Aircraft aviation risk assessment method and system

Abstract

The invention belongs to the technical field of risk level assessment, and provides an aircraft aviation risk assessment method and system, and the method comprises the steps: obtaining and analyzing a force element; a plurality of different algorithms are utilized, based on the extracted key force elements, working condition danger degree sorting is carried out, correlation analysis is carried out on all results, according to correlation analysis results, a plurality of final algorithms are screened, and a preliminary heterogeneous algorithm is formed; processing the data subjected to disturbance by using a heterogeneous algorithm, carrying out cross comparison on the sorting performance of each algorithm under the same disturbance condition, and carrying out secondary screening to form a final heterogeneous algorithm; the final heterogeneous algorithm is utilized to carry out aircraft aviation risk assessment, and an assessed risk sorting result is obtained; and fusing the risk sorting results obtained by evaluation by using different methods to obtain two fused sorting results, and carrying out weighted fusion on the two fused sorting results. The sorting accuracy can be improved.

Description

Technical Field

[0001] This invention belongs to the field of hazard assessment technology, specifically relating to a method and system for assessing aviation risks of aircraft. Background Technology

[0002] The statements in this section are merely background information related to the present invention and do not necessarily constitute prior art.

[0003] The current field of multi-condition risk assessment for aircraft structures generally faces three major technical bottlenecks: First, traditional methods rely on binary safety thresholds for condition screening, lacking the ability to continuously quantify risk. This results in a large number of subcritical states being coarsely classified, failing to accurately identify gradient risk differences. Second, cross-sectional force element analysis often employs single-algorithm static evaluation models (such as finite element stress benchmarking, where detailed models are too labor-intensive and coarse models are prone to failing to accurately assess condition risks), making it difficult to integrate multi-dimensional force element coupling mechanisms (such as bending-torsional nonlinear interactions). This leads to a significant amplification of risk prediction biases for complex load conditions. Third, under the existing technical framework, flight environment parameters and structural response characteristics have long been treated separately, lacking cross-model collaborative optimization mechanisms, resulting in excessive consumption of computational resources on low-risk condition verification. These shortcomings lead to systemic problems in the industry, such as insufficient risk assessment accuracy, low computational efficiency, and a high rate of missed detection of potential hazards.

[0004] In addition, the current aircraft structural risk assessment system has a bottleneck in dynamic expansion. The traditional architecture lacks an intelligent assimilation interface for incremental operating conditions. New operating conditions require repeated full-process calculations, which cannot reuse existing analysis models and risk rules, nor can it access historical risk maps for offset comparison, resulting in a significant increase in assessment delays and extremely low resource reuse rates. Summary of the Invention

[0005] To address the aforementioned problems, this invention proposes a method and system for assessing aircraft aviation risks. This invention achieves multi-physics field coupled aircraft aviation risk ranking.

[0006] According to some embodiments, the present invention adopts the following technical solution: An aircraft aviation risk assessment method includes the following steps: The aircraft is divided into regions, and the FBD section under static equilibrium constraints is extracted for the region where the flight hazard exceeds the set value. The six force elements of the section boundary, as well as the resultant internal force and resultant moment, are obtained. Based on the acquired data, the contribution weight of each section force to the stress distribution change is quantified, and the forces are sorted according to their contribution weights to determine the key force elements. Using multiple different algorithms, the degree of danger of the working conditions is ranked based on the key force elements. Correlation analysis is performed on all results. Based on the correlation analysis results, multiple algorithms are selected to form a preliminary heterogeneous algorithm. The data after perturbation is processed using heterogeneous algorithms. The ranking performance of each algorithm under the same perturbation conditions is cross-compared, and a second selection is performed to form the final heterogeneous algorithm. The aircraft aviation risk assessment was performed using the final heterogeneous algorithm, and the risk ranking results were obtained after the assessment. The risk ranking results obtained from the assessment are fused using a genetic optimization fusion algorithm and an adversarial solution optimization model, respectively, to obtain two fused ranking results. Then, the two fused ranking results are weighted and fused. A three-party voting mechanism is initiated in the disputed area, and a heterogeneous ranking method is used to vote to obtain the final ranking result.

[0007] As an alternative implementation method, the process of quantifying the contribution weight of each section force to the stress distribution change based on the acquired data includes: intelligently learning the complex coupling law between the section force element components and the stress response of key components based on the decision tree algorithm, and introducing the importance measure of OOB disturbance characteristics to quantify the contribution weight of each section force to the stress distribution change, forming an interpretable force element transmission path sensitivity ranking.

[0008] As an alternative implementation method, the process of ranking the degree of work condition hazard based on the key force elements using various algorithms includes: ranking the degree of work condition hazard using methods such as multiple regression, principal component analysis, entropy weight method, hazard weight prediction, K-means, deep learning, and random forest prediction, wherein: Among them, an anomaly measurement model for multivariable force element working conditions is constructed based on principal component analysis. The main variation patterns of multidimensional cross-sectional forces are extracted through orthogonal transformation, and the contributions of each principal component are fused based on eigenvalue weighting to achieve a global ranking of the degree of working condition hazard. Based on the information entropy theory, a cross-sectional force weight evaluation system is constructed. By quantifying the dispersion of force data at each cross section, its influence is inferred in reverse, and the comprehensive anomaly degree of multi-dimensional force element working conditions is determined. The hazard weight prediction method establishes a cross-sectional force weighted fusion model for constructing correlation-based analysis. By using pre-set cross-sectional force weight coefficients, it linearly aggregates multi-dimensional force elements into a single hazard index, and quantifies the degree of hazard under the synergistic action of multiple cross-sectional forces through weighted summation. The multiple regression method trains a multiple linear regression model based on labeled working condition data to predict the risk coefficient of all working conditions. The K-means prediction method uses standardized cross-sectional force data, semi-supervised K-means clustering, and distance weight calculation to intelligently determine the hazard level of a work condition. The deep learning prediction method loads operating condition feature data and labels, performs standardized preprocessing on common operating condition data, and integrates deep feature extraction layers and Gaussian kernel support vector regression mechanism to construct a dual-modal prediction model to predict the continuous hazard coefficient of the entire operating condition. The random forest prediction method is used to construct a random forest prediction model for the probability of working condition hazards based on deep neural networks. It determines the probability prediction value of working condition hazards based on feature data and a working condition label table.

[0009] As an alternative implementation method, the process of performing correlation analysis on all results and selecting the final multiple algorithms based on the correlation analysis results includes: performing correlation analysis on all sorting results, selecting algorithms whose correlation analysis results are greater than a set value, and forming a heterogeneous algorithm.

