CN116010889A - An intelligent identification method for abnormal flight status of aviation aircraft - Google Patents

Abstract

The invention relates to the technical field of aviation flight data analysis, in particular to an aviation aircraft abnormal state detection method. The method comprises the following steps: 1) Selecting historical standard flight data of the aviation aircraft by aviation aircraft flight experts, and marking the flight phases of the historical standard flight data; 2) Designing an optimization classification regression decision tree CART algorithm by using a marked data set D marked with the flight phases to construct a flight phase classification model M; 3) Inputting flight data of an abnormal flight state to be identified; 4) Performing flight phase division on the flight data set X by using the constructed classification model M; 5) Identifying abnormal flight states in the forest by using an optimized isolated forest algorithm IF; 6) And outputting the abnormal flight state time series data. According to the method, the historical flight data of the aircraft are utilized, a model of a normal flight state is established by a training algorithm, and the established model is utilized to identify an abnormal flight state.

Description

An intelligent identification method for abnormal flight status of aircraft

Technical Field

The present invention relates to the technical field of aviation flight data analysis, and in particular to a method for detecting abnormal state of an aviation aircraft.

Background Art

The flight state of an aircraft refers to the movement of an aircraft at a certain moment. The normal flight state of an aircraft refers to the state in which the aircraft flies according to a pre-set route or expected state under the control of the flight control system. The abnormal flight state of an aircraft refers to the flight state of an aircraft that deviates greatly from the expected flight state due to the failure of a component or the interference and influence of the external environment. For example, the actual flight route deviates greatly from the set route, and the given attitude angle, speed, and heading deviate greatly from the corresponding measured values for a long time. These abnormal flight states of aircraft will bring flight risks. If they cannot be identified and handled in time, they may cause major flight safety accidents. Therefore, the identification of abnormal flight states of aircraft is of great significance to ensuring the flight safety of aircraft. The identification of abnormal flight states of aircraft provides decision support for real-time control decisions and health management of aircraft, and facilitates aircraft support personnel to find the causes of abnormal events in advance, avoid flight accidents, and ensure flight safety.

Aiming at the problem of automatic identification of flight status, Xie Chuan et al. constructed a flight status identification method based on expert knowledge base and knowledge inference engine. This identification method is limited by the expert knowledge base. In the literature, Meng Guanglei et al. took the flight parameter data corresponding to the maneuvering action of flight simulation training as the research object, constructed a dynamic Bayesian network model for maneuvering action identification, and identified the flight status with the help of an intelligent method of recursive reasoning based on the network model. Zhou Chao et al. mainly proposed a flight status identification method based on an improved dynamic time warping algorithm, mainly targeting the characteristics of strong randomness and different lengths of tactical maneuvering action data. This method uses different characteristic parameters of the flight status to set different contribution degrees, calculates the frame matching distance between the flight parameter data and the standard template data, and identifies the flight status according to the distance. These documents have realized the automatic identification of certain flight states in a specific application scenario, and are not suitable for the scenario of abnormal flight status identification of aircraft.

Therefore, there is an urgent need for an intelligent identification method for abnormal flight status of aircraft that can solve the above-mentioned technical problems.

Summary of the invention

The purpose of the present invention is to overcome the shortcomings of the above-mentioned prior art and provide a method for intelligently identifying abnormal flight states of aircraft. The method defines multivariate characteristic parameters of abnormal attitude, abnormal speed, abnormal track and abnormal actuator control in different flight stages of aircraft, introduces a grid search method to optimize the parameters of the classification decision tree, realizes accurate classification of the flight stages, adds the abnormal influence of the characteristic parameters to construct an isolated forest, and improves the accuracy of abnormal flight state identification. The intelligent method proposed in the present invention has important application value in aircraft flight monitoring and is of great significance to improving the intelligent monitoring capability of flight safety.

