A QAR data preprocessing method and system based on simulator backdrive

Abstract

The present invention belongs to the field of data processing technology and relates to a QAR data preprocessing method and system based on simulator backdrive. The method aims to address the problems of poor data quality, insufficient parameter coverage, and weak dynamic adaptability encountered in traditional methods. The method comprises: collecting raw QAR data, calculating initial credibility, and screening key parameters; utilizing key parameters to backdrive a full-motion simulator, combining deep reinforcement learning to optimize a PID controller and Monte Carlo simulation to obtain response data; constructing numerical, temporal, and physical three-dimensional difference features based on the raw QAR data and response data, and determining rationality through Gaussian mixture modeling and Bayesian decision making; fusing data based on rational functions and information entropy dynamic weights, optimizing iterative parameters with multiple objectives, and outputting verified high-precision QAR data that meets aviation standards. The present invention obtains full calibration data through a QAR key parameter backdrive simulator, enabling multi-dimensional and precise identification of data anomalies.

Description

Technical Field

[0001] The present invention belongs to the technical field of data processing, and in particular relates to a QAR data preprocessing method and system based on simulator back-drive. Background Art

[0002] As a core device for aviation safety monitoring, flight data recorders (QARs) collect hundreds of critical data items in real time, including aircraft flight status, engine parameters, and control system instructions. These data provide crucial insights for flight performance evaluation, fault diagnosis, maintenance decision-making, and accident analysis. As the aviation industry continues to increase its demands for operational safety and economic efficiency, the accuracy, integrity, and physical consistency of QAR data have become fundamental prerequisites for subsequent data analysis. However, existing QAR data preprocessing technologies still suffer from the following significant drawbacks:

[0003] 1) Data quality issues are prominent, and correction accuracy is limited. The QAR system relies on various sensors distributed throughout the aircraft. Due to mechanical wear and environmental disturbances, the collected data often experiences jumps, gaps, and drift. Traditional filtering algorithms can only handle simple interference such as Gaussian white noise and cannot meet the requirements of high-precision analysis. Furthermore, differences in sampling frequency and transmission delays between different sensors lead to misalignment of time series data.

[0004] 2) The QAR system's sensor configuration is limited by airframe space and cost, making it impossible to directly measure many key parameters (such as wing aerodynamic loads, fuselage structural stresses, and internal engine temperature fields). Traditional methods use empirical formulas to estimate these parameters, but the discrepancy between simplified models and actual flight conditions makes it difficult to perform applications that rely on a complete parameter set, such as structural fatigue analysis and powertrain performance evaluation, and to cover all phases of flight.

[0005] 3) Existing preprocessing systems often use fixed thresholds to identify abnormal data, which is prone to errors. They also require a large amount of labeled data to train the model, and have poor generalization capabilities for new aircraft models or special flight scenarios (such as extreme weather).

[0006] Therefore, there is an urgent need for a QAR data preprocessing method that can break through sensor limitations and dynamically adapt to flight conditions to meet the aviation industry's urgent demand for high-quality flight data. Summary of the Invention

[0007] In order to solve the above-mentioned problems in the prior art, namely, the problems of poor data quality, insufficient parameter coverage, and weak dynamic adaptability in traditional QAR data preprocessing, the first aspect of the present invention proposes a QAR data preprocessing method based on simulator backdrive, the method comprising the following steps:

[0008] Collect raw QAR data and perform preliminary preprocessing;

[0009] Screening key parameter data from the pre-processed raw QAR data, and using the key parameter data to reversely drive the full-motion simulator to obtain full-scale simulation response data;

[0010] Performing a three-dimensional difference analysis on the full-scale simulation response data and the original QAR data to generate a three-dimensional difference analysis result; the three-dimensional difference analysis includes a numerical difference analysis, a time series difference analysis, and a physical relationship difference analysis;

[0011] Performing Gaussian mixture modeling based on the three-dimensional difference analysis results to generate dynamic margin thresholds and data category probabilities;

[0012] According to the data category probability and the dynamic margin threshold, the full amount of simulated response data and the original QAR data are weightedly fused to correct errors in the original data and fill in missing parameters to obtain QAR fitting time series data;

[0013] Perform physical constraint verification and digital twin verification on QAR fitting time series data, and output high-precision QAR data that meets aviation standards.

[0014] In some preferred embodiments, preliminary pretreatment is performed, the method of which is:

[0015] Collect raw QAR data synchronously through dual channels, record GPS timestamps and sensor status codes;

[0016] Standardize the format and unify the units of the original QAR data;

[0017] Calculate the initial credibility of the data based on the sensor failure rate and calibration status;

[0018] Suspicious erroneous data is screened based on the initial credibility and statistical characteristics of the data to generate a tag list.

[0019] In some preferred embodiments, the key parameter data is used to reverse drive a full-motion simulator to obtain full-scale simulation response data, and the method is as follows:

[0020] Screening a subset of key parameters from the pre-processed raw QAR data. The screening is based on a hierarchical analysis of parameter completeness, accuracy, temporal consistency, and initial credibility, determining the weight of each evaluation indicator, calculating the quality factor of each parameter according to a preset parameter quality evaluation model, and screening parameters whose quality factors exceed a threshold as the key parameter subset;

[0021] Inputting the subset of key parameters into a high-precision flight simulator, which is built based on aerodynamics and flight mechanics principles and contains a full physical model of the aircraft;

[0022] Deep reinforcement learning is used to optimize PID controller parameters, and Monte Carlo simulation is used to generate full-scale simulated response data with uncertainty assessment.

