Method, device and electronic device for analyzing pilot landing operations

Abstract

The application provides a kind of pilot landing operation analysis method, device and electronic equipment, it is related to flight evaluation analysis technical field, the method comprises: the target aircraft corresponding target flight data in landing phase is input into target Markov decision process, obtains the target aircraft corresponding pilot landing operation analysis result that target Markov decision process exports.This application technical scheme constructs Markov decision process by the descent rate of aircraft, expected descent rate, pitch angle change value and the aircraft state parameters recorded by rapid access recorder, and then substitutes target flight data to obtain pilot landing operation analysis result.Because the analysis of pilot operation in the present application uses the data of the aircraft, compared with the method of obtaining the nerve and muscle parameters of the pilot in the prior art, it is more accurate and easy to implement, in addition, determining the pilot landing operation analysis result based on the data of the aircraft is also sufficient to reflect the operation mentality of the pilot.

Description

Technical Field

[0001] This invention relates to the field of flight evaluation and analysis technology, and in particular to an analysis method, apparatus, and electronic device for pilot landing operations. Background Technology

[0002] Flight safety data from the past few decades shows that over half of all flight safety incidents occur during the landing process (approach and landing), with nearly one-third of these incidents occurring during landing. During landing, pilots need to use precise and robust maneuvers to level the aircraft and smoothly decelerate to landing in hazardous environments such as wind shear. In this process, pilots exhibit three distinct personal characteristics: a trade-off between the magnitude of control operations and achieving the required rate of descent, a trade-off between maintaining a stable rate of descent and landing at the target speed, and perception delay. Identifying and analyzing these three personal characteristics of pilots can facilitate targeted training and development, thereby improving flight safety levels.

[0003] Previous analyses of pilot landing maneuvers have remained largely hypothetical, typically starting with the pilot's physiological information. This requires pilots to wear neurological monitoring devices to acquire neural and muscular data, which is then used to create models for analyzing landing maneuvers. However, the use of such devices can compromise flight safety, making practical application difficult. Furthermore, these methods suffer from several problems: the models are often complex and cumbersome, containing numerous neural and muscular parameters that require prior human factors experiments; they also tend to focus on the pilot's physiological mechanisms rather than the more decisive psychological aspects of perception and control, thus hindering direct guidance for pilot training and development, and lacking a degree of abstract observation. Summary of the Invention

[0004] This invention provides a method, apparatus, and electronic device for analyzing pilot landing operations. It addresses the shortcomings of existing technologies that rely on implanting neural monitoring devices in pilots, which are difficult to apply and require prior human factors experiments to provide neural and muscle parameters, lacking abstract observation of the pilot's psychological state. The invention constructs a Markov decision process using the aircraft's descent rate, expected descent rate, pitch angle change, and aircraft state parameters recorded by the fast access recorder. This process is then substituted with target flight data to obtain the pilot's landing operation analysis results. Because this application uses aircraft data for pilot operation analysis, it is more accurate and feasible than existing methods that obtain pilot neural and muscle parameters. Furthermore, determining the pilot's landing operation analysis results based on aircraft data is sufficient to reflect the pilot's operational mindset.

[0005] This invention provides an analysis method for pilot landing operations, comprising the following steps.

[0006] The target flight data corresponding to the landing phase of the target aircraft is input into the target Markov decision process to obtain the pilot landing operation analysis results corresponding to the target aircraft output by the target Markov decision process; the state of the target Markov decision process is the aircraft state parameters; the action of the target Markov decision process is the change value of the aircraft pitch angle; the reward function of the target Markov decision process is determined based on the flight fast record dataset and the descent rate, expected descent rate, and pitch angle change value of the aircraft during the landing phase; the transition probability of the target Markov decision process is determined based on the flight fast record dataset; the flight fast record dataset represents the aircraft state parameters recorded by the fast access recorder; the data types included in the target flight data correspond to the data types included in the flight fast record dataset.

[0007] According to the analysis method for pilot landing operations provided by the present invention, the target Markov decision process is constructed based on the following: An aircraft system dynamics model is constructed based on the aforementioned rapid flight record dataset; the aircraft system dynamics model characterizes the transition probabilities of the target Markov decision process; Based on the descent rate, expected descent rate, and pitch angle change values ​​at each moment during the landing phase, multiple quadratic terms of the function are determined. Based on all the quadratic terms of the function and the aircraft system dynamic model, the reward function of the target Markov decision process is constructed. The target Markov decision process is determined based on the state, the action, the transition probability, the reward function, and the preset discount factor.

[0008] According to the analysis method for pilot landing operations provided by the present invention, the step of constructing an aircraft system dynamic model based on the flight rapid record dataset includes: The change in the aircraft's rate of descent is determined as the dependent variable, and the aircraft's pitch angle, airspeed, rate of descent, change in pitch angle, headwind and tailwind conditions, wind shear, and two moments of the aircraft in the non-vertical direction are determined as independent variables. Based on the dependent variable and all the independent variables, an initial linear function is constructed. Substitute the flight rapid recording dataset into the initial linear function, and use the least squares method to determine the dependent variable coefficient and the independent variable coefficient corresponding to each independent variable. Substituting the dependent variable coefficients and the independent variable coefficients into the initial linear function yields the aircraft system dynamic model.

[0009] According to the analytical method for pilot landing operations provided by the present invention, the step of constructing the reward function of the target Markov decision process based on all the quadratic terms of the function and the aircraft system dynamics model includes: Construct the initial reward function for the objective Markov decision process based on all the quadratic terms of the function; Based on the aircraft system dynamic model and the initial reward function, determine the coefficients of the quadratic terms corresponding to the quadratic terms of each function in the initial reward function; Substituting the coefficients of the quadratic terms corresponding to the quadratic terms of each of the aforementioned functions into the initial reward function yields the reward function of the target Markov decision process.

[0010] According to the analysis method for pilot landing operations provided by the present invention, the quadratic term of the function includes a target landing quadratic term, a landing phase quadratic term, and an operation smoothing quadratic term; The determination of multiple quadratic terms of a function based on the descent rate, expected descent rate, and pitch angle change values ​​at various moments during the landing phase includes: The difference between the descent rate at the last moment of the landing phase and the expected descent rate is defined as the target landing quadratic term; The quadratic term for the landing phase is determined based on the difference between the descent rate and the expected descent rate at each time point other than the final time point. The operational stability quadratic term is determined based on the pitch angle changes at various times during the landing phase of the aircraft.

[0011] According to the analysis method for pilot landing operations provided by the present invention, the quadratic term coefficients include the target landing coefficient corresponding to the target landing quadratic term, the landing phase coefficient corresponding to the landing phase quadratic term, and the operation stability coefficient corresponding to the operation stability quadratic term; The determination of the quadratic coefficients corresponding to each quadratic term in the initial reward function based on the aircraft system dynamic model and the initial reward function includes: For each preset perception delay, the predicted optimal pitch angle change value corresponding to each time point is determined based on the aircraft system dynamic model, the initial reward function, and the aircraft state parameters corresponding to each time point during the landing phase corresponding to the preset perception delay; the preset perception delay represents the time required for the pilot to perceive the aircraft state. With the condition of minimizing the mean square error between the pitch angle change value corresponding to each of the aforementioned times and the predicted optimal pitch angle change value, a grid search algorithm is used to traverse the parameter space to determine the target landing coefficient and the operation stability coefficient under the optimal perception delay in each of the aforementioned preset perception delays; the parameter space is the parameter space corresponding to the target landing coefficient and the operation stability coefficient. The preset coefficient value is determined as the landing phase coefficient.