The invention provides an intelligent identification method for abnormal flight status of an aircraft, which is special in that it comprises the following steps:

1) Aviation flight experts select historical standard flight data of aviation aircraft and mark the flight phases thereof;

Specifically, flight experts select historical standard flight data and mark the six basic flight stages of takeoff, climb, level flight, turn, descent and landing. The marked flight data set is recorded as D = {D ₁ ,D ₂ ,......,D ₆ };

2) Using the labeled data set D with labeled flight stages, we design and optimize the Classification And Regression Tree (CART) algorithm to build a flight stage classification model M;

Specifically:

之间的高度变化率Δh、俯仰角变化率Δα、滚转角变化率Δβ、偏航角变化率Δγ、航向角变化率Δδ，计算数据集D的基尼系数公式(1)Step 1: For the standard action template data D = {D ₁ , D ₂ , ..., D ₆ }, calculate the n-element feature parameter vector of the standard flight state D _r at the u-th time point

The height change rate Δh, pitch angle change rate Δα, roll angle change rate Δβ, yaw angle change rate Δγ, and heading angle change rate Δδ are used to calculate the Gini coefficient formula of data set D (1):

In the formula, _pk represents the proportion of the kth category in the data set D, which is simpler than the method of calculating the logarithm using the information gain rate. For the data set D, the Gini coefficient formula for calculating the feature parameter c is (2):

Where |D| represents the number of all time series vectors in the data set D, E represents the total number of value categories of the feature parameter c in the data set D, |D ^e | represents the total number of vectors in the data set D where the feature c value is e, and Gini(D ^e ) is the Gini coefficient calculated according to formula (1).

Step 2: Data set D is divided into two parts according to the characteristic parameter C with the smallest Gini coefficient value to obtain its left and right nodes, denoted as D _left and D _right ;

Step 3: There are two conditions for stopping the construction of the decision tree. One is to determine the relationship between the threshold and the current Gini coefficient. If the current Gini coefficient is less than the threshold, the current node stops building the decision tree subtree, otherwise it continues to build. The second is to check whether there is a classification feature in the feature parameter set D that can continue to decompose the set. If not found, stop building the decision tree.

Step 4: Recursively execute steps 1 to 4 for the left and right child nodes D _left and D _right of the D node until the condition of step 3 is met and the decision tree training model M is returned.

Step 5: To avoid the overfitting problem of the decision tree model, prune the decision tree. Pruning means deleting the left and right child nodes of the non-leaf node T _i corresponding to the α _i with the smallest surface error rate gain value among the non-leaf nodes {T ₁ ,T ₂ ,T ₃ ,......,T _n }. The calculation formula (3) of α _i is:

Where R(i) represents the error generated after the leaf node replaces the i-th non-leaf node, and the calculation formula is:

R(i)＝r(i)p(i), (4)

Where r(i) represents the error rate of node i, p(i) represents the percentage of samples on node i to the total number of samples in the training set, and R(T _i ) represents the sum of the errors of all leaf nodes on subtree T _i when node i is not pruned, that is,

Replace the non-leaf node _Ti with the leaf node of the subtree, repeat the process until there is no non-leaf node to replace, pruning is completed, and return to the decision tree Tree;

Step 6: Set the target loss function (accuracy, precision, recall or _F1 score), place all the hyperparameters of the decision tree in the network, and use the grid search algorithm to adjust the decision tree parameters to make the decision tree model M optimally classified. At the same time, in order to avoid the impact of the division of training sets and test sets, use data cross-validation to improve the classification performance of the decision tree model M.