[0010] As an alternative implementation method, heterogeneous algorithms are used to process the data after perturbation, and the ranking performance of each algorithm under the same perturbation conditions is cross-compared. The process of secondary screening includes: ensuring that there is no consistency deviation exceeding the allowable deviation threshold in all test conditions through global comparison, and then scanning the original data matrix row by row, focusing on a single row of data each time to apply random perturbation, generating multiple sets of random interference factors for each row independently, and limiting the perturbation amplitude to occur uniformly within a preset symmetrical interval. Organize the output structure according to the data row number dimension, and establish the positional mapping relationship between the perturbation data and the source data; The corresponding sorting methods are applied in parallel to all perturbed data, and the sorting performance of different algorithms under the same perturbation conditions is cross-compared.

[0011] As an alternative implementation method, the process of fusing the risk ranking results obtained from the evaluation using a genetic optimization fusion algorithm includes: randomly generating an initial population containing multiple algorithm configuration information, and setting weight parameters, mutation intensity base values ​​and their adjustment period parameters; For each individual, the algorithm consensus evaluation degree and physical constraint satisfaction degree are calculated synchronously, and the two scores are fused by dynamic weights that are updated linearly in algebra to form the core fitness index; Based on fitness scores, a tournament selection strategy is used to screen parent individuals, prioritizing individuals with the best comprehensive evaluation of algorithmic consensus and physical constraint satisfaction to participate in subsequent crossover operations; The mutation intensity is automatically adjusted according to the number of generations of evolution and a preset sine curve: the global exploration ability is enhanced during peak periods, the local optimization ability is strengthened during trough periods, and the phase parameter is compensated according to the population convergence speed. A two-stage crossover is performed on the parent generation, first inheriting the continuous gene segment with the highest fitness from both parents, and then scanning and repairing the key feature sites missing in the offspring to ensure that there is no loss of key information in gene recombination. Eliminate individuals who violate the constraints of the predicted dangerous relationship, and give high-potential individuals K opportunities for repair; At the end of each generation, the overlap between the current best solution and the historical Pareto front, the zero-symmetric feature of the fitness window standard deviation across generations, and the rate of decay of fitness improvement momentum between generations are detected simultaneously. Evolution is terminated only when all three conditions are met. Extract the optimal equilibrium solution from the final generation Pareto front.

[0012] As an alternative implementation method, the process of fusing the risk ranking results obtained by the evaluation using an adversarial optimization model includes: calculating the global ranking trend correlation and micro-ranking conflict intensity among the ranking groups of heterogeneous algorithms in parallel. Based on the consensus index, a negative credibility active filtering rule is designed to identify and zero out the decision weights of low-quality algorithms; a non-uniform distribution scheme that conforms to the strength of group consensus is generated through a dynamic normalization function. Establish order-preserving transformation rules from discrete sorting to continuous space, fully preserve the structural relationship of the original algorithm ranking, and simultaneously analyze the engineering physical constraints into an inequality constraint system on continuous dimensions; An optimization model containing continuous hazard variables and constraints is constructed. The main optimization layer of the optimization model uses the interior point method to perform core solution space search, and the auxiliary correction layer uses gradient projection technology to directionally repair constraint violation points. By integrating dynamic weighting, rule mapping, and two-stage solution results, the consistency of weight allocation and consensus degree, as well as the compliance of the scanning solution vector with engineering constraints, is verified, an interpretable and traceable decision report is generated, and finally, the working condition ranking result is formed.

[0013] As an alternative implementation method, the process of obtaining two fused ranking results and then weighting and fusing the two fused ranking results includes: extracting the rank displacement difference features based on the ranking sequences generated by the two algorithms themselves. The sliding window technique is used to scan the sorted sequence, calculate the statistical divergence of the position displacement within the window in real time, set a detection threshold, and when the divergence value abnormally exceeds the set threshold, it is determined to be a real disputed segment; otherwise, it is determined to be a normal area. In regular areas, a standard fusion strategy is used, based on the adaptive weight allocation coefficient analysis of genetic algorithms and adversarial models, and the weighted and fused ranking results are adopted; in disputed areas, a three-party voting mechanism is initiated, using heterogeneous ranking methods to vote, so as to achieve intelligent adjudication that is accurately adapted to the scenario.

[0014] An aircraft aviation risk assessment system, comprising: The force element acquisition module is configured to divide the aircraft into regions, extract the FBD section under static equilibrium constraints in the region where the flight hazard exceeds the set value, and obtain the six-component force elements, resultant internal force and resultant moment of the section boundary. The key force element extraction module is configured to quantify the contribution weight of each section force to the stress distribution change based on the acquired data, sort them according to the contribution weight, and determine the key force elements; The initial selection module for the sorting algorithm is configured to use multiple different algorithms to sort the degree of hazard of the working conditions based on the key force elements, perform correlation analysis on all results, and select multiple final algorithms based on the correlation analysis results to form a preliminary heterogeneous algorithm. The sorting algorithm secondary selection module is configured to process the perturbed data using heterogeneous algorithms, cross-compare the sorting performance of each algorithm under the same perturbation conditions, perform secondary screening, and form the final heterogeneous algorithm. The parallel module for the ranking algorithm is configured to perform aircraft aviation risk assessments using the final heterogeneous algorithms to obtain the risk ranking results after the assessment. The ranking result fusion module is configured to use a genetic optimization fusion algorithm and an adversarial solution optimization model to fuse the risk ranking results obtained from the evaluation, resulting in two fused ranking results. Then, the two fused ranking results are weighted and fused. In disputed areas, a three-party voting mechanism is initiated, using a heterogeneous ranking method to vote, and the final ranking result is obtained.

[0015] An electronic device includes a memory and a processor, as well as computer instructions stored in the memory and running on the processor, wherein the computer instructions, when executed by the processor, perform the steps in the method described above.

[0016] Compared with the prior art, the beneficial effects of the present invention are as follows: This invention achieves continuous risk classification capability. By establishing a dynamic sorting mechanism driven by cross-sectional force, it breaks through the limitations of binary judgment in traditional safety assessment, realizes continuous quantitative identification of subcritical risk states of aircraft structures, and effectively captures the evolution characteristics of potential gradient risks.

[0017] This invention constructs a multi-algorithm collaborative optimization framework, which dynamically coordinates the conflicting outputs of multiple computational models based on genetic adversarial fusion technology. By introducing constraints and an automated selection method, it forms an adaptive risk decision consensus, significantly enhancing the reliability of risk assessment in complex coupled load scenarios.

[0018] This invention establishes a rapid response interface for incremental operating conditions. The developed dynamic risk network architecture can automatically call database models, reuse end-to-end analysis processes for newly accessed operating conditions, achieve efficient assessment, and output comprehensive analysis reports.

[0019] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, preferred embodiments are described below in detail with reference to the accompanying drawings. Attached Figure Description

[0020] The accompanying drawings, which form part of this invention, are used to provide a further understanding of the invention. The illustrative embodiments of the invention and their descriptions are used to explain the invention and do not constitute an improper limitation of the invention.