[0023] In some preferred embodiments, a three-dimensional difference analysis is performed on the full simulated response data and the original QAR data to generate a three-dimensional difference analysis result, and the method is as follows:

[0024] Calculate the numerical difference between the original QAR data and the full simulated response data, including absolute error and relative error;

[0025] The dynamic time warping algorithm is used to calculate the timing difference between the original QAR data and the full simulated response data;

[0026] Calculate the physical relationship difference between the original QAR data and the full simulation response data based on the flight dynamics model;

[0027] The numerical difference, time sequence difference and physical relationship difference are used to form a three-dimensional difference feature vector, and a difference feature matrix is ​​generated as a three-dimensional difference analysis result.

[0028] In some preferred embodiments, Gaussian mixture modeling is performed based on the three-dimensional difference analysis results to generate dynamic margin thresholds and data category probabilities, and the method is as follows:

[0029] The difference feature matrix is ​​clustered using the expectation maximization algorithm to establish a Gaussian mixture model, and the difference feature space is divided into K Gaussian components, each Gaussian component corresponds to a data state category; the data state categories include normal, abnormal, and suspicious;

[0030] The optimal number of clusters is determined based on the Bayesian information criterion, and a dynamic margin threshold is generated according to the mean and standard deviation of each Gaussian component;

[0031] Calculate the posterior probability that each data point belongs to the normal class as the data class probability.

[0032] In some preferred embodiments, the QAR fitting time series data is obtained by:

[0033] Construct a rational function model, calculate the credibility weights of the original QAR data and the full amount of simulated response data based on information entropy, and dynamically adjust the credibility weights according to the data category probability to form an adaptive fusion weight;

[0034] Based on the adaptive fusion weights, outlier data in the original QAR data is weightedly corrected or filled with null values, and an anti-overflow protection mechanism is added;

[0035] The adaptive fusion weight is used to perform weighted fusion on the original QAR data and the simulator back-drive data, and the fusion parameters are optimized by gradient descent until the termination condition is met to obtain the QAR fitting time series data.

[0036] In some preferred embodiments, the QAR fitting time series data is:

[0037] ;

[0038] in, Fitting time series data to QAR, is the adaptive fusion weight of the simulator data, is the characteristic response function calculated based on the simulator backdrive data, is the credibility weight of QAR data, ; is the adaptive fusion weight, ; is the data category probability, is the information entropy of the original QAR data, The information entropy of the simulator backdrive data, Weight adjustment coefficient.

[0039] In some preferred embodiments, the credibility weight is dynamically adjusted according to the data category probability, and the method is as follows:

[0040] When the data category probability is greater than the first threshold, the credibility weight of the original QAR data is increased;

[0041] When the data category probability is less than the second threshold, the credibility weight of the original QAR data is reduced;

[0042] When the data category probability is between the first threshold and the second threshold, the final fusion weight is adjusted according to the product of the data category probability and the credibility weight.

[0043] In some preferred embodiments, an anti-overflow protection mechanism is added, and the method is as follows:

[0044] Monitor the calculation results of the denominator of the rational function;

[0045] When the absolute value of the denominator is less than the preset minimum value, the denominator is forced to be the preset minimum value;

[0046] When the absolute value of the denominator is greater than a preset minimum value, a protection term proportional to the absolute value of the denominator is added to the denominator, and the coefficient of the protection term is determined through experimental optimization.

[0047] The second aspect of the present invention provides a QAR data preprocessing system based on simulator back-drive, the system comprising:

[0048] A data acquisition module configured to collect raw QAR data and perform preprocessing;

[0049] A parameter screening module is configured to screen key parameter data from the pre-processed raw QAR data and input the data into the simulator back-drive module;

[0050] A simulator back-drive module is configured to back-drive the full-motion simulator using the key parameter data to obtain full-scale simulation response data;

[0051] a data analysis module configured to perform a three-dimensional difference analysis on the full-scale simulated response data and the original QAR data to generate a three-dimensional difference analysis result; the three-dimensional difference analysis includes a numerical difference analysis, a time series difference analysis, and a physical relationship difference analysis; and perform Gaussian mixture modeling based on the three-dimensional difference analysis result to generate a dynamic margin threshold and a data category probability;

[0052] The data fusion module is configured to calculate dynamic weights based on information entropy, dynamically adjust weights based on the category probability of the Gaussian mixture model, complete missing data and add anti-overflow protection, optimize fusion parameters through gradient descent, and output fused time series data and optimized model parameters;

[0053] The verification module is configured to perform physical constraint verification and digital twin verification on the QAR fitting time series data, and output high-precision QAR data that meets aviation standards.

[0054] Beneficial effects of the present invention:

[0055] 1. The present invention uses the collected QAR data as input, screens reliable key parameters, drives the full-motion simulator, obtains the simulator response data through adaptive PID control, and then outputs accurate and reliable QAR data through multi-dimensional comparison and intelligent fusion optimization, realizing full automation of data preprocessing, reducing labor costs and processing cycles;

[0056] 2. Establish a simulator backdrive calibration mechanism. This mechanism uses the QAR key parameter backdrive simulator to obtain full calibration data. Simulator backdrive technology is used to generate parameters not directly collected by the original QAR (such as wing loading and fuselage bending moment). This fills the gaps in analysis dimensions left by traditional methods due to sensor limitations, establishes a data comparison benchmark, and overcomes the limitations of traditional reliance on a single sensor.

