Civil aviation multi-scene coupled intelligent auxiliary model construction method and system

Abstract

The invention belongs to the technical field of aviation safety, and discloses a civil aviation multi-scene coupled intelligent auxiliary model construction method and system, and the method comprises the steps: 1, constructing a risk scene coupled flight simulation model based on risk scene and coupling action mechanism research, and generating basic data for artificial intelligence learning training; 2, for each group of coupling risk scenes, setting a triggering condition and a triggering action which conform to the actual situation of the scene and contain randomness, and generating coupling scene data; and step 3, by constructing a hybrid deep learning model which combines a long short-term memory network and a full convolutional neural network and is used for time sequence modeling and feature extraction, training and verification of an identification model of the coupling risk scene are carried out, and the identification success rate needs to be greater than 95%. According to the method, the problem that the traditional identification and early warning research aiming at the risk scene generally only aims at a single scene and does not consider the coupling condition of multiple risk scenes is solved, and the pilot is assisted through the identification and early warning of the coupling risk.

Description

Technical Field

[0001] This invention belongs to the field of aviation safety and artificial intelligence technology application technology, specifically relating to a method and system for constructing intelligent auxiliary models that couple multiple scenarios in civil aviation. Background Technology

[0002] Flight safety has always been an important research direction in the aviation field. When civil aircraft are flying in the air, various abnormal wind field scenarios may occur due to the influence of people, machines, environment and other factors. Traditional research on the identification and early warning of risk scenarios usually only focuses on a single scenario and does not consider the coupling of multiple risk scenarios. As a result, misjudgment or insufficient judgment may occur, leading to ineffective handling methods and flight safety problems.

[0003] The rapid development of artificial intelligence technology has made its application prospects in the aviation field increasingly apparent. Artificial intelligence deep learning models can process time-series data and identify complex scenarios, thereby effectively assisting pilots in judging the current flight status and detecting abnormal situations as early as possible.

[0004] The present invention proposes a method for constructing a multi-scenario coupled intelligent auxiliary model in civil aviation. This method involves studying the coupling characteristics of multiple scenarios, constructing a flight simulation program to generate data that can be used to train an intelligent recognition model, and training and verifying the intelligent recognition model based on an artificial intelligence model oriented towards time-series data. The generated intelligent recognition model can effectively identify coupled scenarios, achieve effective assistance to pilots, and provide a reference for the effective application of artificial intelligence technology in the field of flight safety. Summary of the Invention

[0005] This invention addresses the problem that traditional risk scenario identification and early warning research typically focuses on a single scenario and does not consider the coupling of multiple risk scenarios, which can lead to misjudgments or insufficient judgments. It provides a method and system for constructing an intelligent auxiliary model for multiple coupled scenarios in civil aviation. Based on the training of an intelligent identification model for coupled risk scenarios, it provides assistance to pilots by identifying and issuing early warnings for coupled risks.

[0006] The technical solution of this invention is implemented as follows: In a first aspect, the present invention provides a method for constructing an intelligent auxiliary model that couples multiple scenarios in civil aviation, comprising the following steps: Step 1: Based on the research of risk scenarios and coupling mechanisms, construct a flight simulation model of risk scenario coupling to generate basic data for artificial intelligence learning and training; Step 2: For each set of coupled risk scenarios, set triggering conditions and triggering actions that conform to the actual situation of the scenario and include randomness, and generate coupled scenario data; Step 3: By constructing a hybrid deep learning model that combines a long short-term memory network and a fully convolutional neural network for temporal modeling and feature extraction, we will conduct training and validation of the identification model for coupled risk scenarios. The identification success rate should be greater than 95%.

[0007] As a further aspect of the present invention: step 1 specifically comprises: Construction of flight simulation models with coupled risk scenarios: Based on the research on risk scenarios and coupling mechanisms, the existence and combination mechanism of risk scenarios is clarified, and a flight simulation model with coupled risk scenarios is constructed. The flight simulation program generates basic data for artificial intelligence learning and training. The combination of scenarios includes six types of situations: "single engine failure + low approach energy", "all engines failure + low approach energy", "single engine failure + course deviation", "single engine failure + course intrusion", "icing + control surface failure", and "single engine failure + control surface failure".