3) Input flight data of abnormal flight status to be identified;

Specifically, the flight data set of an aircraft at t time points is input and recorded as X = {X ₁ , X ₂ , ..., X _t }

4) Use the constructed classification model M to divide the flight data set X into flight stages;

Specifically, the classification model M is used to automatically divide the flight data set X = {X ₁ , X ₂ , …, X _t } at t time points into flight phases, and the flight phases of each time period are marked;

5) Using the optimized Isolation Forest (IF) algorithm to identify abnormal flight status;

Specifically:

Step 1: For X with marked flight stages, calculate the abnormal characteristic parameter values of each flight stage according to Table 1, and the characteristic parameter set of each flight stage is recorded as S;

Taking the six most basic flight stages of takeoff, climb, level flight, turn, descent and landing as an example, each flight stage can be characterized by characteristic parameters such as flight speed, altitude, attitude angle and heading angle. At the same time, each flight stage corresponds to a different flight control law, and the main characteristic parameters of the flight stage are established;

Among them, the abnormal characteristic parameters are classified into: abnormal attitude, abnormal speed, abnormal track and abnormal actuator control. The calculation methods of various parameters are as follows:

(1) Abnormal posture characteristic parameters

The abnormal flight attitude of an aircraft refers to the deviation of the measured values of the attitude angles (pitch angle θ, yaw angle ψ, roll angle φ) from the given values (given pitch angle θ_ref, given yaw angle ψ_ref, given roll angle φ_ref) for a long time. The mean deviation of each attitude angle at N time points (E(Δθ), E(Δψ) and E(Δφ)) is calculated as the characteristic parameter of the attitude abnormality, that is,

Among them, θ _i , ψ _i , φ _i respectively represent the measured values of the pitch angle, yaw angle and roll angle at the i-th (i=1, 2, ..., N) time point, and θ_ref _i , ψ_ref _i , φ_ref _i respectively represent the given values of the pitch angle, yaw angle and roll angle at the i-th (i=1, 2, ..., N) time point.

(2) Speed anomaly characteristic parameters

The speed anomaly of an aircraft refers to the fact that the measured value V of the flight speed deviates from the given speed value V_ref for a long time. The mean deviation of the speed at N time points E(ΔV) is calculated as the speed anomaly characteristic parameter, that is,

Wherein, V _i and V_ref _i represent the measured value and given value of the flight speed of the aircraft at the i-th (i=1, 2, ..., N) time point, respectively.

(3) Track anomaly characteristic parameters

The abnormal flight path of an aircraft refers to the deviation of the actual flight route from the set route. The route is composed of several waypoints. The mean distance deviation E(Δd) of the waypoints at N time points is calculated as the characteristic parameter of the abnormal flight path, that is,

Among them, the measured values and given values of the longitude, latitude and altitude of the waypoint position at the i-th (i=1, 2, ..., N) moment are represented as _Pi ( _xi , _yi , _hi ) and _{P_refi} ( _{x_refi} , _{y_refi} , _{h_refi} ), respectively.

Calculate the mean value of the heading angle deviation E(ΔPSI) of the waypoints at N time points as the track anomaly characteristic parameter, that is,

Wherein, PSI _i and PSI_ref _i represent the measured value and given value of the heading angle at the i-th (i=1, 2, ..., N) moment respectively.

(4) Abnormal actuator control

The abnormal control of the actuator of an aircraft refers to the deviation of the actuator displacement (aileron displacement dtx, rudder displacement dty, elevator displacement dtz, flaperon displacement dtjy, and front wheel displacement dtw) from the actuator command (aileron command dtxc, rudder command dtyc, elevator command dtzc, flaperon command dtjyc, and front wheel command dtwc). The deviation mean of the actuator commands at N time points (E(Δdtx), E(Δdty), E(Δdtz), E(Δdtjy), and E(Δdtw)) is calculated as the characteristic parameter of the actuator control abnormality, that is,

Wherein, dtx _i , dty _i , dtz _i , dtjy _i , dtw _i represent the measured values of aileron displacement, rudder displacement, elevator displacement, flaperon displacement, and nosewheel displacement at the i-th (i=1, 2, ..., N) time point, respectively; dtxc _i , dtyc _i , dtzc _i , dtjyc _i , dtwc _i represent the given values of aileron command, rudder command, elevator command, flaperon command, and nosewheel command at the i-th (i=1, 2, ..., N) time point, respectively;

Step 2: Select the feature parameter with the greatest abnormal impact from the abnormal feature parameter set S and select the data point closest to the mean as the root node of the tree.

