Artificial intelligence-based flight control quality evaluation method and system

Abstract

The invention relates to the technical field of flight control quality evaluation, and discloses a flight control quality evaluation method and system based on artificial intelligence, and the method comprises the following steps: S101, obtaining multi-source flight data in real time; s102, extracting flight state characteristics, cognitive load characteristics of a pilot and environment state characteristics according to the multi-source flight data; s103, determining key control parameters and key flight state parameters, and constructing a space-time diagram; s104, recognizing the cognitive load state through a clustering algorithm, and dynamically adjusting the weight of a corresponding edge in the space-time diagram to generate a dynamic space-time diagram; s105, space-time sensitive features of the nodes are extracted, a comprehensive feature vector is obtained through calculation, and a flight control quality score is output; and S106, grading, and taking corresponding operation measures. According to the invention, real-time, accurate and dynamic flight control quality evaluation is realized, and flight safety and control quality evaluation scientificity are effectively improved.

Description

Technical Field

[0001] The present invention belongs to the technical field of flight handling quality assessment, and particularly relates to an artificial intelligence-based flight handling quality assessment method and system. Background Art

[0002] Flight handling quality assessment is a key link to ensure flight safety and improve pilots' operation levels. Traditional assessment methods mainly rely on the threshold monitoring of flight parameters and post-flight data analysis. During the flight, thresholds of key parameters are preset in advance. When a certain parameter exceeds the set range, the system records the event for post-flight analysis. This method has the following deficiencies:

[0003] 1. Lack of real-time performance: Traditional methods mainly analyze data after the flight and cannot provide immediate feedback during the flight, making it difficult to timely warn of potential risks; 2. Insufficient data utilization: Existing methods have limited comprehensive analysis capabilities for multi-source data and are difficult to comprehensively evaluate by combining control parameters, flight state parameters, physiological signals, and environmental factors, resulting in insufficient comprehensiveness and accuracy of the assessment results; 3. Low level of intelligence: Lack of intelligent data processing and analysis means, unable to automatically identify key influencing factors and provide optimization suggestions, which limits the initiative and adaptability of the assessment system. Summary of the Invention

[0004] The present invention provides an artificial intelligence-based flight handling quality assessment method and system to solve the technical problems of poor real-time performance, insufficient multi-source data fusion ability, and lack of intelligent analysis in related technologies.

[0005] The present invention provides an artificial intelligence-based flight handling quality assessment method and system, including the following steps:

[0006] S101, acquiring multi-source flight data, including control parameters, flight state parameters, pilots' physiological signals, and environmental state data;

[0007] S102, extracting flight state features, pilots' cognitive load features, and environmental state features according to the multi-source flight data;

[0008] S103, determining key control parameters affecting the flight state by using grey relational analysis based on historical flight data, and determining key flight state parameters based on the prior knowledge of aircraft dynamics. Defining the key control parameters and key flight state parameters as nodes, and constructing a spatio-temporal graph;

[0009] S104. Based on the cognitive load characteristics, use a clustering algorithm based on dynamic time warping to identify the cognitive load state. According to the cognitive load state, extract the key manipulation parameters and key flight state parameters from the preset node sensitivity database, and dynamically adjust the weights of the corresponding edges in the spatio-temporal graph to generate a dynamic spatio-temporal graph;

[0010] S105. Extract the spatio-temporal sensitive features of the nodes based on the graph attention mechanism introducing a time decay coefficient, fuse the flight state features, the cognitive load features of the pilot, and the environmental state features, obtain a comprehensive feature vector using a weighted fusion method, and output a flight handling quality score using a support vector regression algorithm;

[0011] S106. According to the flight handling quality score, divide it according to the preset scoring grade standard, and take corresponding operation measures based on the scoring grade.

[0012] Furthermore, the manipulation parameters include: joystick displacement, stick force, rudder deflection angle; the flight state parameters include: pitch angle, roll angle, yaw angle, control surface position, and attitude angular velocity; the physiological signals of the pilot include: heart rate, brain wave α / β ratio, and saccade count; the environmental state data include: wind speed, turbulence intensity level, air temperature, and air pressure.