[0008] As a further aspect of this invention: First, it is necessary to construct seven basic risk scenario models, including single-engine failure, all-engine failure, icing, low approach energy, course deviation, course intrusion, and control surface failure. The triggering logic and triggering mechanism for each scenario are as follows: (1) When a single engine failure scenario is triggered, the input engine parameters of the corresponding engine simulation module are set to 0, so that the output engine thrust is 0, which serves as the triggering mechanism for the single engine failure scenario. (2) When the all-engine failure scenario is triggered, the input parameters of all engine channels in the engine simulation module are set to 0, so that the engine thrust output of all channels is 0, which serves as the triggering mechanism for the all-engine failure scenario. (3) When the icing scene is triggered, the relevant aerodynamic parameters are modified according to the formula based on the set icing severity, so that the aerodynamic parameters output by the aerodynamic calculation module change, which serves as the triggering mechanism for the icing scene; (4) When approaching a low-energy scenario, the trigger operation is performed by manipulating the output terminal. For low potential energy triggering, the standard glide path is manipulated downwards. For low kinetic energy triggering, the speed is reduced to guide the trigger judgment conditions. (5) When the flight path deviation scenario is triggered, the aircraft is set to fly along the set flight path. At the trigger time, random turning maneuvers are performed until the aircraft deviates from the allowed distance and angle from the flight path. (6) When the flight path intrusion scenario is triggered, the position, altitude and speed of the intruding aircraft are initialized, and the flight target of the intruding aircraft is set to the current position of the local aircraft until the position and speed relationship between the two aircraft corresponding to the scenario judgment condition is reached to trigger the scenario. (7) When the rudder surface failure scenario is triggered, the rudder deflection position input by the aerodynamic calculation module is set to a constant value according to the set trigger rudder surface type, which is regarded as the occurrence of rudder surface jamming, and is used as the triggering mechanism of the rudder surface failure scenario.

[0009] Once the above 7 basic scenario models are constructed, the triggering mechanisms of the corresponding two scenarios will be activated simultaneously based on the combination of multiple scenario couplings, so as to achieve simultaneous triggering of coupled scenarios.

[0010] The basic scenarios for the coupled scenarios are seven: single-engine failure, all-engine failure, low approach energy, flight path deviation, flight path intrusion, icing, and control surface failure. There are six combinations of coupled scenarios: "single-engine failure + low approach energy," "all-engine failure + low approach energy," "single-engine failure + flight path deviation," "single-engine failure + flight path intrusion," "icing + control surface failure," and "single-engine failure + control surface failure." Each of these six coupled scenarios generates at least 1000 sets of data, with each set requiring a simulation time of at least 30 seconds and at least 50 data points per second. Feature parameters include aircraft position, altitude, velocity, attitude angles, angular rates, and scenario parameters.

[0011] As a further aspect of the present invention: step 2 specifically comprises: Identify the collection and preprocessing of training data: For each set of coupled risk scenarios, set triggering conditions and triggering actions that conform to the actual situation of the scenario and include randomness, generate coupled scenario data, and include flag parameters indicating whether the scenario has occurred.

[0012] As a further aspect of the present invention: the collection of training data is specifically as follows: Each combined scenario generates 1,000 or more sets of data. The simulation time for each set of data is no less than 30 seconds, and the simulation step size is 0.01 seconds. Therefore, each set of data contains at least 3,000 data points, and each data point involves 38 characteristic parameters of the aircraft. When generating data, random numbers were set for different coupling scenarios, by randomizing the aircraft's initial position, altitude, speed, and heading angle parameters.

[0013] As a further aspect of the present invention: the preprocessing of the identification training data specifically includes: After the data is generated, it is necessary to perform sliding window preprocessing on the data. Each window has no less than 50 data entries and a sliding step size of 20. Finally, the array generated by the sliding window is used as the input for each model recognition. At the same time, corresponding scene flags are generated. The scene flags are divided into 7 states, which are the 7 basic scenes of the aircraft. When all 7 state values ​​are 0, it means that the aircraft is in a normal state. When the coupled scene is triggered, the state values ​​of the two corresponding basic scenes are 1.

[0014] As a further aspect of the present invention: step 3 specifically comprises: Training and validation of the recognition model: Based on the PyTorch framework, a hybrid deep learning model combining a long short-term memory network and a fully convolutional neural network for temporal modeling and feature extraction is constructed to train a recognition model for coupled risk scenarios. The training effect is improved by adjusting hyperparameters, and the trained artificial intelligence model is validated. The recognition success rate should be greater than 95%.

