A method and system for constructing a pilot operation behavior prediction model

Abstract

The application provides a pilot operation behavior prediction model construction method and system, comprising: obtaining flight sample data of multiple flights of a given aircraft type; obtaining response characteristics of pilots to risk environments based on reverse learning of the flight sample data; constructing an influence graph model according to the response characteristics of pilots to risk environments and multiple environmental parameters and behavior parameters; obtaining potential states and behaviors of pilots by transforming and solving the influence graph model; and training and verifying a preset neural network by using the flight sample data, the response characteristics of pilots to risk environments and the potential states and behaviors of pilots to generate a pilot operation behavior prediction model. The application solves the problem that the prior art is difficult to predict operation behaviors according to external factors and decisions of pilots.

Description

Technical Field

[0001] This invention relates to the field of behavior prediction technology, and in particular to a method and system for constructing a pilot operation behavior prediction model. Background Technology

[0002] In the context of human-machine collaboration, machines cannot completely replace human operations in complex environments, especially in critical situations such as aircraft accidents. Therefore, studying sequential decision-making behavior in uncertain environments is crucial, as it affects both the design of intelligent machines and the performance of human-machine collaboration.

[0003] Pilot-controlled sequential decision-making can be viewed as consisting of two parts: an input system and an output system. During a flight mission, the sensory system collects background information from the environment and information on changes in the flight system's state, providing input to the decision-making system. This includes the visual system, vestibular system, tactile system, and proprioceptive system. The decision-making system then formulates actions based on the perceived information. Existing research on pilot control behavior includes control theory, human physiology, fuzzy control models, and neural network models. These models assume the pilot is the controller, and the aircraft state is the primary intermediate variable in the interaction between the pilot and the aircraft. The pilot perceives environmental information, makes operational decisions, and executes them, thus influencing the aircraft state.

[0004] Previous research on sequential decision-making has focused on finding the optimal solution to the sequential decision problem by maximizing the performance function or minimizing the cost function. However, there has been little analysis of the irrational, non-optimal, and unconscious characteristics of human behavior. In reality, people tend to rely on common sense and intuition to process known information and make decisions, rather than based on rigorous logical calculations and optimization of utility goals. This makes it difficult to provide auxiliary decision-making suggestions for aircraft. Summary of the Invention

[0005] This invention provides a method and system for constructing a pilot operational behavior prediction model, which solves the problem that existing technologies are unable to predict operational behavior based on external factors and pilot decisions.

[0006] This invention provides a method for predicting pilot operational behavior, comprising:

[0007] Obtain flight sample data for multiple flights of a given aircraft type;

[0008] Inverse learning is performed based on the flight sample data to obtain the pilots' response characteristics to risky environments;

[0009] An impact diagram model is constructed based on the pilots' response characteristics to the risky environment and multiple environmental and behavioral parameters;

[0010] By transforming and solving the influence diagram model, the pilot's potential state and behavior can be obtained;

[0011] The pilot's operational behavior prediction model is generated by training and validating a pre-set neural network using the flight sample data, the pilot's response characteristics to the risky environment, and the pilot's potential state and behavior.

[0012] The present invention provides a method for constructing a pilot operational behavior prediction model.

[0013] The acquisition of flight sample data for multiple flights of a given aircraft type specifically includes:

[0014] Acquire environmental variables, flight status variables, and pilot operation variables for multiple flights;

[0015] Data cleaning was performed on the environmental variables, flight status variables, and pilot operation variables to remove samples with missing landing points and problematic data for each flight, thus obtaining valid sample data.

[0016] The present invention provides a method for constructing a pilot operational behavior prediction model.

[0017] Inverse learning is performed based on the flight sample data to obtain the pilots' response characteristics to risky environments, specifically including:

[0018] Extract current environmental information from the flight sample data;

[0019] To obtain the optimal decision made by the pilot based on the current environmental information, a Markov decision process is established.

[0020] In the process of establishing Markov decision-making, the pilot's response characteristics to risky environments are obtained through inverse learning.

[0021] The present invention provides a method for constructing a pilot operational behavior prediction model.

