An aging aircraft aviation safety prediction method based on an improved WCGAN model

Abstract

The application provides an aging aircraft aviation safety prediction method based on an improved WCGAN model, which utilizes small sample data in an open source database in an aviation safety voluntary reporting system to predict aviation safety of aging aircraft; the method comprises the following steps: cleaning aircraft aviation failure event data in the open source database to obtain cleaned data; converting unstructured text report data in the cleaned data into structured data; introducing a gradient penalty GP mechanism and a NeuralProphet algorithm to optimize the WCGAN model to obtain a WCGAN-GP model; optimizing parameters according to the error minimization principle; inputting the small sample data into the trained WCGAN-GP model to generate aviation accident prediction values of the aging aircraft at future time points. The application introduces the gradient penalty GP mechanism and the NeuralProphet algorithm to optimize and improve the WCGAN model, overcomes defects of the traditional WCGAN model, such as gradient disappearance, mode collapse and lack of time sequence characteristic mining ability in the training process, improves the stability and convergence speed of the improved WCGAN model, and makes the improved WCGAN model more effectively adapt to the characteristics of aviation accident data.

Description

Technical Field

[0001] This invention belongs to the field of aviation safety technology, specifically relating to a method for predicting aviation safety of aging aircraft based on an improved WCGAN model. Background Technology

[0002] Accurate prediction of aviation accidents is of great significance for promoting proactive management of aviation safety and improving industry efficiency. In particular, as aircraft enter the aging stage (referring to aircraft that have been in use for more than 75% of their lifespan), failures such as fatigue cracks and corrosion of the aircraft airframe structure occur frequently, seriously endangering flight safety.

[0003] Although the frequency of major aviation accidents worldwide has remained low in recent years due to innovations in aviation safety management concepts and technologies—according to publicly available data from the International Civil Aviation Organization (ICAO), since 2019, the number of fatal aviation accidents globally has not exceeded six per year—unlike typical low-damage, high-frequency aviation safety incidents, aviation accidents, once they occur, are extremely difficult to recover from and cause enormous loss of life and property.

[0004] Therefore, preventing aviation accidents, especially those with serious consequences, is a major issue of long-term and overall importance, highlighting the urgency and importance of conducting efficient aviation accident prediction work.

[0005] After decades of development, the industry has largely focused on statistical analysis of aviation safety incidents. The theoretical development of aviation safety prediction is not yet fully mature, and there is still significant room for improvement in accurate prediction and intelligent decision-making regarding aviation safety. From engineering practice and publicly available literature, common prediction methods currently include time series forecasting, econometric modeling, and machine learning. In recent years, with the rise of cutting-edge technologies such as artificial intelligence and big data, machine learning prediction has become more widely used. Deep learning has become an important branch of machine learning. Using complex neural networks as its main architecture, it leverages its powerful self-learning and nonlinear fitting capabilities to efficiently process complex data samples, thereby more fully uncovering the numerical characteristics and patterns inherent in the samples.

[0006] In deep learning, representative models include deep belief network (DBN), recurrent neural network (RNN), long short-term memory (LSTM), and Transformer. Through a review of relevant literature on aviation safety prediction, it was found that the existing aviation safety prediction studies mainly focus on samples of unsafe events or accident symptoms with abundant data and significant numerical characteristics, and have achieved good prediction results through case verification. However, the existing aviation safety prediction studies pay less attention to sparse and highly random aviation accident samples. This is because mainstream prediction methods, represented by deep learning, have high requirements for data quality. Only by continuously optimizing model parameters from sufficient historical samples can a network structure with stable performance and strong generalization ability be obtained. However, small sample aviation accident data has the characteristics of uncertainty, complex causal relationships and strong temporal characteristics: (1) The characteristic of small sample aviation accident data is that the sample size is limited, which leads to high uncertainty in statistical analysis and model building. This uncertainty stems not only from insufficient data volume but also from potential biases during data collection; (2) the causes of aviation accidents are usually the result of multiple factors interacting, manifesting as a complex causal network. Small sample data often fails to reveal these complex relationships, making it difficult for traditional analysis methods to identify key influencing factors; (3) aviation accident data exhibits significant time-series characteristics, with data at different time points showing clear dependencies and trends. The complex time-series characteristics of small sample data necessitate that prediction models fully consider the impact of time factors on accident occurrence. These characteristics, to some extent, limit the computational efficiency of deep learning prediction models.

