A flight safety risk assessment modeling and analysis method

Abstract

The present application relates to the technical field of flight data analysis, and particularly relates to a flight safety risk assessment modeling and analysis method. The technical scheme comprises constructing a concept framework, spatially decomposing flight parameters, analyzing the causal correlation of observed state variables, constructing a Bayesian network to quantize the coupling relationship between observed state variables, determining the spatiotemporal correlation relationship between hidden states and observed state variables, and an evaluation algorithm based on hidden states. The present application quantizes the multi-dimensional coupling relationship of flight parameters by integrating the spatial three-dimensional, time dimension and causal dimension of the model, and quantizes the multi-dimensional coupling relationship of flight parameters by combining the Bayesian network, significantly improving the comprehensiveness and accuracy of flight safety risk assessment. The hidden state is dynamically predicted and optimized based on the Viterbi algorithm, real-time adjustment of flight operation is realized, risk level evaluation and optimization suggestions are output, and the initiative and scientificity of flight safety management are comprehensively improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of flight data analysis, and particularly relates to a flight safety risk assessment modeling and analysis method. BACKGROUND

[0002] Flight attitude is the core state information of an aircraft in the flight process, including flight speed parameters and attitude angle parameters such as pitch angle, roll angle and yaw angle. The dynamic changes of these flight attitude parameters are affected by multiple factors such as spatial position, time evolution, flight operation, aircraft performance and external environment. In the prior art, dynamic equation models are usually used to model and control flight attitude, but these models usually lack the causal, spatiotemporal coupling relationship between flight attitude, flight operation and other flight parameters, and it is difficult to effectively analyze attitude abnormalities and multi-factor influences under complex conditions, and the close relationship between hidden states (flight attitude stability) and observed state variables (flight parameters) is not clearly quantified.

[0003] Therefore, there is an urgent need for a flight safety risk assessment modeling and analysis method of "causal spatiotemporal-five-dimensional-two-state", which can comprehensively represent the change mechanism of flight attitude in the dimensions of time, space and causal factors, and provide new ideas for flight safety risk assessment, anomaly detection and flight control optimization. SUMMARY

[0004] The purpose of the present application is to solve the problem in the background art that the existing models usually lack the causal, spatiotemporal coupling relationship between flight attitude, flight operation and other flight parameters, and it is difficult to effectively analyze attitude abnormalities and multi-factor influences under complex conditions, and a flight safety risk assessment modeling and analysis method is proposed.

[0005] The technical scheme of the present application: a flight safety risk assessment modeling and analysis method, comprising the following steps:

[0006] A concept framework is constructed to define the dimensional information including spatial three dimensions, time dimension and causal dimension, and the relationship between observed state variables and hidden states;

[0007] The flight parameters are spatially decomposed, and the three-dimensional components of the flight attitude class, flight operation class, flight environment class and aircraft performance class variables are calculated based on the body coordinate system and the inertial coordinate system;

[0008] The causal relationship of the observed state variables is analyzed, and a causal influence model of the flight operation, flight environment and aircraft performance class parameters on the flight attitude parameters is established;

[0009] A Bayesian network is constructed to quantify the coupling relationship between the observed state variables, and the quantitative analysis of the variable correlation is realized through discretization, network structure modeling and probability reasoning;

[0010] Determine the spatio-temporal correlation between the hidden state and the observation state variable, and based on the stability evaluation criterion of the flight phase division, screen the associated observation state variable related to the hidden state;

[0011] Based on the hidden state evaluation algorithm, predict the hidden state sequence combined with the observation state variable sequence, and dynamically optimize the flight attitude stability according to the prediction result.

[0012] Optionally, the hidden state evaluation algorithm is a Viterbi algorithm, comprising:

[0013] Based on the emission matrix and the observation state variable sequence, recursively calculate the optimal path probability of the hidden state, backtrack to generate the hidden state sequence, and optimize the low-probability hidden state by adjusting the observation state variable.

[0014] Optionally, the spatial decomposition of the flight parameter includes: calculating the body speed component of the airspeed based on the body coordinate system, and converting to the ground speed component in the inertial coordinate system through the rotation matrix corresponding to the pitch angle, roll angle and yaw angle.

[0015] Optionally, in the causal influence model:

[0016] The flight operation instruction includes pitch operation, roll operation, yaw operation and throttle instruction, which respectively corresponds to the control of pitch angle, roll angle, yaw angle and airspeed;

[0017] The wind speed component of the flight environment affects the ground speed component through the inertial coordinate system;

[0018] The aircraft weight affects the attitude angle and airspeed through the center of mass distribution.