Step 3: Set the abnormal feature parameter value that is less than the current split point as the left node Left of the binary tree, and set the abnormal feature parameter value that is greater than or equal to the current split point as the right node Right of the binary tree.

Step 4: Repeat steps 9 and 10 for nodes Left and Right to recursively construct the tree until only one data cannot be constructed further or the tree has reached a limited height, then stop constructing.

Step 5: Calculate the anomaly score s(x,n) of sample x:

Where h(x) is the path length of the tree where sample x is located, and c(n) is the average path length of each tree ^[197] :

Where: H(n-1) = ln(n-1) + ζ (ζ is the Euler constant, generally 0.58). For data samples x whose abnormal score s(x,n) is close to 1, they are abnormal data points.

6) Output abnormal flight status time series data.

Specifically, the abnormal flight state time series in the aviation aircraft flight data set X = {X ₁ , X ₂ , ..., X _t } is output.

The intelligent identification method of abnormal flight status of an aircraft of the present invention can identify the abnormal flight status of an aircraft, provide decision support for real-time control decision-making and health management of the aircraft, facilitate aircraft support personnel to find the cause of abnormal events, avoid the occurrence of flight accidents, and ensure flight safety.

BRIEF DESCRIPTION OF THE DRAWINGS

FIG1 is a flow chart of a method for intelligently identifying abnormal flight status of an aircraft.

DETAILED DESCRIPTION

The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

Example 1

A method for intelligently identifying abnormal flight status of an aircraft in this embodiment, as shown in FIG1 , includes the following steps:

1) Aviation flight experts select historical standard flight data of aviation aircraft and mark the flight phases thereof;

Specifically, flight experts select historical standard flight data and mark the six basic flight stages of takeoff, climb, level flight, turn, descent and landing. The marked flight data set is recorded as D = {D ₁ ,D ₂ ,......,D ₆ };

2) Using the labeled data set D with labeled flight stages, we design and optimize the classification regression decision tree CART algorithm to build a flight stage classification model M;

Specifically:

之间的高度变化率Δh、俯仰角变化率Δα、滚转角变化率Δβ、偏航角变化率Δγ、航向角变化率Δδ，计算数据集D的基尼系数公式(1)Step 1: For the standard action template data D = {D ₁ , D ₂ , ..., D ₆ }, calculate the n-element feature parameter vector of the standard flight state D _r at the u-th time point

The height change rate Δh, pitch angle change rate Δα, roll angle change rate Δβ, yaw angle change rate Δγ, and heading angle change rate Δδ are used to calculate the Gini coefficient formula of data set D (1):

In the formula, _pk represents the proportion of the kth category in the data set D, which is simpler than the method of calculating the logarithm using the information gain rate. For the data set D, the Gini coefficient formula for calculating the feature parameter c is (2):

Where |D| represents the number of all time series vectors in the data set D, E represents the total number of value categories of the feature parameter c in the data set D, |D ^e | represents the total number of vectors in the data set D where the feature c value is e, and Gini(D ^e ) is the Gini coefficient calculated according to formula (1).

Step 2: Data set D is divided into two parts according to the characteristic parameter C with the smallest Gini coefficient value to obtain its left and right nodes, denoted as D _left and D _right ;

Step 3: There are two conditions for stopping the construction of the decision tree. One is to determine the relationship between the threshold and the current Gini coefficient. If the current Gini coefficient is less than the threshold, the current node stops building the decision tree subtree, otherwise it continues to build. The second is to check whether there is a classification feature in the feature parameter set D that can continue to decompose the set. If not found, stop building the decision tree.