[0057] 3. During the calibration process, a three-dimensional dynamic calibration system is established to compare the original and simulated data from the numerical, time series, and physical dimensions. Accurate calibration is achieved by combining dynamic margin judgment, and multi-dimensional and accurate identification of data anomalies is achieved, which is more accurate than single-dimensional calibration.

[0058] 4. Intelligent data fusion is achieved based on rational functions and dynamic weights of information entropy. Multi-objective optimization takes into account both fitting accuracy and safety margin. The fused data conforms to historical flight patterns and meets flight safety constraints, providing dual protection for aviation safety monitoring. By introducing flight mechanics equations and digital twins to verify physical consistency, blockchain evidence storage ensures data traceability, forming a complete and trusted closed loop. BRIEF DESCRIPTION OF THE DRAWINGS

[0059] Other features, objects and advantages of the present application will become more apparent upon reading the detailed description of non-limiting embodiments made with reference to the following drawings:

[0060] Figure 1 This is a flow chart of a QAR data preprocessing method based on simulator back-drive in an embodiment of the present invention. DETAILED DESCRIPTION

[0061] The present application will be further described in detail below with reference to the accompanying drawings and examples. It should be understood that the specific embodiments described herein are intended only to illustrate the relevant invention and are not intended to limit the invention. It should also be noted that, for ease of description, only portions relevant to the relevant invention are shown in the accompanying drawings.

[0062] It should be noted that, in the absence of conflict, the embodiments and features of the embodiments in this application can be combined with each other. The present application will be described in detail below with reference to the accompanying drawings and in combination with the embodiments.

[0063] The present invention provides a QAR data preprocessing method based on simulator backdrive. By constructing a closed-loop technical system of "key parameter screening-simulator backdrive-multi-dimensional comparison-intelligent fusion optimization", this method effectively solves the problems of poor data quality, insufficient parameter coverage, and weak dynamic adaptability existing in traditional QAR data preprocessing.

[0064] A QAR data preprocessing method based on simulator back-drive of the present invention comprises the following steps:

[0065] S1, collect original QAR data and perform preliminary preprocessing;

[0066] S2. Filtering key parameter data from the pre-processed raw QAR data, and using the key parameter data to reversely drive the full-motion simulator to obtain full-scale simulation response data;

[0067] S3. Perform a three-dimensional difference analysis on the full-scale simulation response data and the original QAR data to generate a three-dimensional difference analysis result; the three-dimensional difference analysis includes a numerical difference analysis, a time series difference analysis, and a physical relationship difference analysis;

[0068] S4. Performing Gaussian mixture modeling based on the three-dimensional difference analysis results to generate a dynamic margin threshold and data category probability;

[0069] S5. Based on the data category probability and the dynamic margin threshold, perform weighted fusion on the full simulated response data and the original QAR data, correct errors in the original data and fill in missing parameters to obtain QAR fitting time series data;

[0070] S6. Perform physical constraint verification and digital twin verification on the QAR fitting time series data, and output high-precision QAR data that meets aviation standards.

[0071] In order to more clearly illustrate the QAR data preprocessing method based on simulator back-drive of the present invention, the following Figure 1 Each step in the embodiment of the present invention is described in detail.

[0072] The QAR data preprocessing method based on simulator back-drive according to the first embodiment of the present invention includes steps S1 to S6, each of which is described in detail as follows:

[0073] S1. Collect the original QAR data and perform preliminary preprocessing. In this embodiment, the preferred preliminary preprocessing method is:

[0074] Collect raw QAR data synchronously through dual channels, record GPS timestamps and sensor status codes;

[0075] Standardize the format and unify the units of the original QAR data;

[0076] Calculate the initial credibility of the data based on the sensor failure rate and calibration status;

[0077] Suspicious erroneous data is screened based on the initial credibility and statistical characteristics of the data to generate a tag list.

[0078] Preferably, the original QAR data is collected synchronously through dual channels using a standardized protocol (ARINC429 / 664), and the sensor status code and GPS timestamp are recorded synchronously to form a time-stamped data sequence. .in is the parameter value, is the timestamp, Is the status code.

[0079] Preferably, the format of the original QAR data is standardized and the units are unified as follows:

[0080] Perform format parsing on the collected data, converting it into a unified structured format, completing unit standardization conversion (feet → meters, knots → meters / second), and calibrating the device clock to the GPS time reference, with timestamp accuracy ≤1ms;

[0081] Mark missing data as NaN, invalid value as -9999, calculate initial credibility based on sensor failure rate and calibration cycle :

[0082] ;

[0083] in, sensor i The failure rate per unit time and reliability It is a negative exponential relationship, with a coefficient of 0.2, which represents the engineering experience that for every 1% increase in failure rate, the reliability decreases by about 0.2%; is the time interval since the last calibration, is the standard calibration cycle of the sensor, and the ratio of the two is the calibration time interval ratio, which reflects the degree to which the equipment deviates from the calibration state. The coefficient 0.1 reflects the attenuation weight of the calibration cycle on the credibility.

[0084] By constructing a standardized data sequence with time stamps, the foundation is laid for subsequent time series analysis. The credibility quantification model is used to achieve preliminary screening of data quality and reduce invalid data interference. The dual-channel acquisition and GPS clock calibration mechanism ensures the consistency of multi-parameter time series.

