A multi-source flight test data quality evaluation method and system

Abstract

This invention belongs to the field of flight test data processing technology and discloses a method and system for quality assessment of multi-source flight test data based on physical-temporal constraints. The method includes the following steps: data preprocessing and benchmark calibration; physical-temporal joint verification; multi-source spatiotemporal alignment; scenario adaptive evaluation; and comprehensive quality scoring and feedback optimization. This invention solves the problems of insufficient correlation and inconsistent spatiotemporal benchmarks in existing flight test multi-source data, achieving full-dimensional and high-precision quality assessment of multi-source flight test data.

Description

Technical Field

[0001] This invention belongs to the field of flight test data processing technology, specifically involving a multi-dimensional data quality assessment method and system based on physical dynamic constraints and multi-source spatiotemporal synchronization. It is applicable to the quality verification of multi-source heterogeneous data such as airborne sensor data, GNSS (Global Navigation Satellite System) data, and video image data in flight tests of aircraft (including human-machine, unmanned aerial vehicle, test aircraft, etc.). Background Technology

[0002] Flight test data is the core basis for aircraft performance verification, fault diagnosis, and design improvement. With the increasing dimensionality of flight test parameters (such as aerodynamic loads, engine status, and avionics system signals), the need for multi-source data fusion applications is becoming increasingly urgent. However, existing technologies still face the following bottlenecks in data quality assessment: 1. Insufficient correlation of multi-source data: Existing solutions mostly focus on isolated analysis of single data sources, lacking modeling and analysis of physical correlations such as airspeed-lift and thrust-fuel consumption, making it difficult to identify hidden contradictions between parameters (for example, airborne sensors show normal lift but GNSS velocity measurement is abnormal, which may be caused by angle-of-attack sensor drift).

[0003] 2. Inconsistent spatiotemporal references: Multi-source data (such as airborne equipment timestamps, GNSS absolute time, and video frame timestamps) suffer from clock drift and synchronization errors. For example, there may be millisecond-level deviations between airborne equipment clocks and GNSS time, and video frame timestamps may be out of sync with actual event times due to recording delays, making it impossible to accurately correlate multi-source data for critical events (such as takeoff and engine failure).

[0004] 3. Static Evaluation System: Existing methods often use fixed thresholds (such as ±5% of the exceedance) to judge data quality, without considering the dynamic differences in flight phases (takeoff, cruise, and landing). For example, the altitude change rate during takeoff can reach 5-10 m / s², while it is only 0.1-0.3 m / s² during cruise. Fixed thresholds can easily lead to misjudging normal fluctuations as abnormalities during takeoff, or missing minute but critical sensor drifts during cruise. Summary of the Invention

[0005] This invention addresses the problems of insufficient correlation between multi-source flight test data and inconsistent spatiotemporal references in existing flight tests by proposing a three-dimensional evaluation framework of "physical law constraints, spatiotemporal reference synchronization, and scene dynamic adaptation." By integrating dynamic models, spatiotemporal alignment algorithms, and scene adaptive mechanisms, it achieves full-dimensional and high-precision quality evaluation of multi-source flight test data.

[0006] The technical solution of this invention is implemented as follows: In a first aspect, the present invention provides a method for evaluating the quality of multi-source flight test data, which includes the following steps: Data preprocessing and benchmark calibration steps: Collect and unify the formats of airborne test data, GNSS differential data and video data; perform preliminary data cleaning, and use GNSS time as a benchmark to calibrate the timestamps of airborne equipment using a piecewise linear correction model to achieve spatiotemporal benchmark alignment of multi-source data; The physical-temporal joint verification steps are as follows: calculate the mutual information entropy between multi-source parameters, construct a parameter correlation matrix to identify strongly correlated links; verify the consistency of parameters on strongly correlated links based on physical dynamic formulas, and conduct root cause analysis on the discovered contradictions in conjunction with video data; Multi-source spatiotemporal alignment steps: Define the core events in the flight test and extract their features in multi-source data; calculate the similarity of the descriptions of the same event in multi-source data, and use the sliding window least squares method to compensate for timescale drift in real time to achieve accurate association of key events; Scenario-adaptive evaluation steps: Based on multi-source features, clustering algorithms are used to automatically identify flight phases; the weights of data quality evaluation indicators are dynamically adjusted according to the characteristics of different flight phases. Comprehensive quality scoring and feedback optimization steps: Based on the dynamic weights, the completeness, accuracy, consistency and timeliness of the data are weighted and calculated to obtain a comprehensive quality score; based on the score results, a feedback mechanism is triggered to supplement data collection, parameter calibration or algorithm optimization.