[0019] Optionally, the construction of the Bayesian network comprises:

[0020] Taking the flight operation, flight environment and aircraft performance parameters as the root nodes, and the flight attitude parameters as the leaf nodes, the continuous variables are interval discretized, and the node dependency relationship and conditional probability table are generated based on the causal correlation.

[0021] Optionally, the hidden state is the flight attitude stability, and the associated observation state variable thereof in the take-off phase includes:

[0022] The ground speed, airspeed and yaw angle at the moment of lifting the nose wheel, and the pitch angle and ground speed at the moment of the main wheel leaving the ground.

[0023] Optionally, the observation state variable includes flight speed, attitude angle, flight operation instruction, wind speed component and aircraft weight, and the spatial decomposition is based on a three-dimensional coordinate system, and the causal correlation analysis covers the coupling of spatio-temporal and multi-dimensional parameters.

[0024] Compared with the prior art, the present application has at least one of the following beneficial technical effects:

[0025] By integrating the three-dimensional space, time dimension and causal dimension through the "five-dimensional two-state" model, a full-scale dynamic analysis of flight attitude parameters is achieved, effectively revealing the multi-factor coupling mechanism, and significantly improving the comprehensiveness and accuracy of risk assessment.

[0026] Based on the Viterbi algorithm and emission matrix, the hidden state sequence is predicted in real time, and the flight operation is dynamically adjusted in combination with the time and space dimensions to optimize flight posture anomalies and reduce safety risks.

[0027] Bayesian networks are used to quantitatively model the causal relationships between flight operations, environment, performance parameters and attitude variables, enhancing the interpretability of the model and supporting data-driven risk assessment and decision optimization.

[0028] Divide the hidden state evaluation criteria according to the flight phase, screen key observation state variables in a targeted manner, and improve the evaluation efficiency and pertinence in different scenarios.

[0029] The present invention integrates the three-dimensional spatial dimension, the temporal dimension, and the causal dimension through a model, and combines the Bayesian network to quantitatively model the multi-dimensional coupling relationship of flight parameters, significantly improving the comprehensiveness and accuracy of flight safety risk assessment. Based on the Viterbi algorithm, it dynamically predicts and optimizes the latent state to achieve real-time adjustment of flight operations; outputs risk level assessment and optimization suggestions, and comprehensively improves the initiative and scientific nature of flight safety management. BRIEF DESCRIPTION OF THE DRAWINGS

[0030] Figure 1 This is a conceptual diagram of the "Causal Spacetime-Five-Dimensional Two-State" model;

[0031] Figure 2 It is a model diagram of the “causal space-time” association between observed variables and hidden states;

[0032] Figure 3 is the causal relationship model diagram of the observed variables;

[0033] Figure 4 Quantify the flow chart for observed state variables;

[0034] Figure 5 This is a framework diagram of flight safety risk assessment based on the “causal space-time-five-dimensional two-state” model. DETAILED DESCRIPTION

[0035] The technical solution of the present invention is further described below with reference to the accompanying drawings and specific embodiments.

[0036] Example

[0037] The present invention proposes a flight safety risk assessment modeling and analysis method, which is described in detail below.

[0038] Step one: Conceptual framework elaboration: Elaborate the expected goals of model establishment, the components of the model, the logical framework, and the conceptual understanding therein, respectively.

[0039] Model achieves expected goals: Through the model based on "causal space-time-five-dimensional two-state", the evaluation of flight attitude stability with time series is realized, and flight operation adjustment can be made on the flight attitude with poor evaluation results in a certain space-time, so as to make the flight attitude stability state optimal.

[0040] "Five-dimensional two-state" refers to the five dimensions and two states involved in building the model. The five dimensions include spatial three dimensions, time dimension and causal dimension, which represent the position of the object in three-dimensional space, the change process of time and the causal relationship between variables. The two states are "observed state variables" and "hidden states" in the system, wherein the observed state variables refer to specific state variables that can be measured or observed, while the hidden states are potential states that cannot be directly observed and can only be understood through indirect inference or reasoning prediction. According to the above analysis, the "causal space-time-five-dimensional two-state" model conceptual diagram is shown in Figure 1 .

[0041] According to the above understanding, the variables of the system during the flight of the aircraft can be divided into "observed state variables" and "hidden states". The observed state variables refer to those variables that can be directly perceived and recorded by instruments or other methods. The analysis of these variables involves spatial three dimensions and causal correlation, that is, we can extract useful information from the spatial distribution and mutual relationship of these data. The hidden state is a variable that cannot be directly observed or measured by instruments. This type of variable is usually potential and cannot be directly obtained. We can infer from the existing observation data and use the model to indirectly calculate or qualitatively evaluate these hidden states.