[0015] As a further aspect of the present invention: the coupled risk scenario dataset is divided into a training set and a test set in a 7:3 ratio, the LSTM-FCN algorithm is used to perform offline learning on the existing coupled risk scenario data training set, and the trained model is applied to online coupled scenario anomaly detection; For each complete flight data segment, its sliding window is sampled and used as the input to the model, and supervised classification training is performed. During inference, a data buffer is designed to store the flight data sampled in real time, and the length of the buffer is the same as the length of the sliding window. After each sampling, the LSTM-FCN algorithm classifies the buffer data. When the same abnormal result is output multiple times in a row, the LSTM-FCN algorithm will determine that the aircraft is in an abnormal state and change the status flag.

[0016] As a further aspect of this invention: the LSTM-FCN algorithm, specifically: Perceiving the same time-series input through two different views; For the FCN part, its fully convolutional block consists of three stacked temporal convolutional blocks with filter sizes of 128, 256 and 128 respectively. Each block consists of a temporal convolutional layer, accompanied by batch normalization, followed by a ReLU activation function. After the third temporal convolutional block, global average pooling is applied to form a global pooling layer. For the LSTM part, the time series input is passed through the dimension rearrangement layer, and the transformed time series from the dimension rearrangement is then passed to the LSTM block, which consists of ordinary layers or interest layers, followed by the dropout layer. Finally, the outputs of the FCN and LSTM parts are concatenated and passed to the SoftMax classification layer.

[0017] The artificial intelligence model used is an "LSTM+FCN" structure. For each set of completed flight scene data, a sliding window will be used to sample the data, which will then be used as the input of the model for supervised classification training. The trained artificial intelligence recognition model can effectively identify coupled scenes.

[0018] Secondly, the present invention provides an intelligent auxiliary model construction system for multi-scenario coupling in civil aviation, comprising: Flight simulation model construction module: Based on the research of risk scenarios and coupling mechanisms, the existence and combination mechanism of risk scenarios is clarified, and a flight simulation model of risk scenario coupling is constructed. The flight simulation program generates basic data for artificial intelligence learning and training. The combination of scenarios includes six types of situations: "single engine failure + low approach energy", "all engines failure + low approach energy", "single engine failure + course deviation", "single engine failure + course intrusion", "icing + control surface failure", and "single engine failure + control surface failure". Data acquisition and preprocessing module: For each set of coupled risk scenarios, set triggering conditions and triggering actions that conform to the actual situation of the scenario and include randomness, generate coupled scenario data, and include flag parameters for whether the scenario has occurred; Training and Validation Module: Based on the PyTorch framework, a hybrid deep learning model combining a long short-term memory network and a fully convolutional neural network is constructed for temporal modeling and feature extraction. The model is used to train a recognition model for coupled risk scenarios. The training effect is improved by adjusting hyperparameters. The trained artificial intelligence model is then validated, and the recognition success rate must be greater than 95%.

[0019] Compared with the prior art, the beneficial effects of the present invention are as follows: 1. This invention targets multiple civil aviation risk scenarios and, based on their multi-risk coupling, trains a multi-scenario risk identification model using artificial intelligence algorithms for real-time identification and early warning of multiple risk categories.

[0020] 2. This invention designs, trains, and constructs an intelligent recognition model for coupled scenarios, which can achieve effective recognition under coupled risk scenarios, meet the intelligent recognition and early warning needs of civil aircraft in complex states, and provide an effective early warning measure reference for pilot assistance and flight safety management.

[0021] 3. The LSTM-FCN algorithm of this invention performs well overall, can accurately analyze various aircraft states, has a certain degree of robustness, and can effectively identify coupled scenarios.

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

[0023] Figure 1 A flowchart illustrating the overall research process for constructing an intelligent auxiliary model that couples multiple scenarios in civil aviation. Figure 2 This is a framework diagram of the "LSTM+FCN" intelligent model. Detailed Implementation

[0024] To make the objectives, technical solutions, and advantages of this invention clearer, the technical solutions in the embodiments of this invention will be described in more detail below with reference to the accompanying drawings.

[0025] In the accompanying drawings, the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. The described embodiments are some, but not all, of the embodiments of the present invention.

[0026] The embodiments described below with reference to the accompanying drawings are exemplary and intended to explain the present invention, and should not be construed as limiting the present invention. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without inventive effort are within the scope of protection of the present invention.

[0027] The following is in conjunction with the appendix Figure 1-2 The embodiments of the present invention will be described in detail below.

