Pilot landing operation analysis method and device, electronic equipment and storage medium
By analyzing pilots' attention allocation patterns using neural networks and Markov decision models, the problem of improper attention allocation during aircraft landing was solved, thus improving flight safety.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- TSINGHUA UNIVERSITY
- Filing Date
- 2024-12-31
- Publication Date
- 2026-05-29
AI Technical Summary
Pilots may find it difficult to maintain focus on the critical operational status of the aircraft during landing, leading to delayed risk perception and operational adjustments, and increasing the risk of accidents.
By employing neural network models and partially observable Markov decision models, combined with external environment and aircraft state parameters, this study analyzes the pilot's attention allocation patterns, characterizes the psychological mechanism of attention allocation through latent states, predicts actions, and determines the optimal control law, thereby achieving accurate analysis of the pilot's landing operations.
It enables accurate and objective analysis of pilots' landing operations, improves flight safety, and reduces the difficulty of risk management.
Smart Images
Figure CN119886829B_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of computer technology, and in particular to a method, apparatus, electronic device, and storage medium for analyzing pilot landing operations. Background Technology
[0002] Landing an aircraft is a complex and challenging task, with over a quarter of flight safety incidents occurring during this phase, which accounts for only 1% of flight time. During this stage, pilots face changes in the external environment and aircraft status, requiring them to appropriately allocate their attention and perform suitable operations on the throttle, pitch, and roll sticks to perceive and control the risky environment and the complex aircraft system.
[0003] However, due to limited cognitive resources, pilots sometimes find it difficult to maintain attention to important aircraft operating conditions, such as yaw angle, altitude, and airspeed, and therefore cannot perceive risks and adjust operations in a timely manner, leading to potential accidents.
[0004] Therefore, finding an accurate and objective method for analyzing pilot landing operations has become a current research hotspot, in order to facilitate targeted risk management and improve pilot flight safety. Summary of the Invention
[0005] This invention provides a method, device, electronic device, and storage medium for analyzing pilot landing operations, enabling accurate and objective analysis of pilot landing operations, thereby facilitating targeted risk management and improving pilot flight safety.
[0006] This invention provides a method for analyzing pilot landing operations. The method includes: acquiring external environmental parameters and aircraft operational state parameters generated during the pilot's landing operation; calling a pre-trained neural network model and inputting the external environmental parameters and the aircraft operational state parameters into the neural network model to obtain a state transition matrix at each time point output by the neural network model, wherein the neural network model is used to obtain the corresponding state transition matrix at each time point based on the external environmental parameters and the aircraft operational state parameters; and obtaining the hidden states corresponding to the state transition matrices based on the state transition matrices to complete the analysis of the pilot's landing operations. The hidden states are used to characterize various states of the pilot's attention allocation pattern, including any one or more of the following: a state without attention deficit, a state with attention deficit regarding yaw angle, a state with attention deficit regarding altitude, and a state with attention deficit regarding airspeed.
[0007] According to a pilot landing operation analysis method provided by the present invention, after obtaining the hidden state corresponding to the state transition matrix, the method further includes: based on the hidden state, obtaining observed values of target operating state parameters formed during the pilot's operation of the aircraft landing process corresponding to the hidden state, wherein the target operating state parameters include any one or more of yaw angle, altitude, and airspeed; based on the observed values of the target operating state parameters, obtaining the predicted actions of the pilot during the aircraft landing process; and based on the predicted actions, determining the flight risk of the pilot during the aircraft landing process.
[0008] According to a pilot landing operation analysis method provided by the present invention, before obtaining the predicted actions of the pilot during the aircraft landing operation based on the observed values of the target operating state parameters, the method further includes: determining the optimal control law of the pilot during the aircraft landing operation; the step of obtaining the predicted actions of the pilot during the aircraft landing operation based on the observed values of the target operating state parameters specifically includes: obtaining the predicted actions of the pilot during the aircraft landing operation based on the observed values of the target operating state parameters and the optimal control law.
[0009] According to a pilot landing operation analysis method provided by the present invention, the optimal control law of the pilot during the aircraft landing process is determined in the following manner: A pre-constructed partially observable Markov decision model corresponding to the pilot's aircraft landing process is invoked, wherein the partially observable Markov decision model is obtained by modeling the pilot's aircraft landing process based on the partially observable Markov decision process; a cost function is obtained, wherein the cost function is a linear exponential quadratic function for maintaining the aircraft at a stable descent rate and landing at a target speed; with the objective of the pilot's aircraft landing trajectory approximating an ideal trajectory and the pilot performing smooth operation during the aircraft landing process, the optimal control law of the pilot during the aircraft landing process is determined based on the cost function.
[0010] According to a pilot landing operation analysis method provided by the present invention, the partially observable Markov decision model is constructed in the following manner: based on the partially observable Markov decision process, the pilot's aircraft landing operation process is modeled to obtain the partially observable Markov decision model, and the state variables, decision actions, state transition probabilities, observations, and cost functions of the partially observable Markov decision model are defined respectively. Specifically, the state variables are determined based on the aircraft operating state parameters; the decision actions are determined based on the pitch and roll stick positions; the state transition probabilities are determined based on the linear system state model corresponding to the aircraft operating state parameters; the observations are determined based on the observed values of the target operating state parameters; and a cost function is constructed to maintain the aircraft landing at a stable descent rate and landing at the target speed.