[0007] As a further technical solution of the present invention: in the data preprocessing and benchmark calibration steps, the piecewise linear correction model is: ; in, For the calibrated timestamp, This is the original timestamp of the airborne equipment. The start time of the joint test. The clock drift coefficient is calibrated through a synchronization test. It is a fixed offset.

[0008] As a further technical solution of the present invention: in the physical-spatiotemporal joint verification step, a parameter correlation matrix is ​​constructed: ; in, The multi-source parameter correlation matrix is ​​the core matrix that stores the degree of correlation between all parameters. It is an n-order real matrix. To normalize mutual information entropy, The mutual information entropy among multiple source parameters; When two parameters and Normalized mutual information entropy When the link is determined to be strongly correlated, the verification of strongly correlated links allows for an error of ±20% between the calculation results based on physical dynamics formulas and the measured values.

[0009] As a further technical solution of the present invention: in the multi-source spatiotemporal alignment step, the similarity calculation model is as follows: ; in, The angle between the video feature vector and the GNSS feature vector. The dynamic time-warped distance between the video data sequence and the sensor data sequence is represented by the weighting coefficients α=0.6 and β=0.4.

[0010] As a further technical solution of the present invention: In the scene adaptive evaluation step, the DBSCAN clustering algorithm is used for flight phase identification, and its neighborhood radius... Set to 1.2 times the standard deviation of the eigenvectors, i.e. Minimum number of samples Set it to 50.

[0011] As a further technical solution of the present invention: In the scene adaptive evaluation step, a reinforcement learning strategy gradient algorithm is used to dynamically adjust the evaluation weights, and its state space is defined as follows: ,in For flight phase identification, to Scoring is done for each quality component; the reward function is designed as follows: ,in and These are the quality scores for core parameters and non-core parameters, respectively.

[0012] As a further technical solution of the present invention: in the comprehensive quality scoring and feedback optimization step, the comprehensive quality score... The calculation formula is: ; in, To score for completeness, For accuracy scoring, For consistency scoring, Rate the timeliness; to The weights are dynamically adjusted according to the flight phase.

[0013] Secondly, the present invention provides a multi-source flight test data quality assessment system for implementing the above-described method, comprising: The data preprocessing and calibration module is used to receive multi-source heterogeneous data and perform format unification, preliminary cleaning and spatiotemporal reference alignment. The physics-spatiotemporal joint verification module is used to construct the parameter correlation matrix and perform multi-level anomaly detection and verification based on physical dynamic constraints and video data; The multi-source spatiotemporal alignment module is used to define core events, calculate the similarity of multi-source data, and compensate for timescale drift in real time. The scenario-adaptive evaluation module is used to identify flight phases and dynamically configure the weights of data quality evaluation indicators. The comprehensive scoring and feedback module is used to calculate the comprehensive quality score and generate feedback optimization instructions based on the scoring results.

[0014] As a further technical solution of the present invention: the data preprocessing and calibration module integrates a piecewise linear correction unit for performing timestamp calibration calculation; the physical-temporal joint verification module integrates an association matrix calculation unit and a three-level verification logic unit.

[0015] As a further technical solution of the present invention: the scene adaptive evaluation module integrates a DBSCAN clustering algorithm unit and a weight optimization unit based on reinforcement learning; the comprehensive scoring and feedback module integrates a weighted scoring calculation unit and a feedback triggering mechanism unit.