Step 4: Recursively execute steps 1 to 4 for the left and right child nodes D _left and D _right of the D node until the condition of step 3 is met and the decision tree training model Tree is returned;

Step 5: To avoid the overfitting problem of the decision tree model, prune the decision tree. Pruning means deleting the left and right child nodes of the non-leaf node T _i corresponding to the α _i with the smallest surface error rate gain value among the non-leaf nodes {T ₁ ,T ₂ ,T ₃ ,......,T _n }. The calculation formula (3) of α _i is:

Where R(i) represents the error generated after the leaf node replaces the i-th non-leaf node, and the calculation formula is:

R(i)＝r(i)p(i), (4)

Where r(i) represents the error rate of node i, p(i) represents the percentage of samples on node i to the total number of samples in the training set, and R(T _i ) represents the sum of the errors of all leaf nodes on subtree T _i when node i is not pruned, that is,

Replace the non-leaf node _Ti with the leaf node of the subtree, repeat the process until there is no non-leaf node to replace, pruning is completed, and return to the decision tree Tree;

Step 6: Set the target loss function (accuracy, precision, recall or _F1 score), place all the hyperparameters of the decision tree in the network, and use the grid search algorithm to adjust the decision tree parameters to make the decision tree model M optimally classified. At the same time, in order to avoid the impact of the division of the training set and the test set, use data cross-validation to improve the classification performance of the decision tree model M;

3) Input flight data of abnormal flight status to be identified;

Specifically, the flight data set of an aircraft at t time points is input and recorded as X = {X ₁ , X ₂ , ..., X _t }

4) Use the constructed classification model M to divide the flight data set X into flight stages;

Specifically, the classification model M is used to automatically divide the flight data set X = {X ₁ , X ₂ , …, X _t } at t time points into flight phases, and the flight phases of each time period are marked;

5) Use the optimized isolation forest algorithm IF to identify abnormal flight status;

Specifically:

Step 1: For X with marked flight stages, calculate the abnormal characteristic parameter values of each flight stage according to Table 1, and the characteristic parameter set of each flight stage is recorded as S;

Taking the six most basic flight stages of takeoff, climb, level flight, turn, descent and landing as examples, each flight stage can be characterized by characteristic parameters such as flight speed, altitude, attitude angle and heading angle. At the same time, each flight stage corresponds to a different flight control law, and the main characteristic parameters of the flight stage are established, as shown in Table 1.

Table 1 Characteristic parameters of aircraft flight phases

The abnormal characteristic parameters in the table are classified into: abnormal attitude, abnormal speed, abnormal track and abnormal actuator control. The calculation methods of various parameters are as follows:

(1) Abnormal posture characteristic parameters

The abnormal flight attitude of an aircraft refers to the deviation of the measured values of the attitude angles (pitch angle θ, yaw angle ψ, roll angle φ) from the given values (given pitch angle θ_ref, given yaw angle ψ_ref, given roll angle φ_ref) for a long time. The mean deviation of each attitude angle at N time points (E(Δθ), E(Δψ) and E(Δφ)) is calculated as the characteristic parameter of the attitude abnormality, that is,

Among them, θ _i , ψ _i , φ _i respectively represent the measured values of the pitch angle, yaw angle and roll angle at the i-th (i=1, 2, ..., N) time point, and θ_ref _i , ψ_ref _i , φ_ref _i respectively represent the given values of the pitch angle, yaw angle and roll angle at the i-th (i=1, 2, ..., N) time point.

(2) Speed anomaly characteristic parameters

The speed anomaly of an aircraft refers to the fact that the measured value V of the flight speed deviates from the given speed value V_ref for a long time. The mean deviation of the speed at N time points E(ΔV) is calculated as the speed anomaly characteristic parameter, that is,

Wherein, V _i and V_ref _i represent the measured value and given value of the flight speed of the aircraft at the i-th (i=1, 2, ..., N) time point, respectively.

(3) Track anomaly characteristic parameters

The abnormal flight path of an aircraft refers to the deviation of the actual flight route from the set route. The route is composed of several waypoints. The mean distance deviation E(Δd) of the waypoints at N time points is calculated as the characteristic parameter of the abnormal flight path, that is,

Among them, the measured values and given values of the longitude, latitude and altitude of the waypoint position at the i-th (i=1, 2, ..., N) moment are represented as _Pi ( _xi , _yi , _hi ) and _{P_refi} ( _{x_refi} , _{y_refi} , _{h_refi} ), respectively.

