A method and system for accident reproduction based on QAR data and simulator reverse drive

Abstract

This application belongs to the field of aviation safety technology and relates to an accident reproduction method and system based on QAR data and simulator-based anti-flight maneuvering, aiming to solve the problems of low reproduction accuracy and poor compliance in existing technologies. The method includes: acquiring an accident QAR dataset; locking the accident timeline nodes and filtering a subset of key parameters; using an anti-flight simulator and optimizing the simulator's anti-flight control strategy based on the key parameter subset and timeline nodes to generate preliminary reproduction data; calculating the timeline and physical deviations between the preliminary reproduction data and the accident QAR dataset, and dynamically fusing the two deviations based on the timeline nodes to generate dynamic correction coefficients; calibrating the simulator's anti-flight process using the dynamic correction coefficients to obtain calibrated reproduction data. This application, by constructing a technical system of data preprocessing, anti-flight optimization, and closed-loop dynamic correction, achieves deep coupling between QAR data and the simulator, improving the accuracy, objectivity, and compliance of accident reproduction.

Description

Technical Field

[0001] This application belongs to the field of aviation safety technology and relates to flight data reproduction methods, particularly to an accident reproduction method and system based on QAR data and simulator reverse flight. Background Technology

[0002] Flight accident reconstruction is a crucial part of accident investigation, causal analysis, and safety risk assessment. By accurately reconstructing the accident process, investigators can deeply analyze the causes, development, and key moments of the accident, thereby developing effective preventative measures and improving aviation safety.

[0003] In existing technologies, flight accident reconstruction mainly relies on investigators conducting manual analysis and deduction based on data from the flight data recorder (commonly known as the "black box"), or using general-purpose flight simulators for scenario simulation. However, in practical applications, existing technologies have some technical shortcomings:

[0004] First, the reproduction accuracy is insufficient; general simulation logic struggles to accurately recreate extremely complex accident conditions such as stalls and engine failures, leading to significant deviations between the reproduced results and the actual accident process, such as excessive timing errors and deviations in physical parameters. Second, flight data processing is relatively crude; existing methods fail to establish a strong correlation between accident types and key flight parameters. The massive amount of redundant data not only interferes with the reproduction process but also makes it difficult to accurately pinpoint key time points of the accident. Third, there is a lack of a dynamic closed-loop deviation calibration mechanism; existing reproduction technologies fail to employ differentiated correction strategies for different stages of the accident, such as its initiation, development, and termination, causing the reproduction process to easily deviate from the essence of the accident.

[0005] Therefore, how to achieve a high-precision and highly compliant method for reproducing flight accidents is a technical problem that urgently needs to be solved in this field. Summary of the Invention

[0006] To address the aforementioned problems in the prior art, namely low accuracy and insufficient compliance in accident reproduction, the first aspect of this application proposes an accident reproduction method based on QAR data and simulator reverse drive, the method comprising the following steps:

[0007] Obtain the accident QAR dataset associated with the accident type, the accident QAR dataset containing flight status parameters and event markers;

[0008] Accident-oriented preprocessing is performed on the QAR dataset to lock the accident time sequence nodes, and a subset of key accident parameters is selected based on the accident type and the accident time sequence nodes.

[0009] The subset of key accident parameters and accident timeline nodes are used as inputs to execute a reverse drive control strategy to drive the simulator to recreate the accident process and generate preliminary reproduction data.

[0010] Based on the accident QAR dataset and preliminary reproduction data, the timing deviation and physical deviation are calculated respectively, and the timing deviation and physical deviation are dynamically fused according to the accident timing nodes to generate dynamic correction coefficients;

[0011] The output parameters of the simulator during the reverse drive process are calibrated using the dynamic correction coefficient to obtain calibrated reproducible data.

[0012] In some preferred embodiments, the step of performing incident-oriented preprocessing on the QAR dataset to lock the incident timeline nodes specifically includes:

[0013] The flight status parameters are analyzed over time to identify abrupt changes in parameters where the rate of change exceeds a preset threshold.

[0014] Retrieve the timestamps of the event markers in the QAR dataset, and determine the start frame, development frame, and termination frame based on the timestamps of the event markers and the parameter mutation points, which serve as the event timeline nodes.

[0015] In some preferred embodiments, a subset of key accident parameters is selected based on the accident type and the accident time sequence nodes, specifically as follows:

[0016] Determine parameter association rules based on accident type;

[0017] Based on the accident timeline nodes and parameter association rules, the physical correlation between parameters and accidents, the criticality of the parameters, and the synchronicity between the parameters and accident timeline nodes are calculated and weighted to quantify the accident priority of each flight state parameter, and a subset of key accident parameters is selected based on the accident priority.

[0018] In some preferred embodiments, executing a reverse drive control strategy to drive the simulator to recreate the accident process includes: optimizing the simulator's reverse drive control strategy based on the accident type;

[0019] The optimized reverse drive control strategy for the simulator specifically includes:

[0020] A state space containing accident type codes is constructed, and a reward function is set based on reproduction accuracy, control stability, and accident condition compliance; wherein, the accident condition compliance is determined based on the degree of matching between the physical characteristics of the simulator output state and the flight mechanics model adapted to the accident scenario.

[0021] Based on the state space and the reward function, the PID controller parameters of the simulator are iteratively optimized using a deep reinforcement learning algorithm.

[0022] In some preferred embodiments, the calculation of timing and physical deviations based on the accident QAR dataset and preliminary reproduction data includes:

[0023] Based on the dynamic time warping algorithm, the temporal deviation between the accident QAR dataset and the preliminary reproduction data is calculated;

[0024] Based on a preset standard flight dynamics model, and using a flight mechanical model adapted to the accident scenario, the physical deviation between the preliminary reproduction data and the output of the flight mechanical model is calculated.

[0025] In some preferred embodiments, the method further includes: constructing a quantified accident recurrence causal chain map using a graph neural network model based on the calibrated recurrence data and the accident QAR dataset.

[0026] A second aspect of this application proposes an accident reproduction system based on QAR data and simulator reverse drive, comprising:

[0027] The data processing module is configured to acquire an accident QAR dataset associated with the accident type, wherein the accident QAR dataset includes flight status parameters and event markers;

[0028] The data filtering module is configured to perform accident-oriented preprocessing on the QAR dataset, lock the accident time sequence nodes, and filter out a subset of key accident parameters based on the accident type and the accident time sequence nodes.