[0016] Compared with the prior art, the beneficial effects of the present invention are as follows: 1. Improved correlation verification capability: The parameter correlation matrix based on mutual information entropy can identify more than 90% of implicit data contradictions, and the anomaly detection accuracy has been improved from 60% to 95%.

[0017] 2. Improved spatiotemporal alignment accuracy: The piecewise linear correction model reduces the timestamp alignment error from ±50ms to ±30ms, and increases the success rate of key event matching from 75% to 92%.

[0018] 3. Enhanced scenario adaptability: DBSCAN clustering and reinforcement learning weight optimization reduce evaluation error by 40%, with a false negative rate of ≤3% during takeoff and a false positive rate of ≤2% during landing.

[0019] 4. Closed-loop optimization capability: The feedback mechanism can automatically correct problems in the data collection and processing process, increasing data availability from 70% to 90%.

[0020] The present invention will be further described in detail below with reference to the accompanying drawings and embodiments. Attached Figure Description

[0021] Figure 1 This is an overall flowchart of a method for evaluating the quality of multi-source flight test data based on physical-temporal constraints according to the present invention. Detailed Implementation

[0022] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the present invention... All other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this invention.

[0023] The features and illustrative embodiments of various aspects of the present invention will now be described in detail. Numerous specific details are set forth in the following detailed description to provide a thorough understanding of the invention. However, it will be apparent to those skilled in the art that the invention may be practiced without requiring some of these specific details. The following description of embodiments is merely intended to provide a better understanding of the invention by illustrating examples of the invention. The invention is by no means limited to any specific setups and methods set forth below, but covers any improvements, substitutions, and modifications to structures, methods, and devices without departing from the spirit of the invention. Well-known structures and techniques are not shown in the drawings and the following description to avoid unnecessarily obscuring the invention.

[0024] It should be noted that, unless otherwise specified, the embodiments of the present invention and the features thereof can be combined with each other, and the various embodiments can be referenced and cited from each other.

[0025] The following is in conjunction with the appendix Figure 1 The embodiments of the present invention will be described in detail below.

[0026] Example 1 like Figure 1 As shown: This invention discloses a method for evaluating the quality of multi-source flight test data based on physical-temporal constraints, which includes the following steps: 1.1 Data Preprocessing and Benchmark Calibration 1.1.1 Multi-source data acquisition and format unification: Acquire three types of raw data streams and convert them into a unified structured format for storage. Airborne test data A (sensor parameters, sampling rate ≥100Hz); GNSS differential data G (positioning accuracy ≤10cm); The unified storage format for video data V (resolution ≥ 720P, frame rate ≥ 30fps) is: "timestamp (UTC milliseconds) - parameter type - parameter value (physical unit)", using IEEE 754 double-precision floating-point encoding (e.g., "1620000000000-a-2.5" means that the acceleration is 2.5m / s² at 1620000000000ms).

[0027] 1.1.2 Preliminary data cleaning and establishment of a multidimensional anomaly detection rule base: Airborne parameters: based on the sensor range ±3σ criterion; Video parameters: Inter-frame brightness change rate threshold detection; Timestamp: GNSS time reference calibration error ≤ 1ms; 1.1.3 Spatiotemporal reference alignment employs a piecewise linear correction model to correct airborne equipment clock drift: (Equation 1) Where: : Original timestamp of airborne equipment (ms); : Calibrated timestamp (ms); : Joint test start time (ms); Clock drift coefficient (calibrated through GNSS / airborne time synchronization tests); : Fixed offset (residual after calibration).

[0028] 1.2 Physics-Spatiotemporal Joint Verification 1.2.1 Constructing the Parameter Correlation Matrix: Calculate the mutual information entropy (MI) among multiple source parameters to quantify the degree of correlation between parameters and construct the correlation matrix. (Equation 2), when When two parameters are determined to be strongly correlated (e.g., airspeed v and lift L satisfy...), the relationship is considered to be strong. The theoretical correlation should be close to 1.