[0007] In aviation safety prediction, few studies have attempted to construct deep learning models with good applicability. For other practical engineering problems involving prediction, considering the high uncertainty, poor generalization ability, and weak interpretability of existing small-sample prediction research, as well as the weak interpretability of directly constructing analytical mapping models, some scholars have introduced novel generative adversarial networks (GANs) deep learning models. GAN deep learning models expand small-sample datasets and improve their generalization ability by generating realistic data in unsupervised learning through adversarial training of the generator and discriminator. In recent years, GAN deep learning models have achieved good results for small-sample prediction tasks.

[0008] How to design a small-sample prediction model based on the actual characteristics of aviation accident data to predict aviation safety for aging aircraft has become an urgent problem to be solved. Summary of the Invention

[0009] The purpose of this invention is to address the shortcomings of existing technologies in predicting aircraft safety based on small samples. It provides a method for predicting the safety of aging aircraft based on an improved WCGAN model. This method introduces a gradient penalty (GP) mechanism and the NeuralProphet algorithm to optimize the improved WCGAN model, overcoming the deficiencies of the traditional WCGAN model during training, such as gradient vanishing, pattern collapse, and insufficient ability to mine temporal characteristics. This improves the stability and convergence speed of the improved WCGAN model, enabling it to more effectively adapt to the characteristics of aviation accident data.

[0010] To achieve the above objectives, the technical solution provided by this invention is:

[0011] A method for predicting aviation safety of aging aircraft based on an improved WCGAN model includes: predicting aviation safety of aging aircraft using small sample data from an open-source database in an aviation safety voluntary reporting system; specifically including:

[0012] Data processing: Clean the aircraft aviation failure event data in the open source database to obtain cleaned data; convert the unstructured text report data in the cleaned data into structured data;

[0013] Improved WCGAN model: The WCGAN model is optimized by introducing a gradient penalty GP mechanism and the NeuralProphet algorithm, resulting in the WCGAN-GP model;

[0014] Training the WCGAN-GP model: The WCGAN-GP model traverses the data and optimizes the parameters based on the principle of minimizing error;

[0015] Testing the WCGAN-GP model: Input the small sample data into the well-trained WCGAN-GP model, which generates aviation accident predictions for aging aircraft at future times.

[0016] As a further limitation of the present invention, the data processing specifically includes:

[0017] Using open-source data from the Voluntary Reporting System for Aviation Safety, the aviation accident data required for the research is screened and statistically analyzed; the aviation accident data is cleaned data.

[0018] The time frame of the aviation accident data required for the research was further filtered and statistically analyzed monthly, thereby converting the unstructured text report data into structured text report data.

[0019] As a further limitation of the present invention, the improved WCGAN model specifically includes:

[0020] The WCGAN-GP model includes a generator and a discriminator. The cleaned data includes historical aviation accident records and data on precipitating events that influenced aviation accidents. The historical aviation accident records are represented as real data x. r The data of causal events affecting aviation accidents are represented as conditional input y; random noise z is concatenated with the conditional input y and then input into the generator, which outputs an aviation accident prediction value G(z|y); the aviation accident prediction value G(z|y) is compared with the actual data x. r The input is to the discriminator, which not only compares the predicted aviation accident value G(z|y) with the actual data x r The discriminator further determines whether the predicted aviation accident value G(z|y) satisfies the constraint of the conditional input y, based on the similarity level. Specifically:

[0021] The objective function of a game containing conditional probability distributions is expressed as:

[0022]

[0023] In the formula, p represents the real data sample r The expected value of (x), Represents the generated data sample The expected value, D(.), represents the output of the discriminator;

[0024] The Wasserstein distance is expressed as:

[0025]

[0026] In the formula, express With p r The set of all possible joint distributions of (x), ∫ x×y d(x,y)dγ(x,y) represents the expected value of the sample distance, and is obtained by taking the lower bound of the expected value. With p r Wasserstein distance between (x) and (x).