[0011] According to the pilot landing operation analysis method provided by the present invention, the system state linear model corresponding to the aircraft operating state parameters is constructed in the following manner: based on historical flight rapid recording data, each of the aircraft operating state parameters is linearly modeled to obtain the system state linear model corresponding to the aircraft operating state parameters. The system state linear model is used to characterize the model formed by the aircraft operating state parameters being affected by any one or more of other aircraft operating state parameters, decision actions, and random terms.
[0012] According to a pilot landing operation analysis method provided by the present invention, the neural network model is pre-trained in the following manner: A training dataset is acquired, wherein the training dataset includes multiple training data, including external environment training parameters and aircraft operating state training parameters; based on the external environment training parameters and aircraft operating state training parameters, predicted action training data of the pilot during the aircraft landing operation is obtained, corresponding to the external environment training parameters and aircraft operating state training parameters; actual action training data of the pilot during the aircraft landing operation is acquired, corresponding to the external environment training parameters and aircraft operating state training parameters; a loss function is determined based on the mean square error of the predicted action training data and the actual action training data; and the neural network model is pre-trained based on the loss function to obtain the trained neural network model.
[0013] The present invention also provides a pilot landing operation analysis device, the device comprising: an acquisition module for acquiring external environmental parameters and aircraft operating state parameters generated during the pilot's operation of the aircraft landing; an invocation module for invoking a pre-trained neural network model and inputting the external environmental parameters and the aircraft operating state parameters into the neural network model to obtain a state transition matrix at each time point output by the neural network model, wherein the neural network model is used to obtain the corresponding state transition matrix at each time point based on the external environmental parameters and the aircraft operating state parameters; and a processing module for obtaining the hidden states corresponding to the state transition matrices based on the state transition matrices to complete the analysis of the pilot's landing operation, wherein the hidden states are used to characterize the various states of the pilot's attention allocation pattern, and the hidden states include any one or more of the following: a state without attention deficit, a state with attention deficit regarding yaw angle, a state with attention deficit regarding altitude, and a state with attention deficit regarding airspeed.
[0014] The present invention also provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the pilot landing operation analysis method as described above.
[0015] The present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the pilot landing operation analysis method as described above.
[0016] The present invention also provides a computer program product, including a computer program that, when executed by a processor, implements the pilot landing operation analysis method as described above.
[0017] The present invention provides a method, device, electronic device, and storage medium for analyzing pilot landing operations. It acquires external environmental parameters and aircraft operational state parameters generated during the pilot's landing operation; calls a pre-trained neural network model and inputs the external environmental parameters and aircraft operational state parameters into the neural network model to obtain a state transition matrix at each time point; based on the state transition matrix, it obtains the hidden states corresponding to the state transition matrix. The hidden states characterize various states of the pilot's attention allocation pattern, including any one or more of the following: a state without attention deficit, a state with attention deficit regarding yaw angle, a state with attention deficit regarding altitude, and a state with attention deficit regarding airspeed. This completes the analysis of the pilot's attention allocation pattern during landing operations, enabling accurate and objective analysis of the pilot's landing operations, thereby facilitating targeted risk management and improving the pilot's flight safety level. Attached Figure Description
[0018] To more clearly illustrate the technical solutions in this invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of this invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.
[0019] Figure 1 This is one of the flowcharts of the pilot landing operation analysis method provided by the present invention.
[0020] Figure 2 This is the second flowchart of the pilot landing operation analysis method provided by the present invention.
[0021] Figure 3 This is a flowchart illustrating the process of determining the optimal control law for a pilot during aircraft landing, as provided by the present invention.
[0022] Figure 4 This is a schematic diagram of the process for training a neural network model provided by the present invention.
[0023] Figure 5 This is a schematic diagram of the pilot landing operation analysis device provided by the present invention.
[0024] Figure 6 This is a schematic diagram of the structure of the electronic device provided by the present invention. Detailed Implementation
[0025] To make the objectives, technical solutions, and advantages of this invention clearer, the technical solutions of this invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of this invention. All other embodiments obtained by those skilled in the art based on the embodiments of this invention without creative effort are within the scope of protection of this invention.
[0026] This invention provides a pilot landing operation analysis method that proposes to indirectly measure the allocation pattern of attention, an important cognitive resource in aircraft operation, through pilot operational behavior. Based on decision theory, it uses hidden states to characterize the psychological mechanism of pilot attention allocation pattern switching. This invention provides a new perspective for indirectly inferring attention allocation patterns, which can be cross-calibrated with direct measurements of cognitive load.
[0027] The pilot landing operation analysis method provided by this invention applies the achievements of sequential decision-making, machine learning, and deep learning. It employs a Hidden Markov Model (HMM) to model the transition patterns of hidden states during the pilot's landing operation, and utilizes a deep neural network to learn the state transition matrix of the hidden states from external environment and aircraft state data. The pilot's hidden state at each time point corresponds to their current attention allocation pattern, thereby determining their actual observation of the aircraft's operational status.
[0028] Figure 1 This is one of the flowcharts of the pilot landing operation analysis method provided by the present invention.
[0029] The following will combine Figure 1 The process of the pilot landing operation analysis method provided by the present invention will be described.
[0030] In an exemplary embodiment of the present invention, combined with Figure 1 As can be seen, the pilot landing operation analysis method may include steps 110 to 130, and each step will be described below.
[0031] In step 110, external environmental parameters and aircraft operating status parameters generated during the pilot's operation of the aircraft landing are acquired.