[0022] An impact diagram model is constructed based on the pilot's response characteristics to a risky environment and multiple environmental and behavioral parameters, specifically including:

[0023] The pilot's response characteristics to the risky environment, multiple environmental parameters, and behavioral parameters are used as nodes in the influence diagram;

[0024] In an influence graph, nodes represent key variables, and directed arcs represent various relationships between variables. An influence graph model is built based on the relationships between variables at the nodes.

[0025] The present invention provides a method for constructing a pilot operational behavior prediction model.

[0026] The process of transforming and solving the influence diagram model to obtain the pilot's potential state and behavior specifically includes:

[0027] The influence graph is converted into a probability graph. Through the transformation of probability relationships, the influence graph model of pilot behavior decision-making is transformed into a hidden Markov model.

[0028] The hidden Markov model is solved by a pre-defined neural network to obtain the pilot's potential state and behavior.

[0029] The present invention provides a method for constructing a pilot operational behavior prediction model.

[0030] The process involves training and validating a pre-defined neural network using flight sample data, pilot response characteristics to risky environments, and pilot potential states and behaviors to generate a pilot operational behavior prediction model. Specifically, this includes:

[0031] The effective dataset is composed of flight sample data, pilot response characteristics to risky environments, and pilot potential states and behaviors.

[0032] The effective dataset is divided into a test set and a validation set. The neural network is trained and its parameters are continuously adjusted using the test set. The trained neural network is validated using the validation set. By dynamically adjusting the neural network, a behavior prediction model is obtained.

[0033] A prediction method for a pilot operational behavior prediction model provided by the present invention includes:

[0034] Known environmental variables, aircraft state variables, and pilot environmental response characteristics are used as inputs to the behavior prediction model;

[0035] Predict pilot operational behavior under different background environmental factors and output the prediction results.

[0036] The prediction method for a pilot operational behavior prediction model provided by the present invention further includes:

[0037] The actual flight operation behavior of the pilot is compared with the predicted operation behavior to represent the probability of the pilot's potential state;

[0038] By combining background environmental factors at different times, we can identify the influencing factors of potential pilot control state transitions, assess the performance of specific types of pilots in specific environmental factors, and evaluate the impact of pilots' personal behavioral characteristics on flight operations.

[0039] The present invention also provides a system for constructing a pilot operational behavior prediction model, the system comprising:

[0040] The data acquisition module is used to acquire flight sample data of multiple flights of a given aircraft type;

[0041] The response feature extraction module is used to obtain the pilot's response features to the risky environment by performing reverse learning based on the flight sample data.

[0042] The impact diagram modeling module is used to construct impact diagram models based on the pilots' response characteristics to risky environments, as well as multiple environmental and behavioral parameters.

[0043] The potential state determination module is used to obtain the pilot's potential state and behavior by transforming and solving the influence diagram model;

[0044] The prediction model generation module is used to train and validate a preset neural network using the flight sample data, the pilot's response characteristics to the risky environment, and the pilot's potential state and behavior, to generate a pilot operation behavior prediction model.

[0045] 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 operation behavior prediction model construction method as described above.

[0046] This invention provides a method and system for constructing a pilot operational behavior prediction model. By constructing an influence graph model, the influence graph is transformed into a probability graph. Through the transformation of probability relationships, it is constructed into a hidden Markov model, which is then solved using a neural network. This establishes a behavior prediction model that can analyze the pilot's potential state, predict the pilot's operational behavior during landing, and achieve the assessment and long-term tracking of the pilot's operational behavior. This assists in pilot training and improves the aviation safety of the flight fleet. Attached Figure Description

[0047] 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.

[0048] Figure 1 This is one of the flowcharts illustrating a method for constructing a pilot operational behavior prediction model provided by the present invention;

[0049] Figure 2 This is the second flowchart of a method for constructing a pilot operational behavior prediction model provided by the present invention;

[0050] Figure 3 This is the third flowchart of a method for constructing a pilot operational behavior prediction model provided by the present invention;

[0051] Figure 4 This is the fourth flowchart of a method for constructing a pilot operational behavior prediction model provided by the present invention;

[0052] Figure 5 This is the fifth flowchart of a method for constructing a pilot operational behavior prediction model provided by the present invention;

[0053] Figure 6 This is the sixth flowchart of a method for constructing a pilot operational behavior prediction model provided by the present invention;

[0054] Figure 7 This is the seventh flowchart of a method for constructing a pilot operational behavior prediction model provided by the present invention;

[0055] Figure 8 This is a schematic diagram of the module connections of a pilot operational behavior prediction model construction system provided by the present invention;

[0056] Figure 9 This is a schematic diagram of the structure of the electronic device provided by the present invention.

