Aircraft approach landing process risk prediction method, system and program product

Abstract

The invention belongs to the technical field of flight safety, and particularly discloses an aircraft approach landing process risk prediction method and system and a program product, and the method comprises the steps: collecting the actual flight QAR data of an aircraft in the approach landing flight process of the aircraft, and starting from the physical motion essence of the aircraft; and selecting and extracting flight state parameters closely related to flight safety, inputting the flight state parameters into the trained and tested WOA-LightGBM model to carry out aircraft approach landing process risk prediction so as to obtain a corresponding risk prediction result, and finally determining the approach landing flight safety risk level of the aircraft. The method can achieve the efficient and real-time risk prediction of the approaching landing process of the aircraft, and improves the timeliness and precision of the risk prediction of the approaching landing process of the aircraft.

Description

Technical Field

[0001] The present invention belongs to the technical field of flight safety, and particularly relates to a method, a system and a program product for predicting risks during the approach and landing process of an aircraft. Background Art

[0002] The approach and landing phase is a critical phase of aircraft operation. To ensure the safe and stable landing of the aircraft, the pilot controls the deflection of each aircraft control surface and the engine power by operating the side stick (or control column), foot pedals and throttle lever, etc., to ensure that the aircraft speed, attitude, heading and engine thrust are always in a stable state. An unstable flight state means that there is an abnormality in the aircraft energy. If not controlled in time, it may cause an accidental release of energy, which may in turn lead to a flight accident. Therefore, it is crucial for flight safety to identify and analyze abnormal parameters of the approach and landing flight state so that the flight crew can intervene in flight state deviations in advance and eliminate the risks of flight accidents.

[0003] To reduce flight operation deviations and lower the risks of approach and landing, airlines have designed standard operating procedures (SOPs) with reference to civil aviation regulations, flight crew operation manuals (FCOM), etc., achieving the quantification and standardization of flight procedures. At the same time, flight quality monitoring technology is used to analyze data through QAR (Quick Access Recorder) to improve flight quality. Airlines use the above two means to standardize and guide the operation behaviors of flight personnel from the flight operation level, which has played an important role in ensuring flight safety, but little attention has been paid to the impact of changes in aircraft flight states on flight safety.

[0004] Some studies have adopted methods such as BP neural networks and hidden Markov models to analyze the flight state of an aircraft by constructing a flight attitude recognition and prediction model. Such flight state analysis based on dynamic modeling is a prerequisite for studying the essence of flight safety accidents, and the accuracy of its analysis is positively correlated with the reliability of data and models. With the continuous improvement of the quality and availability of flight data, at the same time, big data technology also provides the potential for transformation for the challenging problems faced by traditional analysis methods. Some studies have begun to use big data analysis technology to study the approach trajectory of an aircraft, landing parameters, and changes in flight state during the approach and landing process, and have obtained the energy envelope during the approach phase of the aircraft to explain the energy management during the landing process. Some studies have used the QAR data of B737 aircraft to determine that the indicated airspeed (IAS) and beacon deviation are important factors causing the instability of the aircraft state during the approach phase; some have developed an abnormal flight state recognition program using deep learning methods through analyzing QAR data to improve the accuracy of flight state classification. Such data-driven flight state analysis only analyzes individual important flight state parameters and aircraft energy through data characteristics, and there is little further research on the impact relationship between flight state and flight safety risks. From the above studies, domestic and foreign research has achieved a series of results in studying aircraft state from the perspectives of mechanism modeling and big data, but there is little research on analyzing the flight safety risks brought by the deviation of flight state from the essence of aircraft movement.

[0005] To sum up, unstable flight state is the key factor leading to flight accidents, and unsafe control behaviors of pilots, harsh operating environment conditions, etc. are important reasons for causing unstable flight state. At present, although there are many studies on flight state recognition, there is little analysis of the impact relationship between flight state and flight safety risks and the analysis of flight safety risks brought by the deviation of flight state from the essence of aircraft movement. In addition, the existing flight safety risk prediction models established by simulation means cannot fully simulate the real flight process, and their prediction results may be different from the actual situation. Therefore, there is an urgent need for an effective method that can accurately predict the flight safety risks during the approach and landing of an aircraft from the essence of aircraft movement. Summary of the Invention

[0006] The purpose of the present invention is to provide a method, system, and program product for predicting risks during the approach and landing process of an aircraft to solve the above problems existing in the prior art.

[0007] To achieve the above purpose, the present invention adopts the following technical solutions: In the first aspect, a method for predicting risks during the approach and landing process of an aircraft is provided, including: Collecting the actual flight QAR data of the aircraft during the approach and landing flight process of the aircraft; Extracting a set of flight state parameters from the actual flight QAR data; Input the flight state parameter set into the pre-set risk prediction model for the aircraft approach and landing process for model analysis, and obtain the risk prediction result for the aircraft approach and landing process. The risk prediction model for the aircraft approach and landing process uses the trained and tested WOA-LightGBM model; Determine the approach and landing flight safety risk level of the aircraft according to the risk prediction result of the aircraft approach and landing process, and output the approach and landing flight safety risk level of the aircraft.

[0008] In a possible design, the flight state parameter set includes flight speed, flight altitude, horizontal displacement, lateral track offset, pitch angle, roll angle, and yaw angle.