[0012] According to the analysis method for pilot landing operations provided by the present invention, the step of determining the predicted optimal pitch angle change value corresponding to each moment of the landing phase based on the aircraft system dynamic model, the initial reward function, and the aircraft state parameters corresponding to each moment of the landing phase corresponding to the preset perception delay includes: Starting from the last moment of the landing phase, traverse all moments and substitute the aircraft state parameters corresponding to the previous moment into the aircraft system dynamic model to obtain the state transition equation corresponding to the moment. For each of the stated times, the state transition equation corresponding to that time is substituted into the initial reward function to obtain the performance index function corresponding to that time; based on the performance index function corresponding to that time and the partial derivative of the pitch angle change value, the optimal pitch angle change expression corresponding to that time is obtained. Solve for the optimal pitch angle change expression at the given time to obtain the feedback gain matrix at the given time; based on the feedback gain matrix at the given time, determine the optimal control law at the given time. The actual time corresponding to each of the aforementioned times is determined based on the preset perception delay. For each actual time, the predicted optimal pitch angle change value for that time is determined based on the product of the aircraft state parameters corresponding to that actual time and the optimal control law.

[0013] According to the analysis method for pilot landing operations provided by the present invention, the flight rapid record dataset is determined based on the following method: Acquire the aircraft status parameters recorded by the multiple fast access recorders, and the pilot operation status corresponding to each of the aircraft status parameters; The aircraft state parameters in which the number of operations corresponding to multiple pilots is greater than the threshold are defined as mixed operation aircraft state parameters. The flight rapid record dataset is determined based on the remaining aircraft state parameters, excluding all of the hybrid operation aircraft state parameters.

[0014] The present invention also provides an analysis device for pilot landing operations, comprising the following modules: The landing operation analysis module is used to input the target flight data corresponding to the target aircraft during the landing phase into the target Markov decision process, and obtain the pilot landing operation analysis results corresponding to the target aircraft output by the target Markov decision process; the state of the target Markov decision process is the aircraft state parameters; the action of the target Markov decision process is the change value of the aircraft pitch angle; the reward function of the target Markov decision process is determined based on the flight fast record dataset and the descent rate, expected descent rate, and pitch angle change value of the aircraft during the landing phase; the transition probability of the target Markov decision process is determined based on the flight fast record dataset; the flight fast record dataset represents the aircraft state parameters recorded by the fast access recorder; the data types included in the target flight data correspond to the data types included in the flight fast record dataset.

[0015] 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 analysis method for pilot landing operations as described above.

[0016] 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 analysis method for pilot landing operations as described above.

[0017] The present invention also provides a computer program product, including a computer program that, when executed by a processor, implements the analysis method for pilot landing operations as described above.

[0018] The present invention provides a method, apparatus, and electronic device for analyzing pilot landing operations. By inputting target flight data corresponding to the landing phase of the target aircraft into a target Markov decision process, the present invention obtains the pilot landing operation analysis results corresponding to the target aircraft output by the target Markov decision process. The state of the target Markov decision process is the aircraft's state parameters; the action of the target Markov decision process is the aircraft's pitch angle change value; the reward function of the target Markov decision process is determined based on the flight rapid record dataset and the aircraft's descent rate, expected descent rate, and pitch angle change value during the landing phase; the transition probability of the target Markov decision process is determined based on the flight rapid record dataset; the flight rapid record dataset represents the aircraft state parameters recorded by the rapid access recorder. The technical solution of this invention constructs a Markov decision process using the aircraft's descent rate, expected descent rate, pitch angle change value, and aircraft state parameters recorded by the rapid access recorder, and then substitutes these into the target flight data to obtain the pilot landing operation analysis results. Since the analysis of pilot operations in this application uses aircraft data, it is more accurate and feasible than the existing method of obtaining pilot nerve and muscle parameters. In addition, the analysis results of pilot landing operations based on aircraft data are sufficient to reflect the pilot's operational mindset. Attached Figure Description

[0019] 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.

[0020] Figure 1 This is a flowchart illustrating the analysis method for pilot landing operations provided by the present invention.

[0021] Figure 2 This is a flowchart illustrating the construction of the target Markov decision process provided by the present invention.

[0022] Figure 3 This is a schematic diagram of the structure of the analysis device for pilot landing operations provided by the present invention.

[0023] Figure 4 This is a schematic diagram of the structure of the electronic device provided by the present invention. Detailed Implementation

[0024] 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.

[0025] To address the aforementioned problems in the prior art, this invention provides an analysis method for pilot landing operations. Figure 1 This is one of the flowcharts illustrating the pilot landing operation analysis method provided by the present invention, such as... Figure 1 As shown, the method includes the following step 110.

[0026] Step 110: Input the target flight data corresponding to the landing phase of the target aircraft into the target Markov decision process to obtain the pilot landing operation analysis results corresponding to the target aircraft output by the target Markov decision process; the state of the target Markov decision process is the aircraft state parameters; the action of the target Markov decision process is the pitch angle change value of the aircraft; the reward function of the target Markov decision process is determined based on the flight fast record dataset and the descent rate, expected descent rate and pitch angle change value of the aircraft during the landing phase; the transition probability of the target Markov decision process is determined based on the flight fast record dataset; the flight fast record dataset represents the aircraft state parameters recorded by the fast access recorder; the data types included in the target flight data correspond to the data types included in the flight fast record dataset.

[0027] Specifically, a Markov decision process (MDF) is a mathematical framework for modeling decision problems in stochastic environments. It comprises five factors: state, action, transition probability, reward function, and discount factor. It is suitable for modeling discrete dynamic systems with Markov properties, meaning that future states and rewards depend only on current states and actions, and the dynamic system can be discretized. Therefore, the technical solution of this invention can pre-establish a target Markov decision process. The pilot's landing phase operation, which this patent focuses on, possesses Markov properties (i.e., given the current state, the state at the next moment is independent of the states before the current moment). In this target Markov decision process, the state refers to the aircraft's state parameters, with the rate of descent being the most critical; the action is the change in the aircraft's pitch angle; the reward function is determined based on a rapid flight record dataset and the aircraft's rate of descent, expected rate of descent, and pitch angle change during the landing phase; and the transition probability is determined based on the rapid flight record dataset. During the landing phase, the pilot's task is to control the aircraft's rate of descent to ensure a smooth descent to ground, avoiding risks such as a hard landing. Since the core of the aircraft's vertical motion under pilot control is the rate of descent, this invention selects aircraft state parameters including the rate of descent as the state. The pilot's control input is transmitted to the aircraft's ailerons, flaps, elevators, and rudder via mechanical and electrical systems, thereby changing the aircraft's attitude. Because the relationship between the pilot's control input and the basic control components (i.e., ailerons, flaps, elevators, and rudder) is complex and highly random, this invention selects intermediate control parameters (pitch angle changes) as the action. During the landing phase, the pilot mainly needs to consider "landing at the desired rate of descent" and "smooth control operation." Based on this, a reward function can be determined using the flight quick access record dataset and the aircraft's rate of descent, desired rate of descent, and pitch angle changes during the landing phase. The flight quick access record dataset consists of aircraft state parameters recorded by the aircraft's Quick Access Recorder (QAR). The QAR is an airborne flight data recorder used to record various parameters during flight, including crew operations, aircraft status, engine operation, and fuel consumption. A flight rapid recording dataset can include multiple QAR data points, such as thousands of pre-collected QAR data points. It should be noted that the flight rapid recording dataset includes QAR data corresponding to the aircraft's landing phase. The data type (aircraft state parameters) should at least include the aircraft's rate of descent change, pitch angle, airspeed, rate of descent, pitch angle change, headwind and tailwind conditions, wind shear, and two moments of the aircraft in the non-vertical direction. Furthermore, since QAR data is typically collected every second, the pitch angle change can be any pitch angle change within one second.

