Aviation dynamic credit evaluation method and device and computer equipment

Abstract

The invention discloses an aviation dynamic credit evaluation method and device and computer equipment. The method comprises the following steps: selecting three types of core data sources, namely airline historical data, real-time flight data and macropolicy data, and comprehensively covering historical operation-real-time operation-external environment dimensions; a TCN-RiskModel is constructed based on a time sequence convolutional network TCN, and a federal transfer learning architecture and a hyper-parameter optimization framework are adopted to train and optimize the model; setting an operation stability-airline potential-policy risk three-dimensional basic weight, introducing a dynamic adjustment mechanism, and obtaining a dynamic credit score through weighted summation of a three-dimensional score * a corresponding dynamic weight, omega1 + omega2 + omega3 = 1, omega1 is an operation stability weight, omega2 is an airline potential weight, and omega3 is a policy risk weight. In this way, a credible quantitative basis can be provided for financing matching, and the performance of the method is obviously superior to that of a traditional model.

Description

Technical Field

[0001] This invention relates to the field of aviation management technology, and in particular to a method, apparatus and computer equipment for dynamic credit evaluation of aviation. Background Technology

[0002] In the aviation sector's route subsidy accounts receivable financing scenario, funders need to accurately assess an airline's debt repayment ability and financing risk through credit evaluation to achieve efficient financing matching. Existing technologies, such as traditional time series models, have significant shortcomings in this scenario: First, they cannot effectively capture the long-term and short-term time-series dependencies of multi-source data in the aviation sector. Aviation data encompasses multiple dimensions, including historical operations, real-time operations, and external policies, and is characterized by rapid dynamic changes and strong time-series characteristics, making comprehensive correlation analysis difficult for traditional models. Second, general risk control models do not specifically incorporate aviation-specific indicators such as route subsidies and flight operations, leading to incompatibility issues and significant discrepancies between scores and actual risks. Third, fixed model weights cannot dynamically respond to real-time flight anomalies, policy changes, and other data changes. Furthermore, new routes lack historical data support, resulting in a data cold start problem, leading to insufficient real-time performance and low confidence in scoring. Therefore, a dynamic credit evaluation method that adapts to the characteristics of the aviation sector and combines real-time performance with accuracy is urgently needed. Summary of the Invention

[0003] In view of this, the purpose of this invention is to propose an aviation dynamic credit evaluation method, device and computer equipment, which can provide a reliable quantitative basis for financing matching and has significantly better performance than traditional models.

[0004] According to one aspect of the present invention, a dynamic credit rating method for airlines is provided, comprising the following steps: S1: Selecting three core data sources, namely, airline historical data, real-time flight data, and macroeconomic policy data, comprehensively covering the dimensions of historical operation, real-time operation, and external environment; S2: Constructing a TCN-RiskModel based on a temporal convolutional network (TCN), and training and optimizing the model using a federated transfer learning architecture and the Optuna hyperparameter optimization framework; S3: Setting basic weights for three dimensions: operational stability, route potential, and policy risk, introducing a dynamic adjustment mechanism, and obtaining a dynamic credit score St by weighted summation of the three-dimensional scores × corresponding dynamic weights, where ω1 + ω2 + ω3 = 1, ω1 is the operational stability weight, ω2 is the route potential weight, and ω3 is the policy risk weight.

[0005] In step S1: the airline's historical data includes financial health indicators, operational performance indicators, and route sustainability indicators. Financial health indicators include historical route subsidy recovery rate, debt-to-equity ratio, and net cash flow. Operational performance indicators include historical flight execution rate, average load factor, and historical bad debt rate. Route sustainability indicators include historical average route operating cycle and profitability of similar routes. The real-time flight data is processed through the distributed stream processing engine Apache. The Flink streaming engine collects data in real time, updating it every preset time interval with a preset delay. This includes real-time operational metrics, asset liquidity metrics, and abnormal event data. Real-time operational metrics include real-time load factor, flight punctuality rate, fuel consumption per aircraft, and ground handling costs. Asset liquidity metrics include real-time available seats, remaining ticket sales progress, and short-term accounts payable. Abnormal event data includes real-time flight cancellation / delay records and the number of sudden mechanical failures. The macroeconomic policy data includes subsidy policy data, industry regulatory policies, and macroeconomic policies. Subsidy policy data includes adjustments to local government route subsidy standards and changes in fiscal budget amounts. Industry regulatory policies include new regulations for route approval by the Civil Aviation Administration of China and flight control policies during public health events. Macroeconomic policies include interest rate adjustments and regional economic support policies.

[0006] In step S2, the TCN-RiskModel is constructed using causal convolution and dilated convolution structures to capture the long- and short-term temporal dependencies of multi-source data. This includes: TCN-RiskModel abandons the linear modeling method of traditional time series models and uses causal convolution to ensure that the modeling process conforms to the temporal logic and avoids future data from interfering with current predictions; it expands the receptive field through dilated convolution to effectively capture long-term temporal data while maintaining computational efficiency, adapting to the characteristics of rapid dynamic changes and strong temporal sequence of aviation data.