[0032] In one embodiment, external environmental parameters generated during the pilot's landing maneuver can be acquired, such as wind speed, wind direction, temperature, and air pressure. Aircraft operational status parameters generated during the landing maneuver can also be acquired, such as airspeed, yaw angle, altitude, pitch angle, roll angle, ground speed, rate of descent, horizontal wind speed, and vertical wind speed. In this embodiment, no specific limitations are imposed on the external environmental parameters and aircraft operational status parameters.
[0033] In step 120, a pre-trained neural network model is invoked, and external environment parameters and aircraft operating state parameters are input into the neural network model to obtain the state transition matrix at each time point output by the neural network model. The neural network model is used to obtain the corresponding state transition matrix at each time point based on the external environment parameters and aircraft operating state parameters.
[0034] In another embodiment, a pre-trained neural network model can be invoked, and external environment parameters and aircraft operating state parameters can be input into the neural network model to obtain the state transition matrix at each time point output by the neural network model. Here, the neural network model is pre-trained, and it can obtain the corresponding state transition matrix at each time point based on the external environment parameters and aircraft operating state parameters.
[0035] In step 130, based on the state transition matrix, the hidden states corresponding to the state transition matrix are obtained to complete the pilot landing operation analysis. The hidden states are used to characterize the various states of the pilot's attention allocation pattern. The hidden states include any one or more of the following: a state without attention deficit, a state with attention deficit to yaw angle, a state with attention deficit to altitude, and a state with attention deficit to airspeed.
[0036] In another embodiment, hidden states can be generated based on a state transition matrix. These hidden states can characterize the various states of a pilot's attention allocation pattern; that is, the hidden states depict the psychological mechanisms of switching between these patterns. In one example, the hidden states may include any one or more of the following: a state without attention deficit, a state with attention deficit regarding yaw angle, a state with attention deficit regarding altitude, and a state with attention deficit regarding airspeed. In this embodiment, a neural network model can be used to learn the sequence of hidden states, thereby enabling the determination of changes in attention allocation during flight.
[0037] In another example, pilots exhibit four latent states, corresponding to four attention allocation patterns: a state with no attention deficit, where all observed values for each operational state are considered true; a state with attention deficit regarding yaw angle, where observed values for altitude and airspeed are true, and observed value for yaw angle is the sum of true value and normal noise; a state with attention deficit regarding altitude, where observed values for yaw angle and airspeed are true, and observed value for altitude is the sum of true value and normal noise; and a state with attention deficit regarding airspeed, where observed values for yaw angle and altitude are true, and observed value for airspeed is the sum of true value and normal noise. The pilot's attention allocation pattern during the landing phase reflects their risk management capabilities.
[0038] The pilot landing operation analysis method provided by this invention acquires external environmental parameters and aircraft operational state parameters generated during the pilot's landing operation; it calls a pre-trained neural network model and inputs the external environmental parameters and aircraft operational state parameters into the neural network model to obtain the state transition matrix at each time point; based on the state transition matrix, it obtains the hidden states corresponding to the state transition matrix, wherein the hidden states are used to characterize the various states of the pilot's attention allocation pattern. The hidden states include any one or more of the following: no attention deficit, attention deficit to yaw angle, attention deficit to altitude, and attention deficit to airspeed. This completes the analysis of the pilot's attention allocation pattern during the landing operation, enabling accurate and objective analysis of the pilot's landing operation, thereby facilitating targeted risk management and improving the pilot's flight safety level.
[0039] Figure 2 This is the second flowchart of the pilot landing operation analysis method provided by the present invention.
[0040] The following will combine Figure 2 The process of another pilot landing operation analysis method is explained.
[0041] In yet another exemplary embodiment of the present invention, combined with Figure 2 As can be seen, after obtaining the hidden state corresponding to the state transition matrix, the pilot landing operation analysis method can also include steps 210 to 230, which will be introduced below.
[0042] In step 210, based on the latent state, the observed values of the target operating state parameters formed during the pilot's operation of the aircraft landing process corresponding to the latent state are obtained, wherein the target operating state parameters include any one or more of yaw angle, altitude and airspeed.
[0043] In one embodiment, the observed values of target operational state parameters formed during the pilot's landing operation can be obtained based on the hidden state. It should be noted that the target operational state parameters can be considered as the aircraft's operational states of interest, derived from aerodynamic analysis, such as yaw angle, altitude, and airspeed. In another example, the target operational state parameters may not be specific and can be adjusted accordingly based on actual conditions.
[0044] In another embodiment, the hidden state can be a state without attention deficit, meaning that the actual observations for each operational state are all true states. In this state, the actual observations for yaw angle, altitude, and airspeed are all true states.
[0045] In another embodiment, the latent state can be a state where attention to the yaw angle is lacking. In this scenario, the actual observed values of altitude and airspeed are the true states, and the observed value of the yaw angle is the sum of the true state and normal noise, where the mean of the normal noise is 0 and the variance is provided by experts in the field of flight.
[0046] In another embodiment, the latent state can be a state where attention to altitude is lacking. In this scenario, the actual observed values of yaw angle and airspeed are the true states, and the observed altitude is the sum of the true state and normal noise, where the mean of the normal noise is 0 and the variance is provided by experts in the field of flight.
[0047] In another embodiment, the hidden state can be a state where attention to airspeed is missing, i.e., the actual observed values of yaw angle and altitude are the true state, and the observed value of airspeed is the sum of the true state and normal noise, where the mean of the normal noise is 0 and the variance is provided by experts in the field of flight.
[0048] In step 220, based on the observed values of the target operating state parameters, the predicted actions of the pilot during the aircraft landing process are obtained.
[0049] In step 230, based on the predicted actions, the flight risks for the pilot during the aircraft landing process are determined.