[0028] After obtaining the target Markov decision process, if it is necessary to analyze the landing operations of the pilot operating the target aircraft, the target flight data corresponding to the landing phase of the target aircraft can be input into the target Markov decision process. Then, the target Markov decision process based on the input target flight data can be solved to obtain the landing operation analysis results of the pilot corresponding to the target aircraft, output by the target Markov decision process. It should be noted that the aircraft landing process can generally be divided into four sub-phases: altitude hold, glide, pull-up, and taxiing. Taking the Airbus A320 series aircraft as an example, the glide phase typically begins when the aircraft reaches 1500 feet above the ground, and the pull-up phase typically begins when the aircraft reaches 50 feet above the ground. Although most commercial aircraft are equipped with automatic instrument landing systems and autopilot systems, conventional visual landing operations below 200 feet still require manual operation by the pilot. Therefore, this invention can define the phase from 200 feet above the ground to touchdown as the landing phase. The target flight data corresponding to the landing phase is also QAR data. Therefore, the target flight data includes at least the aircraft's rate of descent change, pitch angle, airspeed, rate of descent, pitch angle change, headwind and tailwind conditions, wind shear, and two moments of the aircraft in the non-vertical direction. The above-mentioned pilot landing operation analysis results can reflect the pilot's personal characteristics, which may include "landing at the desired rate of descent," "smooth control operation," and "perception delay." The pilot's perception delay refers to the time required for the pilot to perceive the aircraft's attitude, speed, and other states, typically ranging from 50 milliseconds to 400 milliseconds.

[0029] In one embodiment, the rapid flight recording dataset is determined based on the following method: Acquire the aircraft status parameters recorded by the multiple fast access recorders, and the pilot operation status corresponding to each of the aircraft status parameters; The aircraft state parameters in which the number of operations corresponding to multiple pilots is greater than the threshold are defined as mixed operation aircraft state parameters. The flight rapid record dataset is determined based on the remaining aircraft state parameters, excluding all of the hybrid operation aircraft state parameters.

[0030] Specifically, it is possible to first acquire aircraft status parameters recorded by multiple fast access recorders, or to acquire multiple aircraft status parameters recorded by a single fast access recorder. The acquired multiple aircraft status parameters can be aircraft status parameters from different time periods. At the same time, it is also possible to acquire the pilot operation information corresponding to each aircraft status parameter. The pilot operation information refers to the number of operations performed by each of the multiple pilots (usually the captain and first officer).

[0031] Furthermore, aircraft state parameters whose corresponding pilot operations involve multiple pilots performing operations exceeding a threshold can be identified as mixed-operation aircraft state parameters. This threshold can be set as needed, for example, five times. Since mixed-operation aircraft state parameters indicate that multiple pilots performed operations on the aircraft multiple times, they are unlikely to reflect individual pilot characteristics and therefore need to be removed. Consequently, the remaining aircraft state parameters, excluding all mixed-operation aircraft state parameters, can be defined as the flight rapid recording dataset.

[0032] For example, suppose multiple aircraft state parameters include aircraft state parameter A, aircraft state parameter B, and aircraft state parameter C, with a threshold of five operations. If the pilot operation behavior corresponding to aircraft state parameter A indicates that both Captain A and First Officer B performed six operations on the aircraft during the landing phase, then aircraft state parameter A is a mixed operation aircraft state parameter and needs to be discarded. However, based on the pilot operation behavior, aircraft state parameters B and C are not determined to be mixed aircraft state parameters. Therefore, the final determined flight rapid record dataset includes aircraft state parameters B and C.

[0033] In the above embodiments, by judging the pilot's operation, the mixed operation aircraft state parameters in the aircraft state parameters are eliminated, ensuring that the flight rapid recording dataset can better represent the individual characteristics of a single pilot, thereby making the target Markov decision process constructed based on the flight rapid recording dataset more accurate in analyzing the pilot's landing operation.

[0034] The pilot landing operation analysis method provided by this invention inputs the target flight data corresponding to the landing phase of the target aircraft into a target Markov decision process to obtain the pilot landing operation analysis results corresponding to the target aircraft output by the target Markov decision process. The state of the target Markov decision process is the aircraft's state parameters; the action of the target Markov decision process is the aircraft's pitch angle change value; the reward function of the target Markov decision process is determined based on the flight fast record dataset and the aircraft's descent rate, expected descent rate, and pitch angle change value during the landing phase; the transition probability of the target Markov decision process is determined based on the flight fast record dataset; the flight fast record dataset represents the aircraft state parameters recorded by the fast access recorder. The technical solution of this invention constructs a Markov decision process using the aircraft's descent rate, expected descent rate, pitch angle change value, and aircraft state parameters recorded by the fast access recorder, and then substitutes these into the target flight data to obtain the pilot landing operation analysis results. Since the analysis of pilot operations in this application uses aircraft data, it is more accurate and feasible than the existing method of obtaining pilot nerve and muscle parameters. In addition, the analysis results of pilot landing operations based on aircraft data are sufficient to reflect the pilot's operational mindset.

[0035] In one embodiment, Figure 2 This is a flowchart illustrating the construction of a target Markov decision process provided by the present invention, as shown below. Figure 2 As shown, the target Markov decision process is constructed based on the following steps 210 to 230: Step 210: Construct an aircraft system dynamics model based on the flight rapid record dataset; the aircraft system dynamics model characterizes the transition probabilities of the target Markov decision process.

[0036] Specifically, an aircraft system dynamics model can be constructed based on a rapid flight record dataset. It should be noted that this aircraft system dynamics model is a vertical system dynamics model of the aircraft during the landing phase. This aircraft system dynamics model can essentially be a linear function, and it can be used to characterize the transition probabilities of the target Markov decision process.