[0009] In a possible design, before inputting the flight state parameter set into the pre-set risk prediction model for the aircraft approach and landing process for model analysis, the method further includes: Construct a LightGBM model, and use the WOA algorithm to optimize and iterate the LightGBM model to obtain the WOA-LightGBM model; Obtain a training set and a test set, and use the training set to train the WOA-LightGBM model to obtain the trained WOA-LightGBM model. Use the test set to test the trained WOA-LightGBM model to obtain the tested WOA-LightGBM model, and use the trained and tested WOA-LightGBM model as the risk prediction model for the aircraft approach and landing process.

[0010] In a possible design, the obtaining of the training set and the test set includes: Collect a number of flight QAR data samples, and perform data preprocessing on all flight QAR data samples to obtain a number of preprocessed flight QAR data samples; Extract flight state parameter set samples from each preprocessed flight QAR data sample, and perform approach and landing flight safety risk label annotation on each flight state parameter set sample to obtain the corresponding labeled flight state parameter set samples; Use each labeled flight state parameter set sample to form a flight state parameter data set, and divide the flight state parameter data set into a training set and a test set according to a set ratio.

[0011] In a possible design, the performing of data preprocessing on all flight QAR data samples includes: Exclude the flight QAR data samples with missing data among all flight QAR data samples.

[0012] In a possible design, when training the WOA-LightGBM model using a training set, the method further includes: Taking the accuracy, precision, recall, and f 1 -score value of the model as the evaluation index of the WOA-LightGBM model, and iteratively training the WOA-LightGBM model using the training set until the evaluation index of the WOA-LightGBM model reaches the set convergence condition.

[0013] In a possible design, the risk prediction result of the aircraft approach and landing process is 0, 1, 2, or 3, and the approach and landing flight safety risk level is a risk-free level, a low-risk level, a medium-risk level, or a high-risk level. Among them, the risk prediction result of the aircraft approach and landing process corresponding to the risk-free level is 0, the risk prediction result of the aircraft approach and landing process corresponding to the low-risk level is 1, the risk prediction result of the aircraft approach and landing process corresponding to the medium-risk level is 2, and the risk prediction result of the aircraft approach and landing process corresponding to the high-risk level is 3.

[0014] In a second aspect, a risk prediction system for an aircraft approach and landing process is provided, including a data acquisition unit, a data extraction unit, a model prediction unit, and a risk determination unit, where: The data acquisition unit is configured to collect the actual flight QAR data of the aircraft during the aircraft approach and landing flight process; The data extraction unit is configured to extract a set of flight state parameters from the actual flight QAR data; The model prediction unit is configured to input the set of flight state parameters into a pre-set risk prediction model for the aircraft approach and landing process for model analysis to obtain the risk prediction result of the aircraft approach and landing process of the aircraft. The risk prediction model for the aircraft approach and landing process uses the trained and tested WOA-LightGBM model; The risk determination unit is configured to determine the approach and landing flight safety risk level of the aircraft according to the risk prediction result of the aircraft approach and landing process, and output the approach and landing flight safety risk level of the aircraft.

[0015] In a third aspect, a risk prediction system for an aircraft approach and landing process is provided, including: A memory for storing instructions; A processor for reading the instructions stored in the memory and executing any one of the methods in the first aspect according to the instructions.

[0016] Fourthly, a computer-readable storage medium is provided. Instructions are stored on the computer-readable storage medium. When the instructions run on a computer, the computer is caused to execute any one of the methods in the first aspect. Meanwhile, a computer program product is also provided. When the computer program product runs on a computer, it executes any one of the methods in the first aspect.

[0017] Beneficial effects: By collecting the actual flight QAR data of an aircraft during the approach and landing flight process, the present invention selects and extracts flight state parameters closely related to flight safety from the essence of the aircraft's physical movement to input into the trained and tested WOA-LightGBM model for risk prediction during the aircraft approach and landing process, obtains the corresponding risk prediction results, and finally determines the risk level of the aircraft's approach and landing flight safety. It can achieve efficient and real-time risk prediction during the aircraft approach and landing process, and improve the timeliness and accuracy of risk prediction during the aircraft approach and landing process. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] In order to more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the following will briefly introduce the drawings required for use in the description of the embodiments or the prior art. Obviously, the following drawings are only some embodiments of the present invention. For those of ordinary skill in the art, other drawings can be obtained based on these drawings without creative efforts.

[0019] Figure 1 It is a schematic diagram of the steps of the method in Embodiment 1 of the present invention; Figure 2 It is a schematic diagram of the nominal mission profile of aircraft approach and landing; Figure 3 It is a schematic diagram of the force on the aircraft during the approach and landing process; Figure 4 It is a schematic diagram of the test results of the approach and landing flight safety risk of the WOA-LightGBM model; Figure 5 It is a schematic diagram of the change in the prediction accuracy of each model. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0020] It should be noted here that the description of these embodiments is used to help understand the present invention, but does not constitute a limitation to the present invention. The specific structures and functional details disclosed herein are only used to describe the exemplary embodiments of the present invention. However, the present invention can be embodied in many alternative forms and should not be construed as limited to the embodiments described herein.

[0021] It should be understood that unless otherwise clearly specified and limited, the corresponding terms should be understood in a broad sense. For example, "connection" can be a fixed connection, a detachable connection, or an integral connection; it can be directly connected or indirectly connected through an intermediate medium, and can be the communication inside two components. For those of ordinary skill in the art, the specific meanings of the above terms in the embodiments can be understood according to specific situations.

[0022] Specific details are provided in the following description to facilitate a complete understanding of the exemplary embodiments. However, those of ordinary skill in the art should understand that the exemplary embodiments can be implemented without these specific details. For example, the device can be shown in a block diagram to avoid obscuring the example with unnecessary details. In other embodiments, well-known processes, structures, and technologies can be shown without unnecessary details to avoid obscuring the embodiments.

