Self-adaptive route path generation method for line fault driving

Abstract

The invention relates to the technical field of route path planning, discloses a self-adaptive route path generation method for line fault driving, and aims to improve the safety and efficiency of a route. According to the method, historical route operation data are collected and preprocessed, a route fault prediction model is constructed, and the fault occurrence probability and type on a future route are predicted. And acquiring aircraft performance parameters and environmental condition data in real time, inputting the prediction model to acquire fault information, comparing the fault information with the real-time running state data of the route, and marking a fault deviation point. And modeling the route planning process into a Markov decision process, and constructing a route decision model. And performing real-time path planning based on the trained decision model, and selecting an optimal route adjustment action according to the current aircraft state and environmental conditions. According to the method, line faults can be sensed in real time, the route path can be adaptively adjusted, dynamic changes are effectively handled, and the safety and efficiency of route operation are improved.

Description

Technical Field

[0001] This invention relates to the field of route planning technology, specifically to an adaptive route generation method driven by route faults. Background Technology

[0002] With the rapid development of aviation technology and the increasing air traffic volume, the safe and efficient operation of flight routes has become a key focus in the aviation industry. However, flight routes are often affected by various unforeseen factors during operation, among which route failures are one of the main factors affecting route safety and efficiency. Route failures can be caused by a variety of reasons, such as severe weather conditions, aircraft performance problems, and improper route configuration. These failures can not only lead to flight delays but also pose a serious threat to flight safety. Traditional route planning methods often rely on static route configurations and preset flight plans, making it difficult to respond to dynamic changes such as route failures in real time. When a route failure occurs, manual intervention is usually required to adjust the route, which is not only time-consuming and labor-intensive but may also fail to make the optimal decision in the shortest possible time, thus affecting the overall operational efficiency and safety of the route.

[0003] To address this issue, numerous data-driven route planning methods have emerged in recent years. These methods collect and analyze historical route operation data to attempt to uncover patterns and regularities in route operation, thereby providing decision support for route planning. However, most of these methods focus on the analysis and mining of historical data, neglecting the impact of real-time environmental conditions and aircraft performance parameters on route operation. Therefore, they often fail to achieve the desired results in practical applications. Existing route planning methods typically lack effective adaptive mechanisms when facing route failures. When a route failure occurs, these methods often cannot adjust the route in real time to adapt to the changes brought about by the failure. Therefore, there is an urgent need for a method that can detect route failures in real time and adaptively adjust the route according to the failure situation to improve the safety and efficiency of routes. Summary of the Invention

[0004] The purpose of this invention is to provide an adaptive route generation method driven by line faults, so as to solve the problems mentioned in the background art.

[0005] To achieve the above objectives, the present invention provides the following technical solution: an adaptive route generation method driven by line faults, the method comprising:

[0006] Collect historical flight route operation data, including fault records, environmental conditions, aircraft performance parameters, and flight route configuration information; clean and preprocess the collected data to form a feature dataset;

[0007] A route fault prediction model is constructed. This model is based on a feature dataset and combines environmental conditions and aircraft performance parameters to predict the probability and type of faults occurring on the route in the future.

[0008] The system acquires real-time performance parameters and environmental conditions of the aircraft, inputs the acquired data into the line fault prediction model, and receives the predicted fault information output by the model.

[0009] Acquire real-time operational status data of the flight route; process the acquired real-time operational status data and the predicted fault information output by the model separately, align the predicted fault information with the real-time status with timestamps, and compare them in the same analysis framework; set a fault deviation threshold, and determine whether the deviation between the predicted fault information and the real-time status exceeds the threshold. If it exceeds the threshold, mark it as a fault deviation point.

[0010] Construct a route path decision model and model the route path planning process as a Markov decision process;

[0011] Real-time path planning is performed based on a trained flight path decision model. The model collects the aircraft's performance parameters and environmental condition data in real time as input and selects the optimal flight path adjustment action based on the current state and strategy.