[0050] In another embodiment, the predicted actions of the pilot during aircraft landing can be obtained based on the observed values of the target operational state parameters obtained above. Furthermore, based on these predicted actions, the flight risks during the aircraft landing process can be determined. In this embodiment, the assessment of flight risks during aircraft landing can be achieved, thereby laying the foundation for targeted risk management and improving the pilot's flight safety level.
[0051] In another embodiment, since the latter three latent states of the pilot (lack of attention to yaw angle, lack of attention to altitude, and lack of attention to airspeed) reflect improper attention allocation during landing operations, resulting in an inability to perceive risks and adjust operations in a timely manner, leading to potential accidents, the occurrence of the latter three latent states indicates insufficient risk management ability of the pilot, while the first latent state (no lack of attention) indicates a high level of risk management. Defining the pilot's risk management ability as the sum of the probabilities of the first latent state (ideal attention allocation pattern) occurring at all time points allows for comparison of the abilities of different pilots.
[0052] In yet another exemplary embodiment of the present invention, the preceding text continues... Figure 2 Taking the above embodiment as an example, before obtaining the pilot's predicted actions during the aircraft landing process based on the observed values of the target operating state parameters (corresponding to step 220), the pilot landing operation analysis method may further include the following steps:
[0053] Determine the optimal control law for the pilot during the aircraft landing process;
[0054] The predicted actions of the pilot during aircraft landing, based on the observed values of the target's operational state parameters, can be achieved in the following ways:
[0055] Based on the observed values of the target operating state parameters and the optimal control law, the predicted actions of the pilot during the aircraft landing process are obtained.
[0056] In one embodiment, the optimal control law for the pilot during aircraft landing can be determined. Furthermore, based on the observed values of the target operating state parameters and the optimal control law, the predicted actions of the pilot during aircraft landing can be obtained. It should be noted that the predicted actions are generated based on the pilot's actions in response to biased observations, thus effectively determining the flight risks during aircraft landing based on the predicted actions.
[0057] Figure 3 This is a flowchart illustrating the process of determining the optimal control law for a pilot during aircraft landing, as provided by the present invention.
[0058] The following will combine Figure 3 The process of determining the optimal control law for a pilot during aircraft landing is explained.
[0059] In an exemplary embodiment of the present invention, combined with Figure 3 As can be seen, determining the optimal control law for the pilot during the aircraft landing process can include steps 310 to 330, which will be described in detail below.
[0060] In step 310, a pre-built partially observable Markov decision model corresponding to the pilot's operation of the aircraft landing process is invoked. The partially observable Markov decision model is obtained by modeling the pilot's operation of the aircraft landing process based on the partially observable Markov decision process.
[0061] In step 320, the cost function is obtained, wherein the cost function is a linear exponential quadratic function that maintains the aircraft's descent at a stable rate and landing at the target speed.
[0062] In step 330, with the goal of the pilot's flight path during the landing process being close to the ideal trajectory and the pilot performing smooth operations during the landing process, the optimal control law for the pilot during the landing process is determined based on the cost function.
[0063] In one embodiment, a pre-built partially observable Markov decision model corresponding to the pilot's operation of the aircraft landing process can be invoked. Furthermore, the cost function is obtained, and with the goal of the pilot's flight trajectory during the aircraft landing process approaching the ideal trajectory and the pilot performing smooth operations during the aircraft landing process, the optimal control law for the pilot's operation of the aircraft landing process is determined based on the cost function.
[0064] In another embodiment, when the dynamic system is a discrete-time linear system and the control objective is a quadratic cost function, the optimal action (control input) has a closed-form solution and can be written as a linear function of the state. The solution process is as follows: Substituting the system dynamic equations (which can correspond to the linear model of the system state) into the Bellman equations, the optimal solution for the control input at each time point can be obtained through first-order conditions; substituting the optimal control law into the cost function, the iterative relationship of the cost function can be written, resulting in the discrete-time Riccati equations; solving the Riccati equations (using numerical methods) yields the closed-form solution for the optimal action, which is the optimal control law for the pilot during the aircraft landing process.
[0065] It should be noted that in this invention, the optimal control based on POMDP includes two parts: state estimation and implementation. Therefore, the closed-form solution described above cannot be directly applied. However, this invention focuses on the period of pilot attention deficit during landing. During this time, the pilot not only cannot observe the true information about the operational state, but also mistakenly takes biased observations as true information, failing to perform state estimation according to the optimal control mode, and instead directly applying the optimal control law to the observations.
[0066] In yet another exemplary embodiment of the present invention, the description continues with the previously described embodiments, wherein a portion of the observable Markov decision model can be constructed in the following manner:
[0067] Based on a partially observable Markov decision process, a model is constructed to depict the pilot's aircraft landing operation, resulting in a partially observable Markov decision model. The state variables, decision actions, state transition probabilities, observations, and cost functions of this partially observable Markov decision model are defined.
[0068] Determine state variables based on aircraft operational state parameters;
[0069] Based on the pitch angle and roll angle positions, determine the decision-making action;
[0070] Based on the linear system state model corresponding to the aircraft's operating state parameters, the state transition probability is determined.
[0071] Based on the observed values of the target operating state parameters, determine the observations, and
[0072] Construct a cost function to maintain the aircraft's descent rate at a stable rate and its landing at a target speed.