[0023] Embodiment 1: This embodiment provides a method for predicting risks during the approach and landing process of an aircraft, which can be applied to a corresponding flight safety monitoring terminal, such as Figure 1 As shown, the method includes the following steps: S1. Collect the actual flight QAR data of the aircraft during the approach and landing flight process of the aircraft.

[0024] Specifically, during the approach and landing flight process of the aircraft, the actual flight QAR data of the aircraft can be collected through the QAR (Quick Access Recorder, an important on-board device for recording aircraft flight data, often used by civil aviation regulatory departments and airlines for flight accident investigation and monitoring of pilots' flight operation quality) on the aircraft. The QAR data covers various aspects of the aircraft, including speed, altitude, attitude, acceleration, air pressure, temperature, fuel quantity, engine speed, etc.). Then, the actual flight QAR data of the aircraft is transmitted to the flight safety monitoring terminal for data analysis and processing.

[0025] S2. Extract the flight state parameter set from the actual flight QAR data.

[0026] Specifically, the flight safety monitoring terminal can extract the flight state parameter set from the actual flight QAR data. The flight state parameter set includes flight speed, flight altitude, horizontal displacement, lateral track offset, pitch angle, roll angle, and yaw angle. The extraction of these flight state parameters is completely analyzed and selected from the perspective of the essence of the physical movement of the aircraft: The aircraft approach and landing process includes the approach phase and the landing phase. The aircraft's approach phase is from the initial approach fix to the decision height or the minimum descent altitude; the landing phase is from about 50 ft above the runway threshold, starting to descend until the aircraft touches down and comes to a complete stop on the runway. To reduce flight accidents caused by human operation errors, most aircraft currently perform approach flights according to the instrument approach procedures provided by the destination airport and based on instrument indications. According to civil aviation relevant standards and operation manuals, a nominal mission profile for the aircraft's final approach and landing can be established, such as Figure 2 shown. This mission profile mainly includes processes such as the final approach and go-around during the approach phase, and processes such as descent, flare, touchdown, and deceleration taxi during the landing phase.

[0027] According to the approach and landing standards, the aircraft usually intercepts the glide slope signal at the FAP / FAF point (FAP point for precision approach and FAF point for non-precision approach), aligns with the runway, and starts to descend along the nominal glide slope. Generally, the nominal glide angle is 3°. According to the instrument approach procedure and the airline's flight procedure manual, the aircraft should establish a stable flight state at a minimum altitude not lower than 1000 ft (instrument meteorological conditions approach) and 500 ft (visual meteorological conditions approach). If a stable flight state cannot be established at the above altitudes, the approach must be aborted and the go-around procedure must be executed before the Mapt point (go-around point). When the aircraft decelerates and descends along the nominal glide angle to about 50 ft above the runway threshold, the aircraft speed reaches the approach speed Vapp. When descending to about 30 ft above the ground, through the flare operation, the aircraft gradually exits the descending state and turns into a flare and touchdown state. During this process, the glide angle gradually decreases from 3° to 0°, the aircraft pitch attitude and angle of attack gradually increase, the nose gradually changes from a downward state to an upward state, forming a two-point posture, and then evenly sinks to the touchdown area. After the main wheels of the aircraft touch down, the engine reverse thrust and the ground spoilers are opened, the aircraft resistance increases, the taxi speed decreases, and the lift also decreases accordingly. The nose naturally dips until the front wheels touch down, and finally the aircraft decelerates and taxis to a stop.

[0028] As Figure 3 shown, during the aircraft approach and landing process, the aircraft is affected by external forces such as gravity G, thrust T, drag D, and lift L. Through force analysis, a flight state equation set that describes the relationship between the aircraft's aerodynamic force, moment, and flight state parameters (speed, acceleration, position, attitude, etc.) can be established using a six-degree-of-freedom model. The assumed conditions for flight state equation modeling are that the aircraft is a rigid body with constant mass, the fuselage is symmetric about the aircraft's longitudinal axis, the Earth's surface is a plane, the ground coordinate system is an inertial coordinate, and the gravitational acceleration does not change with flight altitude. Based on the above basic assumptions, the free motion of the aircraft in space can be regarded as the synthesis of the translational motion of the rigid body and the rotational motion around the center of mass, that is, the three degrees of freedom of the instantaneous position of the center of mass and the three degrees of freedom of the instantaneous attitude of the rigid body.

[0029] Generally, the more aircraft motion equations there are, the more accurate and complete the description of the aircraft's motion will be, but the greater the difficulty of solving them. In engineering applications, within the allowable range of solution accuracy, approximate methods can be used to simplify the flight motion equations in order to analyze the aircraft's motion using relatively simple equation systems. Since during the approach and landing phase, the aircraft needs to establish the landing configuration and maintain the correct flight path before 1000 ft to establish a stable flight state, that is, the pilot needs to control the aircraft within the nominal flight mission profile as much as possible. Therefore, the aircraft motion equation system is simplified into a longitudinal motion equation system and a lateral motion equation system to analyze the flight motion law of the aircraft during approach and landing.