[0012] Preferably, the definition of the route decision model includes:

[0013] State space: defined as a combination of the aircraft's real-time operational status data, performance parameters, and environmental condition data;

[0014] Action space: Defined as the set of executable route adjustment operations;

[0015] Reward function: Designed based on the goals of route safety and efficiency optimization, it includes reward mechanisms for state stability, path optimization and fault avoidance. It also combines the monitored fault deviation data to give extra rewards for operations that reduce fault deviation and extra penalties for operations that increase fault deviation.

[0016] Output strategy: Defined as a method for selecting the optimal route adjustment action in the current state, which is continuously optimized through reinforcement learning algorithm.

[0017] Preferably, the operational status data includes the aircraft's position, speed, altitude, heading, engine status, and navigation system status; the environmental conditions include weather conditions, wind speed, wind direction, and temperature; and the performance parameters include the aircraft type, maximum range, cruising speed, and payload capacity.

[0018] Preferably, the set of executable route adjustment operations includes: changing flight altitude, adjusting flight speed, changing heading, selecting an alternative route, adjusting takeoff or landing time, performing weight reduction operations, and adjusting flight mode or configuration.

[0019] Preferably, the reward function R consists of multiple sub-reward items, and the total reward value is obtained by weighted summation, specifically using the formula: R = w1 * R_stab + w2 * R_opt + w3 * R_safe + w4 * R_eff + w5 * R_fault; where R_stab represents the reward item for state stability, calculated based on the smoothness of the aircraft's operating state; R_opt represents the reward item for path optimization, calculated based on the efficiency of the flight path and the degree of path optimization; R_safe represents the reward item for safety, calculated based on the improvement of flight path safety; R_eff represents the reward item for energy efficiency improvement, calculated based on the improvement of the aircraft's energy utilization efficiency; R_fault represents the reward or penalty item for reducing fault deviation, based on the monitored fault deviation data, giving positive rewards for operations that reduce deviation and penalties for operations that increase deviation, and its value is set as a function that is inversely or directly proportional to the change in fault deviation; w1, w2, w3, w4, and w5 are the weight coefficients of each sub-reward item, and w1+w2+w3+w4+w5=1.

[0020] Preferably, the line fault prediction model is constructed using the Long Short-Term Memory (LSTM) network algorithm, and its network structure specifically includes:

[0021] Input layer: Used to receive feature datasets as input, which include time-series fault record data, environmental parameter data, aircraft performance data, and time-related features;

[0022] Hidden layer: Contains at least one LSTM hidden layer. Each LSTM hidden layer contains several LSTM units. Each LSTM unit has an input gate, a forget gate, and an output gate to control the flow of information and state updates.

[0023] Output layer: Receives the output of the hidden layer and maps the output of the LSTM network to the predicted fault type and probability through a fully connected layer or classification layer.

[0024] Preferably, the steps for training a line fault prediction model include:

[0025] S1: Initialize the weights and biases of the LSTM network using small random numbers;

[0026] S2: Select a loss function to measure the difference between the model's predicted values ​​and the actual values;

[0027] S3: Using the backpropagation algorithm and gradient descent optimization algorithm, the gradient is calculated based on the loss function, and the weights and biases of the LSTM network are updated. The network is then trained iteratively until the loss function converges.

[0028] S4: Save the trained LSTM network model as a line fault prediction model.

[0029] Preferably, the policy gradient A3C algorithm is used to train the route decision model, and the specific steps include:

[0030] B1. Initialize the policy network and value network: Construct two deep neural networks, one for generating action probability distribution and the other for evaluating state value;

[0031] B2. Initialize global network parameters: Set the parameters of the global policy network and value network for sharing and updating by multiple parallel training threads;

[0032] B3. Start multiple parallel training threads, with each thread executing the following steps independently:

[0033] B301: Initialize the thread's state s;

[0034] B302: Before the round ends, the following steps are executed in a loop: i) The policy network generates an action probability distribution based on the current state s and samples action a; ii) Action a is executed, and the reward r and the next state s' are observed, as well as the termination flag done; iii) Based on the reward r and the next state s', the value network within the thread estimates the value of the state; iv) The policy network parameters within the thread are updated using the policy gradient algorithm to maximize the expected reward; v) The current state s is updated to the next state s'.