[0042] Therefore, the variables that can be observed during flight are called "observed state variables", such as flight speed, pitch angle, roll angle, yaw angle and flight operation. The variables that cannot be observed are called "hidden states", such as the flight attitude stability of the aircraft in a certain space-time. The observed state variables are analyzed using spatial three-dimensional decomposition and causal correlation, and the hidden states are analyzed using time dimension. Taking the flight parameters during the operation of the aircraft as an example, the "causal space-time" correlation model diagram of the observed state variables and hidden states is shown in Figure 2 .

[0043] Step two: Spatial decomposition of observed state variables: Establish the aircraft body coordinate system with the geometric center of the aircraft as the origin, and establish the inertial coordinate system fixed relative to the earth or space with the takeoff point of the aircraft as the origin. Four types of flight parameters are decomposed in space.

[0044] 1. Flight attitude class variable space decomposition: using attitude angle in the body coordinate system to decompose the speed into three body speed components, using the rotation matrix between two coordinate systems to convert the three body speed components into three fixed speed components. The calculation process of body speed component and inertial speed component is as follows:

[0045] Given the pitch angle , the roll angle , the yaw angle , and the airspeed V, the three speed component calculation formula of airspeed V in the body coordinate system is: The three body speed components are obtained:

[0046]

[0047] The three speed component calculation formula of airspeed V in the inertial coordinate system is:

[0048]

[0049] Where the total rotation matrix

[0050] is the pitch rotation matrix,

[0051] is the roll rotation matrix,

[0052] is the yaw rotation matrix, .

[0053] 2. Flight operation class variable space decomposition: flight operation includes attitude angle control and throttle lever operation. According to the decomposition of flight attitude, attitude angle operation includes pitch operation, roll operation and yaw operation.

[0054] 3. Flight environment class variable space decomposition: flight environment includes wind speed and wind direction, wind speed can be decomposed into three wind speed components according to the inertial coordinate system, which are and .

[0055] 4. Aircraft performance class variable space decomposition: aircraft performance mainly includes aircraft weight, the overall weight of the aircraft is replaced by the center of mass of the aircraft, that is, the center of the body coordinate system, which is evenly distributed in each direction along the space coordinate axis.

[0056] Finally, the observation state variable space decomposition result is shown in Table 1.

[0057] ​​​​Table 1: Decomposition of the state variable space

[0058]

[0059] Step three: causal relationship analysis of the state variable: the causal influence of flight operation, flight environment and aircraft performance parameters on the attitude angle and flight speed is added to the decomposition calculation process, and the causal relationship between the variables and their influencing factors is established.

[0060] Flight speed and attitude angle are key variables of flight attitude parameters, and attitude angle includes pitch angle, roll angle and yaw angle. In spatial transformation, flight speed can be decomposed into three independent components: , and In the same space-time dimension, flight attitude parameters are influenced by flight operation parameters, flight environment parameters and aircraft performance parameters. The three attitude angles are controlled by pitch command, roll operation and yaw command in flight operation; and , and are adjusted by throttle command in flight operation. Overall, flight operation variables, flight environment variables and aircraft performance parameters will have an impact on flight attitude variables. After three-dimensional spatial decomposition of the four types of variables, the causal relationship between the variables needs to be analyzed in detail.

[0061] Pitch operation in flight operation affects pitch angle in flight attitude, roll operation in flight operation affects roll angle in flight attitude, yaw operation in flight operation affects yaw angle in flight attitude, and throttle command in flight operation affects flight speed in flight attitude.

[0062] will affect in flight attitude, in flight environment will affect in flight attitude, in flight environment will affect in flight attitude.

[0063] Aircraft weight in aircraft performance will affect variables in each spatial coordinate axis, that is, aircraft weight affects pitch angle , roll angle , yaw angle , and airspeed V.

[0064] The relationship between the decomposed flight attitude variables is analyzed: pitch angle affects , and ​​​Roll angle Influence , and Yaw angle Influence , and ; Influence , and ; Influence

[0065] , and ; Influence , and .

[0066] Step four: quantitative analysis of observation state variables: the observation state causal correlation model is constructed by using Bayesian network to quantitatively study the coupling mechanism between observation state variables.

[0067] 1. Determine the problem and the goal: the risk assessment is carried out for the flight attitude variable, and the network structure diagram is established with the flight operation variable, the flight environment variable and the aircraft performance variable as the root nodes and the flight attitude variable as the leaf nodes.