[0030] (1) Longitudinal motion equation system As can be seen from the nominal flight mission profile, after the FAF / FAP point, the pilot needs to complete the runway alignment and control the aircraft to descend and land along the established vertical profile. Therefore, the aircraft is always moving in the vertical plane along the runway centerline during the approach and landing process, that is, the longitudinal motion. The longitudinal motion equation system of the aircraft derived according to Newtonian mechanics is

[0031] In the formula: V is the flight speed; is the angle of attack of the aircraft, γ is the glide angle, θ is the pitch angle of the aircraft, and there is ; is the pitching moment; is the moment of inertia about the oz b axis of the body coordinate system; L is the lift of the aircraft; T is the thrust; D is the drag; m is the mass of the aircraft, g is the gravity coefficient, t is the time parameter; is the angular rate of rotation about the oz b axis of the body coordinate system; x g 、y g is the displacement of the aircraft's center of mass in the vertical plane.

[0032] (2) Lateral motion equation system During the approach and landing process, when the aircraft is disturbed laterally and deviates, the pilot needs to actively correct the flight state to bring it back to the established flight path and continue to fly or taxi along the required horizontal profile. Therefore, according to Newtonian mechanics, the lateral motion equation system for all stages of the aircraft's approach and landing can be derived as

[0033] In the formula: is the roll angle; ψ is the yaw angle, β is the sideslip angle, is the track azimuth angle, and there is ; is the speed roll angle; Z is the lateral force; is the rolling moment; is the yawing moment; , are respectively the moments of inertia about the body coordinate system ox b , oy b axis; , are respectively the angular rates of rotation about the body coordinate ox b , oy b axis; z g is the displacement of the aircraft's center of mass in the horizontal plane, i.e., the lateral track offset.

[0034] Analysis of flight state risks: According to the global air transportation safety report released by relevant agencies, during approach and landing flight accidents, accidents such as overrunning the runway, tail strike, hard landing, flight out of control, controlled flight into terrain, and landing off the runway occur most frequently, and the proportion of fatal accidents caused is also the highest. And the occurrence of these accident events is directly related to flight state deviations. Therefore, combined with the flight state equations, select X ={ V, x g 、y g , z g , θ, ψ, } as the flight state parameter set, that is, the flight speed, flight altitude, horizontal displacement, lateral track offset, pitch angle, roll angle, and yaw angle can be selected to form the flight state parameter set.

[0035] (1) Longitudinal motion risk During the glide phase, the main performance characteristic parameters of the aircraft's descent are the glide angle γ and the descent rate . Since the glide angle of the aircraft, its rate of change, and the angle of attack of the aircraft are all small, it can be approximately considered that , and , after substituting into the aircraft's longitudinal motion equations, the equations can be simplified. According to the simplified equations, the glide angleγ is closely related to state parameters such as V flight speed and y flight altitude, and there are and ; while the descent rate is also a function of the glide angle and flight speed. If the influence of wind speed is ignored at different approach altitudes, the greater the flight speed, the farther the flight distance, and the smaller the glide angle, the farther the touchdown point is from the runway threshold, and it may cause overshooting the runway during landing; conversely, if the flight speed is small, the flight distance is shorter, the glide angle is larger, and it may lead to a controlled flight into terrain or landing outside the runway. If descending at a constant glide angle, the greater the flight speed, the greater the descent rate, and an excessive descent rate may cause an excessive vertical acceleration, thus resulting in a hard landing.

[0036] During the flare phase, the nose of the aircraft will gradually change from a downward position to an upward position, forming a two-point landing posture and floating horizontally to the touchdown point. During this period, the glide angle γ of the aircraft will gradually decrease, and the pitch angle θ will gradually increase. As known from , the angle of attack α of the aircraft will gradually increase with the increase of the pitch angle. If the angle of attack exceeds the critical angle of attack value, a flight stall will occur, which may lead to an out-of-control event in the air; if the pitch angle is still too large at touchdown, it may lead to a tailstrike event. In addition, with the increase of the angle of attack, the flight drag will also increase rapidly. If the engine throttle remains unchanged, the flight speed will decrease. At that time, the lift is less than the gravity, and the descent rate of the aircraft increases, which may result in a hard landing. And from the longitudinal motion equations of the aircraft, if the flare altitude is too high, the flight speed will not be effectively reduced, and it may cause overshooting the runway due to a long floating distance; if the flare altitude is too low, the flight speed is too large, the descent rate of the aircraft continues to increase, and if the flare operation is performed again, it may lead to a hard landing.

[0037] (2) Lateral motion risks During the approach and landing process, the pilot mainly uses controls such as ailerons and rudders to overcome the adverse effects of crosswinds on the roll, yaw, and side slip of the aircraft, so as to reduce the track deviation after aligning with the runway and keep the wing from having a large slope. Therefore, the roll angle , yaw angle ψ and lateral track offset z are used as risk analysis parameters.

[0038] During the approach and landing process, if the lateral motion of the aircraft is studied independently, then . Assuming that the aircraft has roll motion but no sideslip motion and the pitch angle is small, then there are , , . Substituting into the simplified lateral motion equations shows that the aircraft's yaw is closely related to the roll angle and heading deflection rate; the lateral track offset of the aircraft is a function of the flight speed and track azimuth angle. From it can be seen that when the aircraft has no sideslip, the track azimuth angle is the same as the yaw angle. If the aircraft's fuselage has a slope or heading deflection during the approach and landing, it will cause the aircraft's nose direction and the glide path to be inconsistent, and it may cause the engine to rub during landing. And from the lateral motion equations, the greater the flight speed or yaw angle, the greater the lateral track offset, which will further cause the approach and landing to deviate from the runway centerline, and ultimately lead to accidents such as landing off the runway or landing outside the runway.