[0035] B303: When the round ends, reset state s to the new start state of the round;

[0036] B304: Periodically synchronize the network parameter updates within the thread to the global network parameters;

[0037] B4: Repeat step B3 until a satisfactory performance level is achieved.

[0038] Preferably, in step B303, the value network is updated using a temporal difference learning method, and the specific implementation steps include:

[0039] B303a: Calculate the time series difference error δ, which is given by the formula: δ = r + γ * V(s') - V(s), where r is the current reward, γ is the discount factor, V(s') is the value estimate of the next state, and V(s) is the value estimate of the current state.

[0040] B303b: Update the parameters of the value network using the temporal difference error δ to minimize the value estimation error.

[0041] Preferably, the policy network is updated using the advantage function policy gradient method, and the specific implementation steps include:

[0042] B302a: Calculate the dominance function A, which is given by the formula: A = δ + γ* V(s') - V(s);

[0043] B302b: The policy gradient is calculated using the advantage function A and the action probability distribution, and the parameters of the policy network are updated to maximize the expected reward.

[0044] Compared with the prior art, the beneficial effects of the present invention are:

[0045] This invention constructs a route fault prediction model that can predict the probability and type of faults on flight routes in real time. This allows aviation management departments to be aware of potential fault risks in advance, thereby taking preventative measures and effectively reducing the probability of flight safety accidents. By acquiring aircraft performance parameters and environmental condition data in real time and comparing them with predicted fault information, it is possible to promptly identify and address fault deviations in actual operation, ensuring flight safety.

[0046] Traditional route planning methods often require manual intervention to adjust routes when faced with route failures, which is time-consuming, labor-intensive, and may not yield optimal decisions. This invention constructs a route decision model, modeling the route planning process as a Markov decision process, enabling automatic route adjustments and significantly improving operational efficiency. Real-time collection of aircraft performance parameters and environmental condition data serves as model input, allowing route planning to better reflect actual operating conditions, avoiding unnecessary flight delays and improving overall operational efficiency.

[0047] The method of this invention can detect line faults in real time and adaptively adjust the flight path according to the fault situation, exhibiting strong adaptability and flexibility. It can react quickly to any type of line fault, ensuring the safe and efficient operation of the flight path. By combining historical flight path operation data with real-time environmental conditions and aircraft performance parameters, this invention can comprehensively consider the impact of multiple factors on flight path operation, making flight path planning more comprehensive and accurate.

[0048] By automating and intelligently planning flight routes, this invention can reduce the time and manpower costs required for manual intervention and decision-making, thereby lowering airline operating costs. Furthermore, by improving route operational efficiency and safety, and reducing flight delays and safety incidents, it can further reduce airline operating costs and risks. Attached Figure Description

[0049] Figure 1 This is a flowchart illustrating the steps of the adaptive flight path generation method described in this invention.

[0050] Figure 2 A flowchart for constructing a route decision-making model;

[0051] Figure 3 This is a flowchart for training a route decision model using the policy gradient A3C algorithm. Detailed Implementation

[0052] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0053] Please see Figure 1-3 This invention provides a technical solution: an adaptive route generation method driven by line faults, the method comprising:

[0054] Collect historical flight route operation data, including but not limited to fault records, environmental conditions (such as weather conditions, wind speed, temperature, etc.), aircraft performance parameters (such as speed, range, fuel consumption rate, etc.), and route configuration information (such as waypoints, heading, altitude, etc.). Clean the collected data, removing outliers and missing values ​​to ensure accuracy and completeness. Preprocess the cleaned data, such as normalization and standardization, to form a feature dataset for subsequent model building and analysis.

