of the present application and that numerous modifications, changes, substitutions, and alterations can be made thereto without departing from the spirit and scope of the present application as defined by the appended claims and their equivalents.

Claims

1. A UAV health prediction method based on multi-source time series signal collaborative analysis driving, characterized in that, The unmanned aerial vehicle health prediction method comprises the following steps: S1: For the unmanned aerial vehicle health prediction method, individualized frequency domain health baseline migration is performed, and the structure of the individualized frequency domain health baseline migration is health evaluated, and baseline establishment is controlled, and real-time monitoring and migration calculation are performed, and the unmanned aerial vehicle is health scored; S2: For the unmanned aerial vehicle health prediction method, a dynamic weighted cross-dimension correlation matrix is constructed for fault prediction, and the correlation matrix in the dynamic weighted cross-dimension correlation matrix is constructed, dynamically learned and adjusted, and comprehensive health entropy value calculation is performed to detect whether the overall state of the system is stable; S3: For the unmanned aerial vehicle health prediction method, mixing between a physical model and data driving is performed to predict the remaining useful life, and the trend is anchored according to the physical model, and the data driving is corrected for deviation; S4: For the unmanned aerial vehicle health prediction method, example data and output are performed, and a radar chart is used to display different dimension states. 2.The unmanned aerial vehicle health prediction method based on multi-source time sequence signal collaborative analysis driving according to claim 1, characterized in that: The baseline is established after a new machine or overhaul, and calibration flight is performed under standard environment to collect high-frequency vibration signals of the structure, wherein the sampling rate is greater than or equal to 1 kHz; high-resolution Fourier transform is performed on the signals to generate a standard frequency domain energy distribution map F0 of the unmanned aerial vehicle in a healthy state, and the map is unique. 3.The unmanned aerial vehicle health prediction method based on multi-source time sequence signal collaborative analysis driving according to claim 2, characterized in that: The real-time monitoring and migration calculation also generates a frequency domain energy distribution map Ft of the current flight after each flight; the migration degree D of the current map and the health baseline is calculated, and the spectral correlation density SCD is used to quantify the difference between the two distributions, wherein the index is extremely sensitive to small frequency band energy changes, and the calculation formula is Dt=SCD(F0,Ft).

4. The unmanned aerial vehicle health prediction method based on multi-source time sequence signal collaborative analysis driving according to claim 3, characterized in that: The smaller the migration degree D is, the higher the health degree is, and the structure health score Sstructure is calculated according to the migration degree: Sstructure=5*exp(-λ*Dt); Wherein λ is an attenuation coefficient, which is calibrated according to historical data, and this method can find early imbalance or small structure cracks that cannot be found by traditional time domain indicators.

5. The unmanned aerial vehicle health prediction method based on multi-source time sequence signal collaborative analysis driving according to claim 1, characterized in that: The correlation matrix construction defines a 6x6 health state correlation matrix M, with elements M ij representing the weight of the health state of the i-th dimension on the j-th dimension. The initial weight is determined by domain expert knowledge and historical fault data.

6. The unmanned aerial vehicle health prediction method based on multi-source timing signal collaborative analysis driving according to claim 1, characterized in that: The dynamic learning and adjustment of the matrix weight is not fixed, the system continuously monitors, uses Granger causality test or transfer entropy analysis, dynamically updates the weight matrix M, if the historical data shows that the battery voltage drops record i multiple times, and then the navigation module restarts record j frequently, the system will automatically increase the weight of M ij .

7. The unmanned aerial vehicle health prediction method based on multi-source timing signal collaborative analysis driving according to claim 6, characterized in that: The comprehensive health entropy value H calculation introduces the concept of health entropy to evaluate the overall chaos degree of the system. First, the independent score Si of each dimension is calculated, and then the comprehensive health entropy value H is obtained by modifying the correlation matrix M. The calculation formula is: H=Σ(Pi*log(Pi)), Wherein Pi=(ΣjMji*Sj) / ΣSj; the lower the entropy value H is, the more stable and healthy the overall state of the system is; and the increase of the entropy value indicates that the system is in chaos and the risk of failure increases. 8.The unmanned aerial vehicle health prediction method based on multi-source time sequence signal collaborative analysis driving according to claim 1, characterized in that: The physical model and data driving are used for components with a clear attenuation model of the battery. Instead of simply using long short-term memory network for prediction, a physical model is used to anchor the trend: an empirical degradation model based on an electrochemical model is used to predict the macroscopic attenuation curve, and data driving is used to correct the deviation: a Kalman filter is used to take the health indicators monitored in real time as observation values to correct the prediction results of the physical model in real time.

9. The unmanned aerial vehicle health prediction method based on multi-source timing signal collaborative analysis driving according to claim 1, characterized in that: The example data with output in the output by pre-warning information: detected power system abnormal fluctuation, is expected to be in the future 15 take-off and landing cycle, resulting in a decrease in structural health to the pre-warning level and then maintenance recommendations.

Images