[0013] Furthermore, extract the flight state features, the cognitive load features of the pilot, and the environmental state features from multi-source flight data, including:

[0014] The calculation formula for the flight state features is: where F s represents the flight state features, ΔX represents the change in joystick displacement, RMS(ΔX) represents the root mean square value of the change in joystick displacement, Δθ represents the change in attitude angular velocity, STD(Δθ) represents the standard deviation of the attitude angular velocity, ΔS represents the control surface position, that is, the control surface deflection angle, is used to measure the response efficiency of the control system, and w1, w2, and w3 respectively represent the first weight coefficient, the second weight coefficient, and the third weight coefficient;

[0015] The calculation formula for the cognitive load features is: where F c represents the cognitive load features, HR max , HR min and HR avg respectively represent the maximum value, minimum value, and average value of the heart rate, represents the brain wave α / β ratio, T eb represents the saccade count, T sacIt represents the fixation time, and w4, w5, and w6 respectively represent the fourth weight coefficient, the fifth weight coefficient, and the sixth weight coefficient;

[0016] The calculation formula for the environmental state characteristics is as follows: Among them, F e represents the environmental state characteristics, WS max , WS min and WS avg respectively represent the maximum value, the minimum value, and the average value of the wind speed, L tur represents the turbulence intensity level, ΔP represents the air pressure change amount, A avg represents the average flight altitude, ΔT represents the preset acquisition time window, and w7, w8, w9, and w 10 respectively represent the seventh weight coefficient, the eighth weight coefficient, the ninth weight coefficient, and the tenth weight coefficient.

[0017] Furthermore, the key control parameters affecting the flight state are determined by using grey relational analysis based on historical flight data. The specific steps include:

[0018] S201, Normalize the control parameter sequence and the flight state parameter sequence in the historical flight data;

[0019] S202, Calculate the grey relational coefficients of each control parameter sequence and the flight state parameter sequence;

[0020] S203, Calculate the grey relational degree of each control parameter, and screen out the control parameters whose grey relational degrees are greater than the first preset threshold as the key control parameters.

[0021] Furthermore, the key flight state parameters are determined based on the prior knowledge of aircraft dynamics. The specific steps include:

[0022] S301, Construct a transfer function from the control parameter to the flight state parameter according to the six-degree-of-freedom kinematic equation of the aircraft, and construct an edge between the control parameter and the flight state parameter;

[0023] S302, Integrate the amplitude-frequency response of the transfer function to obtain the correlation strength between the control parameter and the flight state parameter. If the correlation strength exceeds the preset threshold, then regard this flight state parameter as the key flight state parameter;

[0024] S303, Measure the dynamic response delay time of the control parameter to the flight state parameter through a step response experiment, and normalize it as the weight of the edge.

[0025] Further, the spatio-temporal graph includes: nodes, edges, and characteristics of the nodes. The nodes represent key manipulation parameters or key flight state parameters. The edges represent the association relationships between the nodes. The characteristics of the nodes are represented by a vector composed of key manipulation parameter values or key flight state parameter values collected within a first preset time window according to a first preset time window.

[0026] Further, based on the cognitive load characteristics, the specific steps of using a clustering algorithm based on dynamic time warping to identify the cognitive load state include:

[0027] S401, within a first preset time window, normalize the cognitive load characteristics, and use the dynamic time warping method to align the time series;

[0028] S402, use a clustering algorithm to perform clustering analysis on the aligned cognitive load characteristics, and automatically divide the cognitive load state categories of the pilot. Among them, the cognitive load state categories include: low load, medium load, high load, and extreme load;

[0029] S403, calculate the probability distribution of the cognitive load state categories within the current time window, and use the maximum probability principle to select the category with the highest occurrence frequency as the cognitive load state of the pilot in the current time window.

[0030] Further, according to the cognitive load state of the current time window, query the preset node sensitivity database, extract the key manipulation parameters and key flight state parameters in this cognitive load state, and adjust the weights of the edges of the spatio-temporal graph through a preset cognitive load influence coefficient to generate a dynamic spatio-temporal graph.

[0031] Further, based on the graph attention mechanism introducing a time decay coefficient, extract the spatio-temporal sensitive characteristics of the nodes. The calculation formula of the spatio-temporal sensitive characteristics is:

[0032]

[0033] Among them, represents the spatio-temporal sensitive characteristics of node u, N(u) represents the set of neighbor nodes of node u, γ represents the time decay coefficient, which is used to control the speed of time decay, X v,t represents the corresponding value of node v at the t-th moment, X v,t-1 represents the corresponding value of node v at the (t - 1)-th moment, W con represents the transformable weight matrix, h u represents the characteristics of node u, h k represents the characteristics of node k, h vdenote the features of node v, a denote the attention weight vector, || denote the concatenation operation, sigmoid denote the sigmoid activation function, and LeakyReLU denote the LeakyReLU activation function.