[0029] 1.2.2 Anomaly propagation detection employs a three-level verification mechanism: Primary screening: Initial screening of single parameters using 3σ; Secondary verification: Verify the correlation parameters in the strongly correlated links using dynamic formulas (allowing ±20% error); Level 3 verification: Root cause analysis of dynamic contradictions based on video data; 1.3 Multi-source spatiotemporal alignment technology 1.3.1 Definition of Core Event Features: Extract feature sets for six categories of key events (as shown in Table 1): Table 1 Core Event Definitions

[0030] 1.3.2 The similarity calculation model quantifies the similarity of multi-source data: (Equation 3) Where: : The angle between the video feature vector and the GNSS feature vector (cosine similarity). Dynamic time-normalized distance; 1.3.3 Time scale drift compensation uses a sliding window least squares method to update clock drift parameters in real time. (Equation 4) The sliding window length is set to 10s, and it is updated every 0.5s. To ensure real-time performance.

[0031] 1.4 Scenario Adaptive Evaluation System 1.4.1 Flight Phase Identification: The DBSCAN algorithm is used to cluster multi-source features (mean altitude, velocity fluctuation rate, attitude angle change rate) to automatically divide flight phases (takeoff, cruise, landing, etc.). Parameter configuration is as follows: Neighborhood radius: ; Minimum number of samples: ; 1.4.2 Scene Weight Optimization: Based on the reinforcement learning policy gradient algorithm, the evaluation weights for different flight stages are dynamically adjusted. The state space is defined as follows: The reward function (Equation 5) is designed as follows: (Formula 6).

[0032] 1.5 Comprehensive Quality Score and Feedback Optimization 1.5.1 Comprehensive Scoring Model for Multi-Source Data Comprehensive Quality Score Calculated by weighted average of four parts: (Equation 7) Where: Completeness ( ); :accuracy( ); :consistency( ); Timeliness ); 1.5.2 The feedback optimization mechanism marks low-quality data as "pending review" and triggers: Data supplementation: Supplementation is achieved through interpolation from adjacent sensors or replay experiments; Parameter calibration: Correcting sensor zero-point offset and sensitivity coefficient; Algorithm optimization: Update the dynamic model or correlation matrix based on misjudged cases.

[0033] Example 2 This invention discloses a data quality assessment method for the takeoff phase.

[0034] Experimental Background: Data quality assessment was conducted on the entire takeoff and landing process of a lightweight unmanned aerial vehicle (UAV) (maximum takeoff weight 25kg, wing area 1.2m²). The experiment lasted 120 seconds under low-disturbance environmental conditions (wind speed ≤3m / s, air temperature 25℃, air pressure 101.3kPa). A differentiated sampling strategy was designed based on the parameter characteristics during takeoff. (1) Airborne test data A (vertical accelerometer a) z Pitch gyroscope ω y ): It adopts a 128Hz sampling rate to adapt to the dynamic changes in parameters (acceleration / attitude angle change rate reaches 5-10m / s²), ensuring the capture of instantaneous fluctuation characteristics; (2) GNSS differential data G (positioning, altitude): adopts a sampling rate of 20Hz, balances accuracy and storage pressure, and meets the measurement requirements of altitude change rate ≤6.5m / s² during takeoff; (3) Video data V (landing gear status, instrument readings): 30fps frame rate, 1280×720P resolution, to achieve multi-source data time dimension coverage.

[0035] Step 1: Data Preprocessing and Benchmark Calibration 1.1 Raw Data Parameter Specifications Table 2 Original Data

[0036] 1.2 Data Cleaning Execution Airborne parameter selection: based on the ±3σ criterion, a z The 3σ threshold is 0.03 m / s² (10 × 0.1% × 3), ω y The 3σ threshold is 0.03 rad / s (5 × 0.2% × 3); the mean is calculated for every 100 sampling points (≈0.78125 s), and data exceeding the threshold are removed (only 1 location with 8 consecutive abnormal sampling points is identified as vibration interference, accounting for 0.05%). Timestamp calibration: The "GNSS timestamp interpolation alignment" strategy is adopted. The GNSS 20Hz time series is linearly interpolated to generate a 128Hz series, which is then compared and calibrated with the original airborne timestamp. The final timestamp error is ≤1ms, which meets the test accuracy requirements.