[0021] Figure 1 This is a schematic diagram of an aircraft aviation risk assessment method provided in one embodiment. Detailed Implementation

[0022] The present invention will be further described below with reference to the accompanying drawings and embodiments.

[0023] It should be noted that the following detailed description is illustrative and intended to provide further explanation of the invention. Unless otherwise specified, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains.

[0024] It should be noted that the terminology used herein is for the purpose of describing particular embodiments only and is not intended to limit the scope of exemplary embodiments according to the invention. As used herein, the singular form is intended to include the plural form as well, unless the context clearly indicates otherwise. Furthermore, it should be understood that when the terms "comprising" and / or "including" are used in this specification, they indicate the presence of features, steps, operations, devices, components, and / or combinations thereof.

[0025] Where there is no conflict, the embodiments and features described in this application may be combined with each other.

[0026] Example 1 An aircraft aviation risk assessment method, such as Figure 1 As shown, it includes the following steps: The aircraft is divided into regions, and the FBD section under static equilibrium constraints is extracted for the region where the flight hazard exceeds the set value. The six force elements of the section boundary, as well as the resultant internal force and resultant moment, are obtained. Based on the acquired data, the contribution weight of each section force to the stress distribution change is quantified, and the forces are sorted according to their contribution weights to determine the key force elements. Using multiple different algorithms, the degree of danger of the working conditions is ranked based on the key force elements. Correlation analysis is performed on all results. Based on the correlation analysis results, multiple algorithms are selected to form a preliminary heterogeneous algorithm. The data after perturbation is processed using heterogeneous algorithms. The ranking performance of each algorithm under the same perturbation conditions is cross-compared, and a second selection is performed to form the final heterogeneous algorithm. The aircraft aviation risk assessment was performed using the final heterogeneous algorithm, and the risk ranking results were obtained after the assessment. The risk ranking results obtained from the assessment are fused using a genetic optimization fusion algorithm and an adversarial solution optimization model, respectively, to obtain two fused ranking results. Then, the two fused ranking results are weighted and fused. A three-party voting mechanism is initiated in the disputed area, and a heterogeneous ranking method is used to vote to obtain the final ranking result.

[0027] The following is a detailed explanation of each step: Based on the aircraft regional risk analysis strategy, the structural regions are divided using the Hypermesh platform. For high-risk flight areas (such as wing root sections, vertical tail root sections, etc.), FBD sections under static equilibrium constraints are extracted, and the six force elements of the section boundary, as well as the resultant internal forces and resultant moments, are used as the core features of risk transmission.

[0028] Based on the stress values ​​from the coarse-grid model, a cross-sectional force-local stress nonlinear mapper using random forest regression is further established using cross-sectional force data. A decision tree ensemble architecture is used to intelligently learn the complex coupling relationship between cross-sectional force components and the stress response of key components. An OOB (Out-of-Band) perturbation feature importance metric is introduced to quantify the contribution weight of each cross-sectional force to stress distribution changes, forming an interpretable force transmission path sensitivity ranking. This method constructs a force-stress correlation model from the perspective of the main cross-sectional force pathway, providing quantitative results of force contribution for subsequent risk quantification.

[0029] The random forest regression model establishes a nonlinear mapping relationship between cross-sectional internal forces and local stress responses. By calculating the incremental error of the out-of-bag after randomly permuting the cross-sectional internal force characteristics, the impact of individual cross-sectional internal force disturbances on overall prediction performance is evaluated, directly revealing their statistical correlation with the forces. The decision tree splitting process automatically captures the complex coupling between bending-shear-torsion combined loads and stress responses, while its feature sampling-based splitting mechanism implicitly analyzes the nonlinear enhancement / suppression effect of the synergistic action of multiple cross-sectional internal forces on the forces.

[0030] Data mapping relationship construction: ; Predicted stress value, T Total number of decision trees ， For single-tree prediction functions, x This is the internal force vector.

[0031] The increment of the model prediction error after randomly replacing the internal forces of the cross section quantifies the strength of the nonlinear interaction:

[0032] in, It is the mean square error of the t-th tree after replacing the internal forces of the i-th section. This is the baseline error. The importance score represents the internal force of the i-th cross section, and N_t is the total number of decision trees in the random forest.

[0033] Increased importance and stability: ; The importance of the k-th training iteration; then proceed... After normalization, the contribution ranking is obtained.

[0034] Based on the cross-sectional force dataset extracted by FBD, a working condition classification mechanism is constructed: by defining threshold rules for cross-sectional force distribution characteristics (for example, in this embodiment, the eight cross-sectional force elements are sorted according to the contribution of the stress value analysis in the previous step, and the top six are selected. It is assumed that the top six key cross-sectional forces are all located in the top 20% percentile of the entire working condition set and are marked as dangerous working conditions, and are all located in the bottom 20% percentile and are marked as safe working conditions), the efficient and automated labeling of extreme working conditions is achieved.

[0035] To address the modeling challenges posed by high-dimensional, strongly correlated data in multivariate force analysis, this embodiment constructs an algorithm selection architecture adapted to physical mechanisms. By systematically diagnosing the collinearity sensitivities of multiple regression, PCA, entropy weighting, K-means, deep learning, and random forest methods, it designs differentiated anti-interference paths: multiple regression focuses on suppressing the risk of coefficient estimation distortion caused by variable correlation; PCA optimizes the separation efficiency of principal components with independent physical meaning; entropy weighting enhances the robustness of weight allocation under noise interference; K-means improves the cluster center dominance of key discriminant dimensions; deep learning optimizes the information entropy compression efficiency of feature layers; and random forest avoids computational redundancy on strongly correlated variables. Experiments verify the differences in analytical characteristics of different algorithms for key cross-sectional forces / full cross-sectional forces, and an original multi-method adaptive switching rule is established under the dual constraints of physical interpretability and computational robustness. This architecture fundamentally solves the systemic failure risk of traditional single methods in multicollinearity scenarios.

[0036] PCA Prediction Method: Based on Principal Component Analysis (PCA), an anomaly quantification model for multivariable force element working conditions is constructed. The main variation patterns of multidimensional cross-sectional forces are extracted through orthogonal transformation, and the contributions of each principal component are weighted and fused based on eigenvalues ​​to achieve a global ranking of the working condition's hazard level. The algorithm projects extreme force element patterns in the high-dimensional cross-sectional force space onto the low-dimensional principal component space and quantifies the working condition's hazard level by calculating weighted Euclidean distance.

[0037] Entropy weight prediction method: Based on the information entropy theory, a cross-sectional force weight evaluation system is constructed. By quantifying the dispersion of force data at each cross-section, its influence is inferred in reverse, revealing the comprehensive anomaly degree of multi-dimensional force element conditions. The core idea is that when the force data at a certain cross-section exhibits orderliness (low information entropy), it contains richer effective discriminative information and should be given a higher weight; while the cross-sectional forces with highly disordered data (high information entropy) have lower discriminative power and their weights are correspondingly reduced.