[0057] Figure label:

[0058] 110: Data Acquisition Module; 120: Response Feature Extraction Module; 130: Influence Map Modeling Module; 140: Potential State Determination Module; 150: Prediction Model Generation Module;

[0059] 910: Processor; 920: Communication interface; 930: Memory; 940: Communication bus. Detailed Implementation

[0060] 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.

[0061] The following is combined Figures 1-7 The present invention describes a method for constructing a pilot operational behavior prediction model, comprising:

[0062] S100: Obtain flight sample data for multiple flights of a given aircraft type;

[0063] S200. Based on the flight sample data, perform reverse learning to obtain the pilot's response characteristics to the risky environment;

[0064] S300: Construct an impact diagram model based on the pilot's response characteristics to the risky environment and multiple environmental and behavioral parameters;

[0065] S400. By transforming and solving the influence diagram model, the pilot's potential state and behavior are obtained;

[0066] S500: The pilot's operational behavior prediction model is generated by training and validating the preset neural network using the flight sample data, the pilot's response characteristics to the risky environment, and the pilot's potential state and behavior.

[0067] Acquire flight sample data of multiple flights of a given aircraft type within a set time period and perform data preprocessing;

[0068] A Markov decision process is established based on preprocessed flight sample data, and the pilot's response characteristics to the risky environment are obtained through inverse learning.

[0069] An impact diagram model is constructed based on the pilot's response characteristics to the risky environment, environmental information changing over time, the pilot's psychological state, the pilot's actual operational behavior, and overall performance.

[0070] In this invention, an influence diagram model considering both the potential state of humans and the visible state of machines is constructed to reveal the relationship between environmental information and individual cognition and decision-making behavior, uncovering the hidden states behind decision-making behavior and better characterizing human behavioral patterns. The cognitive decision-making of pilots is modeled, considering their potential psychological states, i.e., capturing changes in the pilot's control patterns during flight, such as capture control, tracking control, and leveling control, analyzing key decision information, and providing suggestions for aircraft auxiliary decision-making.

[0071] Obtain flight sample data for multiple flights of a given aircraft type, specifically including:

[0072] S101. Obtain environmental variables, flight status variables, and pilot operation variables for multiple flights;

[0073] S102. Perform data cleaning on the environmental variables, flight status variables, and pilot operation variables to remove samples with missing landing points and data with problems in each flight, and obtain valid sample data.

[0074] In this invention, given flight QAR data for multiple flights of a certain aircraft type over a period of time, where QAR is an onboard device recording flight data and includes multiple flight parameters, environmental variables include visibility of the landing airport, weather, temperature, air pressure, absolute wind direction, and wind speed. Aircraft state variables include radio altitude, ground speed, inertial vertical velocity, vertical overload (acceleration), descent angle, pitch angle, roll angle, yaw angle, landing gear touchdown status (discrete variable), and aircraft weight. Pilot operation variables include the pitch and roll control stick positions of the primary / secondary pilots. The data is cleaned and preprocessed to remove samples with missing landing points and problematic data from each flight (including altitude changes less than 1 during descent and vertical velocities before landing being less than those at landing), resulting in valid samples. Altitude and pitch angle data are averaged per second, while vertical overload data is the maximum value per second. This data cleaning and preprocessing ensures the data's authenticity and meets the requirements of subsequent data processing.

[0075] Inverse learning is performed based on the flight sample data to obtain the pilots' response characteristics to risky environments, specifically including:

[0076] S201. Extract current environmental information from the flight sample data;

[0077] S202. Obtain the optimal decision made by the pilot based on the current environmental information and establish a Markov decision process;

[0078] S203. In the process of establishing Markov decision-making, the pilot's response characteristics to the risky environment are obtained through reverse learning.