[0037] 1.3 Spatiotemporal reference alignment A piecewise linear correction model is adopted (Equation 1: Achieve time synchronization of multi-source data: (1) Start time of joint test =5s (Drone starts taxiing, 640 valid points have been collected onboard, and 100 valid points have been collected by GNSS); (2) Clock drift coefficient k = 0.0012 (airborne clock drifts 1.2ms every 1000ms), fixed offset b = 0.8ms; (3) Calibration effect: The alignment error between the airborne and GNSS timestamps is stable at ±22ms, which meets the key test index requirement of ≤30ms.

[0038] Step 2: Physics-spacetime joint verification 2.1 Construction of the parameter correlation matrix With a z (128Hz) and ω y (128Hz) is the core correlation parameter, and the theoretical relationship between the two is as follows: (g is the acceleration due to gravity, θ is the pitch angle, and v is the gliding speed): (1) Mutual information entropy calculation: 512 airborne sampling points (≈4s, corresponding to 80 GNSS sampling points) were used as the sliding window to ensure complete coverage of the dynamic cycle, and the mutual information entropy MI=0.82 was calculated; (2) Elements of the correlation matrix: Substitute into the formula ,have to (≥0.7, determined as a strongly correlated link).

[0039] 2.2 Anomaly Detection and Localization Anomaly warning triggered: at t=42s (airborne 5376th sampling point, GNSS 840th sampling point), a z The velocity remained at 2.5 m / s² for 12 consecutive sampling points (≈0.09375 s), exceeding the normal range and triggering an early warning. Dynamics verification: Considering the gliding speed v = 18 m / s and the pitch angle θ ≈ 15° at this moment, the theoretical ω is calculated. y ≈0.25 rad / s (allowable error ±20%, i.e., 0.2-0.3 rad / s); measured ω y =15rad / s, the error exceeds the threshold, and the judgment parameters are contradictory; Video verification: Retrieve video frames (frames 10-11) corresponding to t=42.0-42.09375s to confirm that the drone's attitude changes smoothly, consistent with ω. y The discrepancy between the 15 rad / s reading and the "drastic attitude change" was ultimately identified as a gyroscope zero-point drift.

[0040] Step 3: Multi-source spatiotemporal alignment 3.1 Extraction of core event features Focusing on the critical event of "takeoff and liftoff" (t=45s), multi-source features are extracted: Airborne characteristics: Within 128 sampling points (≈1 second) around 45 seconds, a z Stable at 2.5 m / s², ωy Restored to 0.28 rad / s (gyroscope malfunction has been corrected); GNSS characteristics: 45s corresponds to the 900th GNSS sampling point, with the altitude jumping from 0.1m to 0.3m. A 128Hz altitude sequence is generated through linear interpolation, matching the airborne timeline. Video characteristics: 45s corresponds to frame 1350, the landing gear indicator light changes from "green (down)" to "red (retracted)", and the deviation from the GNSS altitude jump time is ≤33ms.

[0041] 3.2 Similarity Calculation and Drift Compensation Similarity calculation: based on Equation 3: The video features are converted into a 128Hz sequence (features are repeated 4 times per frame); the calculation yields... Dynamic time warping distance The final overall similarity S≈0.66 (≥0.6 indicates that the event descriptions are consistent). Timescale drift compensation: 1280 airborne data sampling points (≈10s) are used as a sliding window, and k and b are updated every 0.5s (64 sampling points); within the t=40-50s window, k is finely adjusted to 0.0011, b is reduced to 0.7ms, and the event timestamp deviation is ≤18ms.