[0027] As a further limitation of the present invention, a gradient-penalized GP mechanism is introduced into the loss function of the discriminator to constrain the gradient of the discriminator to a controllable range; the gradient-penalized GP mechanism calculates the output value of the discriminator on the interpolated data. gradient The square of the deviation from the unit length ensures that the gradient value approaches the unit value while maintaining a positive penalty term, with a larger penalty for greater deviation; where:

[0028] The GP expression for the gradient-penalized GP mechanism is:

[0029]

[0030] In the formula, λ represents the adjustment parameter that controls the intensity of the punishment. This represents the linear interpolation between generated samples and real samples. represents the discriminator's output value for the interpolated data, ||.||2 represents the L2 norm, ε represents a random number following a uniform distribution, and x represents the actual data;

[0031] Introducing the gradient-penalized generalization (GP) mechanism, the objective function expression in the WCGAN-GP model is:

[0032]

[0033] As a further limitation of this invention, the expression for the NeuralProphet algorithm is:

[0034]

[0035] In the formula, T(t) represents the trend value, S(t) represents the seasonal effect, E(t) represents the holiday effect, F(t) represents the regression effect of the known exogenous variable in the future, A(t) represents the autoregressive effect of the historical observed variable, and L(t) represents the regression effect of the lagged observation of the exogenous variable.

[0036] The six components, configured individually or in combination, constitute the NeuralProphet analytical model, which contains multiple hidden neural network layers.

[0037] Suppose we have time series data X = [x1, x2, ..., x...] n The input to the hidden layer of the neural network is represented as:

[0038]

[0039] In the formula, H represents the output of the time series data after passing through the hidden layer of the neural network, and ω i Represents the weight, b i represents the bias term, and f(.) represents the activation function;

[0040] The output, after passing through a linear output layer, predicts future aviation accident values, as follows:

[0041]

[0042] In the formula, Y represents the output of the linear layer, ω j H represents the weight. j Indicates the hidden output layer, b j This indicates the bias term.

[0043] As a further limitation of the present invention, the training of the WCGAN-GP model specifically includes:

[0044] The hidden layer number of the network model is traversed in a step search manner, and the learning rate of the neural network is adaptively adjusted by the Adam algorithm. Dropout regularization mechanism is added between the hidden layers of the discriminator to reduce model redundancy.

[0045] Compare the mean square error under different parameter combinations, and select the optimal parameters based on the principle of minimizing error.

[0046] As a further limitation of the present invention, it also includes:

[0047] Evaluation and Prediction: Based on the aforementioned aviation accident prediction values, evaluate the feasibility and effectiveness of aviation safety prediction for aging aircraft; wherein:

[0048] The predictive ability of the WCGAN-GP model is evaluated using mean square root error, mean absolute error, and mean absolute percentage error as evaluation indicators, expressed as follows:

[0049]

[0050] In the formula, RMSE represents the root mean square error index, y i Let y′ represent the i-th sample in the validation dataset. i This represents the predicted value of aviation accidents, where n represents the number of samples, MAE represents the mean absolute error index, and MAPE represents the mean absolute percentage index.

[0051] The advantages of this invention are:

[0052] 1. This invention introduces a gradient penalty (GP) mechanism and the NeuralProphet algorithm to optimize and improve the WCGAN model, overcoming the shortcomings of the traditional WCGAN model such as gradient vanishing, mode collapse, and lack of temporal characteristic mining ability during training. This improves the stability and convergence speed of the improved WCGAN model, making it more effective in adapting to the characteristics of aviation accident data.

[0053] 2. This invention utilizes open-source data from the Voluntary Reporting System for Aviation Safety, highlighting the temporal characteristics of accident data samples within the open-source data. It employs WCGAN-GP model structure optimization and the NeuralProphet algorithm to enhance the identification ability of autocorrelation, trends, and periodicity in the temporal distribution of these temporal characteristics, resulting in more accurate and reliable predictions and effectively improving the prediction accuracy and robustness of aviation accident data.