Calculate the mean value of the heading angle deviation E(ΔPSI) of the waypoints at N time points as the track anomaly characteristic parameter, that is,

Wherein, PSI _i and PSI_ref _i represent the measured value and given value of the heading angle at the i-th (i=1, 2, ..., N) moment respectively.

(4) Abnormal actuator control

The abnormal control of the actuator of an aircraft refers to the deviation of the actuator displacement (aileron displacement dtx, rudder displacement dty, elevator displacement dtz, flaperon displacement dtjy, and front wheel displacement dtw) from the actuator command (aileron command dtxc, rudder command dtyc, elevator command dtzc, flaperon command dtjyc, and front wheel command dtwc). The deviation mean of the actuator commands at N time points (E(Δdtx), E(Δdty), E(Δdtz), E(Δdtjy), and E(Δdtw)) is calculated as the characteristic parameter of the actuator control abnormality, that is,

Wherein, dtx _i , dty _i , dtz _i , dtjy _i , dtw _i represent the measured values of aileron displacement, rudder displacement, elevator displacement, flaperon displacement, and nosewheel displacement at the i-th (i=1, 2, ..., N) time point, respectively; dtxc _i , dtyc _i , dtzc _i , dtjyc _i , dtwc _i represent the given values of aileron command, rudder command, elevator command, flaperon command, and nosewheel command at the i-th (i=1, 2, ..., N) time point, respectively;

Step 2: Select the feature parameter with the greatest abnormal impact from the abnormal feature parameter set S and select the data point closest to the mean as the root node of the tree.

Step 3: Set the abnormal feature parameter value that is less than the current split point as the left node Left of the binary tree, and set the abnormal feature parameter value that is greater than or equal to the current split point as the right node Right of the binary tree.

Step 4: Nodes Left and Right repeat steps 2 and 3 to recursively construct the tree until only one data cannot be constructed further, or the tree has reached a limited height, and then stop constructing.

Step 5: Calculate the anomaly score s(x,n) of sample x:

Where: h(x) is the path length of the tree where sample x is located, and c(n) is the average path length of each tree:

Where: H(n-1) = ln(n-1) + ζ (ζ is the Euler constant, generally 0.58). For data samples x whose abnormal score s(x,n) is close to 1, they are abnormal data points.

6) Output abnormal flight status time series data.

Specifically, the abnormal flight state time series in the aviation aircraft flight data set X = {X ₁ , X ₂ , ..., X _t } is output.

Although embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions and variations may be made to the embodiments without departing from the principles and spirit of the present invention, and that the scope of the present invention is defined by the appended claims and their equivalents.

Claims (3)

1. A method for intelligently identifying abnormal flight status of an aircraft, characterized by comprising the following steps: 1) Aviation flight experts select historical standard flight data of aviation aircraft and mark the flight phases thereof; Specifically, flight experts select historical standard flight data and mark the six basic flight stages of takeoff, climb, level flight, turn, descent and landing. The marked flight data set is recorded as D = {D ₁ ,D ₂ ,......,D ₆ }; 2) Using the labeled data set D with labeled flight stages, we design and optimize the classification regression decision tree algorithm to construct the flight stage classification model M; 3) Input flight data of abnormal flight status to be identified; Specifically, the flight data set of an aircraft at t time points is input and recorded as X = {X ₁ , X ₂ , ..., X _t }; 4) Use the constructed classification model M to divide the flight data set X into flight stages; Specifically, the classification model M is used to automatically divide the flight data set X = {X ₁ , X ₂ , …, X _t } at t time points into flight phases, and the flight phases of each time period are marked; 5) Use the optimized isolation forest algorithm to identify abnormal flight status; 6) Output abnormal flight status time series data; Specifically, the abnormal flight state time series in the aviation aircraft flight data set X = {X ₁ , X ₂ , ..., X _t } is output. 2. According to the method for intelligently identifying abnormal flight status of an aircraft according to claim 1, it is characterized in that step 2) specifically comprises the following steps: Step 1: For the standard action template data D = {D ₁ , D ₂ , ..., D ₆ }, calculate the n-element feature parameter vector of the standard flight state D _r at the u-th time point