[0042] Step 4: Scene Adaptive Evaluation 4.1 Flight Phase Identification The DBSCAN clustering algorithm was used to divide the flight phases, with "mean altitude, speed fluctuation rate, and attitude angle change rate" as feature vectors. The clustering parameters were set as neighborhood radius ε=0.48 and minimum number of samples. To balance the sample size of multi-source data, feature samples were generated every 1 second by taking the average of 128 airborne points and the average of 20 GNSS points. The final partitioning result is as follows: Table 3 Flight Phases

[0043] 4.2 Weighting and Overall Scoring Based on the assessment requirement of "parameter consistency first" during the takeoff phase, the following quality sub-item weights are configured: completeness ,accuracy ,consistency Timeliness Combined formula 7: Calculate the overall score: Integrity The 6s GNSS data was missing due to obstruction during takeoff (corresponding to 768 airborne points). After interpolation, the missing data was restored. ; accuracy :az The mean absolute error is 0.9 m / s². ; consistency After removing fault points, there are 8 × 128 sets of contradictory data. ; Timeliness Average data processing latency is 32ms (maximum allowed latency is 500ms). ; Overall rating: (≥0.9 indicates that the data quality is acceptable).

[0044] With an overall score of 0.93, the completeness, accuracy, consistency, and timeliness all meet the test indicators. The accurate location of the gyroscope fault verifies the engineering practical value of the evaluation system.

[0045] Example 3 This invention discloses a data quality assessment method for the cruise phase.

[0046] Experimental background: The UAV enters level flight (altitude 1000m, speed 80m / s) and collects engine parameters, barometric altimeter data, and GNSS data for 300 seconds.

[0047] Key verification points: Dynamic constraints: Engine thrust F and velocity v satisfy... ; Spatiotemporal alignment: The timestamp error between GNSS and barometric altimeter must be ≤20ms; Scene weighting: The cruise phase focuses on accuracy assessment (Q2 weight 0.5).

[0048] Step 1: Data Preprocessing and Benchmark Calibration (1) Original data parameters: engine parameter sampling rate 100Hz (thrust measurement range 0-500N), barometric altimeter measurement accuracy ±0.5m, GNSS positioning accuracy 9cm, sampling rate 5Hz, all three types of data are converted to a unified format of "UTC milliseconds-parameter type-parameter value" for storage.

[0049] (2) Preliminary data cleaning: Engine parameters (thrust reasonable range 50-150N) are screened based on the sensor range ±3σ criterion. Video data is detected by inter-frame brightness change rate (no abnormal frames). The barometric altimeter timestamp is calibrated based on GNSS time, with a calibration error ≤18ms (meeting the requirement of ≤20ms for key verification points).

[0050] (3) Spatiotemporal reference alignment: The piecewise linear correction model (Equation 1) was used to correct the barometric altimeter clock drift. The start time of the joint test was set to t0 = 10s after the start of the test. After calibration, k = 0.002 and b = 1.2ms, the final timestamp alignment error was stabilized at ±15ms.

[0051] Step 2: Physics-spacetime joint verification (1) Correlation matrix construction: The mutual information entropy MI between engine thrust F and GNSS velocity v is calculated to be 0.85, which is determined to be a strong correlation link (consistent with F=0.5ρv²SC). D The dynamic constraints are theoretically correlated with a coefficient of ≈0.9.

[0052] (2) Anomaly detection: At a certain moment, the measured thrust F = 120 N and the GNSS velocity v = 80.2 m / s. Substitute these values ​​into the dynamic formula to calculate the theoretical thrust F. 理 =0.5×1.225×80.2²×S×C D (Where S is the windward area of ​​the drone, C) D The drag coefficient, S×C, is pre-calibrated. D =0.02), so F 理 ≈118.3N. The error between the measured value and the theoretical value is ≈1.4% (≤±20% allowable error), and the parameters are judged to be normal; at another moment, the barometric altimeter showed an altitude of 998.2m and a GNSS altitude of 1000.1m, with an error of 1.9m (≤ barometric altimeter accuracy threshold), and there is no contradictory data.