[0085] S2. Filter key parameter data from the pre-processed original QAR data, and use the key parameter data to reversely drive the full-motion simulator to obtain full-scale simulation response data.

[0086] Preferably, the key parameter data is used to reversely drive a full-motion simulator to obtain full-scale simulation response data, and the method is as follows:

[0087] S21. Filtering a subset of key parameters from the pre-processed original QAR data. The screening is based on a hierarchical analysis of parameter completeness, accuracy, temporal consistency, and initial credibility, determining the weight of each evaluation indicator, calculating the quality factor of each parameter according to a preset parameter quality evaluation model, and filtering parameters whose quality factors exceed a threshold as the key parameter subset.

[0088] S22. Input the subset of key parameters into a high-precision flight simulator, which is built based on the principles of aerodynamics and flight mechanics and includes a full physical model of the aircraft; use deep reinforcement learning to optimize PID controller parameters, and generate full simulation response data with uncertainty assessment through Monte Carlo simulation.

[0089] Further preferably, the quality factor of each parameter is calculated according to a preset parameter quality assessment model, and parameters whose quality factors exceed a threshold are screened as a key parameter subset, the method of which is:

[0090] S211. Establish a parameter evaluation hierarchy: target layer (parameter importance), criterion layer (completeness Comp ,accuracy Acc , consistency, credibility Trust ), solution layer (various QAR parameters);

[0091] S212. Combine the quality factor model to screen reliable QAR key parameter data:

[0092] ;

[0093] in, is the parameter completeness index (0-1), which is the inverse function of the proportion of missing values ​​(the fewer missing values, the higher the value); is a parameter accuracy indicator, calculated based on the credibility in step S1 and historical calibration data; is an indicator of parameter temporal consistency, measured by the coefficient of difference between adjacent values ​​(the smaller the fluctuation, the higher the value); is the initial reliability of the sensor, from step S1;

[0094] Weight coefficient ; The particle swarm optimization (PSO) algorithm is iteratively determined, and the optimization goal is:

[0095] ;

[0096] in, It is a quality benchmark for parameters annotated based on expert experience;

[0097] S213. Construct a judgment matrix A and solve the weight vector using the eigenvalue method:

[0098] Calculate the eigenvector corresponding to the maximum eigenvalue and normalize it to get the weight matrix ;

[0099] S214, according to the obtained weight vector and Parameters determine a subset of key parameters .

[0100] Further preferably, during the data screening process, spatial anomaly detection and temporal anomaly detection are performed to correct obvious abnormal points / erroneous data in the data.

[0101] Further preferably, the subset of key parameters is input into a high-precision flight simulator, deep reinforcement learning is used to optimize PID controller parameters, and a full amount of simulated response data with uncertainty assessment is generated through Monte Carlo simulation, the method of which is:

[0102] A1. Mark the QAR key parameter data selected by the flight phase After adding the flight phase mark, it is used as the simulator input to drive the simulator to run through the following mapping relationship: ;

[0103] in, Input parameter vector for the simulator, is a parameter mapping function established based on the aircraft manual, including linear transformation and physical unit conversion; in this embodiment, the flight phase is preferably marked as , where 1, 2, and 3 correspond to takeoff / cruise / landing respectively);

[0104] A2. Define the state space, which includes six characteristics: deviation, deviation change rate, and flight phase.

[0105] ;

[0106] in, is the deviation between the measured value and the simulated value of the parameter at time t , is the deviation change rate between the measured value and the simulated value of the parameter at time t, is the height at time t, is the roll angle at time t, is the Mach number at time t, is the flight phase code at time t (takeoff / cruise / landing, etc.), where t is the time control quantity;

[0107] A3. Optimize the simulator PID controller parameters through deep reinforcement learning to achieve high-precision response and define the state transfer function and the reward function for:

[0108] ;

[0109] ;

[0110] in, To control the amount, is the mean square error between the simulated value and the QAR value, which measures the deviation between the simulated value and the QAR value; is the penalty term for control quantity change, which suppresses control jitter. is the uncertainty reward term, which simulates the negative logarithm of the output variance and encourages low variance output;

[0111] A4. Simulate data acquisition to iteratively update parameters, drive the simulator to run, perform a preset number of Monte Carlo simulations, collect simulator response data, and calculate output uncertainty; in this embodiment, 50 times is preferred:

[0112] ;

[0113] ;

[0114] Where, P is the nominal output of the simulator, σ =0.05 is the disturbance coefficient, U(i) is the parameter i The standard deviation of the output (characterizing uncertainty);

[0115] A5. Collect response data and obtain simulator reverse drive data .

[0116] Six-dimensional state space modeling comprehensively reflects the coupling relationship between flight parameters. The control smoothing term in the reward function makes the simulator output more consistent with physical laws and reduces abnormal fluctuations. The uncertainty index generated by Monte Carlo simulation provides a probabilistic judgment basis for subsequent data comparison.

[0117] S3. Perform a three-dimensional difference analysis on the full-scale simulation response data and the original QAR data to generate a three-dimensional difference analysis result; the three-dimensional difference analysis includes numerical difference analysis, time series difference analysis, and physical relationship difference analysis.