[0039] According to the above longitudinal and lateral flight motion risk analysis, during the approach and landing process of the aircraft, the flight speed V , flight altitude y , horizontal displacement x , lateral track offset z and attitude angles (including pitch angle θ , roll angle , yaw angle ψ ) and other flight state parameters deviate from the nominal values, it may cause flight accidents such as overrunning the runway, hitting the tail of the aircraft, hard landing, flight out of control, controlled flight into terrain, and landing outside the runway. Since the descent rate is comprehensively determined by parameters such as flight speed and flight altitude, it can be equivalently regarded as a flight state parameter. Therefore, the flight speed, flight altitude, horizontal displacement, lateral track offset, pitch angle, roll angle, and yaw angle during the approach and landing flight process are selected as the approach and landing flight safety risk monitoring indicators. The specific monitoring points of each flight state parameter and the hazards that may be caused by abnormal situations are shown in Table 1 below.

[0040] Table 1

[0041] For the determined approach and landing flight safety risk monitoring indicators, the approach and landing flight safety risk levels can be divided into 4 levels: level 0, level 1, level 2, and level 3. Among them, level 0 is the risk-free level, level 1 is the low-risk level, level 2 is the medium-risk level, and level 3 is the high-risk level. The quantitative judgment criteria for the risk levels of each risk monitoring indicator are shown in Table 2 below.

[0042] Table 2

[0043] S3. Input the flight state parameter set into the pre-set risk prediction model for the aircraft approach and landing process for model analysis to obtain the risk prediction result for the aircraft approach and landing process. The risk prediction model for the aircraft approach and landing process uses the trained and tested WOA-LightGBM model.

[0044] In specific implementation, since the simulation method for establishing the flight safety risk prediction model cannot fully simulate the real flight process, there may be differences between the prediction results and the actual situation. Therefore, it is necessary to use real QAR data to analyze and predict the approach and landing flight safety risks. At present, the massive flight QAR data available to airlines has not been effectively utilized, which provides an analysis basis for the research of flight safety risks using machine learning methods. Due to the differences in the learning abilities of different machine learning models, and the weak prediction effect and generalization ability, although deep learning can improve the prediction accuracy, it requires a complex training process and has problems such as gradient disappearance and difficulty in analyzing multi-dimensional data. The ensemble learning method synthesizes the advantages of multiple machine learning algorithms through a certain strategy to construct a learning system that better meets the actual data analysis requirements, and has currently become a hot topic in data mining technology. In this regard, an ensemble learning method based on WOA and LightGBM is proposed to predict the approach and landing flight safety risks of aircraft, in order to provide auxiliary decision-making support for the flight safety risk control of airlines.

[0045] Before inputting the flight state parameter set into the pre-set risk prediction model for the aircraft approach and landing process for model analysis, it is necessary to first construct, train, and test to obtain the risk prediction model for the aircraft approach and landing process. This process includes: (1) Data preprocessing Extract QAR data samples of flights performed at a certain airport from the system of a certain airline, totaling 1027 copies. Through data structure analysis, the acquisition frequency of all flight parameter data is 1 Hz, that is, data is sampled once per second. However, due to problems with the sensitivity of aircraft sensors and system data decoding in some samples, some fields and even some data are missing. For this reason, 117 QAR data samples with missing data are directly excluded. The remaining 910 samples all meet the requirements of prediction analysis. The retained original QAR data samples are shown in Table 3 below.

[0046] Table 3

[0047] Then, according to the approach and landing flight safety risk monitoring indicators, select the maximum and minimum approach speeds from 500 ft to 50 ft, the maximum descent rate from 500 ft to 50 ft, the maximum landing speed at 50 ft and below, the approach runway port height, the flight distance from 50 ft to touchdown, the touchdown elevation angle at the moment of main wheel touchdown, the maximum landing roll angle from 50 ft to all wheels touchdown, the maximum yaw angle from 50 ft to all wheels touchdown, etc. from each sample data to establish a sample of the flight state parameter dataset, and establish a training set and a test set according to a ratio of 9:1, that is, select 819 samples as the training set and 91 samples as the test set. At the same time, according to the risk level quantization criteria of each indicator in Table 2, use the highest risk level of each flight monitoring indicator as the approach and landing flight safety risk label of the flight for sample annotation. To enhance the generalization ability of the WOA-LightGBM model, accelerate the convergence speed of the model, and ensure the training and prediction effects of the model, the original QAR data can be normalized to solve the problems of inconsistent flight parameter data units and large differences in numerical ranges, so as to achieve consistency analysis.

[0048] (2) Construct a LightGBM prediction model Construct a decision tree according to the GBDT algorithm, and use the histogram algorithm and the leaf growth strategy with depth limit to construct a LightGBM (Light Gradient Boosting Machine) model. Among them, set the model random_state to 5, the maximum number of iterations to 300, and both verbose_eval and early_stopping_rounds to 100.

[0049] Light Gradient Boosting Machine (LightGBM) is a gradient boosting framework based on the decision tree algorithm. It mainly adopts Gradient Boosting Decision Tree (GBDT) and gradually improves the prediction ability of the model by iteratively training decision trees. This model adopts the Leaf-wise growth strategy and the exclusive feature bundling technology, which can maximize the retention of samples beneficial to information gain calculation to more efficiently and accurately fit the data, thereby improving the accuracy and generalization ability of the model.

[0050] For a dataset with n samples , the LightGBM prediction result can be expressed as:

[0051] In the formula: f ( x ) is the output value of each decision tree in the model w (x ) The weighted sum obtained through stacking training.