[0055] Based on a feature dataset and considering environmental conditions and aircraft performance parameters, a route fault prediction model is constructed. The model employs machine learning algorithms to train and predict the probability and type of faults occurring on flight routes within a future timeframe. Model parameters are optimized through cross-validation and accuracy evaluation to improve prediction accuracy.

[0056] The system acquires real-time performance parameters and environmental condition data of the aircraft, such as through onboard sensors and weather stations. This data is then input into a line fault prediction model, which outputs predicted fault information, including the probability of fault occurrence, fault type, and estimated occurrence time.

[0057] Acquire real-time operational status data for the flight path, such as the aircraft's current position, speed, and heading. Process the acquired real-time operational status data and the predicted fault information output by the model separately, aligning the predicted fault information with the real-time status using timestamps to ensure data consistency. Within the same analytical framework, compare the predicted fault information with the real-time status and analyze the differences between them.

[0058] Set a fault deviation threshold and determine a reasonable threshold range based on the safety and efficiency requirements of the flight route. Determine whether the deviation between the predicted fault information and the real-time status exceeds the threshold. If it does, mark it as a fault deviation point, indicating that there is a significant difference between the actual operating status and the prediction, and the flight route may need to be adjusted.

[0059] A flight path decision model is constructed, modeling the flight path planning process as a Markov decision process (MDP). Here, the state represents the aircraft's current position and state, the action represents possible route adjustment actions, and the reward function is set based on the safety and efficiency of the flight path operation. A reinforcement learning algorithm is used to train the flight path decision model, learning a strategy to select the optimal action under different states.

[0060] Real-time path planning is performed based on a pre-trained route decision-making model. Real-time data on aircraft performance parameters and environmental conditions is collected and used as model input. Based on the current state and strategy, the model selects the optimal route adjustment actions, such as changing heading or adjusting altitude, to adapt to changes caused by route failures and improve route safety and efficiency.

[0061] The present invention will be further described below with reference to Examples 1 to 3:

[0062] Example 1: The definition of the route decision model includes:

[0063] State space: The state space is defined as a combination of real-time operational status data, performance parameters, and environmental condition data of an aircraft. Real-time operational status data includes, but is not limited to, the aircraft's position, speed, heading, and altitude. Performance parameters include the aircraft's fuel consumption rate, engine status, and onboard equipment status. Environmental condition data includes weather conditions (such as wind speed, wind direction, temperature, and humidity), airspace restrictions, and the positions of other aircraft or obstacles. This data is acquired in real time through sensors on the aircraft, weather stations, air traffic control systems, etc., and serves as input to the flight path decision model.

[0064] Maneuver space: The maneuver space is defined as the set of executable flight path adjustments. These operations include, but are not limited to, changing course, adjusting altitude, accelerating or decelerating, and circumnavigating a specific area. The size and complexity of the maneuver space are determined based on factors such as aircraft performance, airspace restrictions, and flight path requirements.

[0065] Reward Function: The reward function is designed based on the optimization goals of route safety and efficiency. It includes multiple sub-reward items, and the total reward value is obtained by weighted summation. The specific formula is: R = w1 * R_stab + w2 * R_opt + w3 * R_safe + w4 * R_eff + w5 * R_fault.

[0066] Output strategy: The output strategy is defined as the method for selecting the optimal flight path adjustment action in the current state. Through reinforcement learning algorithms (such as Q-learning, deep reinforcement learning, etc.), the generated adjustment actions are continuously optimized, enabling the aircraft to make optimal decisions in complex and ever-changing environments.

[0067] The specific implementation of the reward function includes:

[0068] State stability reward (R_stab): Calculated based on the smoothness of the aircraft's operating state.