[0079] In this invention, pilots will exhibit different behaviors under different environments. In risky environments, experienced pilots will make the best choice based on a comprehensive evaluation. The pilot is simulated as an optimal controller, assuming that the pilot is well-trained and has extensive flight experience, capable of making optimal decisions based on the current environment to achieve the decision objective. A Markov decision process is established, with state variables being aircraft state (such as vertical speed, pitch angle, etc.) and decision variables being pilot operational behaviors (such as stick positions for pitch angle, roll angle, etc.). Parameters are introduced to characterize the pilot's response characteristics to risky environments, such as risk preference parameters, flight proficiency, and trade-off parameters between landing speed and process stability. Based on the obtained flight sample data, inverse reinforcement learning can be used to obtain the pilot's response characteristic parameters to risky environments.

[0080] An impact diagram model is constructed based on the pilot's response characteristics to a risky environment and multiple environmental and behavioral parameters, specifically including:

[0081] The pilot's response characteristics to the risky environment, multiple environmental parameters, and behavioral parameters are used as nodes in the influence diagram;

[0082] In an influence graph, nodes represent key variables, and directed arcs represent various relationships between variables. An influence graph model is built based on the relationships between variables at the nodes.

[0083] In this invention, the influence graph is an acyclic directed graph composed of nodes and directed arcs. Nodes represent the main variables in the problem under study, and directed arcs represent the various relationships between variables. It is an intuitive graphical representation of the problem structure based on the decision-maker's (or client's) description of the problem and combined with expert knowledge. The graph clearly reveals the relationships between variables, especially the conditional independence and information flow. The influence graph itself has two levels: the first level is the graph itself, and the second level is the data structure of each node. When using the influence graph to build a problem model, interpretable influence graphs can be defined based on interrelationships, functional relationships, and numerical relationships. Influence graphs can also be used for computer information storage and formal numerical calculations.

[0084] Based on the perception, cognition, and decision-making behavior cycle, an influence diagram is used for modeling. The variables included in the multiple environmental and behavioral parameters are: pilot response characteristics to the risky environment that do not change over time; environmental information that changes over time; pilot psychological states at different times, i.e., the choice of three flight control strategies, including capture control, tracking control, and leveling decision states; actual operational behaviors of the pilot based on the current environmental information and their chosen flight strategy; the aircraft's vertical dynamic state (vertical velocity) as a function of aircraft flight performance, where the aircraft state is observable and the previous period's flight performance affects the next; the unseen performance resulting from both pilot psychological states and actual operational behaviors; and the overall performance obtained by summing the observable aircraft flight performance and the unobservable pilot state-behavioral performance.

[0085] By transforming and solving the influence diagram model, the pilot's potential state and behavior are obtained, specifically including:

[0086] The influence graph is converted into a probability graph. Through the transformation of probability relationships, the influence graph model of pilot behavior decision-making is transformed into a hidden Markov model.

[0087] The hidden Markov model is solved by a pre-defined neural network to obtain the pilot's potential state and behavior.

[0088] This involves converting the influence diagram into a probability diagram, and through probabilistic relationship transformation, converting the influence diagram model of pilot behavioral decisions into a Hidden Markov Model, including:

[0089] By converting the influence diagram into a probability diagram and then transforming it through probabilistic relationships, the influence diagram model of pilot behavior and decision-making can be converted into a hidden Markov model, which can characterize the influence of background environmental factors, including aircraft status, environmental information, and pilot environmental response characteristics, on the pilot's potential state and decision-making behavior.

[0090] The hidden Markov model is solved using a pre-defined neural network to obtain the pilot's potential state and behavior, including:

[0091] S301. Input flight sample data, pilot response characteristics to risky environments, and environmental information into the neural network. After passing through two layers of neural network, output the parameters of the hidden Markov model.

[0092] S302, The parameters of the hidden Markov model include state behavior matrix elements and state transition matrix elements;

[0093] S303. Maximize the probability of the behavior sequence derived from the Hidden Markov Model based on the elements of the state behavior matrix and the elements of the state transition matrix, and determine the pilot's potential state and behavior based on the probability of the behavior sequence.