[0037] In one embodiment, constructing the aircraft system dynamics model based on the rapid flight record dataset includes: The change in the aircraft's rate of descent is determined as the dependent variable, and the aircraft's pitch angle, airspeed, rate of descent, change in pitch angle, headwind and tailwind conditions, wind shear, and two moments of the aircraft in the non-vertical direction are determined as independent variables. Based on the dependent variable and all the independent variables, an initial linear function is constructed. Substitute the flight rapid recording dataset into the initial linear function, and use the least squares method to determine the dependent variable coefficient and the independent variable coefficient corresponding to each independent variable. Substituting the dependent variable coefficients and the independent variable coefficients into the initial linear function yields the aircraft system dynamic model.

[0038] Specifically, the change in the aircraft's rate of descent can be determined as the dependent variable, while the aircraft's pitch angle, airspeed, rate of descent, change in pitch angle, headwind and tailwind conditions, wind shear, and two moments in the non-vertical direction can be determined as independent variables. Based on the dependent variable and all independent variables, an initial linear function is constructed. It is easy to understand that this initial linear function is not a complete linear function model; it still lacks the coefficients corresponding to the dependent variable and each independent variable.

[0039] Furthermore, the flight rapid recording dataset can be substituted into the initial linear function, and the least squares method can be used to determine the dependent variable coefficient and the independent variable coefficients corresponding to each independent variable. The least squares method is an optimization method that determines model parameters by minimizing the sum of squared errors. After determining the model structure (i.e., the initial linear function), the least squares method can make the model's prediction results as close as possible to the actual data, learning suitable model parameters from the data. This invention uses the least squares method to estimate and determine the model coefficients corresponding to the initial linear function. The method of determining the coefficients in reverse using the least squares method after determining the initial linear function is essentially a process of inverse reinforcement learning. Inverse reinforcement learning is the opposite of reinforcement learning; it aims to learn the objective function from behavioral data, rather than pre-determining the objective function based on expert behavior.

[0040] For example, the flight rapid recording dataset includes 3000 QAR data points. Substituting each of the 3000 QAR data points into the initial linear function, since the QAR data includes the aircraft's rate of descent change, pitch angle, airspeed, rate of descent, pitch angle change, headwind and tailwind conditions, wind shear, and two moments of the aircraft in the non-vertical direction, the least squares method can be used to obtain the dependent variable coefficient and the independent variable coefficients corresponding to each independent variable based on the results of substituting each of the 3000 QAR data points into the initial linear function.

[0041] Furthermore, by substituting the dependent variable coefficients and independent variable coefficients obtained using the least squares method into the initial linear function, a complete linear function, i.e., the dynamic model of the aircraft system, can be obtained.

[0042] In addition, during the modeling process of the aircraft system dynamics model, for the convenience of calculation, the corresponding headwind and tailwind conditions and wind shear of the aircraft can be fixed, and the ground speed can be approximated as the airspeed.

[0043] In the above embodiments, the dependent and independent variables of the aircraft system dynamics model are first determined, and then the coefficients of the dependent variable and the coefficients of the independent variables are determined by the least squares method. This allows for a more reasonable determination of the aircraft system dynamics model in the vertical direction during the landing phase, thereby providing transition probabilities for the target Markov decision process and laying the foundation for solving the reward function.

[0044] Step 220: Determine multiple quadratic terms of a function based on the descent rate, expected descent rate, and pitch angle change values ​​at each moment during the landing phase of the aircraft. Construct the reward function of the target Markov decision process based on all the quadratic terms of the function and the aircraft system dynamics model.

[0045] Specifically, during the landing phase, pilots need to employ appropriate stick position operations to achieve three objectives: landing at the desired rate of descent, a smooth glide path, and smooth control operations. To avoid risks such as hard landings, landing at the desired rate of descent is the primary objective of the landing phase. Simultaneously, a smooth glide path (the smaller the difference between the rate of descent and the desired rate of descent at each moment during the landing phase) and smooth control operations (smaller pitch angle changes indicate smoother control) are also crucial. Frequent adjustments not only affect the flight experience but may also increase the risks during the landing phase.

[0046] Therefore, this invention determines multiple quadratic terms of a function based on the descent rate, expected descent rate, and pitch angle change values ​​at various moments during the landing phase. Each quadratic term corresponds to one of three operational objectives (landing at the expected descent rate, smooth glide path, and smooth control operation). After determining the multiple quadratic terms, a reward function for the objective Markov decision process can be constructed based on all the quadratic terms and the aircraft system dynamics model. Since the data acquisition frequency of QAR data is typically 1 second, the time interval between different moments during the landing phase can be 1 second.

[0047] In one embodiment, the quadratic term of the function includes a target landing quadratic term, a landing phase quadratic term, and an operational smoothness quadratic term; The determination of multiple quadratic terms of a function based on the descent rate, expected descent rate, and pitch angle change values ​​at various moments during the landing phase includes: The difference between the descent rate at the last moment of the landing phase and the expected descent rate is defined as the target landing quadratic term; The quadratic term for the landing phase is determined based on the difference between the descent rate and the expected descent rate at each time point other than the final time point. The operational stability quadratic term is determined based on the pitch angle changes at various times during the landing phase of the aircraft.

[0048] Specifically, the quadratic term of the function can include a target landing quadratic term, a landing phase quadratic term, and an operational stability quadratic term. It is easy to understand that the operational objective corresponding to the target landing quadratic term is "landing at the expected descent rate," the operational objective corresponding to the landing phase quadratic term is "smooth glide path," and the operational objective corresponding to the operational stability quadratic term is "smooth control operation."

[0049] Furthermore, the difference between the descent rate at the final moment of the landing phase and the expected descent rate can be determined, and this difference can be defined as the target landing quadratic term. If the difference between the descent rate at the final moment and the expected descent rate is too large, it may cause the aircraft to make a hard landing. Alternatively, the landing phase quadratic term can be determined based on the difference between the descent rate and the expected descent rate at each moment other than the final moment. It should be noted that the landing phase quadratic term can be the sum of the differences between the descent rate and the expected descent rate at each moment other than the final moment; this embodiment of the invention does not impose specific limitations here. Furthermore, the operational stability quadratic term can be determined based on the pitch angle changes at each moment of the landing phase. It should be noted that the operational stability quadratic term can be the sum of the pitch angle changes at each moment; this embodiment of the invention does not impose specific limitations here.

[0050] In the above embodiments, based on the descent rate, expected descent rate, and pitch angle change values ​​at various moments during the landing phase, a target landing quadratic term is determined, with "landing at the expected descent rate" as the operational objective; a landing phase quadratic term is determined, with "smooth glide path" as the operational objective; and an operational smoothness quadratic term is determined, with "smooth control operation" as the operational objective. This ensures that each function quadratic term reflects the pilot's operational objectives, thereby enabling a more accurate evaluation and analysis of the pilot's actions.

[0051] In one embodiment, constructing the reward function for the target Markov decision process based on all the quadratic terms of the function and the aircraft system dynamics model includes: Construct the initial reward function for the objective Markov decision process based on all the quadratic terms of the function; Based on the aircraft system dynamic model and the initial reward function, determine the coefficients of the quadratic terms corresponding to the quadratic terms of each function in the initial reward function; Substituting the coefficients of the quadratic terms corresponding to the quadratic terms of each of the aforementioned functions into the initial reward function yields the reward function of the target Markov decision process.