[0068] 2. Determine the variable: the random variable in the problem is determined, and the random variable is discretized to determine the possible node state. The numerical variable is divided into three intervals according to the numerical value, and the attribute variable is divided into two or three intervals according to the type. The division results are shown in Table 2.

[0069] Table 2: interval division of observation state variables

[0070]

[0071] 3. Determine the dependence relationship between variables: according to the observation state variable space decomposition results and the observation state variable causal correlation analysis, the correlation relationship with an arrow direction is established between the variables, from one variable to another variable, which indicates that the change of the starting variable will affect or determine the change of the arriving variable, and the arriving variable depends on the starting variable.

[0072] 4. Build network structure diagram: according to the determined variables and the mutual dependence relationship, the network structure diagram of the observation state variable is drawn, and the established observation state variable causal correlation model diagram is shown in Figure 3 .

[0073] 5. Specify conditional probability table: determine the conditional probability table for each network node. For nodes without parent nodes, use the prior probability P(X). For nodes with parent nodes, use the conditional probability P(X|Parents(X)).

[0074] 6. Inference and validation: divide the actual data into two groups, one group is used to train the network model, and the other group is used to validate the network model.

[0075] The final observation state variable quantization flow chart is shown in Figure 4 .

[0076] Step five: determination of the "causal spatio-temporal" relationship between hidden state and observation state variable: hidden state depends on one or more observation state variables, and different stages of flight attitude stability have different evaluation criteria. For different stages of hidden state, determine the related observation state variables.

[0077] In different stages of aircraft taxiing, taking off, climbing, cruising, descending, approaching and landing, there are different observation state variables that need to be concerned. The observation state variables that need to be concerned at different time points of a flight stage are not completely the same. To evaluate the flight attitude stability of a stage, the key observation state variables at different time points of the stage need to be analyzed one by one. Taking the aircraft take-off stage as an example, according to the flight quality monitoring related regulations (see Table 3), the following analysis is made.

[0078] Table 3: Airplane flight quality monitoring items and requirements

[0079]

[0080] According to the flight quality monitoring related regulations, in the take-off stage, the monitoring points include the moment of lifting the front wheel and the moment of the main wheel leaving the ground, so the main time points of the flight attitude stability hidden state include the moment of lifting the front wheel t VR and the moment of the main wheel leaving the ground t V2 . Then analyze the observation state variables corresponding to the monitoring items at different monitoring points, and summarize and arrange as shown in Table 4.

[0081] Table 4: Observation state variables representing hidden state

[0082]

[0083] Step six: hidden state evaluation: establish the emission matrix from hidden state to observation state variable, use the Viterbi algorithm to realize the prediction of observation state variable to hidden variable, adjust the observation state variable under the same space-time for the hidden state with poor prediction result, and realize the optimization of hidden state.

[0084] In the case of multiple observation variable sequences, the Viterbi algorithm is used for different observation variable sequences to find the respective probability maximum hidden state sequence, and finally the optimal hidden state sequence of multiple observation sequences is obtained by using weighted average method.

[0085] Objective of Viterbi algorithm: Given an observation sequence and parameters λ=(Π,A,B), where Π is the initial state probability, A is the state transition matrix, and B is the emission matrix, find the most likely hidden state sequence such that the probability of the hidden state sequence is maximum given the observation sequence O: .

[0086] Steps of Viterbi algorithm: The core of the Viterbi algorithm is to recursively calculate the probability of the optimal path of each hidden state at each time t, and get the optimal hidden state sequence by backtracking. Specifically, the Viterbi algorithm mainly calculates through the following steps:

[0087] 1. Define the recursive relationship: for each time t and hidden state , we define as the probability of the optimal path of state at time t. It represents the maximum probability from the starting state to time t and finally in hidden state i at time t. The recursive formula is:

[0088]

[0089] Where: is the probability of the optimal path of state at time t−1;

[0090] is the transition probability from state to state ;

[0091] is the emission probability of observation in state .

[0092] 2. Initialization: At the initial time t=1, the optimal path probability can be calculated by the initial state probability and the emission probability of the first observation : ;

[0093] 3. Record path: In order to be able to trace back the optimal path, the Viterbi algorithm also needs to record the arrival of a certain state at each time t The optimal forerunner state. The state at time t is When , the predecessor state of the optimal path is: ;

[0094] 4. Recursive calculation: At each time t=2,3,…,T, calculate according to the recursive formula and , until the final moment T is reached.