[0052] The objective function of the LightGBM model is expressed as:

[0053] In the formula: is the loss function, f m is the complexity of the m-th tree, represents the regularization term.

[0054] The LightGBM model uses the one-sided gradient sampling technique to sample the data, retaining all subsets of samples with large gradients A , and randomly sampling subsets of samples with small gradients B . Divide the samples into subsets A and B The variance gain is:

[0055] In the formula: n is the total number of samples; j is the splitting feature used; d is the splitting point of the sample feature; and are the numbers of samples whose splitting feature values used by the model are less than and greater than d respectively; and are the large and small gradient samples of the left child node of the splitting point respectively; and are the large and small gradient samples of the right child node of the splitting point respectively; G is the sample gradient; a is the sampling rate of large gradient samples; c is the sampling rate of small gradient samples. The sample gradient is:

[0056] In the formula: y is the target variable; s is the sample prediction value; p is the number of iterations; L is the loss function (3) Population initialization The Whale Optimization Algorithm (WOA) is a heuristic optimization algorithm based on the predation behavior of whales. This algorithm establishes a bubble net search strategy by simulating the predation behavior of humpback whales, thereby achieving the purpose of finding the optimal solution.

[0057] The feeding behavior patterns of humpback whales are divided into three stages: encircling prey, bubble-net feeding, and searching for prey. Among them, searching for prey includes two forms: narrowing the search range and random searching. In the WOA, the position of each humpback whale represents a potential solution. By continuously updating the positions of whales in the solution space, the global optimal solution is finally obtained. This model has the advantages of simple mechanism, few parameters, and strong optimization ability.

[0058] (3.1) Encircling prey The search range of whales is the global solution space, and the position of the prey needs to be determined first for encirclement. Since the position of the optimal design in the search speed is not known a priori, the WOA algorithm assumes that the current best candidate solution is the target prey or close to the optimal solution. After defining the best search agent, other search agents will try to update their positions towards the best search agent. This behavior can be represented by the following equation:

[0059]

[0060] In the formula: X represents the position vector of the humpback whale; is the optimal solution reached after each iteration of the model; t is the number of iterations; A and C are coefficients, and their calculation methods are as follows:

[0061]

[0062] In the formula: a The value of takes a linear decrease from 2 to 0 according to the reciprocal of the number of iterations; r is a random value between 0 and 1.

[0063] (3.2) Bubble-net feeding There are mainly two mechanisms for humpback whales to prey: encircling prey and bubble-net feeding. When using bubble-net feeding, the position update between the humpback whale and the prey is expressed by the logarithmic spiral equation:

[0064]

[0065] In the formula: is the distance between the current search individual and the current optimal solution; b is the range of the spiral of the humpback whale's feeding behavior; l is a random value between -1 and 1.

[0066] (3.3) Searching for prey To ensure that all whales can search sufficiently in the solution space, WOA updates the positions according to the distances between whales to achieve the purpose of random search. Therefore, when |A|≥1, the searching individual will swim towards a random whale, and the specific equation is as follows:

[0067]

[0068] In the formula: is the distance between the current searching individual and the random individual; is the position of the current random individual.

[0069] Set the number of the whale population, randomly initialize the positions of the whale population, and calculate the fitness of each individual by the following formula:

[0070]

[0071] In the formula: Fitness is the fitness of the individual; Acc is the model accuracy; is the number of samples in the i th row and j th column of the confusion matrix. At the same time, select the optimal solution of all current individuals as the initial optimal solution, and the corresponding fitness as the initial fitness.

[0072] (4)Whale optimization Take the maximum tree depth max_depth, the number of leaf nodes num_leaves, the learning rate learning_rate, the minimum number of samples in a leaf min_data_in_leaf, the ratio of randomly selected features feature_fraction, and the ratio of selected data to all data bagging_fraction of the LightGBM model as the objects to be optimized by WOA, set the number of searching whales to 30, and at the same time complete the initialization of the positions of the whale population and the fitness of individuals. By iteratively updating the spatial positions of the whale population, calculate the fitness of all individuals to obtain the optimal hyperparameter combination for the current iteration.

[0073] (5)Loop iteration Judge whether the maximum number of iterations is reached. If it is reached, output the corresponding position and parameter values of the optimal whale individual, and use them as the best hyperparameter combination; if not, continue to iterate until it is reached. The present invention iteratively updates the spatial positions and fitness of the whale population. After multiple iterations, when the number of iterations reaches the predetermined 300 times or the fitness is 0.95, stop the iteration and output the obtained best hyperparameter combination. The finally obtained best hyperparameters are shown in Table 4 below.

[0074] Table 4

[0075] (6) Training and evaluation of the WOA-LightGBM model Use the optimal hyperparameters in Table 4 as the LightGBM model parameters, and use the training set to train the WOA-LightGBM model. At the same time, observe the stability of the model by increasing the number of model iterations until the evaluation indicators of the WOA-LightGBM model reach the set convergence conditions. When each evaluation indicator tends to be stable, the model can converge well. The influence of the number of iterations on the performance of the WOA-LightGBM model is shown in Table 5 below.