[0069] For example, the rate of change of state parameters such as the aircraft's position, speed, and heading can be calculated. The smaller the rate of change, the more stable the state, and the greater the reward.

[0070] The reward for route optimization (R_opt) is calculated based on the efficiency of the flight path and the degree of route optimization.

[0071] For example, the ratio of the straight-line distance from the aircraft's current position to the target point to the length of the actual flight path can be calculated. The larger the ratio, the more optimized the path, and the greater the reward.

[0072] Safety bonus (R_safe): Calculated based on improvements in route safety.

[0073] For example, a safe distance threshold can be set, and when the distance between the aircraft and other aircraft or obstacles is greater than the threshold, a positive reward is given; otherwise, a penalty is given.

[0074] Energy efficiency improvement incentive (R_eff): Calculated based on the degree of improvement in the aircraft's energy utilization efficiency.

[0075] For example, the fuel consumption rate or electrical energy consumption rate of an aircraft can be calculated. The lower the consumption rate, the higher the energy efficiency, and the greater the reward.

[0076] Rewards or penalties for reducing fault deviation (R_fault): Based on monitored fault deviation data, operations that reduce deviation are rewarded positively, while operations that increase deviation are penalized. Its value is set as a function that is inversely or directly proportional to the change in fault deviation. For example, a positive reward is given when fault deviation decreases, and a penalty is given when fault deviation increases.

[0077] Weighting coefficients are set as follows: w1, w2, w3, w4, and w5 are the weighting coefficients for each sub-reward item, and w1 + w2 + w3 + w4 + w5 = 1. The specific values ​​of the weighting coefficients are determined based on factors such as the importance of route operation safety, efficiency, energy efficiency, and reduction of failure deviations. The weighting coefficients can be optimized through experimental or simulation methods to make the reward function better reflect the actual needs and objectives of route operation.

[0078] Suppose an aircraft encounters severe weather conditions during flight, such as strong wind shear. In this situation, the flight path decision model needs to select the optimal flight path adjustment action based on real-time operational status data, performance parameters, and environmental condition data. The state space includes the aircraft's position, speed, heading, altitude, and environmental condition data such as wind speed and direction. The action space includes executable flight path adjustment operations such as changing heading and adjusting altitude. The reward function calculates the total reward value based on sub-reward items such as state stability, path optimization, safety, energy efficiency improvement, and failure deviation reduction. The output strategy uses a reinforcement learning algorithm to select the optimal flight path adjustment action, such as adjusting altitude to avoid areas of strong wind shear, ensuring the aircraft's safety and efficiency.

[0079] Example 2:

[0080] This invention employs the Asynchronous Advantage Actor-Critic (A3C) algorithm to train the flight path decision model. The specific implementation steps include:

[0081] B1. Initialize the policy network and value network:

[0082] Two deep neural networks are constructed, one for generating action probability distributions and the other for evaluating state value. The policy network is responsible for generating the probability distributions of each possible action based on the current state, while the value network is responsible for estimating the value of the current state, i.e., the sum of expected future rewards.

[0083] B2. Initialize global network parameters:

[0084] The parameters of the global policy network and value network are set, and these parameters will be shared and updated by multiple parallel training threads. The global network parameters act as a bridge for communication between threads, ensuring that each thread can work together to improve model performance.

[0085] B3. Start multiple parallel training threads, with each thread executing the following steps independently:

[0086] B301: Initialize the state s within the thread: At the start of each training thread, randomly or according to specific rules, initialize the state s within the thread as the starting point for training.

[0087] B302: Before the round ends, repeat the following steps:

[0088] i) The policy network generates an action probability distribution based on the current state s, and samples an action a: The policy network takes the current state s as input and outputs the probability distribution of each possible action. Then, it randomly samples an action a based on this probability distribution as the action to be executed.

[0089] ii) Execute action a, observe the reward r and the next state s', and whether the termination flag is done: Execute action a in the environment, observe the environmental feedback reward r, the next state s', and whether the termination condition is met (such as the aircraft reaching the target point or a malfunction occurring).