[0007] In step S2, the application of the federated transfer learning architecture includes, under the premise of protecting airline data privacy, transferring historical data of existing routes to the model training process of new routes after differential privacy processing. This includes: adopting a federated transfer learning architecture, storing the data of each airline locally, and only using the feature information after differential privacy processing to participate in model training without disclosing the original data; extracting common credit features from the existing route data and transferring them to the new route model to make up for the lack of historical data for the new routes.

[0008] The Optuna framework is used to automatically tune the kernel size, dilation rate, and number of hidden layers of the TCN-RiskModel. This includes setting the kernel size, dilation rate, and number of hidden layers of the TCN-RiskModel as tuning parameters, taking the accuracy of the three-dimensional scoring of operational stability, route potential, and policy risk as the optimization objective, and automatically searching for the optimal parameter combination through the Bayesian optimization algorithm of the Optuna framework.

[0009] In step S3, the dynamic adjustment mechanism involves introducing a volatility index and adjusting the values ​​of ω1, ω2, and ω3 in real time based on changes in the volatility index. This includes: focusing the basic weights on the core risks of aviation financing, setting operational stability ω1 as the highest weight, which is directly related to debt repayment ability; route potential ω2 reflects future profit support, and policy risk ω3 reflects external uncertainty; when the volatility index rises, the weights of ω1 and ω3 are increased, while the weight of ω2 is decreased; when the volatility index falls, the weight of ω2 is appropriately increased to dynamically adapt to changes in risk.

[0010] In step S3, the dynamic credit score St is updated in the following way: each time a new batch of real-time flight data or macroeconomic policy data is added, the TCN-RiskModel updates the parameters through incremental training to achieve real-time iteration of the dynamic credit score St.

[0011] In step S3, the dynamic credit score St is quantified into at least two credit levels, the financing conditions corresponding to each level of risk are clarified, and a direct link between the score, level, and financing plan is established, so that the funder can quickly apply the score results without additional conversion.

[0012] According to another aspect of the present invention, an aviation dynamic credit rating device is provided, comprising: a data source selection module, a training optimization module, and a scoring configuration module; the data source selection module is used to select three types of core data sources, namely, airline historical data, real-time flight data, and macroeconomic policy data, comprehensively covering the dimensions of historical operation, real-time operation, and external environment; the training optimization module is used to construct a TCN-RiskModel based on a temporal convolutional network (TCN), and to train and optimize the model using a federated transfer learning architecture and a hyperparameter optimization Optuna framework; the scoring configuration module is used to set the basic weights of three dimensions: operational stability, route potential, and policy risk, and to introduce a dynamic adjustment mechanism, obtaining a dynamic credit score St by weighted summation of the three-dimensional scores multiplied by their corresponding dynamic weights, where ω1+ω2+ω3=1, ω1 is the operational stability weight, ω2 is the route potential weight, and ω3 is the policy risk weight.

[0013] According to another aspect of the present invention, a computer device is provided, comprising: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, the instructions being executed by the at least one processor to enable the at least one processor to perform the aviation dynamic credit evaluation method as described in any of the preceding claims.

[0014] According to another aspect of the present invention, a computer-readable storage medium is provided storing a computer program that, when executed by a processor, implements the aviation dynamic credit evaluation method as described in any of the preceding claims.

[0015] It can be observed that the core shortcomings of the existing technologies mentioned above are incomplete data source coverage, inability of the model to adapt to the temporal characteristics of aviation data, and fixed weights. This solution addresses the issue of single data dimension in general models by comprehensively incorporating aviation-specific historical operational, real-time operational, and policy data through S1; S2 employs a Temporal Convolutional Network (TCN) model, whose causal and dilated convolutional structures can accurately capture the long- and short-term temporal dependencies of multi-source data, combined with federated transfer learning and hyperparameter optimization (Optuna tuning), solving the cold start problem for new route data while ensuring optimal model accuracy; S3 integrates dynamic weight adjustment and three-dimensional scoring, enabling the scoring to flexibly respond to data changes. Through these methods, the solution ultimately achieves continuous updates at preset times, shortening response delays to emergencies, effectively reducing the scoring error rate, and effectively controlling the bad debt rate, providing credible quantitative evidence for financing matching, significantly outperforming traditional models.

[0016] Furthermore, the existing solutions lack specific data indicators designed for the aviation sector, resulting in a lack of targeted scoring. This solution refines the core indicators of three types of data sources, and uses the distributed stream processing engine Apache Flink to collect real-time flight data (Xr) with low latency, ensuring that the data comprehensively covers the core risk points of aviation operations while maintaining real-time accuracy. Specifically, aviation-specific indicators such as historical route subsidy recovery rate and real-time load factor are directly related to the airline's solvency and route profitability, solving the problem of generalization in common model indicators. Multi-dimensional indicators corroborate each other, avoiding scoring distortion caused by single data biases, significantly improving the credibility of the model's input data, laying a data foundation for subsequent accurate scoring, and further reducing the scoring error rate.