[0054] 3. This invention has a higher predictive effect on small sample aviation accident data, and can more accurately predict aviation accidents. It also has a better performance in solving the uncertainty, complex causal relationships and strong time series characteristics of small sample data.

[0055] Additional aspects and advantages of the invention will be set forth in part in the description which follows, and in part will be obvious from the description, or may be learned by practice of the invention. Attached Figure Description

[0056] The above and / or additional aspects and advantages of the present invention will become apparent and readily understood from the description of the embodiments taken in conjunction with the following drawings, in which:

[0057] Figure 1 This invention provides a process for predicting aviation safety of aging aircraft based on an improved WCGAN model. Figure 1 ;

[0058] Figure 2 This invention provides a sample table of security report samples extracted from ASRS.

[0059] Figure 3 The monthly statistical results of aviation safety data collected in this invention are illustrated in the figure.

[0060] Figure 4 The main parameter value table of the NeuralProphet algorithm provided by this invention;

[0061] Figure 5 The device failure event prediction result diagram provided by this invention;

[0062] Figure 6 The diagram showing the prediction results of improper program events provided by this invention;

[0063] Figure 7 The communication breakdown event prediction result diagram provided by this invention;

[0064] Figure 8 The scenario perception bias event prediction result diagram provided by this invention;

[0065] Figure 9 This invention provides a process for predicting aviation safety of aging aircraft based on an improved WCGAN model. Figure 2 ;

[0066] Figure 10 : Comparison results of the small-sample aviation accident prediction method provided by this invention;

[0067] Figure 11 Comparison table of RMSE, MAPE, and MAE for different methods provided in this invention;

[0068] Figure 12 The diagram shows the generator and discriminator structures in the improved WCGAN model provided by this invention. Detailed Implementation

[0069] The embodiments of the present invention are described in detail below. These embodiments are exemplary and intended to explain the present invention, and should not be construed as limiting the present invention.

[0070] This invention discloses a method for predicting aviation safety of aging aircraft based on an improved WCGAN model, comprising: predicting aviation safety of aging aircraft using small sample data from an open-source database in an aviation safety voluntary reporting system; please refer to... Figure 1 The aviation safety prediction method for aging aircraft in this embodiment of the invention specifically includes: (1) data processing, (2) improving the WCGAN model, (3) training the WCGAN-GP model, and (4) testing the WCGAN-GP model.

[0071] in:

[0072] (1) The data processing specifically involves cleaning the aircraft aviation failure event data in the open source database to obtain cleaned data; and converting the unstructured text report data in the cleaned data into structured data.

[0073] More specifically, the above data processing includes: using open-source data from the Aviation Safety Reporting System (ASRS), filtering and statistically analyzing the aviation accident data required for the study based on the "events" attribute value, with the aviation accident data being cleaned data; further filtering the time range of the aviation accident data required for the study based on the "date" attribute value, and performing monthly statistics, thereby converting the unstructured text report data into structured text report data.

[0074] More specifically, the research data in this embodiment of the invention is selected from voluntary report text records from the Aviation Safety Voluntary Reporting System (ASRS), including two types of samples: accident data records causing functional damage to aircraft (hereinafter referred to as aviation accidents) and unsafe event data records. The text data is structured by month based on the "time" attribute in the original text records, with the number of aviation accidents serving as the output indicator. Unsafe events are risk sources that induce accidents; although they do not directly cause predetermined losses, they have a causal relationship with accidents. This embodiment of the invention uses the number of unsafe events as the input indicator. Please refer to [link to relevant documentation]. Figure 2 , Figure 2 The table showing examples of security report samples extracted from ASRS demonstrates some attribute information of voluntary report samples.

[0075] This invention embodiment can extract monthly statistics on aviation accidents from the ASRS database from July 2017 to June 2021, such as... Figure 3 As shown.

[0076] (2) The improvement of the WCGAN model is as follows: the gradient penalty GP mechanism and the NeuralProphet algorithm are introduced to optimize the WCGAN model, resulting in the WCGAN-GP model.