[0029] The simulator reverse drive module is configured to take the subset of key accident parameters and accident timing nodes as input, reverse drive the simulator to recreate the accident process, and optimize the simulator's reverse drive control strategy based on the accident type to generate preliminary reproduction data;

[0030] The closed-loop correction module is configured to calculate the timing deviation and physical deviation based on the accident QAR dataset and the preliminary reproduction data, and to dynamically fuse the timing deviation and physical deviation according to the accident timing nodes to generate dynamic correction coefficients; it is also configured to use the dynamic correction coefficients to calibrate the output parameters of the simulator during the reverse drive process to obtain calibrated reproduction data.

[0031] Compared with the prior art, the technical solutions in the embodiments provided in this application have at least the following beneficial effects:

[0032] 1) By constructing a full-process technical system of "data preprocessing, optimization and reverse drive, and closed-loop dynamic correction", the deep coupling of QAR data and simulator is realized, which aims to make the temporal and physical characteristics of the reproduction results highly consistent with the real situation, and overcome the technical problems such as low reproduction accuracy and insufficient compliance in the existing technology.

[0033] 2) Introduce dynamic correction coefficients based on Gaussian mixture model (GMM), and perform closed-loop control of the simulator's reverse drive by real-time fusion of physical deviation and temporal deviation to reduce physical deviation, so that the reproduction process can closely match the actual trajectory of the accident while satisfying the laws of flight mechanics.

[0034] 3) Based on the accident type, key dimensions are automatically identified from massive parameters, effectively eliminating the interference of irrelevant parameters on the reverse drive control logic. By using the combination criteria of parameter mutation rate and discrete fault code, the accident start, development and termination frames are accurately locked, providing accurate time fixation for subsequent reverse drive and causal chain analysis.

[0035] 4) Furthermore, by using graph neural networks (GNNs) to model parameter mutations, out-of-limit events and operating condition nodes, the causal strength index is automatically calculated, reducing the bias caused by the subjective experience of investigators. This transforms qualitative accident investigation into a quantifiable, traceable, and repeatable scientific calculation process, providing objective technical support for accident liability determination. Attached Figure Description

[0036] Other features, objects, and advantages of this application will become more apparent from the following detailed description of non-limiting embodiments with reference to the accompanying drawings:

[0037] Figure 1 This is a flowchart of an accident reproduction method based on QAR data and simulator reverse drive provided in one embodiment of the present invention;

[0038] Figure 2 This is a flowchart of simulator reverse drive control optimization and dynamic closed-loop calibration in an accident reproduction method provided by an embodiment of the present invention;

[0039] Figure 3 This is a schematic diagram of the structure of a computer system used to implement the methods, apparatus, and electronic devices of this application. Detailed Implementation

[0040] The present application will now be described in further detail with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are for illustrative purposes only and are not intended to limit the invention. Furthermore, it should be noted that, for ease of description, only the parts relevant to the invention are shown in the accompanying drawings.

[0041] It should be noted that, unless otherwise specified, the embodiments and features described in this application can be combined with each other. This application will now be described in detail with reference to the accompanying drawings and embodiments.

[0042] To address the technical problems of low accuracy and poor realism in accident reproduction in existing technologies, this application provides an accident reproduction method based on QAR data and simulator reverse-drive. The method includes: acquiring an accident QAR dataset containing flight state parameters; locking the accident timeline nodes and using an accident parameter priority evaluation model to filter out a subset of key accident parameters; optimizing the simulator's reverse-drive control strategy based on the key parameter subset and timeline nodes to generate preliminary reproduction data; calculating the timeline deviation and physical deviation between the preliminary reproduction data and the accident QAR dataset, and dynamically fusing the two deviations according to the accident timeline nodes to generate a dynamic correction coefficient; and using the dynamic correction coefficient to perform closed-loop calibration of the simulator's reverse-drive process to obtain calibrated reproduction data.

[0043] To more clearly explain the accident reproduction method based on QAR data and simulator reverse drive in this application, the following will combine... Figure 1 The steps in the embodiments of this application are described in detail.

[0044] A method for accident reproduction based on QAR data and simulator reverse drive according to the first embodiment of the present invention includes steps S1 to S5, such as... Figure 1 As shown, each step is described in detail below:

[0045] S1. Obtain the accident QAR dataset associated with the accident type. The accident QAR dataset contains flight status parameters and event markers.

[0046] Preferably, raw QAR data for the accident period associated with the accident type is collected to ensure the integrity and accuracy of the data. The collected data includes not only flight status parameters such as airspeed, altitude, angle of attack, and thrust during the accident period and GPS timestamps (time accuracy ≤1ms), but also accident-related time markers, including: fault codes (such as stall warning codes and engine failure codes), over-limit event records, and parameter mutation markers (for example, when the parameter time change rate exceeds a preset change threshold, such as 50% / s, the marker is triggered).

[0047] These tags constitute the event tags in the accident QAR dataset, providing a crucial basis for accurately pinpointing the accident's timeline nodes.

[0048] Furthermore, in this embodiment, QAR raw data during the accident period is preferably acquired synchronously through dual channels using the ARINC429 / 664 standardized protocol.

[0049] S2. Perform accident-oriented preprocessing on the QAR dataset, lock the accident time sequence nodes, and filter out a subset of key accident parameters based on the accident type and the accident time sequence nodes.

[0050] Preferably, the step of performing accident-oriented preprocessing on the QAR dataset to lock the accident timeline nodes specifically involves:

[0051] The flight status parameters are analyzed over time to identify abrupt changes in parameters where the rate of change exceeds a preset threshold (e.g., the rate of change is consistently higher than the preset threshold, which is determined based on historical data or the flight manual).

[0052] The timestamps of event markers (such as fault codes) in the QAR dataset are retrieved, and the start frame, development frame, and termination frame of the accident are accurately determined based on the timestamps of the event markers and the parameter mutation points, which serve as the accident time sequence nodes.

[0053] Specifically, in this embodiment, the accident timing nodes include:

[0054] The starting frame is the first frame node in the intersection of the timestamps of the two events: the occurrence of the fault code and the first time the critical parameter exceeds the limit.

[0055] A development frame is a sequence of consecutive frame nodes in which the temporal change rate of key parameters is consistently higher than a preset change threshold.

[0056] The terminating frame is the first frame node where the values ​​of all key parameters have returned to their safe thresholds.

[0057] Locking in the accident timeline defines precise time boundaries for the reproduction process and provides a phased basis for subsequent dynamic deviation correction.