[0038] Hazard weighting prediction method: This program constructs a cross-sectional force weighted fusion model based on correlation analysis. Through pre-set cross-sectional force weight coefficients, it linearly aggregates multi-dimensional force components into a single hazard index. Its core idea is to quantify the degree of hazard under the synergistic effect of multiple cross-sectional forces by weighted summation. The weights reflect the differences in the contribution of each cross-sectional force to the overall hazard, and the fluctuation of high-weight cross-sectional forces has a significant amplification effect on the hazard index.

[0039] Multiple Regression Prediction Method: This method trains a multiple linear regression model based on labeled work condition data to predict the hazard coefficients of all work conditions. This program constructs a multiple linear regression model to predict the hazard coefficients of work conditions. Its main processes include: aligning two datasets using work condition numbers; converting text labels to binary labels and normalizing the dataset; creating a 5-fold cross-validation framework to evaluate model performance, training sub-models in each iteration, calculating the mean squared error after denormalizing the prediction results; subsequently training the final model on the complete training set, systematically evaluating key performance indicators such as mean squared error (MSE), root mean square error (RMSE), mean absolute error (MAE), and coefficient of determination (R²); finally, using the optimized model to predict the hazard weight coefficients of all work conditions. The entire process realizes an end-to-end analysis chain from data preprocessing, model training and validation to prediction output.

[0040] K-means Prediction Method: This embodiment constructs a semi-supervised clustering evaluation system that integrates label priors and data-driven approaches. Through standardized cross-sectional force data (z-score), semi-supervised K-means clustering, and distance weight calculation, it achieves intelligent determination of the hazard level of the work condition. A knowledge fusion mechanism is implemented, prioritizing the retention of user-labeled data and intelligently supplementing unlabeled data through clustering. Based on dynamic distance weights, the degree to which the work condition deviates from the cluster center is quantified using Euclidean distance, directly representing the hazard level. Simultaneously, discrete hazard levels and continuous hazard weight coefficients are output.

[0041] Deep learning prediction method: A hazard coefficient prediction system based on deep support vector machine (D-SVM) was constructed. This framework innovatively integrates deep feature extraction layers with Gaussian kernel support vector regression to build a prediction model. Automatic hyperparameter optimization enhances the model's generalization ability, enabling accurate prediction of continuous hazard coefficients under all operating load conditions.

[0042] The Gaussian kernel function is:

[0043] The prediction function is:

[0044] Where γ represents the kernel width parameter, and SV represents the support vector set. and Here, x is a Lagrange multiplier, and x is the sample to be predicted. When measuring sample similarity, b is a bias term.

[0045] Random Forest Prediction Method: A random forest prediction system based on deep neural networks for predicting the probability of working condition hazards was constructed. By reading feature data and working condition label tables, data alignment, feature standardization, and label binarization were first completed. Then, an 8-layer fully connected neural network architecture was designed, and ReLU activation and Dropout regularization were introduced to suppress overfitting. The Adam optimizer was used for training, and finally, the probability prediction value of the "hazard" category (continuous value in the range of 0-1) was output on all working conditions, realizing the quantitative assessment of the health status of the transmission system.

[0046] Correlation analysis is performed on all ranking results. Stress correlation analysis focuses on local similarity within ranking intervals, rather than overall consistency. By examining the overlap of working conditions within the same ranking segment, the stability of the ranking method within a specific ranking range is tested, which is suitable for evaluating the ranking algorithm's performance in terms of local consistency. Suitable algorithms, i.e., those with similarity greater than a set value, are selected to form a heterogeneous algorithm ranking layer.

[0047] Next, we introduce a perturbation.

[0048] Before disturbance analysis, the validity of test conditions is confirmed: global comparison is used to ensure that there is no consistency deviation exceeding the allowable deviation threshold for all test conditions, thus establishing a reliable benchmark for subsequent disturbance experiments.

[0049] This embodiment employs a row-independent perturbation mechanism: the original data matrix is ​​scanned row by row, with random perturbation applied to only a single row at a time. Strict row-limiting operations are used to ensure that the perturbation effect is precisely controlled within the target row.

[0050] Dynamic parameter boundary control is implemented, generating multiple sets of random disturbance factors independently for each row, strictly limiting the disturbance amplitude to occur uniformly within a preset symmetrical interval. The key breakthrough lies in the fact that non-target row data completely inherits the original values ​​during the disturbance process, maintaining the dimensional stability of the overall matrix.

[0051] The output structure is organized by row number, establishing a strict positional mapping between the perturbation data and the source data. Row coordinate binding technology ensures that each perturbation result can be accurately traced back to the original data row.

[0052] Multiple sorting exposure experiments were conducted, applying five sorting methods in parallel to all perturbed data. By cross-referencing the sorting performance of different algorithms under the same perturbation conditions, the noise sensitivity defects of specific algorithms were proactively exposed.

[0053] After multiple sorting exposure experiments revealed the critical perturbation sensitive intervals of specific algorithms, the system immediately initiated the sharp point deep perception process: First, through feature analysis of abnormal fluctuations in sorting, the key numerical boundaries that caused algorithm instability were accurately identified; then, optimization was implemented for the identified sensitive feature space—adaptively switching the anti-disturbance parameter configuration for vulnerable intervals and deploying compensators to fine-tune sorting deviations (if this cannot be achieved, the perturbation-generating algorithm can be excluded); this optimization process is dynamically limited to triggering within the sensitive feature space, and the algorithm maintains its original operating logic within the safe interval, ensuring that while improving local anti-disturbance, the global efficiency of the algorithm is maintained. Finally, all algorithms optimized by the interval participate in the subsequent fusion process, ultimately forming a perturbation-resistant heterogeneous algorithm sorting.

[0054] This step involves a deep optimization of the sorting method, which increases computation time. Therefore, if the operating conditions are within a disturbance-sensitive range, complex calculations and sorting are performed; otherwise, the original sorting method can be used. Since the disturbance is caused by a specific algorithm, if optimization is not possible, that algorithm can be excluded before proceeding with the sorting.

[0055] A constraint-based aircraft condition comparison and identification method based on flight parameter comparison analysis is introduced. This method achieves safety boundary calibration through condition construction constrained by physical information and a reverse inference mechanism for safety factors. Specifically, the following technical approach is adopted: Paired condition groups are constructed that differ significantly only in a single target parameter (e.g., angle of attack), strictly ensuring the consistency of non-target flight parameters (similarity ≥ 95%). When the difference in risk factor exceeds a preset threshold, the differing parameter is determined to be a key risk factor. When the risk factors are significantly different (e.g., 8° angle of attack in condition 1 vs. 18° angle of attack in condition 2), condition 1 is determined to be safer than condition 2.