[0077] The WCGAN model in this embodiment of the invention, based on the original GAN ​​structure, adds a conditional input y, enabling the generator to generate samples of a specific category according to given conditions, thus enhancing the generator's flexibility and practicality. Furthermore, it replaces the JS distance in traditional GANs with the Wasserstein distance as the loss function for model training. This distance can more accurately measure the difference between generated samples and real samples, thereby improving the generator's performance and training stability. Specifically, it includes:

[0078] Please see Figure 12 The generator in this embodiment of the invention includes an input layer, a feedforward neural network 1, a feedforward neural network 2, ..., a feedforward neural network n, and a fully connected layer. The discriminator in this embodiment of the invention includes an input layer, a splicing layer, a feedforward neural network 1, a feedforward neural network 2, ..., a feedforward neural network n, and a fully connected layer.

[0079] This invention traverses the hidden layer values ​​of the network model using a step-by-step search approach, employs the Adam algorithm to adaptively adjust the network learning rate, and adds a dropout regularization mechanism between the hidden layers of the discriminator's neural network to reduce model redundancy. The mean square error (MSE) under different parameter combinations is compared, and the optimal parameters are selected based on the principle of minimizing error. The expression for mean square error is:

[0080]

[0081] In the formula, y i Let y′ represent the i-th sample in the validation dataset. i This represents the predicted value for aviation accidents, and n represents the number of samples.

[0082] In the small-sample aviation accident prediction task, the dataset is extracted from historical aviation accidents involving aging aircraft and their related influencing factors. Specifically, the WCGAN-GP model includes a generator and a discriminator; the cleaned data includes historical aviation accident records and data on precipitating events affecting aviation accidents. The historical aviation accident records are represented as real data x. rThe data of causal events affecting aviation accidents are represented as conditional input y; random noise z is concatenated with the conditional input y and then input into the generator, which outputs the aviation accident prediction value G(z|y); the aviation accident prediction value G(z|y) is compared with the actual data x. r The input discriminator not only compares the predicted aviation accident value G(z|y) with the actual data x r Based on the similarity level, the discriminator also determines whether the predicted aviation accident value G(z|y) satisfies the constraints of the input y; specifically:

[0083] The objective function of a game containing conditional probability distributions is expressed as:

[0084]

[0085] In the formula, p represents the real data sample r The expected value of (x), Represents the generated data sample The expected value, D(.), represents the output of the discriminator;

[0086] When measuring the distributional similarity between generated samples and real samples, traditional GAN ​​models typically use JS divergence as a metric, as shown in the following formula:

[0087]

[0088] Combining the above two equations, the objective function for network training can be rewritten as:

[0089]

[0090] If the generated sample distribution does not overlap with the real sample distribution at all, the JS divergence will be constant, resulting in a gradient of 0 and inability to update normally. To address the shortcomings of JS divergence, this embodiment of the invention uses Wasserstein distance as a metric. The basic idea of ​​Wasserstein distance is to consider the generated sample distribution... Move to the true sample distribution p r (x) is the minimum cost required to measure the similarity between two distributions.

[0091] The Wasserstein distance is expressed as:

[0092]

[0093] In the formula, express With p r The set of all possible joint distributions of (x), ∫ x×yd(x,y)dγ(x,y) represents the expected value of the sample distance, and is obtained by taking the lower bound of the expected value. With p r Wasserstein distance between (x) and (x).

[0094] Furthermore, in this embodiment of the invention, the gradient penalty GP mechanism described above is introduced into the loss function of the discriminator, thereby constraining the gradient of the discriminator to a controllable range; the gradient penalty GP mechanism calculates the output value of the discriminator on the interpolated data. gradient The square of the deviation from the unit length ensures that the gradient value approaches the unit value while maintaining a positive penalty term, with a larger penalty for greater deviation; where:

[0095] The GP expression for the gradient-penalized GP mechanism is:

[0096]