[0058] More preferably, a subset of key accident parameters is selected based on the accident type and the accident time sequence nodes, specifically as follows:

[0059] Parameter association rules are determined based on accident type (including stall, engine failure, structural damage, etc.);

[0060] Based on the accident timeline nodes and parameter association rules, the physical correlation between parameters and accidents, the criticality of the parameters, and the synchronicity between the parameters and accident timeline nodes are calculated and weighted to quantify the accident priority of each flight state parameter; a subset of key accident parameters is selected based on the accident priority.

[0061] Specifically, in order to accurately select key parameters that play a decisive role in the occurrence and development of accidents from massive QAR data, in this embodiment, based on the accident time sequence nodes and parameter association rules, multi-dimensional parameter evaluation indicators are calculated to quantify the accident priority of each flight parameter, and parameters that meet the priority criteria are selected as a subset of key parameters.

[0062] The specific quantitative calculation formula for the accident priority is as follows:

[0063] ;

[0064] ; ;

[0065] in, The accident priority score is assigned to the parameter. This indicates the physical correlation between the parameter and the accident (0-1). It represents the criticality of the parameter (0-1), which is determined based on the over-limit threshold of the accident scenario and reflects the severity of the parameter value deviating from its safety threshold; It indicates the synchronization between parameters and accident timeline nodes, and is calculated by the ratio of the time difference between the parameter timestamp and the accident start frame to measure the consistency between the time point of drastic parameter change and the accident start point. The time difference between the timestamp of the parameter change and the start frame of the incident. This represents the total duration of the accident. Let be the weighting coefficient, satisfying It supports iterative optimization based on historical accident data.

[0066] Preferably, in this embodiment, It is calculated using a pre-defined "accident type-parameter" correlation matrix. For example, in a stall accident, the angle of attack... It can be set to 0.95, while the cabin temperature... Then it is close to 0.1.

[0067] After calculating all parameters After scoring, a filtering threshold is set, and parameters with scores not less than this threshold are selected (for example, if the filtering threshold is 0.8, then...). The selected parameters form a subset of the key accident parameters.

[0068] It should be noted that the preset "accident type-parameter" correlation matrix is ​​an executable, quantifiable, and standardized construction of the physical correlation between different accident types and flight parameters. The specific complete construction method, quantification rules, and specific examples are as follows:

[0069] 1) Based on the ICAO civil aviation accident classification standards, 10 typical accident types were identified, including stall, engine failure, and structural damage;

[0070] 2) Based on the principles of flight dynamics and historical accident database, the correlation parameters of each type of accident are assigned a correlation value of 0-1: key parameters that directly determine the occurrence of the accident are assigned a value of 0.9-1.0, secondary impact parameters of the accident are assigned a value of 0.5-0.8, and parameters that have no physical connection with the accident are assigned a value of 0-0.2.

[0071] 3) Form a two-dimensional matrix, where rows represent accident types, columns represent flight parameters, and matrix values ​​represent the physical correlation between the corresponding parameters and the accident. .

[0072] Specific examples (fragments) of the matrix are shown in Table 1:

[0073] Table 1

[0074]

[0075] By filtering a subset of key accident parameters, a "minimum effective input set" is provided for subsequent simulator reverse drive, thereby eliminating redundant data interference and improving reproduction efficiency and accuracy. Locking the accident timing nodes and filtering key parameters are crucial to this step, and their specific implementation will be described in detail in Example 2.

[0076] Step S3: Using the subset of key accident parameters and accident timing nodes as input, execute the reverse drive control strategy to drive the simulator to recreate the accident process and generate preliminary reproduction data.

[0077] Preferably, executing a reverse-drive control strategy to drive the simulator to recreate the accident process includes:

[0078] The anti-drive control strategy of the simulator is optimized based on the aforementioned accident type, specifically as follows:

[0079] A state space containing accident type codes is constructed, and a reward function is set based on reproduction accuracy, control stability, and accident condition compliance; wherein, the accident condition compliance is determined based on the degree of matching between the physical characteristics of the simulator output state and the flight mechanics model adapted to the accident scenario.

[0080] Based on the state space and the reward function, the PID controller parameters of the simulator are iteratively optimized using a deep reinforcement learning algorithm.

[0081] More preferably, in this embodiment, the state space containing the accident type code is constructed as follows:

[0082] ;

[0083] in, The deviation between the measured and simulated values ​​of the parameter at time t. The rate of change of deviation For flight altitude, For roll angle, Mach number, Code the accident type;

[0084] The reward function is set based on the reproduction accuracy, control stability, and compliance with accident conditions:

[0085] ;

[0086] in, The root mean square error between the reproduced data and the QAR incident data is used to penalize the bias, aiming to improve the accuracy of the reproduction. To smooth out the penalty for control parameters (such as rudder and throttle), for Control input at any time, for The control input at time -1 avoids violent jitters that do not conform to physical laws during the reverse drive process; To determine the degree of conformity to the accident conditions, a flight dynamics model adapted to the accident scenario is used for judgment, aiming to ensure that the reproduced state meets the key physical characteristics of the accident (for example, the stall condition must meet the angle of attack ≥ stall threshold and lift coefficient decrease ≥ 30%). This is a weighting coefficient that is dynamically adjusted according to the timeline of the incident; for example, in the incident development frame, it can be increased. The weighting is adjusted to enhance the relevance to critical accident conditions.

[0087] In this embodiment, the accident type coding preferably adopts a 4-digit coding rule of "major category code + sub-category code" to assign standardized codes to different accident types; for example, the key contents of the coding table are shown in Table 2 below:

[0088] Table 2

[0089]

[0090] Accident type coding The above four-digit code can be used.

[0091] Based on the state space and the reward function, the PID controller parameters of the simulator are iteratively optimized using a deep reinforcement learning algorithm.

[0092] More preferably, this embodiment employs, for example, a Near-Proximal Strategy Optimization algorithm (PPO-Clip) to iteratively optimize the PID controller parameters. The algorithm uses an Actor-Critic dual-network architecture: the Actor network outputs the normal distribution mean and variance of the PID parameters, and the PID parameters are limited to the safety control range of the civil aviation simulator (e.g., ...). The Critic network outputs a value assessment of the current state, providing gradient guidance for policy updates.

[0093] More preferably, at least 50 Monte Carlo simulations are performed to generate multiple sets of reproduced response data, and uncertainty indices are calculated:

[0094] ;

[0095] in, For the first Secondary simulated output. If the reproduced data is deemed valid, output the valid reproduced response data with uncertainty assessment.