The height change rate Δh, pitch angle change rate Δα, roll angle change rate Δβ, yaw angle change rate Δγ, and heading angle change rate Δδ are used to calculate the Gini coefficient formula of data set D (1):

In the formula, _pk represents the proportion of the kth category in the data set D, which is simpler than the method of calculating the logarithm using the information gain rate. For the data set D, the Gini coefficient formula for calculating the feature parameter c is (2):

Where |D| represents the number of all time series vectors in the data set D, E represents the total number of value categories of the feature parameter c in the data set D, |D ^e | represents the total number of vectors in the data set D where the feature c value is e, and Gini(D ^e ) is the Gini coefficient calculated according to formula (1); Step 2: Data set D is divided into two parts according to the characteristic parameter C with the smallest Gini coefficient value to obtain its left and right nodes, denoted as D _left and D _right ; Step 3: There are two conditions for stopping the construction of the decision tree. One is to determine the relationship between the threshold and the current Gini coefficient. If the current Gini coefficient is less than the threshold, the current node stops building the decision tree subtree, otherwise it continues to build. The second is to check whether there is a classification feature in the feature parameter set D that can continue to decompose the set. If not found, stop building the decision tree. Step 4: Recursively execute steps 1 to 4 for the left and right child nodes D _left and D _right of the D node until the condition of step 3 is met and the decision tree training model M is returned. Step 5: To avoid the overfitting problem of the decision tree model, the decision tree is pruned. The so-called pruning is to delete the left and right child nodes of the non-leaf node T _i corresponding to the α _i with the smallest surface error rate gain value among the non-leaf nodes {T ₁ ,T ₂ ,T ₃ ,......,T _n }. The calculation formula (3) of α _i is:

Where R(i) represents the error generated after the leaf node replaces the i-th non-leaf node, and the calculation formula is: R(i)＝r(i)p(i), (4) Where r(i) represents the error rate of node i, p(i) represents the percentage of samples on node i to the total number of samples in the training set, and R(T _i ) represents the sum of the errors of all leaf nodes on subtree T _i when node i is not pruned, that is,

Replace the non-leaf node _Ti with the leaf node of the subtree, repeat the process until there is no non-leaf node to replace, pruning is completed, and return to the decision tree M; Step 6: Set the target loss function, place all the hyperparameters of the decision tree in the network, and use the grid search algorithm to adjust the decision tree parameters to optimize the classification of the decision tree model M. At the same time, in order to avoid the impact of the division of the training set and the test set, the data cross-validation method is used to improve the classification performance of the decision tree model M. 3. The method for intelligently identifying abnormal flight status of an aircraft according to claim 1, wherein step 5) specifically comprises the following steps: Step 1: For X with marked flight stages, calculate the abnormal characteristic parameter values of each flight stage according to Table 1, and the characteristic parameter set of each flight stage is recorded as S; Taking the six most basic flight stages of takeoff, climb, level flight, turn, descent and landing as examples, each flight stage can be characterized by characteristic parameters such as flight speed, altitude, attitude angle and heading angle. At the same time, each flight stage corresponds to a different flight control law, and the abnormal characteristic parameters of the flight stage are established; Among them, the abnormal characteristic parameters are classified into: abnormal attitude, abnormal speed, abnormal track and abnormal actuator control. The calculation methods of various parameters are as follows: (1) Posture abnormality characteristic parameters The abnormal flight attitude of an aircraft refers to the deviation of the measured values of the attitude angles (pitch angle θ, yaw angle ψ, roll angle φ) from the given values (given pitch angle θ_ref, given yaw angle ψ_ref, given roll angle φ_ref) for a long time. The mean deviation of each attitude angle at N time points (E(Δθ), E(Δψ) and E(Δφ)) is calculated as the characteristic parameter of the attitude abnormality, that is,