[0097] In the formula, λ represents the adjustment parameter that controls the intensity of the punishment. This represents the linear interpolation between generated samples and real samples. represents the discriminator's output value for the interpolated data, ||.||2 represents the L2 norm, ε represents a random number following a uniform distribution, and x represents the actual data;

[0098] Introducing the gradient-penalized generalization (GP) mechanism, the objective function expression in the WCGAN-GP model is:

[0099]

[0100] Based on the aforementioned WCGAN model, the NeuralProphet algorithm selected in this embodiment combines the excellent nonlinear fitting capabilities of neural networks with the efficient learning of time series patterns by Prophet, making it suitable for complex time series prediction tasks. Using the NeuralProphet dependency package, the training set from the statistical aviation insecurity data is used to optimize the model structure, thereby establishing a corresponding NeuralProphet time series prediction model for each input indicator. An optimization method combining step search and empirical assignment is employed, ultimately determining the main parameter values ​​for the NeuralProphet algorithm as shown in the table below. Figure 4 As shown. According to Figure 4 The parameters were set, and the loss changes of the NeuralProphet model corresponding to the four input indicators were recorded in each training cycle.

[0101] The expression for the NeuralProphet algorithm described above in this embodiment of the invention is as follows:

[0102]

[0103] In the formula, T(t) represents the trend value, S(t) represents the seasonal effect, E(t) represents the holiday effect, F(t) represents the regression effect of the known exogenous variable in the future, A(t) represents the autoregressive effect of the historical observed variable, and L(t) represents the regression effect of the lagged observation of the exogenous variable.

[0104] The NeuralProphet analytical model is composed of six components, which can be configured individually or in combination. Under default settings, only the trend component and seasonal effect are used. The NeuralProphet analytical model contains multiple hidden neural network layers.

[0105] Suppose we have time series data X = [x1, x2, ..., x...] n The input is fed into the hidden layer of the neural network and is represented as:

[0106]

[0107] In the formula, H represents the output of the time series data after passing through the hidden layer of the neural network, and ω i Represents the weight, b i represents the bias term, and f(.) represents the activation function;

[0108] The output, after passing through a linear output layer, predicts future aviation accident values, as follows:

[0109]

[0110] In the formula, Y represents the output of the linear layer, ω j H represents the weight. j Indicates the hidden output layer, b j This indicates the bias term.

[0111] (3) Training the WCGAN-GP model involves: the WCGAN-GP model iterates through the data and selects the optimal parameters based on the principle of minimizing error.

[0112] Specifically, it includes:

[0113] First, the hidden layer values ​​of the network model are traversed in a step search manner, and the Adam algorithm is used to adaptively adjust the network learning rate. Then, a dropout regularization mechanism is added between the hidden layers of the discriminator to reduce model redundancy. Finally, the mean square error (MSE) under different parameter combinations is compared, and the parameters are optimized based on the principle of minimizing error.

[0114] (4) Testing the WCGAN-GP model involves inputting small sample data into the well-trained WCGAN-GP model, which then generates aviation accident predictions for aging aircraft at future times.

[0115] Test results are as follows Figure 10 As shown, from Figure 10 It can be seen that: (1) Compared with the traditional WCGAN model, the improved WCGAN-GP model proposed in this invention has significantly improved the prediction performance on the test set, and the absolute value and peak trend are more consistent with the test sample. (2) Neural network models (NeuralProphet, LSTM and TCN) and decision tree models (RF, XGboost and LightGBM) of the time series prediction class have large prediction errors on most test samples and are insufficient in describing the changing trend. This indirectly shows that the above comparison methods have shortcomings in the small sample time series prediction problem, such as insufficient learning of time series rules and inaccurate description of causal mapping relationship.

[0116] This invention, using a well-trained NeuralProphet model, calculates predicted values ​​of input metrics at four test time points. The NeuralProphet model, by combining a deep learning architecture with traditional time-series prediction methods, enhances its ability to identify temporal characteristics. Its mechanisms include automatically modeling seasonality and trends, using temporal feature embeddings, employing adaptive learning rates, handling missing and outlier values, considering the interaction effects of multiple features, and providing interpretability. These characteristics enable NeuralProphet to more accurately capture complex temporal relationships during the prediction process, thereby improving the accuracy of the prediction results. The calculation results are as follows: Figures 5-8 As shown.