[0096] By introducing accident-specific features, the control parameters of the simulator are deeply optimized, enabling the reinforcement learning policy network to output different optimal PID parameters for different accident types. This aims to make the reverse-drive process closely resemble real accident scenarios. In practical applications, by loading a preset aerodynamic model and parameter library based on the target aircraft model, the initial PID parameters, reward function weights, and mechanical model coefficients can be automatically adjusted to reproduce accidents for different aircraft models.

[0097] S4. Based on the accident QAR dataset and preliminary reproduction data, calculate the timing deviation and physical deviation respectively, and dynamically fuse the timing deviation and physical deviation according to the accident timing nodes to generate dynamic correction coefficients.

[0098] Preferably, refer to, as Figure 2 As shown, the calculation of timing and physical deviations based on the accident QAR dataset and preliminary reproduction data includes:

[0099] Based on the dynamic time warping algorithm, the temporal deviation between the accident QAR dataset and the preliminary reproduction data is calculated;

[0100] Based on a preset standard flight dynamics model, and using a flight mechanical model adapted to the accident scenario, the physical deviation between the preliminary reproduction data and the output of the flight mechanical model is calculated.

[0101] More preferably, the temporal deviation between the accident QAR dataset and the preliminary reproduction data is calculated based on the Dynamic Time Warping (DTW) algorithm. :

[0102] ;

[0103] in, Let be the parameter value of QAR at time t. The reproducible parameter values ​​after alignment. Set dynamic window constraints for the time mapping function This improves the timing alignment accuracy of critical accident nodes.

[0104] Furthermore, Adjust according to the type of accident, such as a stall accident. W=3 Frame, engine failure accident W=5 frame.

[0105] More preferably, a flight dynamics model adapted to the accident scenario (e.g., a model including stall criterion correction) is used to calculate the physical deviation between the preliminary reproduction data and the theoretical output of the flight dynamics model. :

[0106] ;

[0107] in, It is a theoretical state vector (including thrust, lift coefficient, drag coefficient, etc.) calculated by the flight dynamics model adapted to the accident scenario based on the current control input. It is a state vector (such as a normalized vector containing engine speed, airspeed, pitch angle, etc.) constructed based on preliminary reproduction data. It is the Euclidean norm (L2 norm).

[0108] More preferably, the step of dynamically fusing the timing deviation and the physical deviation based on the accident timing nodes to generate dynamic correction coefficients includes:

[0109] A fusion strategy based on time-series weights is used to weight and fuse the time-series deviation and the physical deviation to generate the dynamic correction coefficient.

[0110] The fusion weights used for the weighted fusion are dynamically adjusted according to the accident timeline node at which the current reproduction process is located. When the accident timeline node is in the development frame, the fusion weight value corresponding to the timeline deviation is greater than the fusion weight value corresponding to the timeline deviation when the accident timeline node is in the start frame or the end frame.

[0111] Specifically, this embodiment preferably uses a time-weighted fusion strategy to fuse time-series deviations and physical deviations to generate dynamic correction coefficients for accident reproduction:

[0112] ;

[0113] in, To integrate weights and enhance deviation calibration during critical stages of an accident; in this embodiment, the weights are integrated. The timing of the incident is dynamically adjusted according to the incident timing node. When the incident timing node is in the "development frame", the weight of the timing deviation can be increased, and its value will be different from the value in the "start frame" or "termination frame". For example, in the start frame... =0.4, Development Frame =0.6, Termination Frame =0.5, to strengthen the deviation calibration of the critical stage of the accident, aiming to make the timing and physical characteristics of the reproduced results highly consistent with the actual situation.

[0114] S5. The output parameters of the simulator during the reverse drive process are calibrated using the dynamic correction coefficient to obtain calibrated reproducible data; specifically including:

[0115] 1) Simulator execution and status update: The simulator updates flight status parameters (altitude, angle of attack, airspeed, etc.) based on the calibrated control variables and generates calibrated reproducible data for the current cycle;

[0116] 2) Closed-loop iterative verification: In the next control cycle, repeat the steps to calculate the dynamic correction coefficient to the simulator execution and state update steps until all accident timing nodes are completed and the physical deviation of the whole process is ≤5%, and finally output complete calibration and reproduction data.

[0117] Specifically, the correction coefficient Real-time feedback is sent to the simulator control terminal, using correction coefficients. Adjust the feedback gain to reduce the deviation, and iterate until the deviation reaches the target (e.g., physical deviation ≤ 5%). For example:

[0118] according to Adjust the output parameters, where, These are the current output parameters of the simulator. These are the calibrated parameters. To update the step size (learning rate) The true target value recorded by QAR. The sign function is used to determine the correction direction (i.e., decreasing when the simulated value is too large and increasing when it is too small) and the correction coefficient. The magnitude is used as a correction for dynamic gain control.

[0119] In the second embodiment of this application, based on the first embodiment, to address the problem of lack of quantitative evidence for causal tracing, the method further includes:

[0120] Based on the calibrated reproducible data and the accident QAR dataset, a quantified accident reproducible causal chain graph is constructed using a graph neural network model.

[0121] Preferably, the step of constructing a quantified causal chain graph of accident recurrence using a graph neural network model includes:

[0122] Extract parameter mutation points, out-of-limit events, and event markers from the calibrated reproducible data and the accident QAR dataset to construct a directed dynamic time series graph containing parameter nodes, event nodes, and operating condition nodes;

[0123] A graph neural network model is used to calculate the causal strength of the edges between nodes, and the nodes and edges are sorted according to the accident timeline to generate an accident recurrence causal chain graph.

[0124] The calculation of the causal strength is based on the parameter change and the parameter reference value under the accident scenario. The parameter reference value is a parameter safety threshold or critical value determined from the flight manual or historical database according to the accident type.

[0125] More preferably, in this embodiment, elements characterizing key aspects of accident development are extracted from the calibrated reproduced data and the original QAR dataset as nodes in the graph. Nodes can be categorized into three types:

[0126] Parameter nodes: These consist of points of dramatic parameter changes, such as "angle of attack exceeding limits", "rapid decrease in lift", and "sudden decrease in thrust".

[0127] Event nodes: Consisting of fault codes or over-limit event markers, such as "stall warning" and "engine failure warning";

[0128] Flight condition node: indicates the stage of flight, such as "cruise" or "approach".

[0129] More preferably, in this embodiment, a two-layer multi-head attention architecture graph attention network (GAT) is used to construct a directed dynamic temporal graph. ,in, It is a set of nodes (including parameter nodes, event nodes, and operating condition nodes). Let be a set of directed edges. This is the node feature matrix. The node features are normalized using the min-max method to... The directed edges in the interval satisfy the following constraints: time constraints (only nodes with prior time are allowed to point to nodes with subsequent time), physical association constraints (edges are only constructed between nodes with aerospace physical associations), and accident association constraints (accident key parameter nodes include nodes with event nodes and operating condition nodes).