[0073] In another embodiment, the pilot's landing phase operation, which is the focus of this invention, involves observations of the aircraft's operational state influenced by attention allocation patterns, exhibiting partially observable characteristics. Furthermore, this operation possesses the Markov property (given the current state, the state in the next period is independent of previous states), can be discretized at 1-second intervals, and has environmental randomness, thus allowing it to be modeled as a POMDP. In application, pilot behavior can be modeled as a finite-period partially observable Markov Decision Process (POMDP), defining actions, state transition probabilities, states, observations, and a cost function. That is, based on the partially observable Markov Decision Process, the pilot's aircraft landing process is modeled to obtain a partially observable Markov decision model, defining the state variables, decision actions, state transition probabilities, observations, and cost function of the partially observable Markov Decision Model.
[0074] In another embodiment, the present invention focuses on the risks of the aircraft landing phase, during which the pilot needs to begin landing from a fixed point while simultaneously meeting the following requirements: maintaining a constant yaw angle and pointing towards the runway; achieving a uniform descent in altitude; and maintaining a uniform descent in airspeed. To characterize the dynamic system encompassing these three key factors, a total of nine system states are considered. Therefore, the aforementioned aircraft operational state parameters can be selected as state variables. That is, state variables are determined based on the aircraft operational state parameters.
[0075] In another embodiment, the pilot completes the landing operation by changing the throttle, pitch, and roll positions. Since the throttle position is sometimes automatically calibrated and does not fully reflect the pilot's actions, the pitch and roll positions are selected as decision variables (actions) for the landing phase. That is, the decision action is determined based on the pitch and roll positions.
[0076] In another embodiment, the pilot's actions during the landing phase satisfy the assumptions of the POMDP and can be modeled using the POMDP. In this case, the discrete time interval in the POMDP is 1 second (the lowest frequency of QAR data), the actions are pitch and roll stick positions; the states are nine items: speed, yaw angle, altitude, pitch angle, roll angle, ground speed, rate of descent, horizontal wind speed, and vertical wind speed; the state transition probabilities are described by a system dynamics model (corresponding to a linear system state model) that includes actions and states; the observations are actual observations based on the attention allocation pattern (corresponding to observations of the target operating state parameters); and the reward (cost function) is a linear exponential quadratic function that simultaneously considers maintaining a stable rate of descent and landing at the target speed (corresponding to the cost function).
[0077] In another embodiment, during landing, the pilot's control objectives include approximating the ideal trajectory and maintaining stable operation. They want to avoid deviations in yaw angle, altitude, and airspeed, as well as excessive maneuvering and turbulent flight paths. Therefore, the LQR cost function is suitable for modeling. The coefficients of the Q, R, and F matrices in the cost function should accurately reflect the pilot's control objectives; that is, the actions predicted by the cost function should closely approximate the actual actions. Therefore, a data-driven approach can be used to determine the cost function, finding the cost function coefficients that best reflect the actual situation while minimizing the mean squared error.
[0078] In another embodiment, parameter estimation can employ a grid search algorithm. The optimal control law is calculated for each parameter combination, and a neural network is trained for each parameter combination (the prediction action is based on the optimal observation law applied to the actual observations), yielding the training error for that parameter combination. This process is repeated for each parameter combination, and the parameter combination corresponding to the smallest training error is selected as the coefficients of the Q, R, and F matrices. Based on this, the cost function is constructed.
[0079] In yet another exemplary embodiment of the present invention, continuing with the previously described embodiments, the linear system state model corresponding to the aircraft operating state parameters can be constructed in the following manner:
[0080] Based on historical flight data, linear modeling is performed on each aircraft operating state parameter to obtain a system state linear model corresponding to the aircraft operating state parameter. The system state linear model is used to characterize the aircraft operating state parameter being affected by any one or more of other aircraft operating state parameters, decision actions, and random terms.
[0081] In another embodiment, based on aerodynamic theory, the system states (corresponding to aircraft operating state parameters) that affect the aircraft operating state that needs attention can be analyzed. Furthermore, a linear model is established for each system state, and the model coefficients are obtained by linear regression and autoregression on more than 4,000 QAR data (corresponding to historical flight rapid record data).
[0082] In another embodiment, an aerodynamic analysis of the aircraft system is performed to determine the system state (corresponding to aircraft operating state parameters) that affects yaw angle, altitude, and airspeed through mechanisms such as generating lift. These parameters may include nine items: airspeed, yaw angle, altitude, pitch angle, roll angle, ground speed, rate of descent, horizontal wind speed, and vertical wind speed.
[0083] In this model, the pilot's actions (control inputs) are pitch and roll stick positions. Through higher-order small-quantity approximations after Taylor expansion, the system dynamics model can be written as a set of linear equations. Each equation describes the pattern of how a system state variable is affected by other system states, actions, and stochastic terms, where the stochastic terms include other unconsidered factors. Horizontal and vertical wind speeds are not affected by other variables and their coefficients can be estimated through autoregression. This embodiment allows for the rapid and accurate acquisition of a linear system state model corresponding to the aircraft's operating state parameters.
[0084] Figure 4 This is a schematic diagram of the process for training a neural network model provided by the present invention.
[0085] The following will combine Figure 4 The process of training a neural network model is explained.
[0086] In an exemplary embodiment of the present invention, combined with Figure 4 As can be seen, training a neural network model may include steps 410 to 450, which will be described in detail below.
[0087] In step 410, a training dataset is obtained, which includes multiple training data, including external environment training parameters and aircraft operating status training parameters.
[0088] In step 420, based on the external environment training parameters and the aircraft operating status training parameters, predictive action training data of the pilot during the aircraft landing process corresponding to the external environment training parameters and the aircraft operating status training parameters are obtained.