[0053] Step 3: Multi-source spatiotemporal alignment (1) Core event feature extraction: The key event during the cruise phase is "stable flight maintenance". The feature dimensions include GNSS speed fluctuation rate (≤0.5m / s), barometric altimeter reading fluctuation (≤1m), and video engine instrument speed stability (±50rpm). Example feature values: speed fluctuation rate = 0.3m / s, altitude fluctuation = 0.8m.

[0054] (2) Similarity calculation: Based on Equation 3, the angle θ between the video feature vector (stable instrument rotation speed) and the GNSS feature vector (stable speed) is calculated. 视觉 =15° (cosθ≈0.966), Dynamic Time Warped Distance (DTW) (S v ,S s The overall similarity was calculated to be S = 0.6 × 0.966 + 0.4 × 0.12 ≈ 0.63, indicating that the multi-source data were consistent in their description of the "stable flight" event.

[0055] (3) Time stamp drift compensation: The 10s sliding window least squares method (Equation 4) is adopted. The clock drift parameters k and b are updated every 0.5s. After the whole time stamp drift compensation, the time synchronization deviation between the barometric altimeter and the GNSS data does not exceed 15ms.

[0056] Step 4: Scene Adaptive Evaluation (1) Flight phase identification: The DBSCAN algorithm is used to cluster the multi-source features (average altitude 1000m, speed fluctuation rate 0.3-0.5m / s, attitude angle change rate ≤0.1rad / s), with a neighborhood radius ε=1.2×std(feature vector)=0.45 and a minimum number of samples MinPts=50. The entire 300s is determined to be the "cruise phase".

[0057] (2) Weight optimization: Based on the reinforcement learning policy gradient algorithm, according to the reward function R=0.8Q 核心 +0.2Q 非核心 (Equation 6) Set the weight of accuracy Q2 to 0.5, the weight of integrity Q1 to 0.2, the weight of consistency Q3 to 0.2, and the weight of timeliness Q4 to 0.1 (the cruise phase focuses on data accuracy).

[0058] (3) Overall score: Data integrity Q1=0.98 (only 2s GNSS data was missing due to signal obstruction, which was supplemented by interpolation), accuracy Q2=0.96 (average absolute error ≤0.8m / 0.5N), consistency Q3=0.97 (number of logarithmic data without contradiction / total number of logarithmic data = 0.03), timeliness Q4=0.99 (average processing delay = 50ms, maximum allowable delay = 500ms), overall score Q=0.2×0.98+0.5×0.96+0.2×0.97+0.1×0.99=0.96 (excellent).

[0059] The cruise phase data, after being evaluated throughout the entire process, met the flight test data quality requirements, demonstrating outstanding accuracy and consistency. This verifies the adaptability of the scenario adaptive evaluation system of this invention to the stable flight phase and also reflects the value of physical and dynamic constraints in verifying the rationality of parameters.

[0060] Thus, the objective of this invention has been achieved.

[0061] The above are merely preferred embodiments of the present invention and are not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for evaluating the quality of multi-source flight test data, characterized in that, Includes the following steps: Data preprocessing and benchmark calibration steps: Collect and unify the formats of airborne test data, GNSS differential data and video data; perform preliminary data cleaning, and use GNSS time as a benchmark to calibrate the timestamps of airborne equipment using a piecewise linear correction model to achieve spatiotemporal benchmark alignment of multi-source data; The physical-temporal joint verification steps are as follows: calculate the mutual information entropy between multi-source parameters, construct a parameter correlation matrix to identify strongly correlated links; verify the consistency of parameters on strongly correlated links based on physical dynamic formulas, and conduct root cause analysis on the discovered contradictions in conjunction with video data; Multi-source spatiotemporal alignment steps: Define the core events in the flight test and extract their features in multi-source data; calculate the similarity of the descriptions of the same event in multi-source data, and use the sliding window least squares method to compensate for timescale drift in real time to achieve accurate association of key events; Scenario-adaptive evaluation steps: Based on multi-source features, clustering algorithms are used to automatically identify flight phases; the weights of data quality evaluation indicators are dynamically adjusted according to the characteristics of different flight phases. Comprehensive quality scoring and feedback optimization steps: Based on dynamic weights, the completeness, accuracy, consistency and timeliness of the data are weighted and calculated to obtain a comprehensive quality score; based on the score results, a feedback mechanism is triggered to supplement data collection, parameter calibration or algorithm optimization.