[0056] Next, we will perform dual-channel fusion.

[0057] The first step is fusion based on a genetic optimization fusion algorithm, which includes the following steps: (1) Population initialization and parameter setting An initial population containing various algorithm configuration information is randomly generated, and dynamic weight parameters (the initial weight ratio of algorithm consensus to physical constraints), mutation intensity base value and its adjustment period parameter are set to provide a basic configuration for adaptive evolution.

[0058] (2) Dynamic bimodal fitness assessment For each individual, the consensus evaluation degree of the algorithm (measuring the coordination of the fusion scheme with the prediction results of heterogeneous algorithms such as PCA anomaly quantification and cross-sectional force entropy weight evaluation) and the physical constraint satisfaction degree (verifying whether the prediction results meet the constraints) are calculated synchronously. The two scores are fused by dynamic weights that are updated linearly in algebra to form the core fitness index.

[0059] (3) Parental individual selection A tournament selection strategy is used to select parent individuals based on fitness scores, prioritizing individuals with the best comprehensive evaluation of algorithmic consensus and physical constraint satisfaction to participate in subsequent crossover operations.

[0060] (4) Time-varying strategy mutation operation The mutation intensity is automatically adjusted according to the number of generations of evolution and a preset sine curve: the global exploration capability is enhanced during peak periods, the local optimization capability is strengthened during trough periods, the phase parameter is compensated in real time according to the population convergence speed, and the exploration and utilization capabilities are dynamically balanced.

[0061] (5) Fragment Recombination-Missing Repair Crossover Two-stage crossover is performed on the parent generation: first, the continuous gene segment with the highest fitness of both parents (such as a sequence configured by a specific algorithm) is inherited, and then the offspring are scanned to repair key feature sites missing (such as the air velocity influence factor) to ensure that there is no loss of key information in gene recombination.

[0062] (6) Constraint perception of elites After merging the previous generation of elites with the intermediate population generated by the current operation, a physical rule verification is enforced: individuals who violate the constraints of dangerous relationship prediction are directly eliminated, and high-potential individuals are given K repair opportunities to balance rule compliance and innovative breakthroughs.

[0063] (7) Triple fusion convergence criterion At the end of each generation, three signals are detected simultaneously: the overlap between the current optimal solution and the historical Pareto front, the zero-symmetric characteristic of the fitness window standard deviation across generations, and the rate of decay of fitness improvement momentum between generations. Evolution terminates only when all three meet the threshold conditions.

[0064] (8) Output optimization fusion scheme The optimal equilibrium solution is extracted from the final generation Pareto front. The analytical generation algorithm integrates the weight configuration (e.g., 32% principal component analysis / 41% cross-sectional force entropy weight) and the constraint verification satisfaction rate (e.g., condition 1 is safer than condition 2), and outputs a fusion model with dual verifiability.

[0065] Then, the fusion of adversarial optimization models includes the following steps: (1) Dynamic quantization mechanism of consensus features of multi-source algorithms By parallel computing the global ranking trend correlation and micro-ranking conflict intensity among heterogeneous algorithm ranking groups, a consensus index weight representing the credibility of individual algorithms is generated. This mechanism accurately captures the inherent reliability differences in algorithm cluster decision-making, distinguishes itself from the inherent pattern of equal weighting of algorithms in traditional methods, lays the theoretical foundation for establishing a non-uniform weighting system, and fundamentally breaks through the homogeneous weighting bottleneck in the field of ranking fusion.

[0066] (2) Noise suppression-oriented adaptive weight allocation mechanism Based on the consensus index, a negative credibility active filtering rule is designed to automatically identify and zero out the decision weights of low-quality algorithms; a non-uniform distribution scheme that conforms to the strength of group consensus is generated through a dynamic normalization function, so that high consensus algorithms can gain enhanced discourse power.

[0067] (3) Domain knowledge-driven physical constraint continuous mapping technology Establish order-preserving transformation rules from discrete sorting to continuous space, fully retain the structural relationship of the original algorithm ranking, and simultaneously analyze the engineering physical constraints into an inequality constraint system on continuous dimensions.

[0068] (4) Robust two-stage solution architecture with strong physical rule protection An optimization model containing continuous risk variables and constraints is constructed, and a pioneering main-auxiliary two-stage collaborative solution protocol is proposed: the main optimization layer uses the interior point method to perform core solution space search, and the auxiliary correction layer uses gradient projection technology to directionally repair constraint violation points.

[0069] (5) Self-verifying full-process closed-loop output system By deeply integrating dynamic weighting, rule mapping, and two-stage solution results, a triple-review verification mechanism is established at the output terminal—auditing the consistency of weight allocation and consensus, scanning the compliance of solution vectors with engineering constraints, and generating an interpretable and traceable report on decision-making, ultimately forming a working condition ranking result with self-verification capabilities.

[0070] Based on the sorting sequences generated by the two algorithms, this embodiment extracts the ranking displacement difference features. By quantifying the magnitude of changes in the working condition position and identifying significant conflict markers of the previous working condition, it achieves the modeling of essential differences that rely solely on the sorting results.

[0071] A sliding window technique is used to scan the sorted sequence and calculate the statistical divergence of the rank displacement within the window in real time. An adaptive detection threshold is set; when the divergence value abnormally exceeds the set threshold, it is determined to be a real disputed segment; otherwise, it is determined to be a normal region.

[0072] Construct a hierarchical decision-making system: In regular areas, a standard fusion strategy is used, based on the adaptive weight allocation coefficient analysis of genetic algorithms and adversarial models, and the weighted and fused ranking results are adopted; in disputed areas, a three-party voting mechanism is initiated, using heterogeneous ranking methods to vote, so as to achieve intelligent adjudication that is accurately adapted to the scenario.

[0073] During the scoring standardization process, a safety strategy is dynamically injected: when a high-risk conflict item is identified, a safety margin offset compensation is made in a conservative direction.

[0074] Design a four-dimensional traceability system: By recording four core dimensions—disputed working conditions, weight allocation parameters, production in disputed areas, and the location where compensation takes effect—a technical closed loop that can be verified throughout the entire process is achieved.