[0089] In one embodiment, a training dataset can be obtained, which may include multiple training data sets, including external environment training parameters and aircraft operational status training parameters. It should be noted that the external environment training parameters and aircraft operational status training parameters correspond to the external environment parameters and aircraft operational status parameters mentioned above, but differ in their application scenarios.
[0090] Furthermore, based on external environment training parameters and aircraft operating status training parameters, a neural network model can be used to obtain predictive action training data for pilots during aircraft landing operations, corresponding to the external environment training parameters and aircraft operating status training parameters.
[0091] In step 430, real action training data of the pilot during the aircraft landing process, corresponding to the external environment training parameters and the aircraft operating status training parameters, are obtained.
[0092] In step 440, the loss function is determined based on the mean square error between the predicted action training data and the actual action training data.
[0093] In step 450, the neural network model is pre-trained based on the loss function to obtain a trained neural network model.
[0094] In another embodiment, the training data includes labels, which are the pilot's actual actions during the aircraft landing process, corresponding to the external environment training parameters and aircraft operating state training parameters. In application, the labels, i.e., the actual action training data, can be acquired, and a loss function can be determined based on the mean squared error between the predicted action training data and the actual action training data. Furthermore, the neural network is trained by minimizing the mean squared error to obtain the true impact of the external environment and aircraft state on the pilot's attention allocation pattern. That is, the neural network model is pre-trained based on the loss function to obtain a trained neural network model, enabling the trained neural network model to accurately reflect the true impact of the external environment and aircraft state on the pilot's attention allocation pattern.
[0095] As described above, the pilot landing operation analysis method provided by this invention acquires external environmental parameters and aircraft operating state parameters generated during the pilot's landing operation; it calls a pre-trained neural network model and inputs the external environmental parameters and aircraft operating state parameters into the neural network model to obtain the state transition matrix at each time point; based on the state transition matrix, it obtains the hidden states corresponding to the state transition matrix, wherein the hidden states are used to characterize the various states of the pilot's attention allocation pattern. The hidden states include any one or more of the following: a state without attention deficit, a state with attention deficit regarding yaw angle, a state with attention deficit regarding altitude, and a state with attention deficit regarding airspeed. This completes the analysis of the pilot's attention allocation pattern during the landing operation, enabling accurate and objective analysis of the pilot's landing operation, thereby facilitating targeted risk management and improving the pilot's flight safety level.
[0096] The pilot landing operation analysis device provided by the present invention is described below. The pilot landing operation analysis device described below and the pilot landing operation analysis method described above can be referred to in correspondence.
[0097] Figure 5 This is a schematic diagram of the pilot landing operation analysis device provided by the present invention.
[0098] The following will combine Figure 5 The structure of the pilot landing operation analysis device is explained.
[0099] In an exemplary embodiment of the present invention, combined with Figure 5As can be seen, the pilot landing operation analysis device may include an acquisition module 510, a recall module 520, and a processing module 530. Each module will be described in detail below.
[0100] The acquisition module 510 can be configured to acquire external environmental parameters and aircraft operating status parameters generated during the pilot's operation of the aircraft landing process;
[0101] The calling module 520 can be configured to call a pre-trained neural network model and input the external environment parameters and the aircraft operating state parameters into the neural network model to obtain the state transition matrix at each time point output by the neural network model. The neural network model is used to obtain the corresponding state transition matrix at each time point based on the external environment parameters and the aircraft operating state parameters.
[0102] The processing module 530 can be configured to obtain the hidden state corresponding to the state transition matrix based on the state transition matrix in order to complete the landing operation analysis of the pilot. The hidden state is used to characterize each state of the pilot's attention allocation mode. The hidden state includes any one or more of the following: a state without attention deficit, a state with attention deficit to yaw angle, a state with attention deficit to altitude, and a state with attention deficit to airspeed.
[0103] In an exemplary embodiment of the present invention, the processing module 530 may further be configured to:
[0104] Based on the hidden state, the observed values of the target operating state parameters formed during the pilot's operation of the aircraft landing process corresponding to the hidden state are obtained, wherein the target operating state parameters include any one or more of yaw angle, altitude and airspeed.
[0105] Based on the observed values of the target operating state parameters, the predicted actions of the pilot during the aircraft landing process are obtained.
[0106] Based on the predicted actions, the flight risks for the pilot during the aircraft landing process are determined.
[0107] In an exemplary embodiment of the present invention, the processing module 530 may further be configured to:
[0108] Determine the optimal control law for the pilot during the aircraft landing process;
[0109] The processing module 530 can obtain the predicted actions of the pilot during the aircraft landing process based on the observed values of the target operating state parameters in the following manner:
[0110] Based on the observed values of the target operating state parameters and the optimal control law, the predicted actions of the pilot during the aircraft landing process are obtained.
[0111] In an exemplary embodiment of the present invention, the processing module 530 may determine the optimal control law for the pilot during the aircraft landing process in the following manner:
[0112] A pre-built partially observable Markov decision model corresponding to the pilot's operation of the aircraft landing process is invoked, wherein the partially observable Markov decision model is obtained by modeling the pilot's operation of the aircraft landing process based on the partially observable Markov decision process;
[0113] Obtain the cost function, wherein the cost function is a linear exponential quadratic function that maintains the aircraft landing at a stable descent rate and at a target speed;
[0114] With the goal of ensuring that the flight path formed by the pilot during the aircraft landing process is close to the ideal trajectory and that the pilot performs smooth operations during the landing process, the optimal control law for the pilot during the aircraft landing process is determined based on the cost function.