[0017] Furthermore, the above solutions, traditional models can only capture short-term temporal dependencies and cannot handle long-term temporal data correlations across cycles and multiple dimensions in the aviation field, resulting in lagging scoring and low accuracy. This solution, through the causal convolution and dilated convolution structure of the Temporal Convolutional Network (TCN), ensures the rigor of temporal logic while efficiently capturing the long-term and short-term correlations between historical operational data and real-time operational data and policy data (such as the linkage between historical subsidy recovery rates and current policy adjustments). This solves the core defect of traditional models' insufficient capture of temporal dependencies, significantly improves the model's adaptability to aviation data, and effectively reduces the scoring error rate compared to traditional models.

[0018] Furthermore, the above solutions suffer from a cold start problem due to the lack of historical data for new routes, hindering model training and preventing timely credit assessment. This solution, through federated transfer learning, resolves the conflict between data privacy protection and data sharing (avoiding the leakage of airlines' core operational data). It also allows for the reuse of existing route data, enabling new routes to obtain high-confidence models without accumulating large amounts of historical data, significantly shortening the implementation cycle of credit assessment for new routes. Simultaneously, differential privacy processing ensures data security and compliance, enhancing the method's practicality and scalability.

[0019] Furthermore, the above solutions rely on manual parameter tuning of traditional models, which is inefficient and struggles to achieve accurate multi-dimensional evaluations, thus limiting model performance. This solution, however, utilizes the Optuna framework for hyperparameter optimization to automatically tune core parameters. This accurately matches the characteristics of aviation data, ensuring optimal accuracy in the three core dimensions of operational stability (solvency), route potential (profit expectations), and policy risk (external impacts). It avoids overall scoring bias caused by single-dimensional optimization, further reducing the scoring error rate and effectively improving accuracy compared to manually tuned models.

[0020] Furthermore, the existing solutions using fixed weights cannot respond to market risks and data changes (e.g., policy risk weights are not adjusted in time when policies change abruptly, leading to distorted scoring). This solution identifies core risk points through basic weights and introduces a volatility index for dynamic adjustment, enabling the scoring to accurately adapt to risk characteristics under different market conditions. For example, when the subsidy policy disbursement period is extended, the ω3 weight increases, increasing the weight of the policy risk dimension in the scoring, preventing the risk from being underestimated under fixed weights; when the market is stable, the weight of route potential is increased, better aligning with the supporting role of profit expectations on credit, significantly improving the targeting and flexibility of the scoring, and further reducing the bad debt rate.

[0021] Furthermore, the above solutions require traditional models to be retrained after batch data processing, resulting in a response delay of over 6 hours and an inability to promptly capture sudden risks (such as large-scale flight delays or policy changes). This solution achieves real-time parameter updates through incremental training; for example, the dynamic credit score St is updated hourly, with a response delay of less than 30 minutes to sudden events, thus addressing the key deficiency of insufficient real-time performance in traditional models. For instance, when a local government suddenly reduces its subsidy budget, the model can quickly incorporate this policy data, adjust the policy risk dimension scoring and weights, enabling funders to promptly grasp risk changes and avoid financing losses due to scoring lag.

[0022] Furthermore, the existing scoring results of the above solutions lack clear application standards, requiring funders to independently assess risk thresholds, resulting in low financing matching efficiency and high subjectivity. This solution, by quantifying credit ratings and financing conditions, provides funders with clear and actionable decision-making criteria, avoiding subjective judgment bias. For example, low-risk airlines (levels 1-2) can be directly matched with low-interest financing, while level 3 and above can hedge risks through guarantees, effectively improving financing matching efficiency while keeping the overall bad debt rate within a manageable range, significantly lower than the industry average, achieving a win-win situation for both funders and airlines. Attached Figure Description

[0023] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0024] Figure 1 This is a flowchart illustrating an embodiment of the aviation dynamic credit evaluation method of the present invention; Figure 2 This is a schematic diagram of the structure of an embodiment of the aviation dynamic credit evaluation device of the present invention; Figure 3 This is a schematic diagram of the structure of an embodiment of the computer device of the present invention. Detailed Implementation

[0025] The present invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be particularly noted that the following embodiments are for illustrative purposes only and do not limit the scope of the invention. Similarly, the following embodiments are only some, not all, embodiments of the present invention, and all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0026] This invention provides a dynamic credit rating method for aviation, which can provide a reliable quantitative basis for financing matching and has significantly outperformed the performance of traditional models.

[0027] Please see Figure 1 , Figure 1 This is a flowchart illustrating an embodiment of the aviation dynamic credit evaluation method of the present invention. It should be noted that if substantially the same result is obtained, the method of the present invention is not necessarily identical. Figure 1 The illustrated process sequence is limited. For example... Figure 1 As shown, the method includes the following steps: S1: Select three core data sources, namely airline historical data Xh, real-time flight data Xr, and macroeconomic policy data Xp, to comprehensively cover historical operations, real-time operations, and external environment dimensions; S2: Construct TCN-RiskModel based on Temporal Convolutional Network (TCN), and train and optimize the model using a federated transfer learning architecture and hyperparameter optimization (Optuna) framework; S3: Set three basic weights for operational stability, route potential, and policy risk, and introduce a dynamic adjustment mechanism. The dynamic credit score St is obtained by weighted summation of the three-dimensional scores × the corresponding dynamic weights, where ω1+ω2+ω3=1, ω1 is the weight of operational stability, ω2 is the weight of route potential, and ω3 is the weight of policy risk.