[0090] iii) Update the value network's estimate of the state value within the thread based on the reward r and the next state s': Update the parameters of the value network using a temporal difference learning method. First, calculate the temporal difference error δ = r + γ * V(s') - V(s), where γ is a discount factor representing the degree of decay of future rewards. Then, update the parameters of the value network using the temporal difference error δ to minimize the value estimation error. The temporal difference error δ is calculated using the formula: δ = r + γ * V(s') - V(s). Update the parameters of the value network using the temporal difference error δ.

[0091] iv) Update the policy network parameters within the thread using the policy gradient algorithm to maximize expected reward: Update the policy network parameters using the advantage function policy gradient method. First, calculate the advantage function A = δ + γ * V(s') - V(s) (Note that the calculation of A here may overlap with the aforementioned temporal difference error δ, but in actual implementations, the calculation of A may include additional adjustments or considerations, such as using more complex estimation methods). Then, calculate the policy gradient using the advantage function A and the action probability distribution, and update the policy network parameters to maximize expected reward. Calculate the advantage function A, which considers immediate reward, the value of future states, and discounting factors for state transitions. Calculate the policy gradient using the advantage function A and the action probability distribution, and update the policy network parameters accordingly.

[0092] v) Update the current state s to the next state s': Update the current state s to the next state s' and continue the next round of training.

[0093] B303: When the round ends, reset state s to the new start state of the round.

[0094] When the termination conditions are met, such as the aircraft reaching the target point or experiencing a malfunction, the state s is reset to the new round start state, ready to begin the next round of training.

[0095] B304: Periodically synchronize network parameter updates within a thread to global network parameters.

[0096] To ensure seamless collaboration between threads, the network parameters within each thread (including those for the policy and value networks) are periodically updated and synchronized to the global network parameters. This allows each thread to leverage the learning outcomes of other threads, collectively improving model performance.

[0097] B4. Repeat step B3 until a satisfactory performance level is achieved:

[0098] Step B3 is repeated continuously, and the parameters of the policy network and value network are continuously optimized through the collaborative work of multiple parallel training threads until a satisfactory performance level is achieved. For example, a specific performance metric (such as route safety, efficiency, etc.) can be set, and training is considered complete when the model reaches a predetermined level on this metric.

[0099] Example 3:

[0100] The line fault prediction model is constructed using the Long Short-Term Memory (LSTM) network algorithm. The following is a detailed implementation of this line fault prediction model:

[0101] The LSTM network structure specifically includes:

[0102] Input Layer: This layer receives the feature dataset as input. The feature dataset includes time-series fault log data, environmental parameter data, aircraft performance data, and time-related features. Fault log data may contain information such as the time, type, and severity of historical faults. Environmental parameter data may include external conditions affecting aircraft operation, such as temperature, humidity, and air pressure. Aircraft performance data may involve performance indicators such as speed, acceleration, and fuel consumption. Time-related features may include time-related information such as timestamps, dates, and seasons.

[0103] Hidden Layer: Contains at least one LSTM hidden layer. Each LSTM hidden layer consists of several LSTM units. Each LSTM unit contains an input gate, a forget gate, and an output gate. These gating mechanisms are used to control the flow of information and state updates. The input gate determines whether new information should be added to the cell state; the forget gate determines which information in the cell state should be discarded; and the output gate determines which parts of the cell state should be output.

[0104] Output layer: Receives the outputs of the hidden layers and maps the LSTM network's outputs to predicted fault types and probabilities through a fully connected layer or classification layer. The fully connected layer linearly combines the outputs of the LSTM units to obtain the raw predicted values ​​for the fault types. The classification layer (such as a softmax layer) transforms the raw predicted values ​​into a probability distribution representing the likelihood of each fault type occurring.