2. The method for evaluating the quality of multi-source flight test data according to claim 1, characterized in that, In the data preprocessing and benchmark calibration steps, the piecewise linear correction model is: ； in, For the calibrated timestamp, This is the original timestamp of the airborne equipment. The start time of the joint test. The clock drift coefficient is calibrated through a synchronization test. It is a fixed offset.

3. The method for evaluating the quality of multi-source flight test data according to claim 1, characterized in that, In the physical-spatiotemporal joint verification step, a parameter correlation matrix is ​​constructed: ； in, It is a multi-source parameter correlation matrix. It is an n-order real matrix. To normalize mutual information entropy, The mutual information entropy among multiple source parameters; When two parameters and Normalized mutual information entropy When the link is determined to be strongly correlated, the verification of strongly correlated links allows for an error of ±20% between the calculation results based on physical dynamics formulas and the measured values.

4. The method for evaluating the quality of multi-source flight test data according to claim 1, characterized in that, In the multi-source spatiotemporal alignment step, the similarity calculation model is as follows: ； in, The angle between the video feature vector and the GNSS feature vector. α represents the dynamic time-warped distance between the video data sequence and the sensor data sequence, and β represents the preset weighting coefficients.

5. The method for evaluating the quality of multi-source flight test data according to claim 1, characterized in that, In the scene adaptive evaluation step, a density clustering algorithm is used for flight phase identification, and its neighborhood radius... Set to 1.2 times the standard deviation of the eigenvectors, i.e. Minimum number of samples Set it to 50.

6. The method for evaluating the quality of multi-source flight test data according to claim 1, characterized in that, In the scene adaptive evaluation step, a reinforcement learning strategy gradient algorithm is used to dynamically adjust the evaluation weights, and its state space is defined as follows: ,in For flight phase identification, to Scoring is done for each quality component; the reward function is designed as follows: ,in and These are the quality scores for core parameters and non-core parameters, respectively.

7. The method for evaluating the quality of multi-source flight test data according to claim 1, characterized in that, In the comprehensive quality scoring and feedback optimization step, the comprehensive quality score... The calculation formula is: ； in, To score for completeness, For accuracy scoring, For consistency scoring, Rate the timeliness; to The weights are dynamically adjusted according to the flight phase.

8. A multi-source flight test data quality assessment system for implementing the method of any one of claims 1 to 7, characterized in that, include: The data preprocessing and calibration module is used to receive multi-source heterogeneous data and perform format unification, preliminary cleaning and spatiotemporal reference alignment. The physics-spatiotemporal joint verification module is used to construct the parameter correlation matrix and perform multi-level anomaly detection and verification based on physical dynamic constraints and video data; The multi-source spatiotemporal alignment module is used to define core events, calculate the similarity of multi-source data, and compensate for timescale drift in real time. The scenario-adaptive evaluation module is used to identify flight phases and dynamically configure the weights of data quality evaluation indicators. The comprehensive scoring and feedback module is used to calculate the comprehensive quality score and generate feedback optimization instructions based on the scoring results.

9. The multi-source flight test data quality assessment system according to claim 8, characterized in that, The data preprocessing and calibration module integrates a piecewise linear correction unit for performing timestamp calibration calculations; the physical-temporal joint verification module integrates an correlation matrix calculation unit and a three-level verification logic unit.

10. A multi-source flight test data quality assessment system according to claim 8, characterized in that, The scene adaptive evaluation module integrates a DBSCAN clustering algorithm unit and a reinforcement learning-based weight optimization unit; the comprehensive scoring and feedback module integrates a weighted scoring calculation unit and a feedback triggering mechanism unit.

Images