[0028] It can be observed that the core shortcomings of existing technologies in this embodiment are incomplete data source coverage, inability of the model to adapt to the temporal characteristics of aviation data, and fixed weights. This solution addresses the problem of single data dimension in general models by comprehensively incorporating aviation-specific historical operational, real-time operational, and policy data in S1; S2 employs a Temporal Convolutional Network (TCN) model, whose causal convolution and dilated convolution structures can accurately capture the long- and short-term temporal dependencies of multi-source data, combined with federated transfer learning and hyperparameter optimization (Optuna tuning), solving the cold start problem for new route data while ensuring optimal model accuracy; S3 integrates dynamic weight adjustment and three-dimensional scoring, enabling the scoring to flexibly respond to data changes; through these methods, the system ultimately achieves continuous updates at preset times, shortens response delays to emergencies, effectively reduces the scoring error rate, and effectively controls the bad debt rate, providing credible quantitative evidence for financing matching, significantly outperforming traditional models.

[0029] In this embodiment, the selection of three core data sources—airline historical data Xh, real-time flight data Xr, and macroeconomic policy data Xp—comprehensively covers historical operations, real-time operations, and the external environment, and may include: The airline's historical data Xh includes financial health indicators, operational performance indicators, and route sustainability indicators. Financial health indicators include historical route subsidy recovery rate, debt-to-equity ratio, and net cash flow. Operational performance indicators include historical flight execution rate, average load factor, and historical bad debt rate. Route sustainability indicators include historical average operating cycle of routes and profitability of similar routes. The real-time flight data Xr is collected in real time through the open-source distributed stream processing engine (Apache Flink) and updated once every preset time interval according to the preset delay. It includes real-time operational indicators, asset liquidity indicators, and abnormal event data. Real-time operational indicators include real-time load factor, flight punctuality rate, fuel consumption per aircraft and ground handling costs. Asset liquidity indicators include real-time available seats, remaining ticket sales progress, and short-term accounts payable. Abnormal event data includes real-time flight cancellation / delay records and the number of sudden mechanical failures. The macroeconomic policy data Xp includes subsidy policy data, industry regulatory policies, and macroeconomic policies. Subsidy policy data includes adjustments to local government route subsidy standards and changes in fiscal budget amounts. Industry regulatory policies include new regulations for route approval by the Civil Aviation Administration and flight control policies during public health emergencies. Macroeconomic policies include interest rate adjustments and regional economic support policies.

[0030] It can be observed that, in this embodiment, existing technologies do not design specific data indicators for the aviation field, resulting in a lack of targeted scoring. This solution refines the core indicators of three types of data sources, and uses the distributed stream processing engine Apache Flink to collect real-time flight data (Xr) with low latency, ensuring that the data comprehensively covers the core risk points of aviation operations while maintaining real-time performance. Among them, aviation-specific indicators such as historical route subsidy recovery rate and real-time load factor are directly related to the airline's debt repayment ability and route profitability, solving the problem of generalization of indicators in general models; multi-dimensional indicators corroborate each other, avoiding scoring distortion caused by single data bias, significantly improving the credibility of the model input data, laying a data foundation for subsequent accurate scoring, and further reducing the scoring error rate.

[0031] In this embodiment, the TCN-RiskModel is constructed using causal convolution and dilated convolution structures to capture long- and short-term temporal dependencies in multi-source data, and may include: TCN-RiskModel abandons the linear modeling approach of traditional time series models and adopts causal convolution to ensure that the modeling process conforms to the chronological logic and avoids future data from interfering with current predictions. By expanding the receptive field through dilated convolution, it can effectively capture long-term time series data while maintaining computational efficiency, adapting to the characteristics of rapid dynamic changes and strong temporal sequence of aviation data.

[0032] It can be observed that the above solutions, traditional models, can only capture short-term temporal dependencies and cannot handle long-term temporal data correlations across cycles and multiple dimensions in the aviation field, resulting in lagging scoring and low accuracy. This solution, through the causal convolution and dilated convolution structure of the Temporal Convolutional Network (TCN), ensures the rigor of temporal logic while efficiently capturing the long-term and short-term correlations between historical operational data and real-time operational and policy data (such as the linkage between historical subsidy recovery rates and current policy adjustments). This solves the core defect of traditional models' insufficient capture of temporal dependencies, significantly improving the model's adaptability to aviation data and effectively reducing the scoring error rate compared to traditional models.

[0033] In this embodiment, the process of migrating historical data of existing routes to new routes after differential privacy processing, while protecting airline data privacy, may include: The federated transfer learning architecture is adopted, with each airline's data stored locally. Only the feature information processed by differential privacy is used in the model training, without leaking the original data. Common credit features are extracted from the existing route data and transferred to the new route model to make up for the lack of historical data for the new route, thus effectively improving the confidence of the new route model.