[0118] Preferably, a three-dimensional difference analysis is performed on the full amount of simulated response data and the original QAR data to generate a three-dimensional difference analysis result, and the method is:

[0119] Calculate the numerical difference between the original QAR data and the full simulated response data, including the absolute error and relative error :

[0120] ;

[0121] ;

[0122] in, The first i The mean of the parameters, The first i The measured value of a parameter (such as the measured value of airspeed at a certain moment);

[0123] Calculate the root mean square error (RMSE) and mean absolute error (MAE) at the same time, ,when When it is 0, the denominator is not 0 to ensure the numerical stability of the relative error calculation;

[0124] The dynamic time warping algorithm is used to calculate the timing difference between the original QAR data and the full amount of simulated response data, and the DTW distance with window constraint is used. , that is, i Timing difference values ​​of parameters:

[0125] ;

[0126] in, is the parameter value at time t in the original QAR data; the window constraint is , W is the window width, is a time alignment mapping function used to establish the time point correspondence between the original QAR data and the full-scale simulated response data; T is the time series length (number of sampling points) of the original QAR data, For the full amount of simulated response data, after mapping function After alignment, the parameter value corresponding to the original data at time t;

[0127] Calculate the physical relationship difference between the original QAR data and the full simulation response data based on the flight dynamics model , that is, i The physical difference value of the parameter;

[0128] ;

[0129] in, is the first i The actual overload value at the moment, For the i Engine thrust at the moment, For the i The angle of attack corresponding to the moment is f It is a mechanical model based on Newton's second law, used to calculate the theoretical N1 speed; For the i Flight altitude data corresponding to each moment;

[0130] The numerical difference, time sequence difference and physical relationship difference constitute a three-dimensional difference feature vector , generate the difference feature matrix as the result of three-dimensional difference analysis.

[0131] S4. Perform Gaussian mixture modeling based on the three-dimensional difference analysis results to generate a dynamic margin threshold and data category probability.

[0132] In some preferred embodiments, Gaussian mixture modeling is performed based on the three-dimensional difference analysis results to generate dynamic margin thresholds and data category probabilities, and the method is as follows:

[0133] The expectation maximization algorithm is used to cluster the difference feature matrix and establish a Gaussian mixture model:

[0134] ;

[0135] in, For samples The joint probability density of , is the Gaussian mixture model weight, ; is the mean and covariance matrix of the kth Gaussian component parameters, K is the number of Gaussian components, For is the mean, is the multivariate Gaussian distribution of the covariance matrix, which is used to describe the k Probability distribution of Gaussian-like component parameters;

[0136] Perform unsupervised clustering on the three-dimensional difference feature vectors, dividing the difference feature space into K Gaussian components, each of which corresponds to a data status category; the data status categories include normal, abnormal, and suspicious;

[0137] The optimal number of clusters K is determined based on the Bayesian information criterion, and a dynamic margin threshold is generated according to the mean and standard deviation of each Gaussian component;

[0138] ;

[0139] Where N is the number of data points, MSE is the mean square error of the model fitting;

[0140] Generate dynamic thresholds for each cluster , and calculate the posterior probability of each data point belonging to the normal category based on the Bayesian posterior probability decision (GMM) , as the data category probability:

[0141] ;

[0142] Among them, the Gaussian mixture model parameters include: (mixing coefficient), (mean vector), (covariance matrix), (standard deviation).

[0143] The three-dimensional difference feature system covers the main dimensions of data comparison and effectively solves the timing deviation problem caused by asynchronous sensor sampling. It also introduces physical constraints through the flight mechanics model to avoid the physical irrationality that may occur in pure data-driven methods. Data with excessive physical differences are automatically marked to reduce errors.

[0144] S5. Based on the data category probability and the dynamic margin threshold, the full simulated response data and the original QAR data are weightedly fused to correct errors in the original data and complete missing parameters to obtain QAR fitting time series data.

[0145] Preferably, the QAR fitting time series data is obtained by:

[0146] Construct a rational function model:

[0147] ;

[0148] in, x Indicates input data, which can be reverse drive data for the simulator or QAR raw data , solve the coefficients by the least squares method 、 , n, n are the orders of the numerator and denominator of the rational function;

[0149] Searching for the optimal order combination using the Bayesian Information Criterion :

[0150] ;

[0151] Based on the information entropy H, the credibility weights of the original QAR data and the full simulated response data are calculated. :

[0152] ;

[0153] ;

[0154] in, is the data distribution probability, is the information entropy of the simulated data; is the information entropy of QAR data, is the weight adjustment coefficient;

[0155] Dynamically adjust the credibility weight according to the data category probability to form an adaptive fusion weight ;

[0156] Based on the adaptive fusion weight , perform weighted correction or null value filling on outlier data in the original QAR data, and add an anti-overflow protection mechanism;

[0157] The adaptive fusion weight is used to perform weighted fusion on the original QAR data and the simulator back-drive data, and the fusion parameters are optimized by gradient descent until the termination condition is met to obtain the QAR fitting time series data.

[0158] Real-time updates are achieved through a multi-gradient optimization mechanism, which dynamically adjusts the proportion of original data and simulator data in the fusion process according to data characteristics to improve the accuracy and reliability of the fusion results.

[0159] Preferably, the credibility weight is dynamically adjusted according to the data category probability, and the method is as follows:

[0160] When the data category probability is greater than the first threshold, the credibility weight of the original QAR data is increased;

[0161] When the data category probability is less than the second threshold, the credibility weight of the original QAR data is reduced;

[0162] When the data category probability is between the first threshold and the second threshold, the final fusion weight is adjusted according to the product of the data category probability and the credibility weight.