[0028] Example 1 This invention discloses a method for constructing an intelligent auxiliary model that couples multiple scenarios in civil aviation, comprising the following steps: Step 1: Construction of a flight simulation model with coupled risk scenarios. Based on the research on risk scenarios and coupling mechanisms, the existence and combination mechanism of risk scenarios is clarified, and a flight simulation model with coupled risk scenarios is constructed. The flight simulation program generates basic data for artificial intelligence learning and training. The scenario combinations include six types of situations: "single engine failure + low approach energy", "all engines failure + low approach energy", "single engine failure + course deviation", "single engine failure + course intrusion", "icing + control surface failure", and "single engine failure + control surface failure".

[0029] Step 2: Identify the collection and preprocessing of training data. For each set of coupled risk scenarios, set triggering conditions and actions that conform to the actual situation of the scenario and include randomness, generate coupled scenario data, and include flag parameters indicating whether the scenario has occurred.

[0030] Step 3: Training and Validation of the Recognition Model. Based on the PyTorch framework, a hybrid deep learning model combining Long Short-Term Memory (LSTM) and Fully Convolutional Neural Network (FCN) for temporal modeling and feature extraction was constructed. The recognition model for coupled risk scenarios was trained, and the training effect was improved by adjusting hyperparameters. The trained artificial intelligence model was then validated, and the recognition success rate was greater than 95%.

[0031] Furthermore, in step 1, based on the research on risk scenarios and coupling mechanisms, the existence and combination mechanisms of risk scenarios are clarified, and a flight simulation model of risk scenario coupling is constructed. First, it is necessary to construct models for seven basic risk scenarios, including single-engine failure, total engine failure, icing, low approach energy, course deviation, course intrusion, and control surface failure. The triggering logic and triggering mechanism for each scenario are as follows: (1) When a single engine failure scenario is triggered, the input engine parameters of the corresponding engine simulation module are set to 0, so that the output engine thrust is 0, which serves as the triggering mechanism for the single engine failure scenario. (2) When the all-engine failure scenario is triggered, the input parameters of all engine channels in the engine simulation module are set to 0, so that the engine thrust output of all channels is 0, which serves as the triggering mechanism for the all-engine failure scenario. (3) When the icing scene is triggered, the relevant aerodynamic parameters are modified according to the formula based on the set icing severity, so that the aerodynamic parameters output by the aerodynamic calculation module change, which serves as the triggering mechanism for the icing scene; (4) When approaching a low-energy scenario, the trigger operation is performed by manipulating the output terminal. For low potential energy triggering, the standard glide path is manipulated downwards. For low kinetic energy triggering, the speed is reduced to guide the trigger judgment conditions. (5) When the flight path deviation scenario is triggered, the aircraft is set to fly along the set flight path. At the trigger time, random turning maneuvers are performed until the aircraft deviates from the allowed distance and angle from the flight path. (6) When the flight path intrusion scenario is triggered, the position, altitude and speed of the intruding aircraft are initialized, and the flight target of the intruding aircraft is set to the current position of the local aircraft until the position and speed relationship between the two aircraft corresponding to the scenario judgment condition is reached to trigger the scenario. (7) When the rudder surface failure scenario is triggered, the rudder deflection position input by the aerodynamic calculation module is set to a constant value according to the set trigger rudder surface type, which is regarded as the occurrence of rudder surface jamming, and is used as the triggering mechanism of the rudder surface failure scenario.

[0032] Once the above 7 basic scenario models are constructed, the triggering mechanisms of the corresponding two scenarios will be activated simultaneously based on the combination of multiple scenario couplings, so as to achieve simultaneous triggering of coupled scenarios.

[0033] Furthermore, in step 2, during the collection and preprocessing of identification data, a total of 1000 sets of data are generated for each combined scenario. The simulation time for each set of data is 30s, and the simulation step size is 0.01s. Therefore, each set of data contains 3000 data points. Each data point involves 38 feature parameters such as the aircraft's position, altitude, speed, attitude angle, angular rate, and control input. In order to increase data diversity, random numbers are set for the specific circumstances of different coupled scenarios during data generation. By randomizing the initial position, altitude, speed, heading angle, and other parameters of the aircraft, the data becomes more diverse, increasing the generalization ability of the model.

[0034] After the data is generated, it is necessary to perform sliding window preprocessing on the data. The data volume of each window is set to 50, and the sliding step size is 20. Finally, the array generated by the sliding window is used as the input for each model recognition. At the same time, the corresponding scene flags are generated. The scene flags are divided into 7 states, which are the 7 basic scenes of the aircraft. When all 7 state values ​​are 0, it means that the aircraft is in a normal state. When the coupled scene is triggered, the state values ​​of the two corresponding basic scenes are 1.