[0052] Specifically, after determining the quadratic terms of each function, an initial reward function for the objective Markov decision process can be constructed based on all the quadratic terms. It's easy to understand that this initial reward function only contains the quadratic terms of the functions, but lacks the coefficients of each quadratic term. Therefore, the coefficients of each quadratic term in the initial reward function can be determined based on the aircraft system dynamics model and the initial reward function. After determining the coefficients of each quadratic term, these coefficients can be substituted into the initial reward function to obtain the complete reward function for the objective Markov decision process.

[0053] In the above embodiments, the method of first solving the reward function framework (i.e., the initial reward function) and then solving the quadratic coefficients of each function's quadratic terms also reflects the idea of ​​inverse reinforcement learning. This embodiment's method, employing the inverse reinforcement learning approach, can learn in complex and dynamic environments without needing to infer the reward function from expert behavior as in traditional reinforcement learning, thus reducing the workload of manual design.

[0054] In one embodiment, the quadratic term coefficients include the target landing coefficient corresponding to the target landing quadratic term, the landing phase coefficient corresponding to the landing phase quadratic term, and the operational stability coefficient corresponding to the operational stability quadratic term; The determination of the quadratic coefficients corresponding to each quadratic term in the initial reward function based on the aircraft system dynamic model and the initial reward function includes: For each preset perception delay, the predicted optimal pitch angle change value corresponding to each time point is determined based on the aircraft system dynamic model, the initial reward function, and the aircraft state parameters corresponding to each time point during the landing phase corresponding to the preset perception delay; the preset perception delay represents the time required for the pilot to perceive the aircraft state. With the condition of minimizing the mean square error between the pitch angle change value corresponding to each of the aforementioned times and the predicted optimal pitch angle change value, a grid search algorithm is used to traverse the parameter space to determine the target landing coefficient and the operation stability coefficient under the optimal perception delay in each of the aforementioned preset perception delays; the parameter space is the parameter space corresponding to the target landing coefficient and the operation stability coefficient. The preset coefficient value is determined as the landing phase coefficient.

[0055] Specifically, this invention can pre-set multiple preset perception delays, each corresponding to pilots with different perception capabilities. In other words, the preset perception delay represents the time required for a pilot to perceive the aircraft's state. The aircraft state parameters corresponding to each moment during the landing phase, based on the preset perception delay, are the aircraft state parameters prior to the corresponding moment by the duration corresponding to the preset perception delay. For example, if one of the multiple preset perception delays is 200 milliseconds, then the aircraft state parameters corresponding to each moment of that preset perception delay are all 200 milliseconds earlier. For instance, at the 8th second, the aircraft state parameters corresponding to this preset perception delay are the same as those at the 7.8th second.

[0056] Furthermore, based on the aircraft system dynamic model, the initial reward function, and the aircraft state parameters corresponding to each moment of the landing phase at the preset perception delay, multiple calculations can be performed to determine the predicted optimal pitch angle change value corresponding to the preset perception delay.

[0057] After determining the predicted optimal pitch angle change value corresponding to each preset sensing delay, the mean square error between the pitch angle change value and the predicted optimal pitch angle change value at each time point can be calculated. Then, with minimizing the mean square error between the pitch angle change value and the predicted optimal pitch angle change value at each time point as a condition, a grid search algorithm is used to traverse the parameter space (the parameter space corresponding to the target landing coefficient and the operational stability coefficient) to determine the target landing coefficient and operational stability coefficient under the optimal sensing delay among the preset sensing delays. Here, the minimum value can also be the cumulative value of the mean square error at each time point; the optimal sensing delay is the preset sensing delay with the smallest mean square error between the corresponding pitch angle change value and the predicted optimal pitch angle change value, that is, the preset sensing delay where the predicted pitch angle change value is closest to the actual pitch angle change value. The grid search algorithm described above is essentially an exhaustive method. It defines a discrete grid in the parameter space based on requirements or prior knowledge, with each point corresponding to a set of parameter combinations. The optimal parameter combination (i.e., the optimal target landing coefficient and operational stability coefficient) is selected by evaluating the model performance under each parameter combination.

[0058] In addition, the landing phase coefficient can be a predetermined preset value, for example, the landing phase coefficient can be set to 1.

[0059] For example, preset sensing delays include 100 milliseconds, 200 milliseconds, and 500 milliseconds. For each of these delays, the mean square error (MSE) between the pitch angle change value and the predicted optimal pitch angle change value can be calculated. The preset sensing delay with the smallest MSE is then determined as the optimal sensing delay. A grid search algorithm is then used to traverse the parameter space to determine the target landing coefficient and operational stability coefficient under the optimal sensing delay. It should be noted that a larger target landing coefficient indicates a higher importance for "landing at the expected descent rate," and a larger operational stability coefficient indicates a higher importance for "smooth control operation." It should also be noted that since the QAR data interval is 1 second, if the preset sensing delay is less than 1 second, QAR data may be missing. In this case, the missing data can be obtained from the known data through interpolation.

[0060] In the above embodiments, by minimizing the mean square error between the actual pitch angle change and the predicted optimal pitch angle change, a grid search algorithm is used to solve for the target landing coefficient and the operational stability coefficient. The grid search algorithm ensures that a globally optimal solution is found within the parameter space. This yields a complete reward function, enabling a thorough analysis and evaluation of the pilot's individual characteristics.

[0061] In one embodiment, determining the predicted optimal pitch angle change value for each of the following moments based on the aircraft system dynamics model, the initial reward function, and the aircraft state parameters corresponding to each moment of the landing phase corresponding to the preset perception delay includes: Starting from the last moment of the landing phase, traverse all moments and substitute the aircraft state parameters corresponding to the previous moment into the aircraft system dynamic model to obtain the state transition equation corresponding to the moment. For each of the stated times, the state transition equation corresponding to that time is substituted into the initial reward function to obtain the performance index function corresponding to that time; based on the performance index function corresponding to that time and the partial derivative of the pitch angle change value, the optimal pitch angle change expression corresponding to that time is obtained. Solve for the optimal pitch angle change expression at the given time to obtain the feedback gain matrix at the given time; based on the feedback gain matrix at the given time, determine the optimal control law at the given time. The actual time corresponding to each of the aforementioned times is determined based on the preset perception delay. For each actual time, the predicted optimal pitch angle change value for that time is determined based on the product of the aircraft state parameters corresponding to that actual time and the optimal control law.

[0062] Specifically, since pilot landing operation analysis can be viewed as a discrete dynamic system, and when the reward function is a quadratic reward function, the prediction of pitch angle change value has a closed-form solution, the optimal pitch angle change value can be determined by the following steps: First, starting from the last moment of the landing phase, traverse all possible moments and substitute the aircraft state parameters from the previous moment into the aircraft system dynamics model to obtain the state transition equation for that moment. This state transition equation describes the evolution of the aircraft's system state. The aircraft system dynamics model can be expressed as follows: in, express Aircraft state parameters at any given time. express Aircraft state parameters at any given time. express The change in pitch angle at time t, and All are preset coefficients.