[0075] Example 2 An aircraft aviation risk assessment system, comprising: The aircraft cross-section force element acquisition and analysis module specifically includes: The force element acquisition module is configured to divide the aircraft into regions, extract the FBD section under static equilibrium constraints in the region where the flight hazard exceeds the set value, and obtain the six-component force elements, resultant internal force and resultant moment of the section boundary. The key force element extraction module is configured to quantify the contribution weight of each section force to the stress distribution change based on the acquired data, sort them according to the contribution weight, and determine the key force elements; The dynamic sorting mechanism module specifically includes: The initial selection module for the sorting algorithm is configured to use multiple different algorithms to sort the degree of hazard of the working conditions based on the key force elements, perform correlation analysis on all results, and select multiple final algorithms based on the correlation analysis results to form a preliminary heterogeneous algorithm. The sorting algorithm secondary selection module is configured to process the perturbed data using heterogeneous algorithms, cross-compare the sorting performance of each algorithm under the same perturbation conditions, perform secondary screening, and form the final heterogeneous algorithm. The parallel module for the ranking algorithm is configured to perform aircraft aviation risk assessments using the final heterogeneous algorithms to obtain the risk ranking results after the assessment. The ranking result fusion module is configured to use a genetic optimization fusion algorithm and an adversarial solution optimization model to fuse the risk ranking results obtained from the evaluation, resulting in two fused ranking results. Then, the two fused ranking results are weighted and fused. In disputed areas, a three-party voting mechanism is initiated, using a heterogeneous ranking method to vote, and the final ranking result is obtained.

[0076] Some embodiments also include a dynamic risk network architecture interface module. After the test condition enters the interface, it first enters the enhanced sorting and fusion of the dynamic sorting mechanism module in interface 1. (If it cannot be achieved through sharp point optimization, then an anti-disturbance optimization sorting method is formed by deleting disturbance-related algorithms. In interface 1, it first searches for cases where the difference between any factor and the existing condition is less than the threshold. If such a case exists, it enters the anti-disturbance optimization sorting method by deleting disturbance-related algorithms. Dissimilar conditions are processed through the standard sorting network model process.) All conditions are then merged into interface 2 to implement multi-threshold risk stratification—a four-level risk topology network is generated through dynamic mutual exclusion verification of the risk matrix sorted by heterogeneous algorithms. Then it flows into the intelligent diagnostic chain, enters interface 3 for similar condition evaluation, interface 4 for condition risk prediction, and interface 5 for condition outlier analysis.

[0077] Interface 1 is module 2 mentioned above. It automatically ranks the working conditions based on the model library by outputting cross-sectional force element data, and generates the ranking of the assessment working conditions in the model.

[0078] Interface 2 Multi-threshold Risk Stratification (1) Risk matrix construction The system receives the ranking results of independent risk assessment algorithms from heterogeneous sources for each working condition. By integrating multi-source ranking data, it automatically constructs a risk location matrix—the matrix's rows correspond to different assessment algorithms, its columns cover all tested working conditions, and each cell records the specific algorithm's ranking of the hazard level for a specific working condition. During this process, a cross-algorithm fluctuation monitoring module is built-in to automatically identify working conditions where there are significant discrepancies in the assessment conclusions of different algorithms.

[0079] (2) Hierarchical decision logic A hierarchical decision tree is used to implement risk zoning: First, it checks whether all algorithms consistently determine that a certain working condition is within the critical danger threshold range. If the condition is met, it is classified into the highest risk red zone. Working conditions that do not meet the standard enter the next level of judgment. If at least one algorithm places it in the critical danger range, it is classified into the gray zone warning zone. The remaining working conditions continue to be screened and sorted. Individuals with drastic fluctuations and potential high-risk tendencies are included in the yellow zone observation list. Finally, working conditions that are all determined to be in the safe range by all algorithms are marked as blue zone, and the rest are classified into the intermediate transition zone.

[0080] (3) Mutual exclusion verification mechanism Strict isolation rules are established between red and gray zones: when a certain operating condition is pushed into the red zone by some algorithms but excluded by others, a multi-dimensional arbitration procedure is initiated to ensure that there is no overlap between the two high-risk areas. Simultaneously, a dynamic buffer adjustment strategy is set up: if a gray zone operating condition exhibits absolutely safe characteristics in most algorithms, it is automatically downgraded to the yellow zone; neighbor monitoring is implemented for red zone boundary operating conditions, forming a gradient monitoring network. The partition boundaries are adaptive, with the judgment threshold being fine-tuned in real time based on the proportion of fluctuating operating conditions in the system, and the stability of risk partitions is predicted through simulated data disturbances.

[0081] Interface 3 Similar Operating Condition Evaluation (1) Construction of cross-sectional force element eigenvectors The system extracts the force element data of the tested aircraft under working conditions at the target section, and simultaneously loads the stress element vectors of all working conditions in the model library into the memory space to construct a unified and computable force element feature matrix.

[0082] (2) Key force element mode identification Based on the force element vector distribution pattern of the tested working condition, the dominant force element type is automatically identified: when the resultant torque M exceeds three orders of magnitude of the resultant force F, it is determined to be a torque-dominant working condition; when a component force in a certain direction (such as Fz) accounts for more than 85% of the resultant force F, it is marked as an axial-dominant working condition. This identification result will dynamically adjust the weighting strategy for subsequent similarity calculations.

[0083] (3) Multi-dimensional distance synchronous calculation Three parallel computation threads are initiated: the first thread calculates the Euclidean distance of the components, focusing on evaluating the differences in force elements of each degree of freedom; the second thread calculates the Manhattan distance of the resultant force space, focusing on the absolute deviation between the resultant force F and the resultant moment M; the third thread calculates the cosine similarity of the force element distribution, analyzing the similarity of the proportional relationships of the force elements in each component. The three computation threads share the same vector standardization processing results.

[0084] (4) Distance fusion and exponential generation Based on the force element modal identification results, three types of distances are dynamically fused: for torque-dominated working conditions, the component Euclidean distance is assigned the highest weight; for axial-dominated working conditions, the weight of the resultant force spatial distance is increased. The fused distance value is converted into a cross-sectional similarity index (SSI) in the 0-1 interval using an S-shaped function, and a working condition is considered valid when the SSI ≥ 0.7.

[0085] (5) Initial screening and ranking of historical operating conditions Sort historical operating conditions in the model library from highest to lowest SSI value, and select the top 20 candidate objects. Perform force element path verification on operating conditions with SSI in the critical range: check the correlation between the force element change curves of its six degrees of freedom and the measured operating condition, and eliminate spurious similar terms with extremely divergent trends.

[0086] (6) Multi-level result output control Design a progressive output strategy: force the output of the benchmark matching case with the highest SSI (at least 1 case); expand the output of effective similar groups (up to 4 cases with SSI ≥ 0.7), and if the similarity is less than 0.7, output the top 4 cases with the highest similarity; specially retain potential similar items (up to 2 cases with SSI ≥ 0.65 and force path matching greater than the threshold).

[0087] Interface 4 Operating Condition Risk Prediction (1) Risk section benchmark generation First, the system divides the model operating conditions into three distinct sections: the front low-risk zone contains the safest operating condition samples, the middle transition zone contains the warning operating conditions that require attention, and the rear high-risk zone gathers dangerous operating conditions. Based on the similar operating condition clusters returned by Interface 3, the system constructs a static distribution sequence of their historical risk values ​​in ascending order to confirm the risk attributes of each similar cluster.