[0105] The steps for training a line fault prediction model include:

[0106] S1: Initialize the weights and biases of the LSTM network: Initialize the weights and biases of the LSTM network using small random numbers to ensure that the network can learn normally during training. Initialization methods can include uniform distribution or normal distribution, among other random initialization methods.

[0107] S2: Choosing a Loss Function: Select an appropriate loss function to measure the difference between the model's predicted values ​​and the true values. For fault type prediction problems, classification loss functions such as cross-entropy loss can be chosen.

[0108] S3: Train the network using backpropagation and gradient descent optimization algorithms: Calculate the gradient based on the loss function and propagate it from the output layer to the input layer using backpropagation. Update the weights and biases of the LSTM network using gradient descent optimization algorithms (such as stochastic gradient descent, Adam, etc.). Iteratively train the network until the loss function converges or the predetermined number of training epochs is reached.

[0109] S4: Save the trained LSTM network model: After training, save the trained LSTM network model as a route fault prediction model. The saved model can include parameters such as network structure, weights, and bias terms, for use in subsequent adaptive route generation.

[0110] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus.

[0111] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. An adaptive route generation method driven by line faults, characterized in that, The method includes: Collect historical flight route operation data, including fault records, environmental conditions, aircraft performance parameters, and flight route configuration information; clean and preprocess the collected data to form a feature dataset; A route fault prediction model is constructed. This model is based on a feature dataset and combines environmental conditions and aircraft performance parameters to predict the probability and type of faults occurring on the route in the future. The system acquires real-time performance parameters and environmental conditions of the aircraft, inputs the acquired data into the line fault prediction model, and receives the predicted fault information output by the model. Acquire real-time operational status data of the flight route; process the acquired real-time operational status data and the predicted fault information output by the model separately, align the predicted fault information with the real-time status with timestamps, and compare them in the same analysis framework; set a fault deviation threshold, and determine whether the deviation between the predicted fault information and the real-time status exceeds the threshold. If it exceeds the threshold, mark it as a fault deviation point. Construct a route path decision model and model the route path planning process as a Markov decision process; Real-time path planning is performed based on a trained flight path decision model. The model collects the aircraft's performance parameters and environmental condition data in real time as input and selects the optimal flight path adjustment action based on the current state and strategy.

2. The adaptive route generation method for line fault-driven operation according to claim 1, characterized in that, The definition of the route decision model includes: State space: defined as a combination of the aircraft's real-time operational status data, performance parameters, and environmental condition data; Action space: Defined as the set of executable route adjustment operations; Reward function: Designed based on the goals of route safety and efficiency optimization, it includes reward mechanisms for state stability, path optimization and fault avoidance. It also combines the monitored fault deviation data to give extra rewards for operations that reduce fault deviation and extra penalties for operations that increase fault deviation. Output strategy: Defined as a method for selecting the optimal route adjustment action in the current state, which is continuously optimized through reinforcement learning algorithm.

3. The adaptive route generation method for line fault-driven approach according to claim 2, characterized in that, The operational status data includes the aircraft's position, speed, altitude, heading, engine status, and navigation system status; the environmental conditions include weather conditions, wind speed, wind direction, and temperature; and the performance parameters include aircraft type, maximum range, cruising speed, and payload capacity.

4. The adaptive route generation method for line fault-driven operation according to claim 2, characterized in that, The set of executable route adjustment operations includes: changing flight altitude, adjusting flight speed, changing heading, selecting an alternative route, adjusting takeoff or landing time, performing weight reduction operations, and adjusting flight mode or configuration.