[0094] In this invention, a neural network is used to solve for relevant parameters. The input of the neural network is the aircraft state, the pilot's response characteristics to the risky environment, and environmental information. After passing through two layers of neural network, the output is the parameters of the hidden Markov model, namely the state-behavior matrix elements and the state transition matrix elements. The neural network maximizes the probability of the behavior sequence obtained from the hidden Markov model and determines the pilot's potential state and behavior based on the probability of the behavior sequence.

[0095] In one specific embodiment, the initial state probability is assumed to be [0.5, 0.5, 0], meaning that within the study range from 200 feet to landing, the probability of the pilot being in the acquisition control state and the tracking control state is 0.5 respectively, and the leveling decision state is performed below 200 feet. No constraints are imposed on the elements of the state transition matrix; during the 200-foot to landing phase, the pilot's actions regarding the vertical pitch angle are represented as discrete variables [-10, -9, -8, -7, -6, -5, -4, -3, -2, -1, 0, 1, 2, 3, 4], and no constraints are imposed on the parameters in the state transition matrix and the state behavior matrix.

[0096] By training and validating a pre-defined neural network using flight sample data, pilot response characteristics to risky environments, and pilot potential states and behaviors, a pilot operational behavior prediction model is generated, specifically including:

[0097] S401. Combine flight sample data, pilot response characteristics to risky environments, and pilot potential states and behaviors into an effective dataset;

[0098] S402. Divide the effective dataset into a test set and a validation set. Train the neural network and continuously adjust the parameters using the test set. Validate the trained neural network using the validation set. Dynamically adjust the neural network to obtain a behavior prediction model.

[0099] By training a neural network, the relationship between aircraft state, environmental information, pilot environmental response characteristics, and pilot latent state and behavior was obtained. Two-thirds of the extracted valid samples were used as the training set, and the remaining flight data were used as the test set. The trained neural network was then used to validate the data in the test set, predicting pilot operational behavior and outputting a behavioral prediction model. The results show that the sequential decision-making model incorporating pilot latent states more accurately characterizes pilot operational behavior. Based on the "perception-cognition-decision" behavioral model, perceived background environmental factors do not directly affect decision-making behavior; rather, it is based on the pilot's cognition of these factors, i.e., changes in their personal state, which in turn influence their actual control operational output.

[0100] The behavior prediction model predicts pilot operational behavior based on multiple real-time data sets, specifically including:

[0101] S501. Use known environmental variables, aircraft state variables, and pilot environmental response characteristics as inputs to the behavior prediction model;

[0102] S502: Predict pilot operational behavior under different background environmental factors and output the prediction results.

[0103] This invention establishes a model considering the impact of pilots' potential states and uses neural networks to train pilot state changes and behavioral patterns under different background environmental factors. Given environmental variables, aircraft state variables, and pilot environmental response characteristics, these can be used as inputs to a behavior prediction model to predict pilot operational behavior under different background environmental factors, thus anticipating and mitigating flight risks. This invention can automatically identify pilots' expected behaviors, thereby establishing a dynamically adaptive control system to better suit human objectives and facilitate the development of decision-making support systems. Furthermore, it can also enable machines to mimic the operational behaviors of human pilots performing specific takeoff missions under identical conditions, simulating human perception, cognition, and decision-making behavior to conduct independent flight missions.

[0104] S601. Compare the pilot's actual flight operation behavior with the predicted operation behavior to represent the pilot's potential state probability;

[0105] S602. In conjunction with background environmental factors at different times, determine the influencing factors of potential control state transitions of pilots, and assess the performance of specific types of pilots in specific environmental factors and the impact of pilots' personal behavioral characteristics on flight operations.