[0088] (2) Precise positioning under single-point working conditions The system generates a three-segment scale (low, medium, and high) based on the risk distribution of similar working condition clusters. It calculates the comprehensive risk value of the current working condition through a dynamic weighted model (calibrated using its force element characteristics and the risk patterns of the 20 most similar samples). Based on the percentile of this value in the risk spectrum of similar clusters, it achieves spatial positioning and realizes risk prediction for the test working condition according to its respective risk attributes.

[0089] (3) Engineering interpretability closed loop The analysis extracts historical operating conditions similar to the evaluation conditions to identify any situations requiring structural optimization. The output includes a location map of the operating conditions within similar clusters and a risk description of the historical operating conditions.

[0090] Interface 5 operating condition anomaly analysis Increased sensitivity to erythrocytes Interface 5 constructs a feature-level safety boundary system, continuously scanning the proximity of the force element characteristics of the test condition to the historical extreme value range of the test condition. When any key force element (such as the wingtip torsional moment) is identified as approaching the historical extreme value range of the test section, the diagnostic process is automatically triggered. (1) Feature tracing: Highlight the abnormal force element and indicate the extent to which it deviates from the historical normal range. If it reaches or exceeds the extreme value range, trigger the analysis and diagnosis. If it is close to the extreme value range, trigger the diagnosis.

[0091] (2) Historical reference closed loop: associate the historical extreme value case library of the feature, output its analysis results, and form a decision chain from feature anomaly to historical reference.

[0092] Based on the evaluation of five interfaces, if interface 1 ranks below the danger threshold, interface 2 is in the red zone, interface 4 is assessed as high-risk, and interface 5 has a certain characteristic value in an extreme state, further evaluation is required. Note that interface 2 is in the gray zone and needs to be added to the watchlist. If interface 5 in the gray zone triggers a diagnostic check (or if other interfaces are assessed as risky), further evaluation is recommended. Meanwhile, interface 2 is in the yellow zone and the same handling method applies; further decisions need to be made based on the situation of other interfaces. All interface information can be adjusted according to different objective conditions such as machine model and environment, and the utilization of interface information can also be analyzed on a case-by-case basis.

[0093] This interface design ensures that the following output format will be generated after the test condition is entered: Interface 1 [Risk Location] is located at the 136th position in the historical working condition risk ranking.

[0094] Interface 2 [Region Division] Red Zone.

[0095] Interface 3 [Similarity Evaluation] The most similar working condition in the history is: Working Condition 246, similarity: 0.785; the second most similar working condition is: Working Condition 136, similarity: 0.458; the third most similar working condition is: Working Condition 149, similarity: 0.421; the fourth most similar working condition is: Working Condition 62, similarity: 0.356.

[0096] Interface 4 [Risk Assessment] The assessment condition is dangerous. Similar condition 246 has been analyzed and found to have risks, requiring optimization.

[0097] Interface 5 [Anomaly Diagnosis] An abnormal force element exists in the test condition, reaching the extreme value range, Fz: 16572000N (or: no abnormal value). The extreme value test condition has been analyzed and there is a risk, which needs to be optimized.

[0098] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of one or more computer-usable storage media (including, but not limited to, disk storage, etc.) containing computer-usable program code. CD - ROM It takes the form of a computer program product implemented on (such as optical memory, etc.).

[0099] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0100] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0101] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0102] The above description is merely a preferred embodiment of the present invention and is not intended to limit the invention. Various modifications and variations can be made to the present invention by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made by those skilled in the art without creative effort within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A method for assessing aviation risks of aircraft, characterized in that, Includes the following steps: The aircraft is divided into regions, and the FBD section under static equilibrium constraints is extracted for the region where the flight hazard exceeds the set value. The six force elements of the section boundary, as well as the resultant internal force and resultant moment, are obtained. Based on the acquired data, the contribution weight of each section force to the stress distribution change is quantified, and the forces are sorted according to their contribution weights to determine the key force elements. Using multiple different algorithms, the degree of danger of the working conditions is ranked based on the key force elements. Correlation analysis is performed on all results. Based on the correlation analysis results, multiple algorithms are selected to form a preliminary heterogeneous algorithm. The data after perturbation is processed using heterogeneous algorithms. The ranking performance of each algorithm under the same perturbation conditions is cross-compared, and a second selection is performed to form the final heterogeneous algorithm. The aircraft aviation risk assessment was performed using the final heterogeneous algorithm, and the risk ranking results were obtained after the assessment. The risk ranking results obtained from the assessment are fused using a genetic optimization fusion algorithm and an adversarial solution optimization model, respectively, to obtain two fused ranking results. Then, the two fused ranking results are weighted and fused. A three-party voting mechanism is initiated in the disputed area, and a heterogeneous ranking method is used to vote to obtain the final ranking result.

2. The aircraft aviation risk assessment method as described in claim 1, characterized in that, Based on the acquired data, the process of quantifying the contribution weight of each section force to the stress distribution change includes: intelligently learning the complex coupling law between the section force element components and the stress response of key components based on the decision tree algorithm, and introducing the importance measure of OOB disturbance characteristics to quantify the contribution weight of each section force to the stress distribution change, forming an interpretable force element transmission path sensitivity ranking.

3. The aircraft aviation risk assessment method as described in claim 1, characterized in that, The process of ranking the hazard levels of operating conditions based on the key force elements using various algorithms includes: ranking the hazard levels of operating conditions using methods such as multiple regression, principal component analysis, entropy weight method, hazard weight prediction, K-means, deep learning, and random forest prediction, respectively. Among them, an anomaly measurement model for multivariable force element working conditions is constructed based on principal component analysis. The main variation patterns of multidimensional cross-sectional forces are extracted through orthogonal transformation, and the contributions of each principal component are fused based on eigenvalue weighting to achieve a global ranking of the degree of working condition hazard. Based on the information entropy theory, a cross-sectional force weight evaluation system is constructed. By quantifying the dispersion of force data at each cross section, its influence is inferred in reverse, and the comprehensive anomaly degree of multi-dimensional force element working conditions is determined. The hazard weight prediction method establishes a cross-sectional force weighted fusion model for constructing correlation-based analysis. By using pre-set cross-sectional force weight coefficients, it linearly aggregates multi-dimensional force elements into a single hazard index, and quantifies the degree of hazard under the synergistic action of multiple cross-sectional forces through weighted summation. The multiple regression method trains a multiple linear regression model based on labeled working condition data to predict the risk coefficient of all working conditions. The K-means prediction method uses standardized cross-sectional force data, semi-supervised K-means clustering, and distance weight calculation to intelligently determine the hazard level of a work condition. The deep learning prediction method loads operating condition feature data and labels, performs standardized preprocessing on common operating condition data, and integrates deep feature extraction layers and Gaussian kernel support vector regression mechanism to construct a dual-modal prediction model to predict the continuous hazard coefficient of the entire operating condition. The random forest prediction method is used to construct a random forest prediction model for the probability of working condition hazards based on deep neural networks. It determines the probability prediction value of working condition hazards based on feature data and a working condition label table.