[0034] The present invention provides an artificial intelligence-based flight handling quality evaluation system, including:

[0035] A data acquisition module, configured to acquire multi-source flight data, including control parameters, flight state parameters, pilot physiological signals, and environmental state data;

[0036] A feature extraction module, configured to extract flight state features, pilot cognitive load features, and environmental state features according to the multi-source flight data;

[0037] A spatio-temporal graph construction module, configured to determine key control parameters affecting the flight state by using grey relational analysis according to historical flight data, and determine key flight state parameters based on prior knowledge of aircraft dynamics, define the key control parameters and flight state parameters as nodes, and construct a spatio-temporal graph;

[0038] A spatio-temporal graph dynamic adjustment module, based on the cognitive load features, uses a clustering algorithm based on dynamic time warping to identify the cognitive load state, and according to the cognitive load state, extracts key control parameters and key flight state parameters from a preset node sensitivity database, and dynamically adjusts the weights of the corresponding edges in the spatio-temporal graph to generate a dynamic spatio-temporal graph;

[0039] A handling quality scoring generation module, based on a graph attention mechanism introducing a time decay coefficient, extracts spatio-temporal sensitive features of nodes, and fuses flight state features, pilot cognitive load features, and environmental state features, obtains a comprehensive feature vector by using a weighted fusion method, and outputs a flight handling quality score by using a support vector regression algorithm;

[0040] A decision execution module, according to the flight handling quality score, divides it according to a preset scoring level standard, and executes corresponding operations based on the scoring level.

[0041] The beneficial effects of the present invention are as follows: The present invention adopts a multi-source data fusion technology, combines flight parameters, physiological signals, and environmental state data, and accurately extracts key features affecting flight handling quality through spatio-temporal graph modeling and graph attention mechanism; compared with the traditional threshold monitoring-based method, this method can evaluate handling quality in real time and dynamically, avoid the lag of post-event analysis, and improve the accuracy and timeliness of evaluation;

[0042] The present invention combines grey relational analysis with prior knowledge of dynamics to automatically screen key control parameters and key flight state parameters, improving the model's ability to adapt to different flight scenarios. At the same time, the cognitive load recognition based on the clustering algorithm can dynamically adjust the evaluation model, enabling it to adaptively optimize according to the changes in the pilot's state, and improving the adaptability and intelligence level of the system to different flight conditions. Brief Description of the Drawings

[0043] Figure 1 is a flowchart of a method for evaluating flight handling qualities based on artificial intelligence according to the present invention. Detailed Embodiments

[0044] Now, the subject matter described herein will be discussed with reference to exemplary embodiments. It should be understood that discussing these embodiments is only to enable those skilled in the art to better understand and thus implement the subject matter described herein. Without departing from the scope of protection of the content of this specification, changes can be made to the functions and arrangements of the elements discussed. Each example can omit, substitute, or add various processes or components as needed. Additionally, the features described relative to some examples can also be combined in other examples.

[0045] As Figure 1 shown, a method and system for evaluating flight handling qualities based on artificial intelligence includes the following steps:

[0046] S101, obtaining multi-source flight data in real time, including control parameters, flight state parameters, pilot physiological signals, and environmental state data;

[0047] S102, extracting flight state features, cognitive load features of the pilot, and environmental state features from the multi-source flight data;

[0048] S103, determining key control parameters affecting the flight state by grey relational analysis based on historical flight data, and determining key flight state parameters based on prior knowledge of aircraft dynamics. Defining the key control parameters and key flight state parameters as nodes, and constructing a spatio-temporal graph;

[0049] S104, based on the cognitive load features, using a clustering algorithm based on dynamic time warping to identify the cognitive load state, and according to the cognitive load state, extracting key control parameters and key flight state parameters from a preset node sensitivity database, and dynamically adjusting the weights of the corresponding edges in the spatio-temporal graph to generate a dynamic spatio-temporal graph;

[0050] S105, extracting the spatiotemporal sensitive features of nodes based on the graph attention mechanism with the time decay coefficient, integrating the flight state features, the pilot's cognitive load features and the environmental state features, obtaining a comprehensive feature vector by weighted fusion, and outputting the flight control quality score by using the support vector regression algorithm;

[0051] S106, according to the flight control quality score, the flight control quality is divided according to a preset score level standard, and corresponding operations are performed based on the score level.

[0052] In one embodiment of the present invention, the control parameters include: joystick displacement, stick force, rudder deflection angle; the flight status parameters include: pitch angle, roll angle, yaw angle, control surface position and attitude angular velocity; the pilot physiological signals include: heart rate, brain wave α / β ratio and number of eye saccades; the environmental status data include: wind speed, turbulence intensity level, temperature and air pressure.