Wherein, θ _i , ψ _i , φ _i represent the measured values of the pitch angle, yaw angle and roll angle at the i-th (i=1, 2, ..., N) time point, respectively, and θ_ref _i , ψ_ref _i , φ_ref _i represent the given values of the pitch angle, yaw angle and roll angle at the i-th (i=1, 2, ..., N) time point, respectively; (2) Speed anomaly characteristic parameters The speed anomaly of an aircraft refers to the fact that the measured value V of the flight speed deviates from the given speed value V_ref for a long time. The mean deviation of the speed at N time points E(ΔV) is calculated as the speed anomaly characteristic parameter, that is,

Wherein, V _i and V_ref _i represent the measured value and given value of the flight speed of the aircraft at the i-th (i=1, 2, ..., N) time point, respectively; (3) Track anomaly characteristic parameters The abnormal flight path of an aircraft refers to the deviation of the actual flight route from the set route. The route is composed of several waypoints. The mean distance deviation E(Δd) of the waypoints at N time points is calculated as the characteristic parameter of the abnormal flight path, that is,

Wherein, the measured values and given values of the longitude, latitude and altitude of the waypoint position at the i-th (i=1, 2, ..., N) moment are expressed as _Pi ( _xi , _yi , _hi ) and _{P_refi} ( _{x_refi} , _{y_refi} , _{h_refi} ), respectively; Calculate the mean value of the heading angle deviation E(ΔPSI) of the waypoints at N time points as the track anomaly characteristic parameter, that is,

Wherein, PSI _i and PSI_ref _i represent the measured value and given value of the heading angle at the i-th (i=1, 2, ..., N) moment respectively; (4) Abnormal actuator control The abnormal control of the actuator of an aircraft refers to the deviation of the actuator displacement (aileron displacement dtx, rudder displacement dty, elevator displacement dtz, flaperon displacement dtjy, and front wheel displacement dtw) from the actuator command (aileron command dtxc, rudder command dtyc, elevator command dtzc, flaperon command dtjyc, and front wheel command dtwc). The deviation mean of the actuator commands at N time points (E(Δdtx), E(Δdty), E(Δdtz), E(Δdtjy), and E(Δdtw)) is calculated as the characteristic parameter of the actuator control abnormality, that is,

Wherein, dtx _i , dty _i , dtz _i , dtjy _i , dtw _i represent the measured values of aileron displacement, rudder displacement, elevator displacement, flaperon displacement, and nosewheel displacement at the i-th (i=1, 2, ..., N) time point, respectively; dtxc _i , dtyc _i , dtzc _i , dtjyc _i , dtwc _i represent the given values of aileron command, rudder command, elevator command, flaperon command, and nosewheel command at the i-th (i=1, 2, ..., N) time point, respectively; Step 2: Select the feature parameter with the greatest abnormal impact from the abnormal feature parameter set S according to the flight expert experience and select the data point closest to the mean as the root node of the tree; Step 3: Set the abnormal feature parameter value that is less than the current split point as the left node of the binary tree, and set the abnormal feature parameter value that is greater than or equal to the current split point as the right node of the binary tree; Step 4: Repeat steps 2 and 3 for nodes Left and Right to recursively construct the tree until only one data cannot be constructed further, or the tree has reached the specified height, and then stop constructing; Step 5: Calculate the anomaly score s(x,n) of sample x:

Where: h(x) is the path length of the tree where sample x is located, and c(n) is the average path length of each tree:

In the formula: H(n-1)=ln(n-1)+ζ (ζ is the Euler constant, with a value of 0.58). The data sample x whose anomaly score s(x,n) is close to 1 is an abnormal data point.

Images