[0034] It can be observed that the above solutions suffer from a cold start problem due to the lack of historical data for new routes, making model training difficult and hindering timely credit evaluation. This proposed solution, through federated transfer learning, resolves the conflict between data privacy protection and data sharing (preventing the leakage of airlines' core operational data). Furthermore, by migrating and reusing existing route data, it enables new routes to obtain high-confidence models without accumulating large amounts of historical data, significantly shortening the implementation cycle for new route credit evaluation. Simultaneously, differential privacy processing ensures data security and compliance, enhancing the method's practicality and scalability.

[0035] In this embodiment, the hyperparameter optimization Optuna framework, used to automatically tune the kernel size, dilation rate, and number of hidden layers of the TCN-RiskModel, may include: The kernel size, dilation rate, and number of hidden layers of the TCN-RiskModel were set as tuning parameters. The optimization objective was to improve the accuracy of the three-dimensional scoring of operational stability, route potential, and policy risk. The optimal parameter combination was automatically searched by the Bayesian optimization algorithm of the Optuna framework without manual intervention.

[0036] It can be observed that the above solutions rely on manual parameter tuning of traditional models, which is inefficient and struggles to achieve accurate multi-dimensional evaluations, thus limiting model performance. This solution, however, utilizes the Optuna framework for hyperparameter optimization to automatically tune core parameters. This accurately matches the characteristics of aviation data, ensuring optimal accuracy in the three core dimensions of operational stability (solvency), route potential (profit expectations), and policy risk (external impact). It avoids overall scoring bias caused by single-dimensional optimization, further reducing the scoring error rate and effectively improving accuracy compared to manually tuned models.

[0037] In this embodiment, the dynamic adjustment mechanism involves introducing a volatility index and adjusting the values ​​of ω1, ω2, and ω3 in real time based on changes in the volatility index. The basic weights can be set as ω1=0.6, ω2=0.3, and ω3=0.1. These basic weights focus on the core risks of aviation financing, setting operational stability (ω1=0.6) as the highest weight, directly related to debt repayment ability; route potential (ω2=0.3) reflects future profit support; and policy risk (ω3=0.1) reflects external uncertainty. When the volatility index rises (market risk intensifies), the weights of ω1 and ω3 are increased, while the weight of ω2 is decreased. When the volatility index falls (market stabilizes), the weight of ω2 is appropriately increased, dynamically adapting to changes in risk.

[0038] It can be observed that the existing technologies using fixed weights cannot respond to market risks and data changes (e.g., policy risk weights are not adjusted in time when policies change abruptly, leading to distorted scoring). This solution identifies core risk points through basic weights and introduces a volatility index for dynamic adjustment, enabling the scoring to accurately adapt to risk characteristics under different market conditions. For example, when the subsidy policy disbursement period is extended, the ω3 weight increases, increasing the weight of the policy risk dimension in the scoring, preventing the risk from being underestimated under fixed weights; when the market is stable, the weight of route potential is increased, better aligning with the supporting role of profit expectations on credit, significantly improving the targeting and flexibility of the scoring, and further reducing the bad debt rate.

[0039] In this embodiment, the dynamic credit score St is updated as follows: Each time a new batch of real-time flight data Xr or macroeconomic policy data Xp is added, the TCN-RiskModel updates its parameters through incremental training to achieve real-time iteration of the dynamic credit score St. This can include: An incremental training mechanism is established so that when new real-time flight data (such as sudden flight cancellations) or macroeconomic policy data (such as adjustments to subsidy standards) are generated, the entire model does not need to be retrained. Instead, the parameters of the TCN-RiskModel are locally updated based on the new data, and a new dynamic credit score St is quickly output.

[0040] It can be observed that the above solutions require traditional models to be retrained after batch data processing, resulting in a response delay of over 6 hours and an inability to promptly capture sudden risks (such as large-scale flight delays or policy changes). This solution achieves real-time parameter updates through incremental training; for example, the dynamic credit score St is updated hourly, resulting in a response delay of less than 30 minutes to sudden events, thus addressing the key deficiency of insufficient real-time performance in traditional models. For instance, when a local government suddenly reduces its subsidy budget, the model can quickly incorporate this policy data, adjust the policy risk dimension scoring and weights, enabling funders to promptly grasp risk changes and avoid financing losses due to scoring lag.

[0041] In this embodiment, the dynamic credit score St is divided into credit levels according to the following criteria: Level 1 (90-100 points, low risk), Level 2 (80-89 points, relatively low risk), Level 3 (60-79 points, medium risk), Level 4 (50-59 points, relatively high risk), and Level 5 (<50 points, high risk). The credit level serves as the basis for financing parties to match financing. Airlines with Level 1-2 credit scores receive lower financing rates, while Level 3 and above require additional collateral.

[0042] In this embodiment, the dynamic credit score St is quantified into 5 credit levels, clarifying the financing conditions corresponding to each level of risk, and establishing a direct link between score, level, and financing plan. This allows funders to quickly apply the score results without additional calculations, achieving standardization and efficiency in financing matching.