[0053] In one embodiment of the present invention, the joystick displacement is obtained by real-time measurement through a joystick displacement sensor; the stick force is obtained by measuring the force applied by the pilot on the joystick through a force sensor; the rudder deflection angle is obtained by measuring the deflection angle of the rudder relative to the neutral position through a rudder angle sensor; the pitch angle, roll angle and yaw angle are calculated by an attitude reference system; the rudder position is represented by the deflection angle, which is measured by the rudder position sensor provided by the flight control system, in units of degrees; the attitude angular velocity is obtained by real-time measurement through an inertial navigation unit, in units of degrees / s; the heart rate is obtained by real-time measurement through an electrocardiogram sensor in the cockpit The wind speed is measured in times / minute. The turbulence intensity level is represented by real number coding, which is divided into level 1 to level 5 turbulence according to existing industry standards and is calculated in real time using airborne meteorological radar. The air temperature and air pressure are measured by airborne environmental sensors, with the temperature unit in °C and the air pressure unit in hPa.

[0054] In one embodiment of the present invention, the maximum and minimum normalization method is used to normalize the multi-source flight data, and the data of different dimensions and units are unified into a dimensionless data scale.

[0055] In one embodiment of the present invention, extracting flight state characteristics, pilot cognitive load characteristics and environmental state characteristics based on multi-source flight data includes:

[0056] The flight status characteristics are obtained by combining the change of joystick displacement, the change of attitude angular velocity and the position of the control surface, and aligning them for weighted fusion calculation;

[0057] Among them, the calculation formula for the flight state characteristics is as follows: Among them, F s represents the flight state characteristics, which are used to evaluate the handling stability and response characteristics. ΔX represents the change in the joystick displacement, RMS(ΔX) represents the root mean square value of the change in the joystick displacement, which measures the stability of flight control. Δθ represents the change in the attitude angular velocity, and STD(Δθ) represents the standard deviation of the attitude angular velocity, which measures the volatility of the flight attitude. ΔS represents the control surface position, that is, the control surface deflection angle, which is used to measure the response efficiency of the control system. w1, w2, and w3 respectively represent the first weight coefficient, the second weight coefficient, and the third weight coefficient, which are used to adjust the weights of handling stability, attitude volatility, and control response efficiency;

[0058] The cognitive load characteristics are obtained by fusing and calculating the pilot's heart rate, brain wave α / β ratio, and saccade count;

[0059] Among them, the calculation formula for the cognitive load characteristics is as follows: Among them, F c represents the cognitive load characteristics, which are used to evaluate the pilot's cognitive load state and judge the current physiological load level of the pilot. HR max 、HR min and HR avg respectively represent the maximum value, minimum value, and average value of the heart rate, represents the brain wave α / β ratio, which is used to reflect the cognitive load. T eb represents the saccade count, T sac represents the fixation time. w4, w5, and w6 respectively represent the fourth weight coefficient, the fifth weight coefficient, and the sixth weight coefficient, which are used to adjust the weights of physiological stress, cognitive load, and attention level;

[0060] The environmental state characteristics are obtained by fusing the wind speed change, air pressure change, and temperature fluctuation and based on a weighted calculation method;

[0061] Among them, the calculation formula for the environmental state characteristics is as follows: Among them, F e represents the environmental state characteristics, which are used to evaluate the influence of environmental factors on flight stability and pilot control load. WS max 、WS min and WS avg respectively represent the maximum value, minimum value, and average value of the wind speed. L tur represents the turbulence intensity level. ΔP represents the air pressure change amount. A avgrepresents the average flight altitude, ΔT represents the preset acquisition time window, w7, w8, w9, and w 10 respectively represent the seventh weight coefficient, the eighth weight coefficient, the ninth weight coefficient, and the tenth weight coefficient, which are respectively used to adjust the weight of the wind speed fluctuation degree, the disturbance situation of the flight environment, the meteorological change situation of the flight altitude, and the meteorological instability.

[0062] In an embodiment of the present invention, the historical flight data includes: a sequence of manipulation parameters and a sequence of flight state parameters. The key manipulation parameters affecting the flight state are determined by using grey relational analysis on the historical flight data. The specific steps include:

[0063] S201, perform normalization processing on the sequence of manipulation parameters and the sequence of flight state parameters in the historical flight data;

[0064] S202, calculate the grey relational coefficients of each sequence of manipulation parameters and the sequence of flight state parameters. Specifically, the grey relational coefficient is used to measure the influence of the manipulation parameter on the flight state parameter. By selecting a flight state parameter as the reference sequence, calculate the absolute difference between each sequence of joystick parameters and the reference sequence, and then calculate the relational coefficient;

[0065] S203, calculate the grey relational degree of each manipulation parameter, and screen out the manipulation parameters whose grey relational degree is greater than the first preset threshold as the key manipulation parameters. The first preset threshold is a preset relational degree threshold.

[0066] By using the grey relational analysis method to determine the key manipulation parameters affecting the flight state, it is possible to accurately measure the influence of the manipulation parameter on the flight state even when the data samples are few or incomplete, improve the stability and applicability of the analysis. This method calculates the grey relational coefficient and the grey relational degree based on the historical flight data, making the extraction of key parameters more efficient and accurate, and at the same time enhancing the interpretability of the results, which helps to optimize the flight control strategy.