[0095] 5. Backtracking the optimal path: After the calculation is completed, the optimal state at the last moment T is obtained :

[0096] , then, by backtracking To get the entire optimal path: , this process starts from t=T, backtracks to t=1, and finally obtains the entire optimal hidden state sequence .

[0097] The hidden states with poor flight attitude stability in the optimal hidden state sequence are optimized, and the probability of flight attitude stability is gradually increased from low to high, ultimately determining the optimal flight operations to be taken under different space-time conditions. The final flight safety risk assessment framework based on the "Causal Space-Time-Five-Dimensional Two-State" model is as follows: Figure 5 shown.

[0098] In summary, the "Causal Space-Time-Five-Dimension Two-State" model integrates the three-dimensional space, time dimension and causal dimension, and combines the Bayesian network to quantitatively model the multidimensional coupling relationship of flight parameters, significantly improving the comprehensiveness and accuracy of flight safety risk assessment; based on the Viterbi algorithm, dynamic prediction and optimization of hidden states are carried out to achieve real-time adjustment of flight operations; at the same time, key observation state variables are screened according to different flight phases, and risk level assessment and optimization suggestions are output to enhance the targetedness and engineering practicality of the assessment, provide intelligent decision-making support for preventing safety hazards, and comprehensively improve the initiative and scientific nature of flight safety management.

[0099] The above specific embodiments are merely several optional embodiments of the present invention. Based on the technical solutions of the present invention and the relevant inspirations of the above embodiments, those skilled in the art may make various alternative improvements and combinations to the above specific embodiments.

Claims

1. A flight safety risk assessment modeling and analysis method, characterized in that: The following steps are involved: Construct a conceptual framework, define dimensional information including spatial, temporal, and causal dimensions, and the relationship between observed variables and latent states; Perform spatial decomposition of flight parameters and calculate the three-dimensional components of flight attitude, flight operation, flight environment, and aircraft performance variables based on the fuselage coordinate system and the inertial coordinate system. Analyze the causal relationship of observed variables and establish a causal impact model of flight operation, flight environment and aircraft performance parameters on flight attitude parameters; Construct a Bayesian network to quantify the coupling relationship between observed variables, and achieve quantitative analysis of variable associations through discretization, network structure modeling, and probabilistic reasoning; Determine the spatiotemporal relationship between the latent state and the observed state variables, and screen the associated observed state variables associated with the latent state based on the stability evaluation criteria for flight phase division. The latent state is flight attitude stability, and its associated observed state variables during takeoff include: ground speed, airspeed, and yaw angle at the time of nose wheel rotation, and pitch angle and ground speed at the time of main wheel liftoff. Based on the hidden state evaluation algorithm, the hidden state sequence is predicted in combination with the observed state variable sequence, and the flight attitude is dynamically optimized according to the prediction results.

2. The flight safety risk assessment modeling and analysis method according to claim 1, characterized in that: The hidden state evaluation algorithm is a Viterbi algorithm, including: Based on the emission matrix and the observation state variable sequence, the optimal path probability of the hidden state is recursively calculated, the hidden state sequence is backtracked, and the low-probability hidden state is optimized by adjusting the observation state variables.

3. The flight safety risk assessment modeling and analysis method according to claim 1, characterized in that: The spatial decomposition of the flight parameters includes: calculating the fuselage velocity component of the airspeed based on the fuselage coordinate system, and converting it to the ground speed component of the inertial coordinate system through the rotation matrix corresponding to the pitch angle, roll angle and yaw angle.

4. A flight safety risk assessment modeling and analysis method according to claim 1 or 2, characterized in that: In the causal influence model: Flight operation commands include pitch operation, roll operation, yaw operation and throttle command, which respectively control the pitch angle, roll angle, yaw angle and airspeed; The wind speed component of the flight environment affects the ground speed component through the inertial coordinate system; Aircraft weight distribution through the center of mass affects attitude angle and airspeed.

5. The flight safety risk assessment modeling and analysis method according to claim 1, characterized in that: The construction of the Bayesian network includes: Taking flight operation, flight environment and aircraft performance parameters as root nodes and flight attitude parameters as leaf nodes, the continuous variables are discretized into intervals, and the node dependency and conditional probability table are generated based on causal correlation.

6. The flight safety risk assessment modeling and analysis method according to claim 1, characterized in that: The observed variables include flight speed, attitude angle, flight operation instructions, wind speed components and aircraft weight. Their spatial decomposition is based on a three-dimensional coordinate system, and the causal correlation analysis covers time, space and multi-dimensional parameter coupling.

Images