[0043] It can be observed that the scoring results of the above solutions lack clear application standards, requiring funders to independently assess risk thresholds, resulting in low financing matching efficiency and high subjectivity. This solution, by quantifying credit ratings and financing conditions, provides funders with clear and actionable decision-making criteria, avoiding subjective judgment bias. For example, low-risk airlines (levels 1-2) can be directly matched with low-interest financing, while level 3 and above can hedge risks through guarantees, effectively improving financing matching efficiency while keeping the overall bad debt rate within a manageable range, significantly lower than the industry average, achieving a win-win situation for both funders and airlines.

[0044] This invention also provides an aviation dynamic credit evaluation device, which can provide credible quantitative basis for financing matching and has significantly better performance than traditional models.

[0045] Please see Figure 2 , Figure 2 This is a schematic diagram of an embodiment of the aviation dynamic credit evaluation device of the present invention. In this embodiment, the aviation dynamic credit evaluation device 20 includes a data source selection module 21, a training optimization module 22, and a scoring configuration module 23.

[0046] The data source selection module 21 is used to select three core data sources: airline historical data Xh, real-time flight data Xr, and macroeconomic policy data Xp, which comprehensively cover the dimensions of historical operations, real-time operations, and external environment.

[0047] The training optimization module 22 is used to construct TCN-RiskModel based on the temporal convolutional network TCN, and to train and optimize the model using a federated transfer learning architecture and the hyperparameter optimization Optuna framework. The scoring configuration module 23 is used to set the basic weights of the three dimensions of operational stability, route potential and policy risk, and introduces a dynamic adjustment mechanism. The dynamic credit score St is obtained by weighted summation of the three-dimensional scores and their corresponding dynamic weights, where ω1+ω2+ω3=1, ω1 is the operational stability weight, ω2 is the route potential weight, and ω3 is the policy risk weight.

[0048] Each unit module of the aviation dynamic credit evaluation device 20 can execute the corresponding steps in the above method embodiment, so the details of each unit module will not be elaborated here. Please refer to the description of the corresponding steps above for details.

[0049] This invention also provides a computer device, such as... Figure 3 As shown, it includes: at least one processor 31; and a memory 32 communicatively connected to at least one processor 31; wherein the memory 32 stores instructions executable by at least one processor 31, the instructions being executed by at least one processor 31 to enable at least one processor 31 to perform the above-described aviation dynamic credit evaluation method.

[0050] The memory 32 and processor 31 are connected via a bus, which may include any number of interconnecting buses and bridges, connecting various circuits of one or more processors 31 and memory 32. The bus can also connect various other circuits, such as peripheral devices, voltage regulators, and power management circuits, which are well known in the art and therefore will not be described further herein. A bus interface provides an interface between the bus and the transceiver. The transceiver can be a single element or multiple elements, such as multiple receivers and transmitters, providing a unit for communicating with various other devices over a transmission medium. Data processed by processor 31 is transmitted over a wireless medium via an antenna, which further receives data and transmits it to processor 31.

[0051] Processor 31 is responsible for managing the bus and general processing, and can also provide various functions, including timing, peripheral interfaces, voltage regulation, power management, and other control functions. Memory 32 can be used to store data used by processor 31 during operation.

[0052] The present invention further provides a computer-readable storage medium storing a computer program. When the computer program is executed by a processor, it implements the above-described method embodiments.

[0053] It can be observed that the core shortcomings of the existing technologies mentioned above are incomplete data source coverage, inability of the model to adapt to the temporal characteristics of aviation data, and fixed weights. This solution addresses the issue of single data dimension in general models by comprehensively incorporating aviation-specific historical operational, real-time operational, and policy data through S1; S2 employs a Temporal Convolutional Network (TCN) model, whose causal and dilated convolutional structures can accurately capture the long- and short-term temporal dependencies of multi-source data, combined with federated transfer learning and hyperparameter optimization (Optuna tuning), solving the cold start problem for new route data while ensuring optimal model accuracy; S3 integrates dynamic weight adjustment and three-dimensional scoring, enabling the scoring to flexibly respond to data changes. Through these methods, the solution ultimately achieves continuous updates at preset times, shortening response delays to emergencies, effectively reducing the scoring error rate, and effectively controlling the bad debt rate, providing credible quantitative evidence for financing matching, significantly outperforming traditional models.

[0054] Furthermore, the existing solutions lack specific data indicators designed for the aviation sector, resulting in a lack of targeted scoring. This solution refines the core indicators of three types of data sources, and uses the distributed stream processing engine Apache Flink to collect real-time flight data (Xr) with low latency, ensuring that the data comprehensively covers the core risk points of aviation operations while maintaining real-time accuracy. Specifically, aviation-specific indicators such as historical route subsidy recovery rate and real-time load factor are directly related to the airline's solvency and route profitability, solving the problem of generalization in common model indicators. Multi-dimensional indicators corroborate each other, avoiding scoring distortion caused by single data biases, significantly improving the credibility of the model's input data, laying a data foundation for subsequent accurate scoring, and further reducing the scoring error rate.