[0117] Please see Figure 9 The aviation safety prediction method for aging aircraft in this embodiment of the invention, in addition to including: (1) data processing, (2) improving the WCGAN model, (3) training the WCGAN-GP model, and (4) testing the WCGAN-GP model, also includes: (5) evaluating the prediction. Wherein:

[0118] (5) The evaluation and prediction specifically involves: evaluating the feasibility and effectiveness of aviation safety prediction for aging aircraft based on aviation accident prediction values; wherein: the mean square root error, mean absolute error, and mean absolute percentage error are used as evaluation indicators to evaluate the predictive ability of the WCGAN-GP model, expressed as follows:

[0119]

[0120] In the formula, RMSE represents the root mean square error index, y iLet y′ represent the i-th sample in the validation dataset. i This represents the predicted value of aviation accidents, where n represents the number of samples, MAE represents the mean absolute error index, and MAPE represents the mean absolute percentage index.

[0121] To more intuitively evaluate the prediction performance of the method disclosed in this invention, WCGAN, NeuralProphet, LSTM, TCN, RF, XGBoost, and LightGBM prediction methods were used as control models for the prediction method of this invention, and the prediction results were calculated under the same experimental environment. The prediction performance comparison is as follows: Figure 10 As shown.

[0122] from Figure 10 It can be seen that (1) compared with the traditional WCGAN model, the improved WCGAN-GP model disclosed in this invention has significantly improved the prediction performance on the test set, and the absolute value and peak trend are more consistent with the test sample (Test). (2) Neural network models (NeuralProphet, LSTM and TCN) and decision tree models (RF, XGboost and LightGBM) of the time series prediction class have large prediction errors on most test samples, and their ability to describe the changing trend is insufficient. This indirectly shows that the above comparison methods have shortcomings in the small sample time series prediction problem, such as insufficient learning of time series rules and inaccurate description of causal mapping relationship.

[0123] To further quantitatively evaluate the model's generalization performance, record the model's RMSE, MAE, and MAPE on each test sample. Please refer to [link to relevant documentation]. Figure 11 ,right Figure 11 Comparative analysis shows that, in terms of prediction model performance, the improved WCGAN-GP disclosed in this embodiment of the invention consistently maintains the best accuracy, demonstrating excellent performance in RMSE, MAE, and MAPE metrics, achieving improvements of over 30% compared to the unimproved WCGAN model. Furthermore, ensemble learning-based decision tree models generally outperform time-series prediction-based neural network models in prediction accuracy.

[0124] Analysis of the prediction model's calculation results shows that the improved WCGAN-GP model disclosed in this invention can effectively predict the changing trends of small-sample aviation accidents, provide early warnings of major dangerous consequences, and help managers take timely and targeted protective measures.

[0125] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any person skilled in the art can easily conceive of various equivalent modifications or substitutions within the scope of the technology disclosed in the present invention, and such modifications or substitutions should all be covered within the scope of protection of the present invention.