[0115] In an exemplary embodiment of the present invention, the processing module 530 may determine the partially observable Markov decision model in the following manner:
[0116] Based on the partially observable Markov decision process, the pilot's aircraft landing operation is modeled to obtain a partially observable Markov decision model. The state variables, decision actions, state transition probabilities, observations, and cost functions of this partially observable Markov decision model are defined, where...
[0117] The state variables are determined based on the aircraft operating state parameters;
[0118] The decision action is determined based on the pitch angle and roll angle positions of the control sticks.
[0119] The state transition probability is determined based on the system state linear model corresponding to the aircraft operating state parameters;
[0120] Based on the observed values of the target operating state parameters, the observations are determined, and
[0121] Construct a cost function to maintain the aircraft's descent rate at a stable rate and its landing at a target speed.
[0122] In an exemplary embodiment of the present invention, the processing module 530 may determine the system state linear model corresponding to the aircraft operating state parameters in the following manner:
[0123] Based on historical flight data, linear modeling is performed on each of the aircraft operating state parameters to obtain a system state linear model corresponding to the aircraft operating state parameters. The system state linear model is used to characterize the model formed by the aircraft operating state parameters being affected by any one or more of other aircraft operating state parameters, decision actions, and random terms.
[0124] In an exemplary embodiment of the present invention, the calling module 520 can pre-train the neural network model in the following manner:
[0125] Obtain a training dataset, wherein the training dataset includes multiple training data, and the training data includes external environment training parameters and aircraft operating status training parameters;
[0126] Based on the external environment training parameters and the aircraft operating status training parameters, predictive action training data of the pilot during the aircraft landing process is obtained, which corresponds to the external environment training parameters and the aircraft operating status training parameters.
[0127] Acquire the pilot’s actual action training data during the aircraft landing process, which corresponds to the external environment training parameters and the aircraft operating status training parameters.
[0128] Based on the mean squared error of the predicted action training data and the real action training data, a loss function is determined.
[0129] The neural network model is pre-trained based on the loss function to obtain the trained neural network model.
[0130] Figure 6 An example is a schematic diagram of the physical structure of an electronic device, such as... Figure 6As shown, the electronic device may include: a processor 610, a communications interface 620, a memory 630, and a communications bus 640, wherein the processor 610, the communications interface 620, and the memory 630 communicate with each other through the communications bus 640. The processor 610 can call logical instructions in the memory 630 to execute a pilot landing operation analysis method. This method includes: acquiring external environmental parameters and aircraft operating state parameters generated during the pilot's landing operation; calling a pre-trained neural network model and inputting the external environmental parameters and aircraft operating state parameters into the neural network model to obtain a state transition matrix at each time point output by the neural network model. The neural network model is used to obtain the corresponding state transition matrix at each time point based on the external environmental parameters and aircraft operating state parameters. Based on the state transition matrix, hidden states corresponding to the state transition matrix are obtained to complete the pilot landing operation analysis. The hidden states are used to characterize various states of the pilot's attention allocation pattern, including any one or more of the following: a state without attention deficit, a state with attention deficit regarding yaw angle, a state with attention deficit regarding altitude, and a state with attention deficit regarding airspeed.
[0131] Furthermore, the logical instructions in the aforementioned memory 630 can be implemented as software functional units and, when sold or used as independent products, can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.
[0132] On the other hand, the present invention also provides a computer program product, which includes a computer program that can be stored on a non-transitory computer-readable storage medium. When the computer program is executed by a processor, the computer can execute the pilot landing operation analysis method provided by the above methods. The method includes: acquiring external environmental parameters and aircraft operating state parameters formed during the pilot's operation of the aircraft landing; calling a pre-trained neural network model and inputting the external environmental parameters and the aircraft operating state parameters into the neural network model to obtain a state transition matrix at each time point output by the neural network model, wherein the neural network model is used to obtain the corresponding state transition matrix at each time point based on the external environmental parameters and the aircraft operating state parameters; and obtaining the hidden state corresponding to the state transition matrix based on the state transition matrix to complete the pilot landing operation analysis, wherein the hidden state is used to characterize the various states of the pilot's attention allocation mode, and the hidden state includes any one or more of the following: a state without attention deficit, a state with attention deficit to yaw angle, a state with attention deficit to altitude, and a state with attention deficit to airspeed.
[0133] In another aspect, the present invention also provides a non-transitory computer-readable storage medium storing a computer program thereon. When executed by a processor, the computer program is implemented to perform the pilot landing operation analysis method provided by the above methods. The method includes: acquiring external environmental parameters and aircraft operating state parameters generated during the pilot's operation of the aircraft landing; calling a pre-trained neural network model and inputting the external environmental parameters and the aircraft operating state parameters into the neural network model to obtain a state transition matrix at each time point output by the neural network model, wherein the neural network model is used to obtain the corresponding state transition matrix at each time point based on the external environmental parameters and the aircraft operating state parameters; and obtaining the hidden state corresponding to the state transition matrix based on the state transition matrix to complete the pilot landing operation analysis. The hidden state is used to characterize the various states of the pilot's attention allocation pattern, and the hidden state includes any one or more of the following: a state without attention deficit, a state with attention deficit regarding yaw angle, a state with attention deficit regarding altitude, and a state with attention deficit regarding airspeed.
[0134] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. Those skilled in the art can understand and implement this without any creative effort.
[0135] Through the above description of the embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus necessary general-purpose hardware platforms, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solutions, in essence or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods described in the various embodiments or some parts of the embodiments.