[0076] Table 5

[0077] When the number of iterations reaches 100, the accuracy, precision, recall, and f 1 score values of the WOA-LightGBM model are the highest and then tend to be stable. The calculation formulas for the evaluation indicators of accuracy, precision, recall, and f 1 score value are as follows:

[0078]

[0079]

[0080]

[0081] After using the training set to iteratively train the WOA-LightGBM model, the effectiveness and stability of the model can be tested using the test set data. The test results are as Figure 4 shown. In the figure, 0, 1, 2, and 3 represent the flight safety risk levels in the flight approach and landing phases of the flight, corresponding to "no risk", "low risk", "medium risk", and "high risk" respectively. According to Figure 4 the confusion matrix in, the accuracy of the WOA-LightGBM model predicting the test set data after training can reach about 95.60%.

[0082] If the test is qualified, the tested WOA-LightGBM model is obtained. Finally, the trained and tested WOA-LightGBM model is used as the risk prediction model for the aircraft approach and landing process. In actual application, input the actually collected and extracted flight state parameter set into the risk prediction model for the aircraft approach and landing process for model analysis to obtain the risk prediction result for the aircraft approach and landing process of the aircraft.

[0083] S4. Determine the approach and landing flight safety risk level of the aircraft based on the risk prediction result of the aircraft approach and landing process, and output the approach and landing flight safety risk level of the aircraft.

[0084] During specific implementation, the risk prediction result of the aircraft approach and landing process is 0, 1, 2, or 3, and the approach and landing flight safety risk level is a risk-free level, a low-risk level, a medium-risk level, or a high-risk level. Among them, the risk prediction result of the aircraft approach and landing process corresponding to the risk-free level is 0, the risk prediction result of the aircraft approach and landing process corresponding to the low-risk level is 1, the risk prediction result of the aircraft approach and landing process corresponding to the medium-risk level is 2, and the risk prediction result of the aircraft approach and landing process corresponding to the high-risk level is 3. The approach and landing flight safety risk level corresponding to the risk prediction result of the aircraft approach and landing process can be determined based on the corresponding relationship, and then the approach and landing flight safety risk level of the aircraft can be output.

[0085] This method can achieve efficient and real-time risk prediction of the aircraft approach and landing process, and improve the timeliness and accuracy of risk prediction of the aircraft approach and landing process. To further verify the performance of the WOA-LightGBM model in predicting the approach and landing flight safety risk, this embodiment continues to use the corresponding training set and test set, and selects the same type of models such as LightGBM, XGBoost, and RandomForest for prediction comparison. Among them: (1) Set the parameters of the LightGBM model as follows: the maximum tree depth max_depth is 5, the number of leaf nodes num_leaves is 5, the learning rate learning_rate is 0.01, the minimum number of samples in a leaf min_data_in_leaf is 5, the ratio of randomly selected features feature_fraction is 0.5, and the ratio of selected data to all data bagging_fraction is 0.5; (2) Set the parameters of the XGBoost model as follows: the maximum tree depth max_depth is 4, the sum of the sample weights of the minimum leaf nodes min_child_weight is 4, the learning rate learning_rate is 0.1, the L1 regularization term reg_alpha is 1, and the minimum decrease value of the loss function gamma is 0.1; (3) Set the parameters of the Random Forest model as follows: the number of decision trees n_estimators is 81, the maximum tree depth max_depth is 4, the minimum number of samples in a leaf min_samples_lef is 5, and the maximum number of features max_features is 6. The above parameter settings are the best parameters after repeated training.

[0086] The comparison of the prediction accuracy changes of the above models is as follows Figure 5 As shown, the accuracy of the WOA-LightGBM model in predicting the approach and landing flight safety risk is higher than that of the LightGBM, XGBoost, and Random Forest models. The performance comparison of the above models using the optimal parameters for prediction is shown in Table 6 below.

[0087] Table 6

[0088] According to the comparison results, it can be seen that the accuracy, precision, recall, and f 1 score values of the WOA-LightGBM model are all higher than those of the traditional LightGBM, XGBoost, and Random Forest models, indicating that the accuracy and prediction effect of using the WOA-LightGBM model to predict the approach and landing flight safety risk are higher than those of other models.

[0089] Example 2: This example provides a risk prediction system for the aircraft approach and landing process, including a data acquisition unit, a data extraction unit, a model prediction unit, and a risk determination unit, where: The data acquisition unit is used to collect the actual flight QAR data of the aircraft during the approach and landing flight of the aircraft; The data extraction unit is used to extract the flight state parameter set from the actual flight QAR data; The model prediction unit is used to input the flight state parameter set into the pre-set risk prediction model for the aircraft approach and landing process for model analysis, and obtain the risk prediction result for the aircraft approach and landing process. The risk prediction model for the aircraft approach and landing process uses the trained and tested WOA-LightGBM model; The risk determination unit is used to determine the approach and landing flight safety risk level of the aircraft according to the risk prediction result for the aircraft approach and landing process, and output the approach and landing flight safety risk level of the aircraft.

[0090] Example 3: This example provides a risk prediction system for the aircraft approach and landing process. At the hardware level, it includes: A data interface for establishing data docking between the processor and external data acquisition devices; A memory for storing instructions; A processor for reading the instructions stored in the memory and executing the risk prediction method for the aircraft approach and landing process in Example 1 according to the instructions.

[0091] Optionally, the system further includes an internal bus through which the processor, the memory, and the data interface can be interconnected with each other. The internal bus can be an ISA (Industry Standard Architecture) bus, a PCI (Peripheral Component Interconnect) bus, an EISA (Extended Industry Standard Architecture) bus, or the like. The bus can be divided into an address bus, a data bus, a control bus, and the like.