[0130] More preferably, the output layer of the graph attention network outputs the causal strength of each directed edge through a sigmoid activation function, calculated as follows:

[0131] ;

[0132] in, The parameter variation can be obtained from the comparison results of simulator reproduction data and QAR data; The parameter reference value is used in the accident scenario (e.g., the angle of attack reference value is the stall threshold in a stall accident); causality strength The range of values ​​is A higher value indicates a stronger causal relationship. Thresholds can be set for grading, such as... Determined to be a strong association, It is a weak correlation; thus transforming the traditional subjective judgment that relies on expert experience into an objective quantitative assessment.

[0133] More preferably, when generating the graph, nodes and associated edges are sorted according to the accident timeline to generate a visual graph that includes a timeline, parameter change curves, and causal logic explanations. The graph format conforms to the requirements for archiving civil aviation accident investigation data. This graph can objectively and quantitatively display the causal relationships and time sequence between various parameters and events during the accident development process, providing a strong and traceable chain of key evidence for the accident investigation.

[0134] More preferably, the method further includes:

[0135] The accident reproduction causal chain graph and the calibrated reproduction data are subjected to compliance verification, which includes at least one of physical constraint verification, digital twin verification and blockchain evidence storage.

[0136] The blockchain-based evidence storage includes:

[0137] Calculate hash values ​​for the accident QAR dataset, anti-drive parameters, deviation correction records, and causal chain graph, and upload the hash values ​​or combinations thereof to the blockchain network.

[0138] Specifically, the compliance verification includes:

[0139] Physical constraint verification: The preset standard flight dynamics model is preferably a six-degree-of-freedom rigid body flight dynamics model. The simulation results of reverse-drive are substituted into the six-degree-of-freedom flight equations to verify the physical rationality of the simulation results. The residuals are not greater than the preset residual threshold (e.g., If the condition is deemed acceptable, but the residual exceeds the limit, the reverse drive parameters are readjusted using the deviation correction model.

[0140] Digital twin verification: Input the reproduction results into the digital twin model of the corresponding aircraft model to verify the accuracy of the reproduced trajectory. The degree of consistency between the reproduced trajectory and the digital twin simulated trajectory is not less than the preset consistency threshold (e.g., ≥90%), and the time difference of the key accident nodes is not greater than the preset time difference threshold (e.g., ≤100ms). The result is considered as qualified.

[0141] Blockchain-based evidence storage: The original QAR accident data, simulator reverse drive parameters, deviation correction records, and causal chain graph are calculated using SHA-256 hash values ​​and uploaded to the blockchain network for evidence storage. The block structure is as follows:

[0142]

[0143] in, The hash value of the previous block. For key data hash value, For evidence storage timestamp, To ensure compliance verification, the identification system enables the reproduction of tamper-proof and traceable data throughout the entire process.

[0144] Output a standardized compliance verification report. The report format complies with the data requirements for accident investigation in the Annex to the Convention on International Civil Aviation. The report clearly marks the key parameters of the simulator's anti-drive, deviation correction records, key conclusions of the causal chain, and the blockchain storage address.

[0145] Preferably, to cover the accident reproduction needs of mainstream civil aviation aircraft models, the method further includes a multi-aircraft model adaptation step: pre-setting aerodynamic models, structural parameter libraries, and typical accident scenario correction rule libraries for mainstream aircraft models such as Boeing 737 / 787 and Airbus A320 / A350; for the aircraft model corresponding to the target accident, loading the corresponding parameter library, automatically adjusting the simulator's reverse drive initial parameters, reward function weights, and mechanical model correction rules, further reducing the adaptation time.

[0146] As a concrete example, the default standard flight dynamics model is preferably a six-degree-of-freedom rigid body flight dynamics model. Using this model as the baseline, a general process of "dimensional correction → parameter orientation correction → compliance verification" is employed to construct an adapted model for different accident types. For instance, when the accident type is a stall accident, the specific correction method for the flight dynamics model is as follows:

[0147] The key correction dimension is the aerodynamic model. A stall criterion is introduced to correct the lift coefficient. For example, considering the unsteady aerodynamic characteristics of airflow separation after stall, the lift coefficient, drag coefficient, and pitching moment coefficient are corrected simultaneously. The formula is as follows:

[0148] Introducing stall criteria to distinguish lift characteristics before and after stall, and calculating lift coefficients. Correction:

[0149] ;

[0150] in, For the angle of attack, The stall angle of attack thresholds are for the corresponding aircraft models (Boeing 737 = 15°, Airbus A320 = 16°). The slope of the lift coefficient decrease after stall is represented by values ​​derived from wind tunnel test data and QAR measurements.

[0151] Introducing additional stall drag increments to calculate the drag coefficient Synchronous correction:

[0152] ;

[0153] in, Zero-lift drag coefficient, The induced drag coefficient, This is the additional drag increment due to stall; hour, ; hour, , The drag growth slope is derived from wind tunnel test data of the aircraft model.

[0154] Introducing a pitch damping attenuation term to calculate the pitch moment coefficient Synchronous correction:

[0155] ;

[0156] in, The zero-lift pitch moment coefficient, The slope of the pitch moment. This is the stall pitch moment increment. This is the pitch damping attenuation term; hour, ; hour, , The damping attenuation slope, It is dynamic pressure.

[0157] The revised lift, drag, and pitch moment coefficients are simultaneously substituted into the force and moment equations of the six-degree-of-freedom flight dynamics model to form a complete flight mechanics model adapted to stall conditions, aiming to make the model fit the unsteady aerodynamic characteristics after stall.

[0158] For other accident types such as engine failure and structural damage, the general process of "correction dimension positioning → parameter orientation correction → compliance verification" is followed to construct the flight dynamics model: the key correction dimensions for engine failure accidents are the thrust model, aerodynamic drag model, and lateral moment model. Based on the number of failed engines and the remaining thrust, the thrust formula, windmill drag increment, and asymmetric yaw / roll moment terms are corrected in a directional manner. The key correction dimensions for structural damage accidents are the aerodynamic derivative model, moment of inertia parameter, and center of gravity position. Based on the damage location and damage area, the aerodynamic derivative, moment of inertia, and center of gravity offset are corrected in a directional manner, and the model is matched with the structural damage database.