[0035] Further, in step 3, the coupled risk scenario dataset is divided into a training set and a test set in a 7:3 ratio. The LSTM-FCN algorithm is used for offline learning on the existing coupled risk scenario data training set, and the trained model is applied to online coupled scenario anomaly detection. For each complete flight data segment, its sliding window sampling is used as the model input, and supervised classification training is performed. During inference, a data buffer is designed to store the real-time sampled flight data, with a buffer length equal to the sliding window length. After each sampling, the algorithm classifies the buffer data. When the same abnormal result is continuously output multiple times, the algorithm determines that the aircraft is in an abnormal state and changes the status flag.

[0036] The proposed model consists of two parts: FCN and LSTM, which perceive the same time-series input from two different perspectives. For the FCN part, its fully convolutional blocks consist of three stacked temporal convolutional blocks with filter sizes of 128, 256, and 128, respectively. Each block comprises a temporal convolutional layer, followed by batch normalization and a ReLU activation function. After the third temporal convolutional block, global average pooling is applied to form a global pooling layer. For the LSTM part, the time-series input is passed through a dimension rearrangement layer. The transformed time-series from the dimension rearrangement is then passed to an LSTM block, which consists of either a normal layer or a focus layer, followed by a dropout layer. Finally, the outputs of the FCN and LSTM parts are concatenated and passed to a SoftMax classification layer.

[0037] The LSTM-FCN algorithm performs well overall, accurately analyzing various aircraft states, exhibiting robustness, and effectively recognizing coupled scenarios.

[0038] Example 2 This invention discloses a method for constructing an intelligent auxiliary model that couples multiple scenarios in civil aviation, comprising the following steps: Step 1: Construction of a flight simulation model with coupled risk scenarios. Based on the research on risk scenarios and coupling mechanisms, the existence and combination mechanism of risk scenarios is clarified, and a flight simulation model with coupled risk scenarios is constructed. The flight simulation program generates basic data for artificial intelligence learning and training. The scenario combinations include six types of situations: "single engine failure + low approach energy", "all engines failure + low approach energy", "single engine failure + course deviation", "single engine failure + course intrusion", "icing + control surface failure", and "single engine failure + control surface failure".

[0039] Step 2: Identify the collection and preprocessing of training data. For each set of coupled risk scenarios, set triggering conditions and actions that conform to the actual situation of the scenario and include randomness, generate coupled scenario data, and include flag parameters indicating whether the scenario has occurred.

[0040] Step 3: Training and Validation of the Recognition Model. Based on the PyTorch framework, a hybrid deep learning model combining Long Short-Term Memory (LSTM) and Fully Connected Convolutional Network (FCN) for temporal modeling and feature extraction was constructed. This model was used to train a recognition model for coupled risk scenarios. Hyperparameter adjustments were made to improve the training effect, and the trained AI model was validated, achieving a recognition success rate greater than 95%.

[0041] Furthermore, in step 1, based on the research on risk scenarios and coupling mechanisms, the existence and combination mechanisms of risk scenarios are clarified, and a flight simulation model of risk scenario coupling is constructed. First, it is necessary to construct seven basic risk scenario models, including single-engine failure, total engine failure, icing, low approach energy, course deviation, course intrusion, and control surface failure. The triggering logic and triggering mechanism for each scenario are as follows: (1) When a single engine failure scenario is triggered, the input engine parameters of the corresponding engine simulation module are set to 0, so that the output engine thrust is 0, which serves as the triggering mechanism for the single engine failure scenario. (2) When the all-engine failure scenario is triggered, the input parameters of all engine channels in the engine simulation module are set to 0, so that the engine thrust output of all channels is 0, which serves as the triggering mechanism for the all-engine failure scenario. (3) When the icing scene is triggered, the relevant aerodynamic parameters are modified according to the formula based on the set icing severity, so that the aerodynamic parameters output by the aerodynamic calculation module change, which serves as the triggering mechanism for the icing scene; (4) When approaching a low-energy scenario, the trigger operation is performed by manipulating the output terminal. For low potential energy triggering, the standard glide path is manipulated downwards. For low kinetic energy triggering, the speed is reduced to guide the trigger judgment conditions. (5) When the flight path deviation scenario is triggered, the aircraft is set to fly along the set flight path. At the trigger time, random turning maneuvers are performed until the aircraft deviates from the allowed distance and angle from the flight path. (6) When the flight path intrusion scenario is triggered, the position, altitude and speed of the intruding aircraft are initialized, and the flight target of the intruding aircraft is set to the current position of the local aircraft until the position and speed relationship between the two aircraft corresponding to the scenario judgment condition is reached to trigger the scenario. (7) When the rudder surface failure scenario is triggered, the rudder deflection position input by the aerodynamic calculation module is set to a constant value according to the set trigger rudder surface type, which is regarded as the occurrence of rudder surface jamming, and is used as the triggering mechanism of the rudder surface failure scenario.