[0063] The above process will then begin. Represented as , Indicates the final moment. That is to say The state transition equation at time t is Then it is The state transition equation corresponding to time t.

[0064] Furthermore, for each time step, the state transition equation corresponding to that time step can be substituted into the initial reward function to obtain the performance index function for that time step. Then, based on the performance index function and the partial derivatives of the pitch angle change value for that time step, the optimal pitch angle change expression for that time step can be obtained. This optimal pitch angle change expression can then be solved. In the process of solving the optimal pitch angle change expression, it can be transformed into a Riccati equation, thus allowing the calculation of the feedback gain matrix for that time step. Furthermore, based on the feedback gain matrix for that time step, the optimal control law for that time step can be determined. The Riccati equation is a nonlinear differential equation that holds an important position in modern control theory.

[0065] Alternatively, the Riccati equations can be used to approximate all optimal pitch angle change expressions to obtain the optimal control law corresponding to all optimal pitch angle change expressions. In other words, in this case, all optimal pitch angle change expressions correspond to one optimal control law.

[0066] Based on a preset sensing delay, the actual time corresponding to each moment is determined. Furthermore, for each actual time, the predicted optimal pitch angle change value for that moment can be determined by multiplying the aircraft state parameters corresponding to that moment with the optimal control law. For example, with a preset sensing delay of 500 milliseconds, the actual time corresponding to the 1st second is 0.5 seconds, and the actual time corresponding to the 2nd second is 1.5 seconds. Therefore, the product of the aircraft state parameters corresponding to the 1.5th second and the optimal control law corresponding to the 1.5th second can be determined, and this product can be used as the predicted optimal pitch angle change value for the 2nd second.

[0067] In the above embodiment, the pitch angle change value is predicted and expressed in reverse from the last moment. Furthermore, the optimal pitch angle change expression can accurately obtain the predicted optimal pitch angle change value corresponding to each preset sensing delay, so that the mean square error of the predicted value and the actual value can be used to construct the coefficient optimization condition.

[0068] Step 230: Determine the target Markov decision process based on the state, the action, the transition probability, the reward function, and the preset discount factor.

[0069] Specifically, the target Markov decision process can be determined based on the state, action, transition probability, reward function, and preset discount factor. The preset discount factor can be set as needed, for example, it can be set to 1.

[0070] In steps 210 to 230 above, a target Markov decision process is constructed. The target Markov decision process, through elements such as state, action, transition probability and reward function, combined with Markov properties and Bellman equations, can help find the optimal strategy, that is, obtain the pilot landing operation analysis results.

[0071] The analysis device for pilot landing operations provided by the present invention will be described below. The analysis device for pilot landing operations described below and the analysis method for pilot landing operations described above can be referred to in correspondence with each other.

[0072] Figure 3 This is a schematic diagram of the structure of the analysis device for pilot landing operations provided by the present invention, as shown below. Figure 3 As shown, the pilot landing operation analysis device 300 includes the following modules: The landing operation analysis module 310 is used to input the target flight data corresponding to the target aircraft during the landing phase into the target Markov decision process to obtain the pilot landing operation analysis results corresponding to the target aircraft output by the target Markov decision process; the state of the target Markov decision process is the aircraft state parameters; the action of the target Markov decision process is the pitch angle change value of the aircraft; the reward function of the target Markov decision process is determined based on the flight fast record dataset and the descent rate, expected descent rate, and pitch angle change value of the aircraft during the landing phase; the transition probability of the target Markov decision process is determined based on the flight fast record dataset; the flight fast record dataset represents the aircraft state parameters recorded by the fast access recorder; the data type included in the target flight data corresponds to the data type included in the flight fast record dataset.

[0073] In one embodiment, the analysis apparatus for pilot landing operations further includes a construction module specifically used for: An aircraft system dynamics model is constructed based on the aforementioned rapid flight record dataset; the aircraft system dynamics model characterizes the transition probabilities of the target Markov decision process; Based on the descent rate, expected descent rate, and pitch angle change values ​​at each moment during the landing phase, multiple quadratic terms of the function are determined. Based on all the quadratic terms of the function and the aircraft system dynamic model, the reward function of the target Markov decision process is constructed. The target Markov decision process is determined based on the state, the action, the transition probability, the reward function, and the preset discount factor.

[0074] In one embodiment, the building module is further configured to: The change in the aircraft's rate of descent is determined as the dependent variable, and the aircraft's pitch angle, airspeed, rate of descent, change in pitch angle, headwind and tailwind conditions, wind shear, and two moments of the aircraft in the non-vertical direction are determined as independent variables. Based on the dependent variable and all the independent variables, an initial linear function is constructed. Substitute the flight rapid recording dataset into the initial linear function, and use the least squares method to determine the dependent variable coefficient and the independent variable coefficient corresponding to each independent variable. Substituting the dependent variable coefficients and the independent variable coefficients into the initial linear function yields the aircraft system dynamic model.

[0075] In one embodiment, the building module is further configured to: Construct the initial reward function for the objective Markov decision process based on all the quadratic terms of the function; Based on the aircraft system dynamic model and the initial reward function, determine the coefficients of the quadratic terms corresponding to the quadratic terms of each function in the initial reward function; Substituting the coefficients of the quadratic terms corresponding to the quadratic terms of each of the aforementioned functions into the initial reward function yields the reward function of the target Markov decision process.

[0076] In one embodiment, the quadratic term of the function includes a target landing quadratic term, a landing phase quadratic term, and an operational stabilization quadratic term; the building module is further specifically used for: The difference between the descent rate at the last moment of the landing phase and the expected descent rate is defined as the target landing quadratic term; The quadratic term for the landing phase is determined based on the difference between the descent rate and the expected descent rate at each time point other than the final time point. The operational stability quadratic term is determined based on the pitch angle changes at various times during the landing phase of the aircraft.

[0077] In one embodiment, the quadratic term coefficients include the target landing coefficient corresponding to the target landing quadratic term, the landing phase coefficient corresponding to the landing phase quadratic term, and the operational stability coefficient corresponding to the operational stability quadratic term; the construction module is further specifically used for: For each preset perception delay, the predicted optimal pitch angle change value corresponding to each time point is determined based on the aircraft system dynamic model, the initial reward function, and the aircraft state parameters corresponding to each time point during the landing phase corresponding to the preset perception delay; the preset perception delay represents the time required for the pilot to perceive the aircraft state. With the condition of minimizing the mean square error between the pitch angle change value corresponding to each of the aforementioned times and the predicted optimal pitch angle change value, a grid search algorithm is used to traverse the parameter space to determine the target landing coefficient and the operation stability coefficient under the optimal perception delay in each of the aforementioned preset perception delays; the parameter space is the parameter space corresponding to the target landing coefficient and the operation stability coefficient. The preset coefficient value is determined as the landing phase coefficient.