5. The adaptive route generation method for line fault-driven operation according to claim 2, characterized in that, The reward function R consists of multiple sub-reward items, and the total reward value is obtained by weighted summation. The specific formula is: R = w1 * R_stab + w2 * R_opt + w3 * R_safe + w4 * R_eff + w5 * R_fault; where R_stab represents the reward item for state stability, calculated based on the smoothness of the aircraft's operating state; R_opt represents the reward item for path optimization, calculated based on the efficiency of the flight path and the degree of path optimization; R_safe represents the reward item for safety, calculated based on the improvement of flight path safety; R_eff represents the reward item for energy efficiency improvement, calculated based on the improvement of the aircraft's energy utilization efficiency; R_fault represents the reward or penalty item for reducing fault deviation. Based on the monitored fault deviation data, positive rewards are given for operations that reduce deviation, and penalties are given for operations that increase deviation. Its value is set as a function that is inversely or directly proportional to the change in fault deviation; w1, w2, w3, w4, and w5 are the weight coefficients of each sub-reward item, and w1+w2+w3+w4+w5=1.

6. The adaptive route generation method for line fault-driven operation according to claim 1, characterized in that, The line fault prediction model is constructed using the Long Short-Term Memory (LSTM) network algorithm, and its network structure specifically includes: Input layer: Used to receive feature datasets as input, which include time-series fault record data, environmental parameter data, aircraft performance data, and time-related features; Hidden layer: Contains at least one LSTM hidden layer. Each LSTM hidden layer contains several LSTM units. Each LSTM unit has an input gate, a forget gate, and an output gate to control the flow of information and state updates. Output layer: Receives the output of the hidden layer and maps the output of the LSTM network to the predicted fault type and probability through a fully connected layer or classification layer.

7. The adaptive route generation method for line fault-driven operation according to claim 6, characterized in that, The steps for training a line fault prediction model include: S1: Initialize the weights and biases of the LSTM network using small random numbers; S2: Select a loss function to measure the difference between the model's predicted values ​​and the actual values; S3: Using the backpropagation algorithm and gradient descent optimization algorithm, the gradient is calculated based on the loss function, and the weights and biases of the LSTM network are updated. The network is then trained iteratively until the loss function converges. S4: Save the trained LSTM network model as a line fault prediction model.

8. The adaptive route generation method for line fault-driven operation according to claim 2, characterized in that, The route decision model is trained using the policy gradient A3C algorithm. The specific steps include: B1. Initialize the policy network and value network: Construct two deep neural networks, one for generating action probability distribution and the other for evaluating state value; B2. Initialize global network parameters: Set the parameters of the global policy network and value network for sharing and updating by multiple parallel training threads; B3. Start multiple parallel training threads, with each thread executing the following steps independently: B301: Initialize the thread's state s; B302: Before the round ends, the following steps are executed repeatedly: i) The policy network generates an action probability distribution based on the current state s and samples action a; ii) Action a is executed, and the reward r and the next state s' are observed, as well as the termination flag done; iii) The value network within the thread estimates the state value based on the reward r and the next state s'; iv) The policy network parameters within the thread are updated using the policy gradient algorithm to maximize the expected reward; v) The current state s is updated to the next state s'. B303: When the round ends, reset state s to the new start state of the round; B304: Periodically synchronize the network parameter updates within the thread to the global network parameters; B4: Repeat step B3 until a satisfactory performance level is achieved.

9. The adaptive route generation method for line fault-driven operation according to claim 8, characterized in that, In step B303, the value network is updated using a temporal difference learning method, and the specific implementation steps include: B303a: Calculate the time series difference error δ, which is given by the formula: δ = r + γ * V(s') - V(s), where r is the current reward, γ is the discount factor, V(s') is the value estimate of the next state, and V(s) is the value estimate of the current state. B303b: Update the parameters of the value network using the temporal difference error δ to minimize the value estimation error.

10. The adaptive route generation method for line fault-driven operation according to claim 9, characterized in that, The policy network is updated using the advantage function policy gradient method, and the specific implementation steps include: B302a: Calculate the dominance function A, which is given by the formula: A = δ + γ* V(s') - V(s); B302b: The policy gradient is calculated using the advantage function A and the action probability distribution, and the parameters of the policy network are updated to maximize the expected reward.

Images