4. The aircraft aviation risk assessment method as described in claim 1, characterized in that, The process of performing correlation analysis on all results and then selecting the final multiple algorithms based on the correlation analysis results includes: performing correlation analysis on all sorting results, selecting algorithms whose correlation analysis results are greater than a set value, and forming a heterogeneous algorithm.

5. The aircraft aviation risk assessment method as described in claim 1, characterized in that, The process of using heterogeneous algorithms to process the data after applying perturbation, cross-comparing the ranking performance of each algorithm under the same perturbation conditions, and performing secondary screening includes: ensuring that there is no consistency deviation exceeding the allowable deviation threshold in all test conditions through global comparison, and then scanning the original data matrix row by row, focusing on a single row of data each time to apply random perturbation, generating multiple sets of random interference factors for each row independently, and limiting the perturbation amplitude to occur uniformly within a preset symmetrical interval. Organize the output structure according to the data row number dimension, and establish the positional mapping relationship between the perturbation data and the source data; The corresponding sorting methods are applied in parallel to all perturbed data, and the sorting performance of different algorithms under the same perturbation conditions is cross-compared.

6. The aircraft aviation risk assessment method as described in claim 1, characterized in that, The process of fusing the risk ranking results obtained from the evaluation using the genetic optimization fusion algorithm includes: randomly generating an initial population containing configuration information for multiple algorithms, and setting weight parameters, mutation intensity base values ​​and their adjustment period parameters; For each individual, the algorithm consensus evaluation degree and physical constraint satisfaction degree are calculated synchronously, and the two scores are fused by dynamic weights that are updated linearly in algebra to form the core fitness index; Based on fitness scores, a tournament selection strategy is used to screen parent individuals, prioritizing individuals with the best comprehensive evaluation of algorithmic consensus and physical constraint satisfaction to participate in subsequent crossover operations; The mutation intensity is automatically adjusted according to the number of generations of evolution and a preset sine curve: the global exploration ability is enhanced during peak periods, the local optimization ability is strengthened during trough periods, and the phase parameter is compensated according to the population convergence speed. A two-stage crossover is performed on the parent generation, first inheriting the continuous gene segment with the highest fitness from both parents, and then scanning and repairing the key feature sites missing in the offspring to ensure that there is no loss of key information in gene recombination. Eliminate individuals who violate the constraints of the predicted dangerous relationship, and give high-potential individuals K opportunities for repair; At the end of each generation, the overlap between the current best solution and the historical Pareto front, the zero-symmetric feature of the fitness window standard deviation across generations, and the rate of decay of fitness improvement momentum between generations are detected simultaneously. Evolution is terminated only when all three conditions are met. Extract the optimal equilibrium solution from the final generation Pareto front.

7. The aircraft aviation risk assessment method as described in claim 1, characterized in that, The process of fusing the risk ranking results obtained from the evaluation using an adversarial optimization model includes: calculating the global ranking trend correlation and micro-ranking conflict intensity among the ranking groups of heterogeneous algorithms in parallel. Based on the consensus index, a negative credibility active filtering rule is designed to identify and zero out the decision weights of low-quality algorithms; a non-uniform distribution scheme that conforms to the strength of group consensus is generated through a dynamic normalization function. Establish order-preserving transformation rules from discrete sorting to continuous space, fully preserve the structural relationship of the original algorithm ranking, and simultaneously analyze the engineering physical constraints into an inequality constraint system on continuous dimensions; An optimization model containing continuous hazard variables and constraints is constructed. The main optimization layer of the optimization model uses the interior point method to perform core solution space search, and the auxiliary correction layer uses gradient projection technology to directionally repair constraint violation points. By integrating dynamic weighting, rule mapping, and two-stage solution results, the consistency of weight allocation and consensus degree, as well as the compliance of the scanning solution vector with engineering constraints, is verified, an interpretable and traceable decision report is generated, and finally, the working condition ranking result is formed.

8. The aircraft aviation risk assessment method as described in claim 1, characterized in that, The process of obtaining the two merged ranking results and then performing weighted fusion of the two merged ranking results includes: extracting the difference features of rank displacement based on the ranking sequences generated by the two algorithms themselves; The sliding window technique is used to scan the sorted sequence, calculate the statistical divergence of the position displacement within the window in real time, set a detection threshold, and when the divergence value abnormally exceeds the set threshold, it is determined to be a real disputed segment; otherwise, it is determined to be a normal area. In regular areas, a standard fusion strategy is used, based on the adaptive weight allocation coefficient analysis of genetic algorithms and adversarial models, and the weighted and fused ranking results are adopted; in disputed areas, a three-party voting mechanism is initiated, using heterogeneous ranking methods to vote, so as to achieve intelligent adjudication that is accurately adapted to the scenario.

9. An aircraft aviation risk assessment system, characterized in that, include: The force element acquisition module is configured to divide the aircraft into regions, extract the FBD section under static equilibrium constraints in the region where the flight hazard exceeds the set value, and obtain the six-component force elements, resultant internal force and resultant moment of the section boundary. The key force element extraction module is configured to quantify the contribution weight of each section force to the stress distribution change based on the acquired data, sort them according to the contribution weight, and determine the key force elements; The initial selection module for the sorting algorithm is configured to use multiple different algorithms to sort the degree of hazard of the working conditions based on the key force elements, perform correlation analysis on all results, and select multiple final algorithms based on the correlation analysis results to form a preliminary heterogeneous algorithm. The sorting algorithm secondary selection module is configured to process the perturbed data using heterogeneous algorithms, cross-compare the sorting performance of each algorithm under the same perturbation conditions, perform secondary screening, and form the final heterogeneous algorithm. The parallel module for the ranking algorithm is configured to perform aircraft aviation risk assessments using the final heterogeneous algorithms to obtain the risk ranking results after the assessment. The ranking result fusion module is configured to use a genetic optimization fusion algorithm and an adversarial solution optimization model to fuse the risk ranking results obtained from the evaluation, resulting in two fused ranking results. Then, the two fused ranking results are weighted and fused. In disputed areas, a three-party voting mechanism is initiated, using a heterogeneous ranking method to vote, and the final ranking result is obtained.

10. An electronic device, characterized in that, It includes a memory and a processor, as well as computer instructions stored in the memory and running on the processor, which, when executed by the processor, perform the steps of the method according to any one of claims 1-8.

Images