[0067] In an embodiment of the present invention, the key flight state parameters are determined based on the prior knowledge of aircraft dynamics. The specific steps include:

[0068] S301, construct a transfer function from the manipulation parameter to the flight state parameter according to the six-degree-of-freedom kinematic equation of the aircraft, and construct an edge between the manipulation parameter and the flight state parameter. Specifically, the six-degree-of-freedom kinematic equation of the aircraft describes how the manipulation parameter affects the flight state parameter, and then a transfer function is established;

[0069] S302. Integrate the amplitude-frequency response of the transfer function to obtain the correlation strength between the control parameter and the flight state parameter. If the correlation strength exceeds the preset threshold, then regard this flight state parameter as a key flight state parameter. Specifically, the effective frequency band range of flight control is fixed from 0.1 rad / s to 10 rad / s, covering the control scenarios from slow adjustment to emergency maneuver. The formula for integrating the amplitude-frequency response of the transfer function is: where φ(v i ,v j ) represents the correlation strength between the control parameter v i and the flight state parameter v j , H i,j (jω) represents the complex form of the transfer function in the frequency domain, j represents the imaginary unit, ω represents the angular frequency, indicating how fast the signal changes. If the correlation strength between the control parameter and the flight state parameter exceeds the preset threshold, then add an edge in the spatio-temporal graph. The preset threshold is determined according to the statistical distribution of historical flight data. Preferably, the preset threshold is set to 1.5;

[0070] S303. Measure the dynamic response delay time of the control parameter to the flight state parameter through a step response experiment, and normalize it as the weight of the edge. Specifically, apply a unit step signal to the node, record the time-domain response curve of the node, and calculate the time required for it to reach 63.2% of the steady-state value, and regard it as the delay time.

[0071] By determining the key flight state parameters based on the prior knowledge of aircraft dynamics, the influence of the control parameter on the flight state can be quantified using the six-degree-of-freedom kinematic equation and the transfer function, ensuring the physical rationality and engineering interpretability of parameter selection. By calculating the correlation strength through amplitude-frequency response integration, the key state parameters that have a significant impact on flight quality can be effectively screened, improving the accuracy of the analysis. At the same time, combining the step response experiment to measure the dynamic response delay time and normalizing it as the weight of the edge makes the correlation relationship between parameters more in line with the actual flight characteristics, thereby enhancing the scientificity and precision of flight control quality assessment.

[0072] In one embodiment of the present invention, the key manipulation parameters determined in S203 and the key flight state parameters determined in S302 are used as the nodes of the spatio-temporal graph. Edges are constructed based on the transfer function established between the manipulation parameters and the flight state parameters, and edges can also be constructed according to the coupling relationship between the flight state parameters. For example, a change in the pitch angle will cause asymmetric aerodynamic forces, which in turn affect the roll angle. Therefore, an edge can be constructed between the pitch angle and the roll angle. In addition, according to the dynamic response delay time between the nodes, the weight of the edge is calculated, and the time series composed of the values of the key manipulation parameters or the key flight state parameters collected by the node within the first preset time window according to the first preset time window is represented as the feature of the node, and finally the spatio-temporal graph is obtained.

[0073] In one embodiment of the present invention, based on the cognitive load characteristics, the specific steps of identifying the cognitive load state by adopting a clustering algorithm based on dynamic time warping include:

[0074] S401, within the first preset time window, normalize the cognitive load characteristics and align the time series data by using the dynamic time warping method;

[0075] S402, adopt the DTW-K-means clustering algorithm to perform clustering analysis on the aligned cognitive load characteristics, and automatically divide the cognitive load state categories of the pilot. Among them, the cognitive load state categories include: low load, medium load, high load, and extreme load. Low load represents the normal flight state, with stable physiological signals and gentle changes in the manipulation input. Medium load represents the existence of slight tension, a slightly elevated heart rate, and an increased fluctuation in the manipulation input. High load represents the state of high pressure, with a significantly elevated heart rate, a reduced number of saccades, and a significantly increased amplitude and frequency of the manipulation input. Extreme load represents the cognitive collapse of the pilot, with abnormal physiological signals and unstable or abnormally increased manipulation input;

[0076] S403, calculate the probability distribution of the cognitive load state categories within the current time window, and adopt the maximum probability principle to select the category with the highest occurrence frequency as the cognitive load state of the pilot in the current time window. Specifically, count the number of occurrences of each cognitive load state category within the current time window, calculate its proportion, form the probability distribution, and select the cognitive load state category with the highest proportion from the probability distribution as the cognitive load state of the current time window.