[0078] In one embodiment, the building module is further configured to: Starting from the last moment of the landing phase, traverse all moments and substitute the aircraft state parameters corresponding to the previous moment into the aircraft system dynamic model to obtain the state transition equation corresponding to the moment. For each of the stated times, the state transition equation corresponding to that time is substituted into the initial reward function to obtain the performance index function corresponding to that time; based on the performance index function corresponding to that time and the partial derivative of the pitch angle change value, the optimal pitch angle change expression corresponding to that time is obtained. Solve for the optimal pitch angle change expression at the given time to obtain the feedback gain matrix at the given time; based on the feedback gain matrix at the given time, determine the optimal control law at the given time. The actual time corresponding to each of the aforementioned times is determined based on the preset perception delay. For each actual time, the predicted optimal pitch angle change value for that time is determined based on the product of the aircraft state parameters corresponding to that actual time and the optimal control law.

[0079] In one embodiment, the analysis apparatus for the pilot's landing operation further includes a dataset determination module, which is specifically used for: Acquire the aircraft status parameters recorded by the multiple fast access recorders, and the pilot operation status corresponding to each of the aircraft status parameters; The aircraft state parameters in which the number of operations corresponding to multiple pilots is greater than the threshold are defined as mixed operation aircraft state parameters. The flight rapid record dataset is determined based on the remaining aircraft state parameters, excluding all of the hybrid operation aircraft state parameters.

[0080] The pilot landing operation analysis device provided by this invention inputs target flight data corresponding to the landing phase of the target aircraft into a target Markov decision process to obtain the pilot landing operation analysis results corresponding to the target aircraft output by the target Markov decision process. The state of the target Markov decision process is the aircraft's state parameters; the action of the target Markov decision process is the aircraft's pitch angle change value; the reward function of the target Markov decision process is determined based on the flight fast record dataset and the aircraft's descent rate, expected descent rate, and pitch angle change value during the landing phase; the transition probability of the target Markov decision process is determined based on the flight fast record dataset; the flight fast record dataset represents the aircraft state parameters recorded by the fast access recorder. The technical solution of this invention constructs a Markov decision process using the aircraft's descent rate, expected descent rate, pitch angle change value, and aircraft state parameters recorded by the fast access recorder, and then substitutes these into the target flight data to obtain the pilot landing operation analysis results. Since the analysis of pilot operations in this application uses aircraft data, it is more accurate and feasible than the existing method of obtaining pilot nerve and muscle parameters. In addition, the analysis results of pilot landing operations based on aircraft data are sufficient to reflect the pilot's operational mindset.

[0081] Figure 4 An example is a schematic diagram of the physical structure of an electronic device, such as... Figure 4 As shown, the electronic device may include: a processor 410, a communications interface 420, a memory 430, and a communication bus 440, wherein the processor 410, the communications interface 420, and the memory 430 communicate with each other via the communication bus 440. The processor 410 can call logical instructions in the memory 430 to execute an analysis method for pilot landing operations, the method including: The target flight data corresponding to the landing phase of the target aircraft is input into the target Markov decision process to obtain the pilot landing operation analysis results corresponding to the target aircraft output by the target Markov decision process; the state of the target Markov decision process is the aircraft state parameters; the action of the target Markov decision process is the change value of the aircraft pitch angle; the reward function of the target Markov decision process is determined based on the flight fast record dataset and the descent rate, expected descent rate, and pitch angle change value of the aircraft during the landing phase; the transition probability of the target Markov decision process is determined based on the flight fast record dataset; the flight fast record dataset represents the aircraft state parameters recorded by the fast access recorder; the data types included in the target flight data correspond to the data types included in the flight fast record dataset.

[0082] Furthermore, the logical instructions in the aforementioned memory 430 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, 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.

[0083] 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 the pilot landing operation analysis method provided by the above methods, the method comprising: The target flight data corresponding to the landing phase of the target aircraft is input into the target Markov decision process to obtain the pilot landing operation analysis results corresponding to the target aircraft output by the target Markov decision process; the state of the target Markov decision process is the aircraft state parameters; the action of the target Markov decision process is the change value of the aircraft pitch angle; the reward function of the target Markov decision process is determined based on the flight fast record dataset and the descent rate, expected descent rate, and pitch angle change value of the aircraft during the landing phase; the transition probability of the target Markov decision process is determined based on the flight fast record dataset; the flight fast record dataset represents the aircraft state parameters recorded by the fast access recorder; the data types included in the target flight data correspond to the data types included in the flight fast record dataset.

[0084] 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, implements an analysis method for pilot landing operations provided by the methods described above, the method comprising: The target flight data corresponding to the landing phase of the target aircraft is input into the target Markov decision process to obtain the pilot landing operation analysis results corresponding to the target aircraft output by the target Markov decision process; the state of the target Markov decision process is the aircraft state parameters; the action of the target Markov decision process is the change value of the aircraft pitch angle; the reward function of the target Markov decision process is determined based on the flight fast record dataset and the descent rate, expected descent rate, and pitch angle change value of the aircraft during the landing phase; the transition probability of the target Markov decision process is determined based on the flight fast record dataset; the flight fast record dataset represents the aircraft state parameters recorded by the fast access recorder; the data types included in the target flight data correspond to the data types included in the flight fast record dataset.

[0085] 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.

[0086] 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.

[0087] 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. An analysis method for pilot landing operations, characterized in that, include: Input the target flight data corresponding to the landing phase of the target aircraft into the target Markov decision process to obtain the pilot landing operation analysis results corresponding to the target aircraft output by the target Markov decision process; The state of the target Markov decision process is the aircraft's state parameters; the action of the target Markov decision process is the change in the aircraft's pitch angle; the reward function of the target Markov decision process is determined based on the flight fast record dataset and the aircraft's descent rate, expected descent rate, and pitch angle change during the landing phase; the transition probability of the target Markov decision process is determined based on the flight fast record dataset; the flight fast record dataset represents the aircraft state parameters recorded by the fast access recorder; the data types included in the target flight data correspond to the data types included in the flight fast record dataset; The target Markov decision process is constructed based on the following method: An aircraft system dynamics model is constructed based on the aforementioned rapid flight record dataset; the aircraft system dynamics model characterizes the transition probabilities of the target Markov decision process; Based on the descent rate, expected descent rate, and pitch angle change values ​​at each moment during the landing phase, multiple quadratic terms of the function are determined. Based on all the quadratic terms of the function and the aircraft system dynamic model, the reward function of the target Markov decision process is constructed. The target Markov decision process is determined based on the state, the action, the transition probability, the reward function, and the preset discount factor. The reward function for constructing the target Markov decision process based on all the quadratic terms of the function and the aircraft system dynamics model includes: An initial reward function for the target Markov decision process is constructed based on all the quadratic terms of the functions; the coefficients of the quadratic terms corresponding to each of the quadratic terms in the initial reward function are determined based on the aircraft system dynamics model and the initial reward function; the coefficients of the quadratic terms corresponding to each of the quadratic terms are substituted into the initial reward function to obtain the reward function for the target Markov decision process. The quadratic term of the function includes a target landing quadratic term, a landing phase quadratic term, and an operational smoothness quadratic term; the determination of multiple quadratic terms based on the descent rate, expected descent rate, and pitch angle change values ​​at each moment during the landing phase includes: The difference between the descent rate at the last moment of the landing phase and the expected descent rate is determined as the target landing quadratic term; the landing phase quadratic term is determined based on the difference between the descent rate and the expected descent rate at each moment other than the last moment; the operational smoothness quadratic term is determined based on the pitch angle change value at each moment of the landing phase. The quadratic term coefficients include the target landing coefficient corresponding to the target landing quadratic term, the landing phase coefficient corresponding to the landing phase quadratic term, and the operational stability coefficient corresponding to the operational stability quadratic term; determining the quadratic term coefficients corresponding to each quadratic term in the initial reward function based on the aircraft system dynamic model and the initial reward function includes: For each preset perception delay, based on the aircraft system dynamic model, the initial reward function, and the aircraft state parameters corresponding to each moment of the landing phase at each preset perception delay, the predicted optimal pitch angle change value corresponding to each moment is determined; the preset perception delay represents the time required for the pilot to perceive the aircraft state; with the condition of minimizing the mean square error between the pitch angle change value corresponding to each moment and the predicted optimal pitch angle change value, a grid search algorithm is used to traverse the parameter space to determine the target landing coefficient and the operational stability coefficient under the optimal perception delay in each preset perception delay; the parameter space is the parameter space corresponding to the target landing coefficient and the operational stability coefficient; the preset coefficient value is determined as the landing phase coefficient.