[0163] Further preferably, in this embodiment, the first threshold is 0.7, the second threshold is 0.3; when P(Normal|x)<0.3, ; When P(Normal|x)>0.7 or when 0.3≤P(Normal|x)≤0.7, .

[0164] In this embodiment, By dynamically adjusting the weights of the parameters set for different flight phases, the effect of entropy differences on the weights is amplified. This application is preferably used for the cruise phase. , landing phase .

[0165] In some preferred embodiments, an anti-overflow protection item is added to the fusion process, and the method is as follows:

[0166] Adding overflow protection:

[0167] ;

[0168] in, To fix the minimum value and prevent the denominator from being zero, β 0.001, dynamic scaling factor, dynamically adjust the protection strength, is a rational function denominator polynomial;

[0169] Monitor the calculation results of the rational function denominator and suppress the denominator overflow through the maximum value of the dynamic scaling factor and the fixed minimum value, specifically:

[0170] When the absolute value of the denominator is less than the preset minimum value, the denominator is forced to be the preset minimum value;

[0171] When the absolute value of the denominator is greater than a preset minimum value, a protection term proportional to the absolute value of the denominator is added to the denominator, and the coefficient of the protection term is determined through experimental optimization.

[0172] Further preferably, the adaptive fusion weight is used to perform weighted fusion on the original QAR data and the simulator back-drive data, and the fusion parameters are optimized by gradient descent until the termination condition is met to obtain the QAR fitting time series data, which is as follows:

[0173] Construct a multi-objective loss function:

[0174] ;

[0175] in, is the error MSE (fusion error) between the fused data and the simulated data, is the error MSE (simulator response error) between the simulated data and the original QAR data, is the safety margin error MSE (dynamic margin error) between the fusion result and the boundary value;

[0176] Define the gradient update:

[0177] ;

[0178] in, , is the learning rate of the tth iteration, , is the initial learning rate, β is the attenuation coefficient, t is the number of iterations; Parameter vector to be optimized (including rational function coefficients, PID controller parameters, etc.) 、 is the weight coefficient of each sub-goal, and ;

[0179] The iteration is terminated when the double threshold condition is met:

[0180] ;

[0181] Convergence occurs when the relative error decreases by more than 5% and the parameter update amount is less than 0.01. Fusion calculations are performed to obtain the QAR fitting time series data:

[0182] ;

[0183] in, Fitting time series data to QAR, is the adaptive fusion weight of the simulator data, is the characteristic response function calculated based on the simulator backdrive data, is the adaptive fusion weight.

[0184] The dynamic weighting of information entropy realizes the adaptive allocation of data quality, and high-credibility data is given higher weight. Through the triple protection of "dynamic allocation of information entropy weight + GMM probabilistic anomaly detection + anti-overflow numerical protection", a complete technical closed loop from anomaly identification to numerical stability is formed, ensuring the preprocessing accuracy under various flight conditions.

[0185] S6. Perform physical constraint verification and digital twin verification on the QAR fitting time series data, and output high-precision QAR data that meets aviation standards. The method is as follows:

[0186] Substitute the fused data into the flight physical constraint equation for verification. If the residual exceeds the threshold, readjust the fusion parameters.

[0187] Input the fused data into the aircraft digital twin model to verify its logical consistency with other relevant parameters;

[0188] Convert the format of the verified data to generate high-precision QAR data that complies with the ARINC717 standard;

[0189] The hash value is calculated for the verified fusion data, the key features of the pre-processed data are stored on the blockchain, and the above-mentioned high-precision QAR data is output as the result.

[0190] Preferably, the hash value is calculated for the verified fusion data by:

[0191] ;

[0192] Among them, F is the standardized fusion data, TimeStamp is the evidence timestamp, and Nonce is a random number, which generates an unalterable evidence record.

[0193] Although the various steps in the above embodiment are described in the above-mentioned order, those skilled in the art will understand that in order to achieve the effect of this embodiment, different steps do not have to be executed in such an order. They can be executed simultaneously (in parallel) or in a reverse order. These simple changes are within the scope of protection of the present invention.

[0194] A QAR data preprocessing system based on simulator back-drive according to a second embodiment of the present invention includes:

[0195] A data acquisition module configured to collect raw QAR data and perform preprocessing;

[0196] A parameter screening module is configured to screen key parameter data from the pre-processed raw QAR data and input the data into the simulator back-drive module;

[0197] A simulator back-drive module is configured to back-drive the full-motion simulator using the key parameter data to obtain full-scale simulation response data;

[0198] a data analysis module configured to perform a three-dimensional difference analysis on the full-scale simulated response data and the original QAR data to generate a three-dimensional difference analysis result; the three-dimensional difference analysis includes a numerical difference analysis, a time series difference analysis, and a physical relationship difference analysis; and perform Gaussian mixture modeling based on the three-dimensional difference analysis result to generate a dynamic margin threshold and a data category probability;

[0199] The data fusion module is configured to calculate dynamic weights based on information entropy, dynamically adjust weights based on the category probability of the Gaussian mixture model, complete missing data and add anti-overflow protection, optimize fusion parameters through gradient descent, and output fused time series data and optimized model parameters;

[0200] The verification module is configured to perform physical constraint verification and digital twin verification on the QAR fitting time series data, and output high-precision QAR data that meets aviation standards.