[0077] In one embodiment of the present invention, according to the identified cognitive load state of the current time window, query the preset node sensitivity database, extract the key manipulation parameters and key flight state parameters in this cognitive load state, and adjust the weight of the edge of the spatio-temporal graph according to the first adjustment formula through the preset cognitive load influence coefficient to generate a dynamic spatio-temporal graph;

[0078] Among them, the cognitive load influence coefficient is determined by the current cognitive load state. Preferably, the original weight is maintained under low load. When the load is medium, the cognitive load influence coefficient is 0.1. When the load is high, the cognitive load influence coefficient is 0.3. When the load is extreme, the cognitive load influence coefficient is -0.5. The first adjustment formula is: represents the weight of the adjusted edge, represents the weight of the edge of the spatio-temporal graph in the initial state, λ c represents the cognitive load influence coefficient.

[0079] In an embodiment of the present invention, the node sensitivity database is constructed based on historical data analysis, flight simulator experiments, and expert knowledge, and includes the importance of each manipulation parameter and flight state parameter to flight quality under different cognitive load states. For example, in the "high load" state, the position of the joystick and the change of stick force may be more sensitive; while in the "low load" state, the change speed of the aircraft attitude and the change of overload may be more sensitive parameters.

[0080] In an embodiment of the present invention, spatio-temporal sensitive features of nodes are extracted based on a graph attention mechanism introducing a time decay coefficient, and the calculation formula of the spatio-temporal sensitive features is:

[0081]

[0082] Among them, represents the spatio-temporal sensitive feature of node u, N(u) represents the set of neighbor nodes of node u, that is, the set of nodes connected to node u by edges, γ represents the time decay coefficient, used to control the speed of time decay. When the value of γ is large, small time changes will also cause great influence. On the contrary, the system is not sensitive to time changes, X v,t represents the corresponding value of node v at the t-th moment, X v,t-1 represents the corresponding value of node v at the (t - 1)-th moment, W con represents the transformable weight matrix, h u represents the feature of node u, h k represents the feature of node k, h v represents the feature of node v, a represents the attention weight vector, used to calculate the attention score of neighbor nodes, that is, a[W con h u ||W con h k , || represents the concatenation operation, sigmoid represents the sigmoid activation function, and LeakyReLU represents the LeakyReLU activation function.

[0083] By extracting the spatio-temporal sensitive features of nodes through a graph attention mechanism based on the introduction of a time decay coefficient, the dynamic relationship between the manipulation parameters and the flight state parameters can be adaptively learned, improving the accuracy of feature extraction. The attention mechanism can automatically assign importance weights to different nodes, enhancing the influence of key parameters and suppressing the interference of noise data, enabling the model to capture the temporal changes and spatial correlations of the flight state. Compared with the traditional method with fixed weights, the graph attention mechanism can more effectively characterize complex non-linear relationships, improving the intelligent level and generalization ability of flight control quality assessment.

[0084] In an embodiment of the present invention, according to the spatio-temporal sensitive features of nodes, flight state features, pilots' cognitive load features, and environmental state features, a comprehensive feature vector is calculated through a weighted fusion method, and a flight control quality score is calculated through a support vector regression algorithm. The calculation steps of the flight control quality score include:

[0085] S501, construct the objective function of the support vector regression algorithm, and the objective function is: The constraint condition is: |S pre -(w * F total + b)| ≤ ∈, where F represents the value of the objective function, min represents the operation of taking the minimum value, w represents the regression weight vector, ||w|| 2 represents the L2 norm of w, that is, the modulus of the regression weight vector, F total represents the comprehensive feature vector, b represents the bias coefficient, ∈ represents the insensitive loss threshold, and S pre represents the predicted flight control quality score;

[0086] S502, select the radial basis kernel as the kernel function, which can enhance the fitting ability of the support vector regression on non-linear data;

[0087] S503, use historical flight data for training;

[0088] S504, after training, use the support vector regression algorithm to calculate the flight control quality score for multi-source flight data. The calculation formula of the flight control quality score is: where o represents the index in the training data, δ o represents the weight coefficient corresponding to the support vector, F o represents the comprehensive feature vector in the training data, F test represents the comprehensive feature vector of multi-source flight data, and σ represents the bandwidth of the kernel function, controlling the attenuation speed of similarity.