2. The analysis method for pilot landing operations according to claim 1, characterized in that, The construction of the aircraft system dynamic model based on the flight rapid record dataset includes: The change in the aircraft's rate of descent is determined as the dependent variable, and the aircraft's pitch angle, airspeed, rate of descent, change in pitch angle, headwind and tailwind conditions, wind shear, and two moments of the aircraft in the non-vertical direction are determined as independent variables. Based on the dependent variable and all the independent variables, an initial linear function is constructed. Substitute the flight rapid recording dataset into the initial linear function, and use the least squares method to determine the dependent variable coefficient and the independent variable coefficient corresponding to each independent variable. Substituting the dependent variable coefficients and the independent variable coefficients into the initial linear function yields the aircraft system dynamic model.

3. The analysis method for pilot landing operations according to claim 1, characterized in that, The step of determining the predicted optimal pitch angle change value for each moment based on the aircraft system dynamic model, the initial reward function, and the aircraft state parameters corresponding to each moment of the landing phase corresponding to the preset perception delay includes: Starting from the last moment of the landing phase, the system sequentially traverses each of the aforementioned moments, substituting the aircraft state parameters corresponding to the previous moment into the aircraft system dynamic model to obtain the state transition equations corresponding to the aforementioned moments. For each of the stated times, the state transition equation corresponding to that time is substituted into the initial reward function to obtain the performance index function corresponding to that time; based on the performance index function corresponding to that time and the partial derivative of the pitch angle change value, the optimal pitch angle change expression corresponding to that time is obtained. Solve for the optimal pitch angle change expression at the given time to obtain the feedback gain matrix at the given time; based on the feedback gain matrix at the given time, determine the optimal control law at the given time. The actual time corresponding to each of the aforementioned times is determined based on the preset perception delay. For each actual time, the predicted optimal pitch angle change value for that time is determined based on the product of the aircraft state parameters corresponding to that actual time and the optimal control law.

4. The method for analyzing pilot landing operations according to any one of claims 1 to 3, characterized in that, The rapid flight log dataset was determined based on the following method: Acquire the aircraft status parameters recorded by the multiple fast access recorders, and the pilot operation status corresponding to each of the aircraft status parameters; The aircraft state parameters in which the number of operations corresponding to multiple pilots is greater than the threshold are defined as mixed operation aircraft state parameters. The flight rapid record dataset is determined based on the remaining aircraft state parameters, excluding all of the hybrid operation aircraft state parameters.

5. An analysis device for pilot landing operations, characterized in that, include: The landing operation analysis module is used to input the target flight data corresponding to the target aircraft during the landing phase into the target Markov decision process, and obtain the pilot landing operation analysis results corresponding to the target aircraft output by the target Markov decision process. The state of the target Markov decision process is the aircraft's state parameters; the action of the target Markov decision process is the change in the aircraft's pitch angle; the reward function of the target Markov decision process is determined based on the flight fast record dataset and the aircraft's descent rate, expected descent rate, and pitch angle change during the landing phase; the transition probability of the target Markov decision process is determined based on the flight fast record dataset; the flight fast record dataset represents the aircraft state parameters recorded by the fast access recorder; the data types included in the target flight data correspond to the data types included in the flight fast record dataset; The construction module is specifically used for: An aircraft system dynamics model is constructed based on the aforementioned rapid flight record dataset; the aircraft system dynamics model characterizes the transition probabilities of the target Markov decision process; Multiple quadratic terms of a function are determined based on the descent rate, expected descent rate, and pitch angle change values ​​at various moments during the landing phase of the aircraft. A reward function for the target Markov decision process is constructed based on all the quadratic terms of the function and the aircraft system dynamics model. The target Markov decision process is determined based on the state, the action, the transition probability, the reward function, and a preset discount factor. The building module is also specifically used for: An initial reward function for the target Markov decision process is constructed based on all the quadratic terms of the functions; the coefficients of the quadratic terms corresponding to each of the quadratic terms in the initial reward function are determined based on the aircraft system dynamics model and the initial reward function; the coefficients of the quadratic terms corresponding to each of the quadratic terms are substituted into the initial reward function to obtain the reward function for the target Markov decision process. The quadratic term of the function includes a target landing quadratic term, a landing phase quadratic term, and an operational stabilization quadratic term; the construction module is specifically used for: The difference between the descent rate at the last moment of the landing phase and the expected descent rate is determined as the target landing quadratic term; the landing phase quadratic term is determined based on the difference between the descent rate and the expected descent rate at each moment other than the last moment. The operational stability quadratic term is determined based on the pitch angle changes at various moments during the landing phase of the aircraft. The quadratic term coefficients include the target landing coefficient corresponding to the target landing quadratic term, the landing phase coefficient corresponding to the landing phase quadratic term, and the operational stability coefficient corresponding to the operational stability quadratic term; the construction module is specifically used for: For each preset perception delay, based on the aircraft system dynamics model, the initial reward function, and the aircraft state parameters corresponding to each moment during the landing phase, the predicted optimal pitch angle change value is determined for each of the preset perception delays. The preset perception delay represents the time required for the pilot to perceive the aircraft state. Using the condition of minimizing the mean square error between the pitch angle change value and the predicted optimal pitch angle change value at each of the preset perception delays, a grid search algorithm is used to traverse the parameter space to determine the target landing coefficient and the operational stability coefficient under the optimal perception delay among the preset perception delays. The parameter space is the parameter space corresponding to the target landing coefficient and the operational stability coefficient. The preset coefficient value is determined as the landing phase coefficient.

6. An electronic device comprising a memory, a processor, and a computer program stored in the memory and running on the processor, characterized in that, When the processor executes the computer program, it implements the analysis method for pilot landing operations as described in any one of claims 1 to 4.

Images