[0042] Once the above 7 basic scenario models are constructed, the triggering mechanisms of the corresponding two scenarios will be activated simultaneously based on the combination of multiple scenario couplings, so as to achieve simultaneous triggering of coupled scenarios.

[0043] Furthermore, in step 2, during the collection and preprocessing of identification data, a total of 1000 sets of data are generated for each combined scenario. The simulation time for each set of data is 30s, and the simulation step size is 0.01s. Therefore, each set of data contains 3000 data points. Each data point involves 38 feature parameters such as the aircraft's position, altitude, speed, attitude angle, angular rate, and control input. In order to increase data diversity, random numbers are set for the specific circumstances of different coupled scenarios during data generation. By randomizing the initial position, altitude, speed, heading angle, and other parameters of the aircraft, the data becomes more diverse, increasing the generalization ability of the model.

[0044] After the data is generated, it is necessary to perform sliding window preprocessing on the data. The data volume of each window is set to 50, and the sliding step size is 20. Finally, the array generated by the sliding window is used as the input for each model recognition. At the same time, the corresponding scene flags are generated. The scene flags are divided into 7 states, which are the 7 basic scenes of the aircraft. When all 7 state values ​​are 0, it means that the aircraft is in a normal state. When the coupled scene is triggered, the state values ​​of the two corresponding basic scenes are 1.

[0045] Furthermore, in step 3, the LSTM-FCN algorithm is used for offline learning on the existing coupled risk scenario dataset, and the trained model is applied to online coupled scenario anomaly detection. For each complete flight data segment, a sliding window is used as the input to the model for supervised classification training. During inference, a data buffer is designed to store the sampled flight data in real time, with a buffer length equal to the sliding window length. After each sampling, the algorithm classifies the data in the buffer. When the same anomalous result is continuously output multiple times, the algorithm determines that the aircraft is in an abnormal state and changes the status flag.

[0046] The coupled risk scenario dataset is divided into training and test sets in a 7:3 ratio. After obtaining the dataset, the rearranged time series are converted into 3D tensors as inputs to the model. Temporal convolution has proven to be an effective learning model for solving time series classification problems. Before classification, global average pooling is used to reduce the number of parameters in the model. In the proposed model, an LSTM block is introduced on top of a fully convolutional block, followed by a dropout layer. The fully convolutional block consists of three stacked temporal convolutional blocks with filter sizes of 128, 256, and 128 respectively. The structure of each convolutional block is the same as that of a convolutional block in a CNN, containing a temporal convolutional layer and a subsequent ReLU activation function. Global average pooling is applied after the final convolutional block. Simultaneously, the time series input is passed through a dimension rearrangement layer. The transformed time series from the dimension rearrangement is then passed to an LSTM block, which consists of ordinary or attention layers, followed by a dropout layer. The outputs of the global pooling layer and the block are concatenated and passed to a SoftMax classification layer. The fully convolutional block and the block perceive the same time series input from two different views.

[0047] The LSTM+FCN algorithm performs well overall, accurately analyzing various aircraft states, exhibiting robustness, and enabling effective recognition of coupled scenarios.

[0048] Example 3 This invention discloses a method for constructing an intelligent auxiliary model that couples multiple scenarios in civil aviation, comprising the following steps: To construct a flight simulation model with coupled risk scenario triggers, the research on risk scenarios and coupling mechanisms is first used to clarify the existence and combination mechanism of risk scenarios. Then, a flight simulation model with coupled risk scenarios is constructed. Based on the completion of the basic risk scenario model, six scenarios are combined: "single engine failure + low approach energy", "all engines failure + low approach energy", "single engine failure + course deviation", "single engine failure + course intrusion", "icing + control surface failure", and "single engine failure + control surface failure".

[0049] For each set of coupled risk scenarios, triggering conditions and actions that conform to the actual situation of the scenario and include randomness are set to generate coupled scenario data, which also includes a flag parameter indicating whether the scenario has occurred. The generated coupled scenario parameters are sampled using a sliding window and used as input to the model. The coupled risk scenario dataset is then divided into training and test sets in a 7:3 ratio.