[0106] Specifically, the actual and predicted flight operation behaviors of pilots are visualized, and the probabilities of potential pilot states are represented. Combined with background environmental factors at different times, the influencing factors of potential control state transitions are revealed, and the performance of specific types of pilots under specific environmental factors is evaluated. After obtaining the pilot's environmental response characteristics through extensive flight data, the predicted pilot operation behaviors under different background environmental factors are compared with the aircraft behavior controlled by existing machine intelligence, assessing the impact of individual pilot behavior characteristics on flight operations. In pilot training, long-term flight data of the same pilot can be collected to observe changes in potential control state transitions. Cluster analysis of the state transition matrix and state behavior matrix is ​​performed to evaluate the pilot's behavioral patterns at different times.

[0107] This invention discloses a method for predicting pilot operational behavior. It constructs an influence graph model based on the pilot's response characteristics to a risky environment, time-varying environmental information, pilot's psychological state, actual operational behavior, and overall performance. This influence graph is then transformed into a probability graph, and through the transformation of probabilistic relationships, it is constructed into a Hidden Markov Model (HMM). This model is solved using a neural network, establishing a behavior prediction model that can analyze the pilot's potential state and predict their operational behavior during landing. This enables the assessment and long-term tracking of pilot operational behavior, assisting in pilot training and improving aviation safety for the flight fleet. The method constructs an influence graph model considering both the pilot's potential state and the machine's visible state, revealing the relationship between environmental information and individual cognition and decision-making behavior, uncovering the hidden states behind decision-making behavior, and better characterizing human behavioral patterns. It models the pilot's cognitive decision-making, considering the pilot's potential psychological state, i.e., capturing changes in the pilot's control mode during flight, analyzing key decision information, and providing suggestions for aircraft auxiliary decision-making.

[0108] refer to Figure 8 The present invention also discloses a pilot operational behavior prediction system, the system comprising:

[0109] The data acquisition module 110 is used to acquire flight sample data of multiple flights of a given aircraft type;

[0110] The response feature extraction module 120 is used to obtain the pilot's response features to the risky environment by performing reverse learning based on the flight sample data.

[0111] The influence diagram modeling module 130 is used to construct an influence diagram model based on the pilot's response characteristics to the risky environment and multiple environmental and behavioral parameters.

[0112] The potential state determination module 140 is used to obtain the pilot's potential state and behavior by transforming and solving the influence diagram model;

[0113] The prediction model generation module 150 is used to train and validate a preset neural network using the flight sample data, the pilot's response characteristics to the risky environment, and the pilot's potential state and behavior, to generate a pilot operation behavior prediction model.

[0114] The data acquisition module acquires environmental variables, flight status variables, and pilot operation variables for multiple flights.

[0115] Data cleaning was performed on the environmental variables, flight status variables, and pilot operation variables to remove samples with missing landing points and problematic data for each flight, thus obtaining valid sample data.

[0116] The response feature extraction module extracts current environmental information from the flight sample data.

[0117] To obtain the optimal decision made by the pilot based on the current environmental information, a Markov decision process is established.

[0118] In the process of establishing Markov decision-making, the pilot's response characteristics to risky environments are obtained through inverse learning.

[0119] The impact diagram modeling module uses the pilot's response characteristics to the risky environment, multiple environmental parameters, and behavioral parameters as nodes in the impact diagram;

[0120] In an influence graph, nodes represent key variables, and directed arcs represent various relationships between variables. An influence graph model is built based on the relationships between variables at the nodes.

[0121] The potential state determination module converts the influence graph into a probability graph, and through the transformation of probability relationships, it converts the influence graph model of pilot behavior decision-making into a hidden Markov model.

[0122] The hidden Markov model is solved by a pre-defined neural network to obtain the pilot's potential state and behavior.

[0123] The predictive model generation module combines flight sample data, pilot response characteristics to risky environments, and pilot potential states and behaviors into an effective dataset.

[0124] The effective dataset is divided into a test set and a validation set. The neural network is trained and its parameters are continuously adjusted using the test set. The trained neural network is validated using the validation set. By dynamically adjusting the neural network, a behavior prediction model is obtained.

[0125] Based on the acquired behavior prediction model, flight operational behaviors can be predicted, and corresponding evaluations can be performed based on the prediction results. Known environmental variables, aircraft state variables, and pilot environmental response characteristics are used as inputs to the behavior prediction model.

[0126] Predict pilot operational behavior under different background environmental factors and output the prediction results.