[0089] In an embodiment of the present invention, when the flight handling quality score is less than 40, it is an unqualified maneuver, and the operation measures are as follows: trigger an automatic warning to prompt the pilot to immediately adjust the maneuvering strategy, pay attention to whether there are abnormalities in the key maneuvering parameters and flight state parameters, and provide adjustment suggestions; when the flight handling quality score is between 40 and 60, it is a general maneuver, and the operation measures are as follows: provide maneuver optimization suggestions to prompt the pilot to pay attention to the key maneuvering parameters; when the flight handling quality score is between 60 and 80, it is a good maneuver, and the operation measures are as follows: appropriately optimize the maneuvering strategy; when the flight handling quality score is between 80 and 100, it is an excellent maneuver, and the operation measures are as follows: save the best maneuvering method and provide further optimization suggestions.

[0090] An embodiment of the present invention further provides an artificial intelligence-based flight handling quality evaluation system, including:

[0091] A data acquisition module for real-time acquisition of multi-source flight data, including maneuvering parameters, flight state parameters, pilot physiological signals, and environmental state data;

[0092] A feature extraction module for extracting flight state features, pilot cognitive load features, and environmental state features according to the multi-source flight data;

[0093] A spatio-temporal graph construction module for determining key maneuvering parameters affecting the flight state by using grey relational analysis based on historical flight data, and determining key flight state parameters based on prior knowledge of aircraft dynamics, defining the key maneuvering parameters and flight state parameters as nodes, and constructing a spatio-temporal graph;

[0094] A spatio-temporal graph dynamic adjustment module, based on the cognitive load features, identifying the cognitive load state through a clustering algorithm, and extracting key maneuvering parameters and key flight state parameters from a preset node sensitivity database according to the cognitive load state, and dynamically adjusting the weights of the corresponding edges in the spatio-temporal graph to generate a dynamic spatio-temporal graph;

[0095] A maneuvering quality score generation module for extracting spatio-temporal sensitive features of nodes based on a graph attention mechanism introducing a time decay coefficient, fusing flight state features, pilot cognitive load features, and environmental state features, obtaining a comprehensive feature vector by using a weighted fusion method, and outputting a flight handling quality score by using a support vector regression algorithm;

[0096] A decision execution module for classifying according to the flight handling quality score according to a preset score level standard, and executing corresponding operations based on the score level.

[0097] The embodiments of the present invention have been described above. However, the present invention is not limited to the above specific embodiments. The above specific embodiments are merely illustrative rather than restrictive. Under the inspiration of this embodiment, those of ordinary skill in the art can also make many forms, all of which fall within the protection scope of this embodiment.

Claims

1. An artificial intelligence-based flight handling quality assessment method, characterized in that, Including the following steps: S101. Obtain multi-source flight data, including control parameters, flight state parameters, pilot physiological signals, and environmental state data; S102. Extract flight state features, pilot cognitive load features, and environmental state features based on the multi-source flight data; S103. Determine the key control parameters affecting the flight state by using grey relational analysis according to historical flight data, and determine the key flight state parameters based on the prior knowledge of aircraft dynamics. Define the key control parameters and key flight state parameters as nodes, and construct a spatio-temporal graph; S104. Based on the cognitive load features, use a clustering algorithm based on dynamic time warping to identify the cognitive load state. According to the cognitive load state, extract the key control parameters and key flight state parameters from a preset node sensitivity database, and dynamically adjust the weights of the corresponding edges in the spatio-temporal graph to generate a dynamic spatio-temporal graph; S105. Extract the spatio-temporal sensitive features of the nodes based on the graph attention mechanism introducing a time decay coefficient, fuse the flight state features, pilot cognitive load features, and environmental state features, obtain a comprehensive feature vector by using a weighted fusion method, and output a flight control quality score by using a support vector regression algorithm; S106. According to the flight control quality score, divide it according to a preset score level standard, and take corresponding operation measures based on the score level.

2. The flight control quality evaluation method based on artificial intelligence according to claim 1, wherein The control parameters include: joystick displacement, stick force, rudder deflection angle. The flight state parameters include: pitch angle, roll angle, yaw angle, control surface position, and attitude angular velocity. The pilot physiological signals include: heart rate, brain wave α / β ratio, and saccade count. The environmental state data include: wind speed, turbulence intensity level, air temperature, and air pressure.

3. The flight control quality assessment method based on artificial intelligence according to claim 2, characterized in that, Extracting flight state features, pilot cognitive load features, and environmental state features based on the multi-source flight data includes: Combining the change of joystick displacement, the change of attitude angular velocity, and the control surface position, and performing weighted fusion calculation after alignment to obtain flight state features; Calculating the cognitive load features by fusing the heart rate, brain wave α / β ratio, and saccade count of the pilot; Fusing the wind speed change, air pressure change, and temperature fluctuation, and obtaining environmental state features based on a weighted calculation method.

