Search

CN-122022984-A - Aviation dynamic credit evaluation method and device and computer equipment

CN122022984ACN 122022984 ACN122022984 ACN 122022984ACN-122022984-A

Abstract

The invention discloses an aviation dynamic credit evaluation method, an aviation dynamic credit evaluation device and computer equipment. The method comprises the steps of selecting three types of core data sources which are respectively airline company historical data, real-time flight data and macro policy data, comprehensively covering historical operation-real-time operation-external environment dimensions, constructing TCN-RiskModel based on a time sequence convolution network TCN, training and optimizing a model by adopting a federal transfer learning architecture and a super-parameter optimization framework, setting three-dimensional basic weights of operation stability, airline potential and policy risk, introducing a dynamic adjustment mechanism, and obtaining dynamic credit scores by weighting and summing corresponding dynamic weights through three-dimensional scores, wherein omega 1 +ω 2 +ω 3 =1,ω 1 is the operation stability weight, omega 2 is the airline potential weight, and omega 3 is the policy risk weight. By the mode, reliable quantitative basis can be provided for financing matching, and the performance of the method is obviously superior to that of a traditional model.

Inventors

  • XU HONGJIANG
  • YANG XIAOBO

Assignees

  • 海南太美航空股份有限公司

Dates

Publication Date
20260512
Application Date
20260121

Claims (10)

  1. 1. The aviation dynamic credit evaluation method is characterized by comprising the following steps of: s1, selecting three types of core data sources, namely historical data, real-time flight data and macroscopic policy data of an airline company, and comprehensively covering the dimensions of historical operation, real-time operation and external environment; S2, constructing TCN-RiskModel based on a time sequence convolutional network TCN, and training and optimizing the model by adopting a federal transfer learning architecture and a super-parameter optimization Optuna framework; And S3, setting three-dimensional basic weights of operation stability, route potential and policy risk, introducing a dynamic adjustment mechanism, and obtaining a dynamic credit score St through weighted summation of the three-dimensional score multiplied by the corresponding dynamic weights, wherein omega 1 +ω 2 +ω 3 =1,ω 1 is the operation stability weight, omega 2 is the route potential weight, and omega 3 is the policy risk weight.
  2. 2. The aerodynamics credit evaluation method according to claim 1, wherein in step S1: The airline company historical data comprises financial health indexes, operation performance indexes and airline duration indexes, wherein the financial health indexes comprise historical airline subsidy recovery rate, asset liability rate and cash flow net amount, the operation performance indexes comprise historical flight execution rate, average passenger seat rate and historical bad account rate, and the airline duration indexes comprise historical airline average operation period and same type of airline profit level; The real-time flight data is collected in real time through a distributed stream processing engine APACHE FLINK stream computing engine, and updated once every preset time according to preset time delay, wherein the real-time flight data comprises real-time operation indexes, asset liquidity indexes and abnormal event data, the real-time operation indexes comprise real-time passenger seat rate, flight standard point rate, single-machine fuel consumption and ground service cost, the asset liquidity indexes comprise real-time available seat numbers, residual passenger ticket sales progress and short-term accounts payable amount, and the abnormal event data comprise real-time flight cancellation/delay records and sudden mechanical failure times; the macro policy data comprises subsidy policy data, industry supervision policies and macro economic policies, wherein the subsidy policy data comprises local government airline subsidy standard adjustment and financial budget limit change, the industry supervision policies comprise civil aviation airline approval new rules and public health event flight management and control policies, and the macro economic policies comprise interest rate adjustment and regional economic support policies.
  3. 3. The aerodynamics credit evaluation method according to claim 1, wherein in step S2, the TCN-RiskModel is constructed by a causal convolution and an expanded convolution structure, and is used for capturing long-short-term time-sequence dependency of multi-source data, and the method comprises the following steps: The TCN-RiskModel abandons a linear modeling mode of a traditional time sequence model, adopts causal convolution to ensure that the modeling process accords with time sequence logic, avoids future data from interfering with current prediction, expands a receptive field through expanding convolution, realizes effective capture of long time sequence data, simultaneously maintains calculation efficiency, and adapts to the characteristics of quick dynamic change and strong time sequence of aviation data.
  4. 4. The aerodynamics credit evaluation method according to claim 1, wherein in step S2, the application of the federal transition learning architecture includes migrating the historical data of the existing airlines to the model training process of the new airlines after the differential privacy processing on the premise of protecting the data privacy of the airlines, and the method comprises the following steps: the method adopts a federal transfer learning architecture, data of each airline company is stored locally, only characteristic information subjected to differential privacy treatment participates in model training, original data is not revealed, general credit characteristics in existing airline data are extracted, and the data are transferred to a new airline model, so that the defect of insufficient historical data of the new airline is overcome.
  5. 5. The aerodynamics credit evaluation method according to claim 1, wherein in step S2, the Optuna framework is configured to automatically tune the convolution kernel size, the expansion rate, and the number of hidden layers of TCN-RiskModel, including: The convolution kernel size, the expansion rate and the number of hidden layers of TCN-RiskModel are set as tuning parameters, the three-dimensional grading precision of operation stability, route potential and policy risk is used as an optimization target, and the optimal parameter combination is automatically searched through a Bayesian optimization algorithm of a Optuna framework.
  6. 6. The method of claim 1, wherein in step S3, the dynamic adjustment mechanism is configured to introduce a volatility index, and adjust the value of ω 1 、ω 2 、ω 3 in real time according to the volatility index value change, and the method comprises: The basic weight focuses on aviation financing core risks, the operation stability omega 1 is set to be the highest weight, the repayment capability is directly related, the route potential omega 2 reflects future profit support, the policy risk omega 3 reflects external uncertainty, when the fluctuation rate index is increased, the weights of omega 1 and omega 3 are increased, the weight of omega 2 is reduced, when the fluctuation rate index is reduced, the weight of omega 2 is properly increased, and the risk change is dynamically adapted.
  7. 7. The method for evaluating aviation dynamic credit according to claim 1, wherein in step S3, the dynamic credit score St is updated by adding a batch of real-time flight data or macro policy data, and TCN-RiskModel realizes real-time iteration of the dynamic credit score St by training update parameters in an increment.
  8. 8. The method for evaluating aviation dynamic credit according to claim 1, wherein in step S3, the dynamic credit score St is quantized to at least 2 credit levels, financing conditions corresponding to risks of each level are defined, and direct correlation of score-level-financing schemes is established, so that a fund party can rapidly apply a score result without additional conversion.
  9. 9. An aeronautical dynamic credit evaluation device, comprising: the system comprises a data source selection module, a training optimization module and a scoring configuration module; the data source selection module is used for selecting three types of core data sources, namely historical data of an airline company, real-time flight data and macro policy data, and comprehensively covering the dimensions of historical operation-real-time operation-external environment; The training optimization module is used for constructing TCN-RiskModel based on the time sequence convolutional network TCN and performing training optimization on the model by adopting a federal transfer learning architecture and a super-parameter optimization Optuna framework; The evaluation allocation module is used for setting three-dimensional basic weights of operation stability, route potential and policy risk, introducing a dynamic adjustment mechanism, and obtaining a dynamic credit score St through weighted summation of the three-dimensional score multiplied by the corresponding dynamic weights, wherein omega 1 +ω 2 +ω 3 =1,ω 1 is the operation stability weight, omega 2 is the route potential weight, and omega 3 is the policy risk weight.
  10. 10. A computer device comprising at least one processor and a memory communicatively coupled to the at least one processor, wherein the memory stores instructions executable by the at least one processor to enable the at least one processor to perform the aerodynamics credit assessment method according to any one of claims 1 to 8.