[0092] The memory can include, but is not limited to, a random access memory (RAM), a read only memory (ROM), a flash memory, a first input first output (FIFO) memory, and / or a first in last out (FILO) memory, etc. The processor can be a general-purpose processor, including a central processing unit (CPU), a network processor (NP), etc.; it can also be a digital signal processor (DSP), an application specific integrated circuit (ASIC), a field-programmable gate array (FPGA), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components.

[0093] Embodiment 4: This embodiment provides a computer-readable storage medium, on which instructions are stored. When the instructions run on a computer, the computer is caused to execute the risk prediction method for the aircraft approach and landing process in Embodiment 1. Among them, the computer-readable storage medium refers to a carrier for storing data, and can include, but is not limited to, a floppy disk, an optical disc, a hard disk, a flash memory, a USB flash drive, and / or a memory stick, etc. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable devices.

[0094] This embodiment also provides a computer program product. When the computer program product runs on a computer, it executes the risk prediction method for the aircraft approach and landing process in Embodiment 1. Among them, the computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable devices.

[0095] Finally, it should be noted that the above are only the preferred embodiments of the present invention and are not intended to limit the protection scope of the present invention. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. A method for predicting the risk of an aircraft approach and landing process, characterized in that: include: Collect the actual flight QAR data of the aircraft during the aircraft approach and landing flight; Extract flight status parameter set from actual flight QAR data; Inputting the flight status parameter set into a preset aircraft approach and landing process risk prediction model for model analysis to obtain an aircraft approach and landing process risk prediction result of the aircraft, wherein the aircraft approach and landing process risk prediction model adopts a trained and tested WOA-LightGBM model; The aircraft's approach and landing flight safety risk level is determined based on the risk prediction results of the aircraft's approach and landing process, and the aircraft's approach and landing flight safety risk level is output.

2. The method for predicting the risk of an aircraft approach and landing process according to claim 1, characterized in that: The flight status parameter set includes flight speed, flight altitude, horizontal displacement, lateral track offset, pitch angle, roll angle and yaw angle.

3. The method for predicting the risk of an aircraft approach and landing process according to claim 1, characterized in that: Before inputting the flight status parameter set into a preset aircraft approach and landing process risk prediction model for model analysis, the method further includes: Construct a LightGBM model, and use the WOA algorithm to optimize and iterate the LightGBM model to obtain the WOA-LightGBM model; A training set and a test set are obtained, and the training set is used to train the WOA-LightGBM model to obtain the trained WOA-LightGBM model. The trained WOA-LightGBM model is tested with the test set to obtain the tested WOA-LightGBM model, and the trained and tested WOA-LightGBM model is used as a risk prediction model for the aircraft approach and landing process.

4. The method for predicting the risk of an aircraft approach and landing process according to claim 3, characterized in that: The obtaining of the training set and the test set comprises: Collecting a number of flight QAR data samples, and performing data preprocessing on all the flight QAR data samples to obtain a number of preprocessed flight QAR data samples; Extracting flight status parameter set samples from each preprocessed flight QAR data sample, and labeling each flight status parameter set sample with an approach and landing flight safety risk label to obtain a corresponding labeled flight status parameter set sample; The flight state parameter data set is composed of the labeled flight state parameter set samples, and the flight state parameter data set is divided into a training set and a test set according to a set ratio.

5. The method for predicting the risk of an aircraft approach and landing process according to claim 4, characterized in that: The data preprocessing of all flight QAR data samples includes: All flight QAR data samples with missing data are removed.

6. The method for predicting the risk of an aircraft approach and landing process according to claim 3, characterized in that: When the WOA-LightGBM model is trained using the training set, the method further includes: The accuracy, precision, recall and f1 score of the model are used as evaluation indicators of the WOA-LightGBM model. The WOA-LightGBM model is iteratively trained using the training set until the evaluation indicators of the WOA-LightGBM model reach the set convergence conditions.

7. The method for predicting the risk of an aircraft approach and landing process according to claim 1, characterized in that: The risk prediction result of the aircraft approach and landing process is 0, 1, 2 or 3, and the approach and landing flight safety risk level is no risk level, low risk level, medium risk level or high risk level, among which the risk prediction result of the aircraft approach and landing process corresponding to the no risk level is 0, the risk prediction result of the aircraft approach and landing process corresponding to the low risk level is 1, the risk prediction result of the aircraft approach and landing process corresponding to the medium risk level is 2, and the risk prediction result of the aircraft approach and landing process corresponding to the high risk level is 3.

8. A risk prediction system for aircraft approach and landing process, characterized in that: It includes a data collection unit, a data extraction unit, a model prediction unit and a risk determination unit, wherein: A data collection unit is used to collect the actual flight QAR data of the aircraft during the aircraft approach and landing flight; A data extraction unit, used to extract a flight status parameter set from actual flight QAR data; A model prediction unit is used to input the flight state parameter set into a preset aircraft approach and landing process risk prediction model for model analysis to obtain an aircraft approach and landing process risk prediction result of the aircraft, wherein the aircraft approach and landing process risk prediction model adopts a trained and tested WOA-LightGBM model; The risk determination unit is used to determine the aircraft's approach and landing flight safety risk level according to the risk prediction results of the aircraft's approach and landing process, and output the aircraft's approach and landing flight safety risk level.

9. A risk prediction system for aircraft approach and landing process, characterized in that: include: A memory for storing instructions; A processor is used to read the instructions stored in the memory and execute the aircraft approach and landing process risk prediction method described in any one of claims 1-7 according to the instructions.

10. A computer program product, characterized in that When the computer program product is run on a computer, the aircraft approach and landing process risk prediction method according to any one of claims 1 to 7 is executed.

Images