4. The flight control quality evaluation method based on artificial intelligence according to claim 1, wherein, Determining the key control parameters affecting the flight state by using grey relational analysis according to historical flight data. The specific steps include: S201. Normalize the control parameter sequence and flight state parameter sequence in the historical flight data; S202. Calculate the grey relational coefficients of each control parameter sequence and flight state parameter sequence; S203. Calculate the grey relational degree of each control parameter, and screen out the control parameters with grey relational degrees greater than a first preset threshold as the key control parameters.

5. The flight control quality evaluation method based on artificial intelligence according to claim 1, wherein Determining the key flight state parameters based on the prior knowledge of aircraft dynamics. The specific steps include: S301. Construct a transfer function from control parameters to flight state parameters according to the six-degree-of-freedom kinematic equation of the aircraft, and construct edges between the control parameters and flight state parameters; S302, Integrate the amplitude-frequency response of the transfer function to obtain the correlation strength between the control parameter and the flight state parameter. If the correlation strength exceeds the preset threshold, then regard this flight state parameter as a key flight state parameter; S303, Measure the dynamic response delay time of the control parameter to the flight state parameter through a step response experiment, and normalize it as the weight of the edge.

6. The flight control quality evaluation method based on artificial intelligence according to claim 1, characterized in that The spatio-temporal graph includes: nodes, edges, and characteristics of the nodes. The nodes represent key control parameters or key flight state parameters, the edges represent the correlation relationships between the nodes, and the characteristics of the nodes are represented by a vector composed of the key control parameter values or key flight state parameter values collected within the first preset time window according to the first preset time window within the first preset time period.

7. A flight control quality assessment method based on artificial intelligence according to claim 1, characterized in that, The specific steps for identifying the cognitive load state using a clustering algorithm based on dynamic time warping based on the cognitive load characteristics include: S401, Within the first preset time window, normalize the cognitive load characteristics and use the dynamic time warping method to align the time series; S402, Use a clustering algorithm to perform clustering analysis on the aligned cognitive load characteristics, and automatically divide the pilot's cognitive load state categories. Among them, the cognitive load state categories include: low load, medium load, high load, and extreme load; S403, Calculate the probability distribution of the cognitive load state categories within the current time window, and use the maximum probability principle to select the category with the highest occurrence frequency as the pilot's cognitive load state in the current time window.

8. The flight control quality evaluation method based on artificial intelligence according to claim 7, wherein, According to the cognitive load state of the current time window, query the preset node sensitivity database, extract the key control parameters and key flight state parameters in this cognitive load state, and adjust the weight of the edges of the spatio-temporal graph through the preset cognitive load influence coefficient to generate a dynamic spatio-temporal graph.

9. The flight control quality assessment method based on artificial intelligence according to claim 1, characterized in that Extract the spatio-temporal sensitive features of the nodes based on the graph attention mechanism introducing a time decay coefficient. The spatio-temporal sensitive features are obtained by fusing the features of the neighbor nodes of the node, the time decay coefficient, and the attention mechanism. Among them, first use the time decay coefficient to perform weighted adjustment on the influence of neighbor nodes at different times, then use the graph attention mechanism to calculate the feature correlation between the neighbor nodes and this node, and normalize the neighbor node features. Finally, fuse the normalized neighbor node features with the features of this node to calculate the spatio-temporal sensitive features of the node.

10. An artificial intelligence-based flight handling quality assessment system, characterized in that, Adopt a flight handling quality evaluation method based on artificial intelligence as described in any one of claims 1-9, including: A data acquisition module for acquiring multi-source flight data, including control parameters, flight state parameters, pilot physiological signals, and environmental state data; A feature extraction module for extracting flight state features, the pilot's cognitive load features, and environmental state features according to the multi-source flight data; A spatio-temporal graph construction module for determining the key control parameters affecting the flight state using grey relational analysis according to historical flight data, and determining the key flight state parameters based on the prior knowledge of aircraft dynamics. Define the key control parameters and flight state parameters as nodes, and construct a spatio-temporal graph; The spatio-temporal graph dynamic adjustment module, based on the cognitive load characteristics, uses a clustering algorithm based on dynamic time warping to identify the cognitive load state, and according to the cognitive load state, extracts key manipulation parameters and key flight state parameters from a preset node sensitivity database, and dynamically adjusts the weights of the corresponding edges in the spatio-temporal graph to generate a dynamic spatio-temporal graph; The handling quality scoring generation module extracts the spatio-temporal sensitive features of nodes based on the graph attention mechanism introducing a time decay coefficient, fuses the flight state features, the cognitive load features of the pilot and the environmental state features, obtains a comprehensive feature vector by using a weighted fusion method, and outputs the flight handling quality score by using a support vector regression algorithm; The decision execution module divides according to the flight handling quality score according to the preset scoring level standard, and executes corresponding operations based on the scoring level.

Images