Description

Aviation dynamic credit evaluation method and device and computer equipment Technical Field The present invention relates to the field of aviation management technologies, and in particular, to an aviation dynamic credit evaluation method, an aviation dynamic credit evaluation device, and computer equipment. Background In the aviation field route subsidy receivables financing scene, the fund party needs to judge the repayment capacity and financing risk of the airlines through accurate credit evaluation so as to realize efficient matching financing. In the prior art, the traditional time sequence model has the obvious defects that firstly, long-period time sequence dependency relationship of multisource data in the aviation field cannot be effectively captured, the aviation data cover multiple dimensions such as historical operation, real-time operation and external policy, and the traditional model is difficult to comprehensively correlate and analyze, secondly, the general wind control model is not pertinently incorporated into aviation exclusive indexes such as airline subsidy and flight operation, and has the problem of water and soil inadequacy, so that the deviation between scoring and actual risk is large, thirdly, the model weight is fixed, the real-time flight abnormality, policy mutation and other data changes cannot be dynamically responded, and a new airline lacks historical data support, so that the problem of data cold starting exists, and the scoring real-time performance is insufficient and the confidence is low. Therefore, a dynamic credit evaluation method which is adaptive to the characteristics of the aviation field and has both real-time performance and accuracy is needed. Disclosure of Invention In view of the above, the invention aims to provide an aviation dynamic credit evaluation method, an aviation dynamic credit evaluation device and computer equipment, which can provide a credible quantitative basis for financing matching and are obviously superior to the performance of the traditional model. According to one aspect of the invention, an aviation dynamic credit evaluation method is provided, which comprises the following steps of S1, selecting three types of core data sources, namely historical data of an airline company, real-time flight data and macroscopic policy data, and comprehensively covering historical operation-real-time operation-external environment dimensions, S2, constructing TCN-RiskModel based on a time sequence convolution network TCN, performing training optimization on a model by adopting a federal transfer learning architecture and a super-parameter optimization Optuna framework, S3, setting three-dimensional basic weights of operation stability, airline potential and policy risk, introducing a dynamic adjustment mechanism, and obtaining a dynamic credit score St by weighting and summing the three-dimensional scores and the corresponding dynamic weights, wherein omega 1+ω2+ω3=1,ω1 is the operation stability weight, omega 2 is the airline potential weight, and omega 3 is the policy risk weight. In step S1, the historical data of the airline company include financial health indexes, operation performance indexes and airline survival indexes, wherein the financial health indexes include historical airline subsidy recovery rate, asset liability rate and cash flow net, the operation performance indexes include historical flight execution rate, average passenger seat rate and historical bad account rate, the airline survival indexes include historical airline average operation period and the same type of airline profit level, the real-time flight data are collected in real time through a distributed stream processing engine APACHE FLINK stream computing engine and updated every preset time according to preset time delay, the real-time flight data comprise real-time operation indexes, asset flowability indexes and abnormal event data, the real-time operation indexes comprise real-time passenger seat rate, flight standard point rate, single-machine fuel consumption and ground service cost, the asset flowability indexes comprise real-time available seat number, residual ticket selling progress, accounts payable amount, the abnormal event data comprise real-time flight cancellation/error records and the number of sudden mechanical faults, the macroscopic policy data comprise the local government policy and the macroscopic economic policy, the local government policy and the policy, the financial policy is regulated, the public economic policy is regulated, the public policy is regulated and the public economic policy is regulated by the public policy is a major policy is regulated by the public and the public economic policy is regulated by the public policy and the public service policy is a major. In step S2, the TCN-RiskModel is constructed through a causal convolution and expansion convolution structure and is used for capturing long-period time sequence dependency