[0127] The actual flight operation behavior of the pilot is compared with the predicted operation behavior to represent the probability of the pilot's potential state;

[0128] By combining background environmental factors at different times, we can identify the influencing factors of potential pilot control state transitions, assess the performance of specific types of pilots in specific environmental factors, and evaluate the impact of pilots' personal behavioral characteristics on flight operations.

[0129] This invention discloses a pilot operational behavior prediction model construction system. By constructing an influence graph model based on the pilot's response characteristics to risky environments, time-varying environmental information, pilot's psychological state, actual operational behavior, and overall performance, the system transforms the influence graph into a probability graph. Through the transformation of probabilistic relationships, it constructs a Hidden Markov Model (HMM) and solves it using a neural network, establishing a behavior prediction model. This model can analyze the pilot's potential state, predict the pilot's operational behavior during landing, and achieve the assessment and long-term tracking of pilot operational behavior, assisting in pilot training and improving the aviation safety of the flight fleet. The system constructs an influence graph model considering both the potential state of the pilot and the visible state of the machine, revealing the relationship between environmental information and individual cognition and decision-making behavior, uncovering the hidden states behind decision-making behavior, and better characterizing human behavioral patterns. It models the pilot's cognitive decision-making, considering the pilot's potential psychological state, i.e., capturing changes in the pilot's control mode during flight, analyzing key decision information, and providing suggestions for aircraft auxiliary decision-making.

[0130] Figure 9 An example is a schematic diagram of the physical structure of an electronic device, such as... Figure 9As shown, the electronic device may include: a processor 910, a communication interface 920, a memory 930, and a communication bus 940, wherein the processor 910, the communication interface 920, and the memory 930 communicate with each other via the communication bus 940. The processor 910 can call logical instructions in the memory 930 to execute a method for constructing a pilot operational behavior prediction model, the method including:

[0131] Obtain flight sample data for multiple flights of a given aircraft type;

[0132] Inverse learning is performed based on the flight sample data to obtain the pilots' response characteristics to risky environments;

[0133] An impact diagram model is constructed based on the pilots' response characteristics to the risky environment and multiple environmental and behavioral parameters;

[0134] By transforming and solving the influence diagram model, the pilot's potential state and behavior can be obtained;

[0135] The pilot's operational behavior prediction model is generated by training and validating a pre-set neural network using the flight sample data, the pilot's response characteristics to the risky environment, and the pilot's potential state and behavior.

[0136] Furthermore, the logical instructions in the aforementioned memory 930 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, essentially, 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.

[0137] On the other hand, the present invention also provides a computer program product, the computer program product comprising a computer program that can be stored on a non-transitory computer-readable storage medium, wherein when the computer program is executed by a processor, the computer is able to execute a method for constructing a pilot operation behavior prediction model provided by the above methods, the method comprising:

[0138] Obtain flight sample data for multiple flights of a given aircraft type;

[0139] Inverse learning is performed based on the flight sample data to obtain the pilots' response characteristics to risky environments;

[0140] An impact diagram model is constructed based on the pilots' response characteristics to the risky environment and multiple environmental and behavioral parameters;

[0141] By transforming and solving the influence diagram model, the pilot's potential state and behavior can be obtained;

[0142] The pilot's operational behavior prediction model is generated by training and validating a pre-set neural network using the flight sample data, the pilot's response characteristics to the risky environment, and the pilot's potential state and behavior.

[0143] In another aspect, the present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, is implemented to perform a method for constructing a pilot operational behavior prediction model provided by the methods described above, the method comprising:

[0144] Obtain flight sample data for multiple flights of a given aircraft type;

[0145] Inverse learning is performed based on the flight sample data to obtain the pilots' response characteristics to risky environments;

[0146] An impact diagram model is constructed based on the pilots' response characteristics to the risky environment and multiple environmental and behavioral parameters;

[0147] By transforming and solving the influence diagram model, the pilot's potential state and behavior can be obtained;

[0148] The pilot's operational behavior prediction model is generated by training and validating a pre-set neural network using the flight sample data, the pilot's response characteristics to the risky environment, and the pilot's potential state and behavior.

[0149] 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.