[0159] The third embodiment of this application, based on the first or second embodiment, further illustrates the specific application of the accident reproduction method based on QAR data and simulator reverse drive. This embodiment will take a Boeing 737-800 flight stall accident during the cruise phase as an example to explain the implementation process of the present invention in detail.

[0160] Accident background: The flight exceeded the angle of attack limit at a cruising altitude of 10,000m, resulting in a rapid decrease in lift and triggering a stall warning. The incident occurred between 14:20:00 and 14:20:30 (30 seconds), with a QAR data sampling frequency of 10Hz (300 frames in total).

[0161] The specific steps to reproduce this are as follows:

[0162] S1. Obtain the accident-related QAR dataset: Extract QAR data from 14:20:00 to 14:20:30, including parameters such as angle of attack, airspeed, altitude, thrust, and engine speed. Mark the stall warning code (0x0023) and angle of attack abrupt change points (abrupt change rate 60% / s), and associate with GPS timestamps (accuracy 1ms) to form a labeled accident QAR dataset.

[0163] S2, Accident-oriented pre-processing:

[0164] 1) Lock timing nodes: The starting frame is 14:20:05 (frame 50, stall alarm code appears and angle of attack exceeds the limit to 17°), the development frames are 14:20:05-14:20:15 (frames 50-150), and the ending frame is 14:20:25 (frame 250, angle of attack returns to the safe threshold).

[0165] 3) Filter the subset of key parameters: For accidents of stall type (code 0102), load the correlation matrix and calculate the key parameters. Value (weighted optimal value) ):

[0166] Angle of attack , , , airspeed ,thrust All met the standards and were included in the key parameter subset.

[0167] S3, Simulator Reverse Drive:

[0168] Constructing the state space Set the weight of the reward function , , (Accident development frame, enhancing compliance with operating conditions).

[0169] Loading a Boeing 737-800 aerodynamic model, optimizing PID parameters through 1000 iterations using the PPO-Clip algorithm, and performing 50 Monte Carlo simulations, with uncertainty indicators... Output valid preliminary reproduction data.

[0170] S4. Calculate the correction factor:

[0171] Using a control period of 10ms, calculate timing and physical deviations in each period: set window constraints. W= 3 frames (stall incident). Substituting the flight dynamics model adapted for stall conditions, stall accidents... .

[0172] Generate correction coefficients: Accident development frame fusion weights Calculated .

[0173] S5, Dynamic Deviation Correction:

[0174] according to The simulator's reverse-drive output parameters are adjusted in real time, and the flight status is updated by inputting the data into the simulator. After iterative correction throughout the entire process, the physical deviation is reduced to 0.04. The reproduction accuracy meets the standard, and the calibrated reproduction data is output.

[0175] Furthermore, based on the calibrated reproduced data and the original QAR dataset, a quantitative causal chain map of accident reproduction is constructed using a graph neural network model, enabling visualization and traceability of the causal relationship of the accident, and providing evidentiary support for the accident investigation.

[0176] 1) Extraction nodes: Angle of attack exceeding limit (parameter node), lift reduction (parameter node), stall alarm (event node), cruise condition (condition node).

[0177] 2) Calculate causal intensity: Angle of attack exceeds limit Lift decrease Lift decreases Stall warning =0.91, all are strong correlations.

[0178] 3) Generate a timeline: including a timeline from 14:20:00 to 14:20:30, and label the causal logic as "angle of attack exceeds limit → lift decreases → stall warning".

[0179] The key causal chain of this stall accident is "exceeding the angle of attack limit → a sudden drop in lift coefficient → stall alarm triggering". The causal correlation strength between exceeding the angle of attack limit and the drop in lift is 0.85, and the causal correlation strength between the drop in lift and the stall alarm is 0.91. Both are strong causal relationships that led to the accident. This clarifies that the key trigger of this accident was aerodynamic stall caused by exceeding the angle of attack limit, providing a quantitative causal analysis basis for the accident investigation.

[0180] Furthermore, compliance verification is performed across three dimensions: physical constraints, digital twins, and blockchain evidence storage. If verification fails, the simulator's anti-flight parameters are readjusted. The final output is a traceable, reproducible result and a standardized report conforming to Annex 13 of the Convention on International Civil Aviation. The compliance verification process includes:

[0181] 1) Physical constraint verification: Substitute the reproduction result into the six-degree-of-freedom equation, the residual is 3.5% (≤5%), which is acceptable.

[0182] 2) Digital twin verification: Input the Boeing 737-800 digital twin model. The consistency between the reproduced trajectory and the twin simulation trajectory is 93%, and the time difference of key nodes is 80ms (≤100ms). It is qualified.

[0183] 3) Blockchain Evidence Storage: Calculate the SHA-256 hash value of the original QAR data, reverse drive parameters, deviation correction records, and causal chain graph, upload it to the blockchain network, and generate an evidence storage address.

[0184] Output a standardized reproduction report and causal chain diagram that conforms to Annex 13 of the Convention on International Civil Aviation.

[0185] This case fully demonstrates the ability of the proposed method to reproduce and trace the root causes of aviation accidents using QAR data.

[0186] The fourth aspect of this application proposes an accident reproduction system based on QAR data and simulator reverse drive, comprising:

[0187] The data processing module is configured to acquire an accident QAR dataset associated with the accident type, wherein the accident QAR dataset includes flight status parameters and event markers;

[0188] The data filtering module is configured to perform accident-oriented preprocessing on the QAR dataset, lock the accident time sequence nodes, and filter out a subset of key accident parameters based on the accident type and the accident time sequence nodes.

[0189] The simulator reverse drive module is configured to take the subset of key accident parameters and accident timing nodes as input, reverse drive the simulator to recreate the accident process, and optimize the simulator's reverse drive control strategy based on the accident type to generate preliminary reproduction data;

[0190] The closed-loop correction module is configured to calculate the timing deviation and physical deviation based on the accident QAR dataset and the preliminary reproduction data, and to dynamically fuse the timing deviation and physical deviation according to the accident timing nodes to generate dynamic correction coefficients; it is also configured to use the dynamic correction coefficients to calibrate the output parameters of the simulator during the reverse drive process to obtain calibrated reproduction data.

[0191] Those skilled in the art will understand that, for the sake of convenience and brevity, the specific working process and related descriptions of the system described above can be found in the corresponding processes in the foregoing method embodiments, and will not be repeated here.