[0055] Furthermore, the above solutions, traditional models can only capture short-term temporal dependencies and cannot handle long-term temporal data correlations across cycles and multiple dimensions in the aviation field, resulting in lagging scoring and low accuracy. This solution, through the causal convolution and dilated convolution structure of the Temporal Convolutional Network (TCN), ensures the rigor of temporal logic while efficiently capturing the long-term and short-term correlations between historical operational data and real-time operational data and policy data (such as the linkage between historical subsidy recovery rates and current policy adjustments). This solves the core defect of traditional models' insufficient capture of temporal dependencies, significantly improves the model's adaptability to aviation data, and effectively reduces the scoring error rate compared to traditional models.

[0056] Furthermore, the above solutions suffer from a cold start problem due to the lack of historical data for new routes, hindering model training and preventing timely credit assessment. This solution, through federated transfer learning, resolves the conflict between data privacy protection and data sharing (avoiding the leakage of airlines' core operational data). It also allows for the reuse of existing route data, enabling new routes to obtain high-confidence models without accumulating large amounts of historical data, significantly shortening the implementation cycle of credit assessment for new routes. Simultaneously, differential privacy processing ensures data security and compliance, enhancing the method's practicality and scalability.

[0057] Furthermore, the above solutions rely on manual parameter tuning of traditional models, which is inefficient and struggles to achieve accurate multi-dimensional evaluations, thus limiting model performance. This solution, however, utilizes the Optuna framework for hyperparameter optimization to automatically tune core parameters. This accurately matches the characteristics of aviation data, ensuring optimal accuracy in the three core dimensions of operational stability (solvency), route potential (profit expectations), and policy risk (external impacts). It avoids overall scoring bias caused by single-dimensional optimization, further reducing the scoring error rate and effectively improving accuracy compared to manually tuned models.

[0058] Furthermore, the existing solutions using fixed weights cannot respond to market risks and data changes (e.g., policy risk weights are not adjusted in time when policies change abruptly, leading to distorted scoring). This solution identifies core risk points through basic weights and introduces a volatility index for dynamic adjustment, enabling the scoring to accurately adapt to risk characteristics under different market conditions. For example, when the subsidy policy disbursement period is extended, the ω3 weight increases, increasing the weight of the policy risk dimension in the scoring, preventing the risk from being underestimated under fixed weights; when the market is stable, the weight of route potential is increased, better aligning with the supporting role of profit expectations on credit, significantly improving the targeting and flexibility of the scoring, and further reducing the bad debt rate.

[0059] Furthermore, the above solutions require traditional models to be retrained after batch data processing, resulting in a response delay of over 6 hours and an inability to promptly capture sudden risks (such as large-scale flight delays or policy changes). This solution achieves real-time parameter updates through incremental training; for example, the dynamic credit score St is updated hourly, with a response delay of less than 30 minutes to sudden events, thus addressing the key deficiency of insufficient real-time performance in traditional models. For instance, when a local government suddenly reduces its subsidy budget, the model can quickly incorporate this policy data, adjust the policy risk dimension scoring and weights, enabling funders to promptly grasp risk changes and avoid financing losses due to scoring lag.

[0060] Furthermore, the existing scoring results of the above solutions lack clear application standards, requiring funders to independently assess risk thresholds, resulting in low financing matching efficiency and high subjectivity. This solution, by quantifying credit ratings and financing conditions, provides funders with clear and actionable decision-making criteria, avoiding subjective judgment bias. For example, low-risk airlines (levels 1-2) can be directly matched with low-interest financing, while level 3 and above can hedge risks through guarantees, effectively improving financing matching efficiency while keeping the overall bad debt rate within a manageable range, significantly lower than the industry average, achieving a win-win situation for both funders and airlines.

[0061] In the several embodiments provided by this invention, it should be understood that the disclosed systems, apparatuses, and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of modules or units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces, or indirect coupling or communication connection of apparatuses or units, and may be electrical, mechanical, or other forms.

[0062] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment, depending on actual needs.

[0063] Furthermore, the functional units in the various embodiments of the present invention can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit.

[0064] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this invention, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) or processor to execute all or part of the steps of the methods of various embodiments of this invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0065] The above description is only a part of the embodiments of the present invention and does not limit the scope of protection of the present invention. Any equivalent device or equivalent process transformation made based on the content of the present invention specification and drawings, or direct or indirect application in other related technical fields, are similarly included within the patent protection scope of the present invention.

Claims

1. A dynamic credit rating method for aviation, characterized in that, Includes the following steps: S1: Select three core data sources: airline historical data, real-time flight data, and macroeconomic policy data, to comprehensively cover historical operations, real-time operations, and external environment dimensions; S2: Construct TCN-RiskModel based on Temporal Convolutional Network (TCN), and train and optimize the model using a federated transfer learning architecture and the Optuna hyperparameter optimization framework; S3: Set three basic weights for operational stability, route potential, and policy risk, and introduce a dynamic adjustment mechanism. The dynamic credit score St is obtained by weighted summation of the three-dimensional scores × the corresponding dynamic weights, where ω1+ω2+ω3=1, ω1 is the weight of operational stability, ω2 is the weight of route potential, and ω3 is the weight of policy risk.