Claims

1. A method for predicting aviation safety of aging aircraft based on an improved WCGAN model, characterized in that, include: Using small sample data from an open-source database in the Voluntary Aviation Safety Reporting System, we predict aviation safety for aging aircraft. Specifically, it includes: Data processing: Clean the aircraft aviation failure event data in the open source database to obtain cleaned data; convert the unstructured text report data in the cleaned data into structured data; Improved WCGAN model: The WCGAN model is optimized by introducing a gradient penalty GP mechanism and the NeuralProphet algorithm, resulting in the WCGAN-GP model; Training the WCGAN-GP model: The WCGAN-GP model traverses the data and optimizes the parameters based on the principle of minimizing error; Testing the WCGAN-GP model: Input the small sample data into the well-trained WCGAN-GP model, which generates aviation accident predictions for aging aircraft at future times; The WCGAN-GP model includes a generator and a discriminator. A gradient-penalized GP mechanism is introduced into the loss function of the discriminator to constrain its gradient to a controllable range. The gradient-penalized GP mechanism calculates the output value of the discriminator on the interpolated data. gradient The square of the deviation from the unit length ensures that the gradient value approaches the unit value while maintaining a positive penalty term, and the larger the deviation, the larger the penalty; where: The GP expression for the gradient-penalized GP mechanism is: In the formula, This represents the adjustment parameter that controls the severity of the punishment. This represents the linear interpolation between generated samples and real samples. This represents the discriminator's output value for the interpolated data. express Norm, This represents a random number that follows a uniform distribution. Represents real data; Introducing the gradient-penalized generalization (GP) mechanism, the objective function expression in the WCGAN-GP model is: ； The expression for the NeuralProphet algorithm is: In the formula, Indicates trend value, Indicating seasonal effects, This indicates the holiday effect. This represents the regression effect of known exogenous variables in the future. This represents the autoregressive effect of historical observed variables. This represents the regression effect of lagged observations of exogenous variables; The six components, configured individually or in combination, constitute the NeuralProphet analytical model, which contains multiple hidden neural network layers. Assuming there is time series data The input to the hidden layer of the neural network is represented as: In the formula, This represents the output of time series data after passing through the hidden layer of a neural network. Indicates the weight. Indicates the bias term. Indicates the activation function; The output, after passing through a linear output layer, predicts future aviation accident values, as follows: In the formula, This represents the output of the linear layer. Indicates the weight. This indicates a layer containing hidden output results. This indicates the bias term.

2. The method for predicting aviation safety of aging aircraft based on an improved WCGAN model according to claim 1, characterized in that, The data processing specifically includes: Using open-source data from the Voluntary Reporting System for Aviation Safety, the aviation accident data required for the research is screened and statistically analyzed; the aviation accident data is cleaned data. The time frame of the aviation accident data required for the research was further filtered and statistically analyzed monthly, thereby converting the unstructured text report data into structured text report data.

3. The method for predicting aviation safety of aging aircraft based on an improved WCGAN model according to claim 1, characterized in that, The improved WCGAN model specifically includes: The cleaned data includes historical aviation accident records and data on precipitating events that influenced aviation accidents. The historical aviation accident records are represented as authentic data. The data on causal events that influence aviation accidents are represented as conditional inputs. Random noise With conditional input The concatenated data is input into the generator, which outputs aviation accident prediction values. The predicted aviation accident value With the real data The input is to the discriminator, which not only determines the aviation accident prediction value. With the real data The discriminator also determines the similarity level of the aviation accident prediction value. Does the condition meet? (Input) Constraints; specifically: The objective function of a game containing conditional probability distributions is expressed as: In the formula, Represents real data samples Expected value Represents the generated data sample Expected value This represents the output of the discriminator; The Wasserstein distance is expressed as: In the formula, express and The set of all possible joint distributions, The expected value of the sample distance is obtained by taking its lower bound from the expected value. and The Wasserstein distance between them.

4. The method for predicting aviation safety of aging aircraft based on an improved WCGAN model according to claim 1, characterized in that, The training of the WCGAN-GP model specifically includes: The hidden layer number of the network model is traversed in a step search manner, and the learning rate of the neural network is adaptively adjusted by the Adam algorithm. Dropout regularization mechanism is added between the hidden layers of the discriminator to reduce model redundancy. Compare the mean square error under different parameter combinations, and select the optimal parameters based on the principle of minimizing error.

5. The method for predicting aviation safety of aging aircraft based on an improved WCGAN model according to claim 1, characterized in that, Also includes: Evaluation and Prediction: Based on the aforementioned aviation accident prediction values, evaluate the feasibility and effectiveness of aviation safety prediction for aging aircraft; wherein: The predictive ability of the WCGAN-GP model is evaluated using mean square root error, mean absolute error, and mean absolute percentage error as evaluation indicators, expressed as follows: In the formula, This represents the mean square root error index. Indicates the first in the validation dataset One sample, This represents the predicted value for aviation accidents. Indicates the number of samples. This represents the mean absolute error index. This represents the average absolute percentage indicator.

Images