[0150] 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.

[0151] 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 constructing a pilot operational behavior prediction model, characterized in that, include: Obtain flight sample data for multiple flights of a given aircraft type; Extract current environmental information from the flight sample data; To obtain the optimal decision made by the pilot based on the current environmental information, a Markov decision process is established. In the process of establishing Markov decision-making, the pilot's response characteristics to the risky environment are obtained through inverse learning; The pilot's response characteristics to the risky environment, multiple environmental parameters, and behavioral parameters are used as nodes in the influence diagram; In an influence graph, nodes represent key variables, and directed arcs represent various relationships between variables. An influence graph model is built based on the relationships between variables at the nodes. The influence graph is converted into a probability graph. Through the transformation of probability relationships, the influence graph model of pilot behavior decision-making is transformed into a hidden Markov model. The hidden Markov model is solved by a pre-defined neural network to obtain the pilot's potential state and behavior. By training and validating a pre-defined neural network using the flight sample data, the pilot's response characteristics to the risky environment, and the pilot's potential state and behavior, a pilot operational behavior prediction model is generated.

2. The method for constructing a pilot operational behavior prediction model according to claim 1, characterized in that, The acquisition of flight sample data for multiple flights of a given aircraft type specifically includes: Acquire environmental variables, flight status variables, and pilot operation variables for multiple flights; Data cleaning was performed on the environmental variables, flight status variables, and pilot operation variables to remove samples with missing landing points and problematic data for each flight, thus obtaining valid sample data.

3. The method for constructing a pilot operational behavior prediction model according to claim 1, characterized in that, The process of training and validating a pre-defined neural network using the flight sample data, pilot response characteristics to risky environments, and pilot potential states and behaviors to generate a pilot operational behavior prediction model specifically includes: The effective dataset is composed of flight sample data, pilot response characteristics to risky environments, and pilot potential states and behaviors. The effective dataset is divided into a test set and a validation set. The neural network is trained and its parameters are continuously adjusted using the test set. The trained neural network is validated using the validation set. By dynamically adjusting the neural network, a behavior prediction model is obtained.

4. A prediction method for a pilot operation behavior prediction model generated based on the pilot operation behavior prediction model construction method according to any one of claims 1-3, characterized in that, include: Known environmental variables, aircraft state variables, and pilot environmental response characteristics are used as inputs to the behavior prediction model; Predict pilot operational behavior under different background environmental factors and output the prediction results.

5. The prediction method of the pilot operation behavior prediction model according to claim 4, characterized in that, Also includes: The actual flight operation behavior of the pilot is compared with the predicted operation behavior to represent the probability of the pilot's potential state; By combining background environmental factors at different times, we can identify the influencing factors of potential pilot control state transitions, assess the performance of specific types of pilots in specific environmental factors, and evaluate the impact of pilots' personal behavioral characteristics on flight operations.

6. A system for constructing a pilot operational behavior prediction model, characterized in that, The system includes: The data acquisition module is used to acquire flight sample data of multiple flights of a given aircraft type; The response feature extraction module is used for: Extract current environmental information from the flight sample data; To obtain the optimal decision made by the pilot based on the current environmental information, a Markov decision process is established. In the process of establishing Markov decision-making, the pilot's response characteristics to the risky environment are obtained through inverse learning; The influence diagram modeling module is used for: The pilot's response characteristics to the risky environment, multiple environmental parameters, and behavioral parameters are used as nodes in the influence diagram; In an influence graph, nodes represent key variables, and directed arcs represent various relationships between variables. An influence graph model is built based on the relationships between variables at the nodes. The potential state determination module is used for: The influence graph is converted into a probability graph. Through the transformation of probability relationships, the influence graph model of pilot behavior decision-making is transformed into a hidden Markov model. The hidden Markov model is solved by a pre-defined neural network to obtain the pilot's potential state and behavior. The prediction model generation module is used to train and validate a preset neural network using the flight sample data, the pilot's response characteristics to the risky environment, and the pilot's potential state and behavior, to generate a pilot operation behavior prediction model.

7. 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 operation behavior prediction model construction method as described in any one of claims 1 to 5.

Images