2. The aviation dynamic credit evaluation method according to claim 1, characterized in that, In step S1: The airline's historical data includes financial health indicators, operational performance indicators, and route sustainability indicators. Financial health indicators include historical route subsidy recovery rate, debt-to-equity ratio, and net cash flow. Operational performance indicators include historical flight execution rate, average load factor, and historical bad debt rate. Route sustainability indicators include historical average operating cycle of routes and profitability of similar routes. The real-time flight data is collected in real time through the distributed stream processing engine Apache Flink streaming computing engine and updated once every preset time interval according to the preset delay. It includes real-time operation indicators, asset liquidity indicators, and abnormal event data. Real-time operation indicators include real-time load factor, flight punctuality rate, single aircraft fuel consumption and ground handling costs. Asset liquidity indicators include real-time available seats, remaining ticket sales progress, and short-term accounts payable amount. Abnormal event data includes real-time flight cancellation / delay records and the number of sudden mechanical failures. The macroeconomic policy data includes subsidy policy data, industry regulatory policies, and macroeconomic policies. Subsidy policy data includes adjustments to local government route subsidy standards and changes in fiscal budget amounts. Industry regulatory policies include new regulations for route approval by the Civil Aviation Administration and flight control policies during public health emergencies. Macroeconomic policies include interest rate adjustments and regional economic support policies.

3. The aviation dynamic credit evaluation method according to claim 1, characterized in that, In step S2, the TCN-RiskModel is constructed using causal convolution and dilated convolution structures to capture long- and short-term temporal dependencies in multi-source data, including: TCN-RiskModel abandons the linear modeling approach of traditional time series models and adopts causal convolution to ensure that the modeling process conforms to the chronological logic and avoids future data from interfering with current predictions. By expanding the receptive field through dilated convolution, it can effectively capture long-term time series data while maintaining computational efficiency, adapting to the characteristics of rapid dynamic changes and strong temporal sequence of aviation data.

4. The aviation dynamic credit evaluation method according to claim 1, characterized in that, In step S2, the application of the federated transfer learning architecture includes, while protecting airline data privacy, transferring historical data from existing routes to the model training process for new routes after differential privacy processing, including: The model employs a federated transfer learning architecture, where each airline's data is stored locally. Only the feature information processed with differential privacy is used in the model training, without disclosing the original data. Common credit features are extracted from existing route data and transferred to the new route model to compensate for the lack of historical data for the new routes.

5. The aviation dynamic credit evaluation method according to claim 1, characterized in that, In step S2, the Optuna framework is used to automatically tune the kernel size, dilation rate, and number of hidden layers of the TCN-RiskModel, including: The kernel size, dilation rate, and number of hidden layers of the TCN-RiskModel were set as tuning parameters. The optimization objective was to use the Bayesian optimization algorithm of the Optuna framework to automatically search for the optimal parameter combination with the three dimensions of operational stability, route potential, and policy risk as the optimization objective.

6. The aviation dynamic credit evaluation method according to claim 1, characterized in that, In step S3, the dynamic adjustment mechanism involves introducing a volatility index and adjusting the values ​​of ω1, ω2, and ω3 in real time based on changes in the volatility index, including: The basic weighting focuses on the core risks of aviation financing, setting operational stability ω1 as the highest weight, which is directly related to debt repayment ability; route potential ω2 reflects future profit support, and policy risk ω3 reflects external uncertainty; when the volatility index rises, the weights of ω1 and ω3 are increased, and the weight of ω2 is decreased; when the volatility index falls, the weight of ω2 is appropriately increased to dynamically adapt to changes in risk.

7. The aviation dynamic credit evaluation method according to claim 1, characterized in that, In step S3, the dynamic credit score St is updated as follows: each time a new batch of real-time flight data or macroeconomic policy data is added, the TCN-RiskModel updates the parameters through incremental training to achieve real-time iteration of the dynamic credit score St.

8. The aviation dynamic credit evaluation method according to claim 1, characterized in that, In step S3: the dynamic credit score St is quantified into at least two credit levels, the financing conditions corresponding to each level of risk are clarified, and a direct link between the score, level, and financing plan is established, so that the funder can quickly apply the score results without additional conversion.

9. An aviation dynamic credit rating device, characterized in that, include: The module includes a data source selection module, a training optimization module, and a scoring configuration module. The data source selection module is used to select three types of core data sources: airline historical data, real-time flight data, and macroeconomic policy data, comprehensively covering historical operations, real-time operations, and external environment dimensions. The training optimization module is used to construct a TCN-RiskModel based on the temporal convolutional network TCN, and to train and optimize the model using a federated transfer learning architecture and the hyperparameter optimization Optuna framework. The scoring configuration module is used to set the basic weights of three dimensions: operational stability, route potential, and policy risk. A dynamic adjustment mechanism is introduced, and the dynamic credit score St is obtained by weighted summation of the three-dimensional scores multiplied by the corresponding dynamic weights, where ω1+ω2+ω3=1, ω1 is the operational stability weight, ω2 is the route potential weight, and ω3 is the policy risk weight.

10. A computer device, characterized in that, include: At least one processor; And a memory communicatively connected to at least one processor; wherein the memory stores instructions executable by at least one processor, the instructions being executed by at least one processor to enable at least one processor to perform the aviation dynamic credit evaluation method as described in any one of claims 1 to 8.

Images