[0201] It should be noted that the QAR data preprocessing system based on simulator backdrive provided in the above embodiment is only illustrated by the division of the above functional modules. In actual applications, the above functions can be assigned to different functional modules as needed, that is, the modules or steps in the embodiments of the present invention can be further decomposed or combined. For example, the modules in the above embodiment can be combined into one module, or further divided into multiple sub-modules to complete all or part of the functions described above. The names of the modules and steps involved in the embodiments of the present invention are only for distinguishing the modules or steps and are not considered to be improper limitations of the present invention.

[0202] Those skilled in the art will clearly understand that, for the convenience and brevity of description, the specific working process and related instructions of the system described above can refer to the corresponding process in the aforementioned method embodiment and will not be repeated here.

[0203] An electronic device according to a third embodiment of the present invention includes:

[0204] at least one processor; and

[0205] a memory communicatively connected to at least one of the processors; wherein,

[0206] The memory stores instructions that can be executed by the processor, and the instructions are used to be executed by the processor to implement the above-mentioned QAR data preprocessing method based on simulator back-drive.

[0207] A fourth embodiment of the present invention provides a computer-readable storage medium, wherein the computer-readable storage medium stores computer instructions, and the computer instructions are used to be executed by a computer to implement the above-mentioned QAR data preprocessing method based on simulator back-drive.

[0208] Those skilled in the art will clearly understand that, for the convenience and brevity of description, the specific working processes and related instructions of the electronic device and computer-readable storage medium described above can refer to the corresponding processes in the aforementioned method embodiments and will not be repeated here.

[0209] Those skilled in the art should be able to appreciate that the modules and method steps of each example described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, computer software, or a combination of the two, and the programs corresponding to the software modules and method steps can be placed in random access memory (RAM), internal memory, read-only memory (ROM), electrically programmable ROM, electrically erasable programmable ROM, registers, hard disks, removable disks, CD-ROMs, or any other form of storage medium known in the art. In order to clearly illustrate the interchangeability of electronic hardware and software, the above description has generally described the components and steps of each example according to their functions. Whether these functions are performed in electronic hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art may use different methods to implement the described functions for each specific application, but such implementation should not be considered to be beyond the scope of the present invention.

[0210] Computer program code for performing the operations of the present application may be written in one or more programming languages, or a combination thereof, including object-oriented programming languages ​​such as Java, Smalltalk, C++, and conventional procedural programming languages ​​such as "C" or similar programming languages. The program code may be executed entirely on the user's computer, partially on the user's computer, as a stand-alone software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In cases involving a remote computer, the remote computer may be connected to the user's computer through any type of network, including a local area network (LAN) or a wide area network (WAN), or may be connected to an external computer (e.g., through the Internet using an Internet service provider).

[0211] The flow charts and block diagrams in the accompanying drawings illustrate the possible architecture, functions and operations of the systems, methods and computer program products according to various embodiments of the present application. In this regard, each box in the flow chart or block diagram can represent a module, program segment or a part of code, and the module, program segment or a part of code contains one or more executable instructions for realizing the specified logical function. It should also be noted that in some alternative implementations, the functions marked in the box can also occur in a different order than that marked in the accompanying drawings. For example, two boxes represented in succession can actually be executed substantially in parallel, and they can sometimes be executed in the opposite order, depending on the functions involved. It should also be noted that each box in the block diagram and / or flow chart, and the combination of the boxes in the block diagram and / or flow chart can be implemented by a dedicated hardware-based system that performs the specified function or operation, or can be implemented by a combination of dedicated hardware and computer instructions.

[0212] The terms "first", "second", etc. are used to distinguish similar objects, rather than to describe or indicate a particular order or sequence.

[0213] The term "comprise" or any other similar term is intended to cover non-exclusive inclusion such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed or inherent to such process, method, article, or apparatus.

[0214] Thus far, the technical solutions of the present invention have been described in conjunction with the preferred embodiments shown in the accompanying drawings. However, it will be readily understood by those skilled in the art that the scope of protection of the present invention is obviously not limited to these specific embodiments. Without departing from the principles of the present invention, those skilled in the art may make equivalent changes or substitutions to the relevant technical features, and the technical solutions after such changes or substitutions will fall within the scope of protection of the present invention.

Claims

1. A QAR data preprocessing method based on simulator back-drive, characterized in that: The method comprises the following steps: Collect raw QAR data and perform preliminary preprocessing; Screening key parameter data from the pre-processed raw QAR data, and using the key parameter data to reversely drive the full-motion simulator to obtain full-scale simulation response data; Performing a three-dimensional difference analysis on the full-scale simulation response data and the original QAR data to generate a three-dimensional difference analysis result; the three-dimensional difference analysis includes a numerical difference analysis, a time series difference analysis, and a physical relationship difference analysis; Performing Gaussian mixture modeling based on the three-dimensional difference analysis results to generate dynamic margin thresholds and data category probabilities; According to the data category probability and the dynamic margin threshold, the full amount of simulated response data and the original QAR data are weightedly fused to correct errors in the original data and fill in missing parameters to obtain QAR fitting time series data; Perform physical constraint verification and digital twin verification on QAR fitting time series data, and output high-precision QAR data that meets aviation standards.

2. The QAR data preprocessing method based on simulator back-drive according to claim 1, characterized in that: Perform preliminary preprocessing as follows: Collect raw QAR data synchronously through dual channels, record GPS timestamps and sensor status codes; Standardize the format and unify the units of the original QAR data; Calculate the initial credibility of the data based on the sensor failure rate and calibration status; Suspicious erroneous data is screened based on the initial credibility and statistical characteristics of the data to generate a tag list.