[0050] A smart recognition model architecture based on LSTM and FCN algorithms was constructed. The smart recognition model was trained using a dataset, and the hyperparameters were adjusted during the training process to improve the training effect. Finally, the trained smart recognition model for coupled scenes was validated, with recognition accuracy as the validation metric and recognition time as the evaluation parameter for model efficiency. The final effect of the recognition model was obtained. In the validation, the model's recognition accuracy was greater than 95%, and the average time was 4.5 milliseconds, indicating that the model has high recognition efficiency and good recognition effect.

[0051] Thus, the objective of this invention has been achieved.

[0052] The above are merely preferred embodiments of the present invention and are not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for constructing an intelligent auxiliary model that couples multiple scenarios in civil aviation, characterized in that, Includes the following steps: Step 1: Based on the research of risk scenarios and coupling mechanisms, construct a flight simulation model of risk scenario coupling to generate basic data for artificial intelligence learning and training; Step 2: For each set of coupled risk scenarios, set triggering conditions and triggering actions that conform to the actual situation of the scenario and include randomness, and generate coupled scenario data; Step 3: By constructing a hybrid deep learning model that combines a long short-term memory network and a fully convolutional neural network for temporal modeling and feature extraction, we will conduct training and validation of the identification model for coupled risk scenarios. The identification success rate should be greater than 95%.

2. The intelligent auxiliary model construction method for multi-scenario coupling in civil aviation according to claim 1, characterized in that, Step 1 specifically includes: Construction of flight simulation models with coupled risk scenarios: Based on the research on risk scenarios and coupling mechanisms, the existence and combination mechanism of risk scenarios is clarified, and a flight simulation model with coupled risk scenarios is constructed. The flight simulation program generates basic data for artificial intelligence learning and training. The combination of scenarios includes six types of situations: "single engine failure + low approach energy", "all engines failure + low approach energy", "single engine failure + course deviation", "single engine failure + course intrusion", "icing + control surface failure", and "single engine failure + control surface failure".

3. The intelligent auxiliary model construction method for multi-scenario coupling in civil aviation according to claim 2, characterized in that, First, it is necessary to construct models for seven basic risk scenarios, including single-engine failure, total engine failure, icing, low approach energy, course deviation, course intrusion, and control surface failure. The triggering logic and mechanism for each scenario are as follows: (1) When a single engine failure scenario is triggered, the input engine parameters of the corresponding engine simulation module are set to 0, so that the output engine thrust is 0, which serves as the triggering mechanism for the single engine failure scenario. (2) When the all-engine failure scenario is triggered, the input parameters of all engine channels in the engine simulation module are set to 0, so that the engine thrust output of all channels is 0, which serves as the triggering mechanism for the all-engine failure scenario. (3) When the icing scene is triggered, the relevant aerodynamic parameters are modified according to the formula based on the set icing severity, so that the aerodynamic parameters output by the aerodynamic calculation module change, which serves as the triggering mechanism for the icing scene; (4) When approaching a low-energy scenario, the trigger operation is performed by manipulating the output terminal. For low potential energy triggering, the standard glide path is manipulated downwards. For low kinetic energy triggering, the speed is reduced to guide the trigger judgment conditions. (5) When the flight path deviation scenario is triggered, the aircraft is set to fly along the set flight path. At the trigger time, random turning maneuvers are performed until the aircraft deviates from the allowed distance and angle from the flight path. (6) When the flight path intrusion scenario is triggered, the position, altitude and speed of the intruding aircraft are initialized, and the flight target of the intruding aircraft is set to the current position of the local aircraft until the position and speed relationship between the two aircraft corresponding to the scenario judgment condition is reached to trigger the scenario. (7) When the rudder surface failure scenario is triggered, the rudder deflection position input by the aerodynamic calculation module is set to a constant value according to the set trigger rudder surface type, which is regarded as the occurrence of rudder surface jamming, and is used as the triggering mechanism of the rudder surface failure scenario. Once the above 7 basic scenario models are constructed, the triggering mechanisms of the corresponding two scenarios will be activated simultaneously based on the combination of multiple scenario couplings, so as to achieve simultaneous triggering of coupled scenarios.

4. The intelligent auxiliary model construction method for multi-scenario coupling in civil aviation according to claim 1, characterized in that, Step 2 specifically includes: Identify the collection and preprocessing of training data: For each set of coupled risk scenarios, set triggering conditions and triggering actions that conform to the actual situation of the scenario and include randomness, generate coupled scenario data, and include flag parameters indicating whether the scenario has occurred.