[0136] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.
Claims
1. A method for analyzing pilot landing operations, characterized in that, The method includes: Acquire external environmental parameters and aircraft operational status parameters generated during the pilot's landing operation; A pre-trained neural network model is invoked, and the external environment parameters and the aircraft operating state parameters are input into the neural network model to obtain the state transition matrix at each time point output by the neural network model. The neural network model is used to obtain the corresponding state transition matrix at each time point based on the external environment parameters and the aircraft operating state parameters. Based on the state transition matrix, hidden states corresponding to the state transition matrix are obtained to complete the landing operation analysis of the pilot. The hidden states characterize various states of the pilot's attention allocation pattern. The hidden states include any one or more of the following: a state without attention deficit, a state with attention deficit regarding yaw angle, a state with attention deficit regarding altitude, and a state with attention deficit regarding airspeed. After obtaining the hidden states corresponding to the state transition matrix, the method further includes: Based on the hidden state, the observed values of the target operating state parameters formed during the pilot's operation of the aircraft landing process corresponding to the hidden state are obtained, wherein the target operating state parameters include any one or more of yaw angle, altitude and airspeed. Based on the observed values of the target operating state parameters, the predicted actions of the pilot during the aircraft landing process are obtained. Based on the predicted actions, the flight risks for the pilot during the aircraft landing process are determined, wherein... Before obtaining the predicted actions of the pilot during aircraft landing based on the observed values of the target operational state parameters, the method further includes: Determine the optimal control law for the pilot during the aircraft landing process; The predicted actions of the pilot during the aircraft landing process, obtained based on the observed values of the target operating state parameters, specifically include: Based on the observed values of the target operating state parameters and the optimal control law, the pilot's predicted actions during the aircraft landing process are obtained.
2. The pilot landing operation analysis method according to claim 1, characterized in that, The optimal control law for the pilot during aircraft landing is determined in the following manner: A pre-built partially observable Markov decision model corresponding to the pilot's operation of the aircraft landing process is invoked, wherein the partially observable Markov decision model is obtained by modeling the pilot's operation of the aircraft landing process based on the partially observable Markov decision process; Obtain the cost function, wherein the cost function is a linear exponential quadratic function that maintains the aircraft's descent at a stable rate and landing at the target speed; With the goal of ensuring that the flight path formed by the pilot during the aircraft landing process is close to the ideal trajectory and that the pilot performs smooth operations during the landing process, the optimal control law for the pilot during the aircraft landing process is determined based on the cost function.
3. The pilot landing operation analysis method according to claim 2, characterized in that, The partially observable Markov decision model is constructed in the following manner: Based on the partially observable Markov decision process, the pilot's aircraft landing operation is modeled to obtain a partially observable Markov decision model. The state variables, decision actions, state transition probabilities, observations, and cost functions of this partially observable Markov decision model are defined, where... The state variables are determined based on the aircraft operating state parameters; The decision action is determined based on the pitch angle and roll angle positions of the control sticks. The state transition probability is determined based on the system state linear model corresponding to the aircraft operating state parameters; Based on the observed values of the target operating state parameters, the observations are determined, and Construct a cost function to maintain the aircraft's descent rate at a stable rate and its landing at a target speed.
4. The pilot landing operation analysis method according to claim 3, characterized in that, The linear system state model corresponding to the aircraft operating state parameters is constructed in the following manner: Based on historical flight data, linear modeling is performed on each of the aircraft operating state parameters to obtain a system state linear model corresponding to the aircraft operating state parameters. The system state linear model is used to characterize the model formed by the aircraft operating state parameters being affected by any one or more of other aircraft operating state parameters, decision actions, and random terms.
5. The pilot landing operation analysis method according to any one of claims 1 to 4, characterized in that, The neural network model was pre-trained using the following method: Obtain a training dataset, wherein the training dataset includes multiple training data, and the training data includes external environment training parameters and aircraft operating status training parameters; Based on the external environment training parameters and the aircraft operating status training parameters, predictive action training data of the pilot during the aircraft landing process is obtained, which corresponds to the external environment training parameters and the aircraft operating status training parameters. Acquire the pilot’s actual action training data during the aircraft landing process, which corresponds to the external environment training parameters and the aircraft operating status training parameters. Based on the mean squared error of the predicted action training data and the real action training data, a loss function is determined. The neural network model is pre-trained based on the loss function to obtain the trained neural network model.
6. A pilot landing operation analysis device, characterized in that, The apparatus is used to implement the pilot landing operation analysis method according to any one of claims 1 to 5, the apparatus comprising: The acquisition module is used to acquire external environmental parameters and aircraft operating status parameters generated during the pilot's operation of the aircraft landing. The calling module is used to call a pre-trained neural network model and input the external environment parameters and the aircraft operating state parameters into the neural network model to obtain the state transition matrix at each time point output by the neural network model. The neural network model is used to obtain the corresponding state transition matrix at each time point based on the external environment parameters and the aircraft operating state parameters. The processing module is used to obtain the hidden states corresponding to the state transition matrix based on the state transition matrix in order to complete the landing operation analysis of the pilot. The hidden states are used to characterize the various states of the pilot's attention allocation mode. The hidden states include any one or more of the following: a state without attention deficit, a state with attention deficit to yaw angle, a state with attention deficit to altitude, and a state with attention deficit to airspeed.
7. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the computer program, it implements the pilot landing operation analysis method as described in any one of claims 1 to 5.
8. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the pilot landing operation analysis method as described in any one of claims 1 to 5.