3. The QAR data preprocessing method based on simulator back-drive according to claim 1, characterized in that: The key parameter data is used to reverse drive the full-motion simulator to obtain full-scale simulation response data, and the method is as follows: Screening a subset of key parameters from the pre-processed raw QAR data. The screening is based on a hierarchical analysis of parameter completeness, accuracy, temporal consistency, and initial credibility, determining the weight of each evaluation indicator, calculating the quality factor of each parameter according to a preset parameter quality evaluation model, and screening parameters whose quality factors exceed a threshold as the key parameter subset; Inputting the subset of key parameters into a high-precision flight simulator, which is built based on aerodynamics and flight mechanics principles and contains a full physical model of the aircraft; Deep reinforcement learning is used to optimize PID controller parameters, and Monte Carlo simulation is used to generate full-scale simulated response data with uncertainty assessment.

4. The QAR data preprocessing method based on simulator back-drive according to claim 1, characterized in that: Performing a three-dimensional difference analysis on the full simulated response data and the original QAR data to generate a three-dimensional difference analysis result is as follows: Calculate the numerical difference between the original QAR data and the full simulated response data, including absolute error and relative error; The dynamic time warping algorithm is used to calculate the timing difference between the original QAR data and the full simulated response data; Calculate the physical relationship difference between the original QAR data and the full simulation response data based on the flight dynamics model; The numerical difference, time sequence difference and physical relationship difference are used to form a three-dimensional difference feature vector, and a difference feature matrix is ​​generated as a three-dimensional difference analysis result.

5. The QAR data preprocessing method based on simulator back-drive according to claim 4, characterized in that: Gaussian mixture modeling is performed based on the three-dimensional difference analysis results to generate dynamic margin thresholds and data category probabilities, and the method is as follows: The difference feature matrix is ​​clustered using the expectation maximization algorithm to establish a Gaussian mixture model, and the difference feature space is divided into K Gaussian components, each Gaussian component corresponds to a data state category; the data state categories include normal, abnormal, and suspicious; The optimal number of clusters is determined based on the Bayesian information criterion, and a dynamic margin threshold is generated according to the mean and standard deviation of each Gaussian component; Calculate the posterior probability that each data point belongs to the normal class as the data class probability.

6. The QAR data preprocessing method based on simulator back-drive according to claim 5, characterized in that: The QAR fitting time series data is obtained as follows: Construct a rational function model, calculate the credibility weights of the original QAR data and the full amount of simulated response data based on information entropy, and dynamically adjust the credibility weights according to the data category probability to form an adaptive fusion weight; Based on the adaptive fusion weights, outlier data in the original QAR data is weightedly corrected or filled with null values, and an anti-overflow protection mechanism is added; The adaptive fusion weight is used to perform weighted fusion on the original QAR data and the simulator back-drive data, and the fusion parameters are optimized by gradient descent until the termination condition is met to obtain the QAR fitting time series data.

7. The QAR data preprocessing method based on simulator back-drive according to claim 6, characterized in that: The QAR fitting time series data is: ； in, Fitting time series data to QAR, is the adaptive fusion weight of the simulator data, is the characteristic response function calculated based on the simulator backdrive data, is the credibility weight of QAR data, ; is the adaptive fusion weight, ; is the data category probability, is the information entropy of the original QAR data, The information entropy of the simulator backdrive data, Weight adjustment coefficient.

8. The QAR data preprocessing method based on simulator back-drive according to claim 6, characterized in that: The credibility weight is dynamically adjusted according to the data category probability, and the method is as follows: When the data category probability is greater than the first threshold, the credibility weight of the original QAR data is increased; When the data category probability is less than the second threshold, the credibility weight of the original QAR data is reduced; When the data category probability is between the first threshold and the second threshold, the final fusion weight is adjusted according to the product of the data category probability and the credibility weight.

9. The QAR data preprocessing method based on simulator back-drive according to claim 6, characterized in that: Add an anti-overflow protection mechanism, the method is: Monitor the calculation results of the denominator of the rational function; When the absolute value of the denominator is less than the preset minimum value, the denominator is forced to be the preset minimum value; When the absolute value of the denominator is greater than a preset minimum value, a protection term proportional to the absolute value of the denominator is added to the denominator, and the coefficient of the protection term is determined through experimental optimization.

10. A QAR data preprocessing system based on simulator backdrive, characterized in that: The system includes: A data acquisition module configured to collect raw QAR data and perform preprocessing; A parameter screening module is configured to screen key parameter data from the pre-processed raw QAR data and input the data into the simulator back-drive module; A simulator back-drive module is configured to back-drive the full-motion simulator using the key parameter data to obtain full-scale simulation response data; a data analysis module configured to perform a three-dimensional difference analysis on the full-scale simulated response data and the original QAR data to generate a three-dimensional difference analysis result; the three-dimensional difference analysis includes a numerical difference analysis, a time series difference analysis, and a physical relationship difference analysis; and perform Gaussian mixture modeling based on the three-dimensional difference analysis result to generate a dynamic margin threshold and a data category probability; The data fusion module is configured to calculate dynamic weights based on information entropy, dynamically adjust weights based on the category probability of the Gaussian mixture model, complete missing data and add anti-overflow protection, optimize fusion parameters through gradient descent, and output fused time series data and optimized model parameters; The verification module is configured to perform physical constraint verification and digital twin verification on the QAR fitting time series data, and output high-precision QAR data that meets aviation standards.

Images