[0192] It should be noted that the accident reproduction system based on QAR data and simulator reverse drive provided in the above embodiments is only an example of the division of the above functional units. In practical applications, the above functions can be assigned to different functional modules as needed, that is, the modules or steps in the embodiments of this application can be further decomposed or combined. For example, the modules in the above embodiments can be merged 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 this application are only for distinguishing the various modules or steps and are not considered as an improper limitation of this application.

[0193] A device according to the fifth embodiment of this application includes:

[0194] At least one processor;

[0195] and a memory communicatively connected to at least one of the processors;

[0196] The memory stores instructions that can be executed by the processor to implement the above-described accident reproduction method based on QAR data and simulator reverse drive.

[0197] A computer-readable storage medium according to a sixth embodiment of this application stores computer instructions that are executed by a computer to implement the above-described accident reproduction method based on QAR data and simulator reverse drive.

[0198] Those skilled in the art will understand that, for the sake of convenience and brevity, the specific working process and related descriptions of the storage device and processing device described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here.

[0199] The following is for reference. Figure 3 It shows a schematic diagram of the structure of a computer system for implementing the methods, apparatus, and electronic devices of this application. Figure 3 The server shown is merely an example and should not impose any limitations on the functionality and scope of use of the embodiments of this application.

[0200] like Figure 3 As shown, the computer system includes a Central Processing Unit (CPU) 301, which can perform various appropriate actions and processes based on programs stored in Read Only Memory (ROM) 302 or programs loaded from storage section 308 into Random Access Memory (RAM) 303. The RAM 303 also stores various programs and data required for system operation. The CPU 301, ROM 302, and RAM 303 are interconnected via a bus 304. An Input / Output (I / O) interface 305 is also connected to the bus 304.

[0201] The following components are connected to I / O interface 305: an input section 306 including a keyboard, mouse, etc.; an output section 307 including a cathode ray tube (CRT), liquid crystal display (LCD), and speakers, etc.; a storage section 308 including a hard disk, etc.; and a communication section 309 including a network interface card such as a LAN (Local Area Network) card and a modem, etc. The communication section 309 performs communication processing via a network such as the Internet. A drive 310 is also connected to I / O interface 305 as needed. Removable media 311, such as a disk, optical disk, magneto-optical disk, semiconductor memory, etc., are installed on drive 310 as needed so that computer programs read from them can be installed into storage section 308 as needed.

[0202] Specifically, according to embodiments of this disclosure, the processes described above with reference to the flowcharts can be implemented as computer software programs. For example, embodiments of this disclosure include a computer program product comprising a computer program carried on a computer-readable medium, the computer program containing program code for performing the methods shown in the flowcharts. In such embodiments, the computer program can be downloaded and installed from a network via communication section 309, and / or installed from removable medium 311. When the computer program is executed by central processing unit (CPU) 301, it performs the functions defined in the methods of this application. It should be noted that the computer-readable medium described above in this application can be a computer-readable signal medium or a computer-readable storage medium, or any combination of the two. A computer-readable storage medium can be, for example,—but not limited to—an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any combination thereof. More specific examples of computer-readable storage media may include, but are not limited to: electrical connections having one or more wires, portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fiber, portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination thereof. In this application, a computer-readable storage medium can be any tangible medium containing or storing a program that can be used by or in connection with an instruction execution system, apparatus, or device. In this application, a computer-readable signal medium may include a data signal propagated in baseband or as part of a carrier wave, carrying computer-readable program code. Such propagated data signals can take various forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination thereof. A computer-readable signal medium can also be any computer-readable medium other than a computer-readable storage medium, which can send, propagate, or transmit a program for use by or in connection with an instruction execution system, apparatus, or device. The program code contained on a computer-readable medium can be transmitted using any suitable medium, including but not limited to: wireless, wire, optical fiber, RF, etc., or any suitable combination thereof.

[0203] Computer program code for performing the operations of this application can be written in one or more programming languages ​​or a combination thereof, including object-oriented programming languages ​​such as Java, Smalltalk, and C++, and conventional procedural programming languages ​​such as the "C" language or similar programming languages. The program code can be executed entirely on the user's computer, partially on the user's computer, as a standalone 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 remote computers, the remote computer can be connected to the user's computer via any type of network—including a local area network (LAN) or a wide area network (WAN)—or can be connected to an external computer (e.g., via the Internet using an Internet service provider).

[0204] The flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of systems, methods, and computer program products according to various embodiments of this application. In this regard, each block in a flowchart or block diagram may represent a module, segment, or portion of code containing one or more executable instructions for implementing a specified logical function. It should also be noted that in some alternative implementations, the functions indicated in the blocks may occur in a different order than those indicated in the drawings. For example, two consecutively indicated blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each block in the block diagrams and / or flowcharts, and combinations of blocks in the block diagrams and / or flowcharts, can be implemented using a dedicated hardware-based system that performs the specified function or operation, or using a combination of dedicated hardware and computer instructions.

[0205] The terms “first”, “second”, etc., are used to distinguish similar objects, not to describe or indicate a specific order or sequence.

[0206] The term "comprising" or any other similar term is intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus / device that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent in such process, method, article, or apparatus / device.

[0207] The technical solutions of this application have been described above with reference to 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 this application is obviously not limited to these specific embodiments. Without departing from the principles of this application, those skilled in the art can make equivalent changes or substitutions to the relevant technical features, and the technical solutions after these changes or substitutions will all fall within the scope of protection of this application.

Claims

1. An accident reproduction method based on QAR data and counter-driving of a simulator, characterized by, Includes the following steps: Obtain the accident QAR dataset associated with the accident type, the accident QAR dataset containing flight status parameters and event markers; Accident-oriented preprocessing is performed on the QAR dataset to lock the accident time sequence nodes, and a subset of key accident parameters is selected based on the accident type and the accident time sequence nodes. The subset of key accident parameters and accident timeline nodes are used as inputs to execute a reverse drive control strategy to drive the simulator to recreate the accident process and generate preliminary reproduction data. Based on the aforementioned accident QAR dataset and preliminary reproduction data, the timing deviation and physical deviation are calculated respectively. Then, based on the accident timing nodes, the timing deviation and physical deviation are dynamically fused to generate dynamic correction coefficients, including: Based on the dynamic time warping algorithm, the temporal deviation between the accident QAR dataset and the preliminary reproduction data is calculated; Based on a preset standard flight dynamics model, and using a flight mechanical model adapted to the accident scenario, the physical deviation between the preliminary reproduction data and the output of the flight mechanical model is calculated. A fusion strategy based on time-series weights is used to weight and fuse the time-series deviation and the physical deviation to generate the dynamic correction coefficient. The fusion weights used for the weighted fusion are dynamically adjusted according to the accident timeline node at which the current reproduction process is located. When the accident timeline node is in the development frame, the fusion weight value corresponding to the timeline deviation is greater than the fusion weight value corresponding to the timeline deviation when the accident timeline node is in the start frame or the end frame. The output parameters of the simulator during the reverse drive process are calibrated using the dynamic correction coefficient to obtain calibrated reproducible data.