5. The intelligent auxiliary model construction method for multi-scenario coupling in civil aviation according to claim 4, characterized in that, The collection of identification training data is specifically as follows: Each combined scenario generates 1,000 or more sets of data. The simulation time for each set of data is no less than 30 seconds, and the simulation step size is 0.01 seconds. Therefore, each set of data contains at least 3,000 data points, and each data point involves 38 characteristic parameters of the aircraft. When generating data, random numbers were set for different coupling scenarios, by randomizing the aircraft's initial position, altitude, speed, and heading angle parameters.

6. The intelligent auxiliary model construction method for multi-scenario coupling in civil aviation according to claim 4, characterized in that, The preprocessing of the training data is as follows: After the data is generated, it is necessary to perform sliding window preprocessing on the data. Each window has no less than 50 data entries and a sliding step size of 20. Finally, the array generated by the sliding window is used as the input for each model recognition. At the same time, corresponding scene flags are generated. The scene flags are divided into 7 states, which are the 7 basic scenes of the aircraft. When all 7 state values ​​are 0, it means that the aircraft is in a normal state. When the coupled scene is triggered, the state values ​​of the two corresponding basic scenes are 1.

7. The method for constructing a multi-scenario coupled intelligent auxiliary model for civil aviation according to claim 1, characterized in that, Step 3 specifically includes: Training and validation of the recognition model: Based on the PyTorch framework, a hybrid deep learning model combining a long short-term memory network and a fully convolutional neural network for temporal modeling and feature extraction is constructed to train a recognition model for coupled risk scenarios. The training effect is improved by adjusting hyperparameters, and the trained artificial intelligence model is validated. The recognition success rate should be greater than 95%.

8. The intelligent auxiliary model construction method for multi-scenario coupling in civil aviation according to claim 7, characterized in that, The coupled risk scenario dataset was divided into a training set and a test set in a 7:3 ratio. The LSTM-FCN algorithm was used to learn the existing coupled risk scenario training set offline, and the trained model was applied to online coupled scenario anomaly detection. For each complete flight data segment, its sliding window is sampled and used as the input to the model, and supervised classification training is performed. During inference, a data buffer is designed to store the flight data sampled in real time, and the length of the buffer is the same as the length of the sliding window. After each sampling, the LSTM-FCN algorithm classifies the buffer data. When the same abnormal result is output multiple times in a row, the LSTM-FCN algorithm will determine that the aircraft is in an abnormal state and change the status flag.

9. The intelligent auxiliary model construction method for multi-scenario coupling in civil aviation according to claim 8, characterized in that, The LSTM-FCN algorithm is as follows: Perceiving the same time-series input through two different views; For the FCN part, its fully convolutional block consists of three stacked temporal convolutional blocks with filter sizes of 128, 256 and 128 respectively. Each block consists of a temporal convolutional layer, accompanied by batch normalization, followed by a ReLU activation function. After the third temporal convolutional block, global average pooling is applied to form a global pooling layer. For the LSTM part, the time series input is passed through the dimension rearrangement layer, and the transformed time series from the dimension rearrangement is then passed to the LSTM block, which consists of ordinary layers or interest layers, followed by the dropout layer. Finally, the outputs of the FCN and LSTM parts are concatenated and passed to the SoftMax classification layer.

10. A civil aviation multi-scenario coupled intelligent auxiliary model construction system, characterized in that, include: Flight simulation model construction module: Based on the research of risk scenarios and coupling mechanisms, the existence and combination mechanism of risk scenarios is clarified, and a flight simulation model of risk scenario coupling is constructed. The basic data for artificial intelligence learning and training is generated through flight simulation program. The combination of scenarios includes 6 types of situations: "single engine failure + low approach energy", "all engines failure + low approach energy", "single engine failure + course deviation", "single engine failure + course intrusion", "icing + control surface failure", and "single engine failure + control surface failure". Data acquisition and preprocessing module: For each set of coupled risk scenarios, set triggering conditions and triggering actions that conform to the actual situation of the scenario and include randomness, generate coupled scenario data, and include flag parameters for whether the scenario has occurred; Training and Validation Module: Based on the PyTorch framework, a hybrid deep learning model combining a long short-term memory network and a fully convolutional neural network is constructed for temporal modeling and feature extraction. The model is used to train a recognition model for coupled risk scenarios. The training effect is improved by adjusting hyperparameters. The trained artificial intelligence model is then validated, and the recognition success rate must be greater than 95%.

Images