2. The method of claim 1, wherein, The step of performing incident-oriented preprocessing on the QAR dataset to lock the incident timeline nodes specifically involves: The flight status parameters are analyzed over time to identify abrupt changes in parameters where the rate of change exceeds a preset threshold. The threshold is determined based on historical data or the flight manual. Retrieve the timestamps of the event markers in the QAR dataset, and determine the start frame, development frame, and termination frame based on the timestamps of the event markers and the parameter mutation points, which serve as the event timeline nodes.

3. The QAR data and simulator counterdriven accident reproduction method according to any one of claims 1 or 2, characterized in that, Based on the accident type and the accident time sequence nodes, a subset of key accident parameters is selected, specifically as follows: Determine parameter association rules based on accident type; Based on the accident timeline nodes and parameter association rules, the physical correlation between parameters and accidents, the criticality of the parameters, and the synchronicity between the parameters and accident timeline nodes are calculated and weighted to quantify the accident priority of each flight state parameter, and a subset of key accident parameters is selected based on the accident priority.

4. The accident reproduction method based on QAR data and simulator reverse drive according to claim 1, characterized in that, Executing a reverse drive control strategy to drive the simulator to recreate the accident process includes: optimizing the simulator's reverse drive control strategy based on the accident type; The optimized reverse drive control strategy for the simulator specifically includes: A state space containing accident type codes is constructed, and a reward function is set based on reproduction accuracy, control stability, and accident condition compliance; wherein, the accident condition compliance is determined based on the degree of matching between the physical characteristics of the simulator output state and the flight mechanics model adapted to the accident scenario. Based on the state space and the reward function, the PID controller parameters of the simulator are iteratively optimized using a deep reinforcement learning algorithm.

5. The accident reproduction method based on QAR data and simulator reverse drive according to claim 1, characterized in that, The dynamic correction coefficient for: ； in, To integrate weights and strengthen deviation calibration during critical stages of an accident; integrate weights Dynamically adjusted according to the timeline of the accident. This is due to timing deviation. This is a physical deviation.

6. The accident reproduction method based on QAR data and simulator reverse drive according to claim 5, characterized in that, When calibrating the output parameters of the simulator during the reverse drive process using the aforementioned dynamic correction coefficients, closed-loop iterative verification is performed, specifically including: The simulator updates flight status parameters based on the calibrated control variables and generates calibrated reproducible data for the current period. In the next control cycle, the steps of calculating the dynamic correction coefficient are repeated until the simulator executes and updates the status, until all accident timing nodes are completed and the physical deviation of the entire process meets the standard, and finally the complete calibration and reproduction data is output.

7. The accident reproduction method based on QAR data and simulator reverse drive according to claim 1, characterized in that, The method further includes: constructing a quantified accident recurrence causal chain map using a graph neural network model based on the calibrated recurrence data and the accident QAR dataset.

8. The accident reproduction method based on QAR data and simulator reverse drive according to claim 7, characterized in that, The construction of a quantified causal chain graph of accident recurrence using a graph neural network model includes: Extract parameter mutation points, out-of-limit events, and event markers from the calibrated reproduction data and the accident QAR dataset to construct a directed dynamic time series graph containing parameter nodes, event nodes, and operating condition nodes; A graph neural network model is used to calculate the causal strength of the edges between nodes, and the nodes and edges are sorted according to the accident timeline to generate an accident recurrence causal chain graph. The calculation of the causal strength is based on the parameter change and the parameter reference value under the accident scenario. The parameter reference value is a parameter safety threshold or critical value determined from the flight manual or historical database according to the accident type.

9. The accident reproduction method based on QAR data and simulator reverse drive according to claim 7, characterized in that, The method further includes: The accident reproduction causal chain graph and the calibrated reproduction data are subjected to compliance verification, which includes at least one of physical constraint verification, digital twin verification and blockchain evidence storage. The blockchain-based evidence storage includes: Calculate hash values ​​for the accident QAR dataset, anti-drive parameters, deviation correction records, and causal chain graph, and upload the hash values ​​or combinations thereof to the blockchain network.

10. An accident reproduction system based on QAR data and simulator reverse drive, characterized in that, include: The data processing module is configured to acquire an accident QAR dataset associated with the accident type, wherein the accident QAR dataset includes flight status parameters and event markers; The data filtering module is configured to perform accident-oriented preprocessing on the QAR dataset, lock the accident time sequence nodes, and filter out a subset of key accident parameters based on the accident type and the accident time sequence nodes. The simulator reverse drive module is configured to take the subset of key accident parameters and accident timing nodes as input, reverse drive the simulator to recreate the accident process, and optimize the simulator's reverse drive control strategy based on the accident type to generate preliminary reproduction data; The closed-loop correction module is configured to calculate the timing deviation and physical deviation based on the accident QAR dataset and the preliminary reproduction data, and to dynamically fuse the timing deviation and physical deviation according to the accident timing nodes to generate dynamic correction coefficients; it is also configured to use the dynamic correction coefficients to calibrate the output parameters of the simulator during the reverse drive process to obtain calibrated reproduction data. Specifically, the timing deviation and physical deviation are calculated based on the aforementioned accident QAR dataset and preliminary reproduction data, as follows: Based on the dynamic time warping algorithm, the temporal deviation between the accident QAR dataset and the preliminary reproduction data is calculated; Based on a preset standard flight dynamics model, and using a flight mechanical model adapted to the accident scenario, the physical deviation between the preliminary reproduction data and the output of the flight mechanical model is calculated. Based on the accident timeline nodes, the timing deviation and the physical deviation are dynamically fused to generate dynamic correction coefficients, specifically as follows: The fusion strategy based on time-series weights performs weighted fusion of the time-series deviation and the physical deviation to generate the dynamic correction coefficient; wherein, the fusion weights used for the weighted fusion are dynamically adjusted according to the accident time-series node in the current reproduction process. When the accident time-series node is in the development frame, the fusion weight value corresponding to the time-series deviation is greater than the fusion weight value corresponding to the time-series deviation when the accident time-series node is in the start frame or the end frame.

Images