An aircraft flight risk real-time early warning method, device, medium and product

Abstract

The application discloses an aircraft flight risk real-time early warning method, device, medium and product, relates to the technical field of aircraft operation safety, and comprises the following steps: firstly, flight data at the current time is acquired; part of the flight data at the current time is used to represent the influence degree of the pilot human factor; then, the trained pilot control mode prediction model is used to process the flight data at the current time, and the control mode type at the current time is output; the aircraft performance data and the flight environment data in the flight data at the current time are input into the corresponding trained flight risk real-time early warning model; the risk result of the target early warning point under the control mode type at the current time is output; finally, when the risk result is that the target early warning point is risky, early warning is performed, the aircraft flight risk real-time early warning is realized from two aspects of the pilot human factor and the selection of the real-time early warning time, the accuracy of the early warning result is improved, and technical support is provided for further guarantee of aviation flight safety.

Description

Technical Field

[0001] The present application relates to the field of aircraft operation safety technology, and in particular to a real-time early warning method, equipment, medium and product for aircraft flight risks based on a pilot control mode. Background Art

[0002] Flight safety is a core concern of ongoing concern in the aviation sector. With the rapid development of sensor and data transmission technologies, flight safety management is gradually shifting from risk assessment based on post-event data to proactive safety control based on real-time data. Real-time flight risk warnings based on flight data will become the primary means of safety management in the era of aviation big data.

[0003] The evolution of flight risk directly impacts the accuracy of real-time flight risk warnings. However, as aircraft reliability continues to improve, pilot human factors are becoming increasingly crucial to the accuracy of real-time warnings. Furthermore, the timing of warnings also impacts the effectiveness of the final real-time warnings. Existing technologies for real-time flight risk warnings fail to consider both pilot human factors and the timing of warnings. Summary of the Invention

[0004] The purpose of this application is to provide a real-time warning method, equipment, medium and product for aircraft flight risks, which can provide real-time warning of aircraft flight risks based on the pilot's human factors and warning timing, thereby improving the accuracy of the warning results.

[0005] To achieve the above objectives, this application provides the following solutions:

[0006] In a first aspect, the present application provides a real-time early warning method for aircraft flight risks, comprising:

[0007] Acquiring current flight data, wherein the flight data includes speed deviation, trajectory deviation, flight altitude, aircraft performance data, and flight environment data, wherein the speed deviation represents the pilot's situational awareness of the aircraft, the trajectory deviation represents the pilot's control decision-making ability for the aircraft, and the flight altitude represents the pilot's control difficulty for the aircraft;

[0008] Inputting the current flight data into a trained pilot control mode prediction model to output the current control mode type, wherein the current control mode type remains unchanged in the current flight mission profile, wherein the flight mission profile refers to a specific flight operation of the aircraft, and the specific flight operation includes takeoff, climb, cruise, descent, and landing;

[0009] Inputting aircraft performance data and flight environment data in the current flight data into a corresponding trained real-time flight risk warning model, and outputting a risk result of a target warning point under the current control mode type, wherein the risk result indicates whether the target warning point is at risk, and the control mode type of the corresponding trained real-time flight risk warning model is the same as the current control mode type;

[0010] When the risk result indicates that the target warning point is at risk, a real-time warning is issued.

[0011] Optionally, the specific training process of the pilot control mode prediction model includes:

[0012] Constructing a first training dataset, wherein the first training dataset includes historical flight data and control mode types of different pilots flying a specific aircraft model under a flight mission profile, the historical flight data including historical speed deviations, historical trajectory deviations, historical flight altitudes, historical aircraft performance data, and historical flight environment data, and the control mode types are obtained by processing the pilots' historical operational control data using a clustering algorithm;

[0013] The historical flight data is used as feature data, and the control mode type is used as label data to train the pilot control mode prediction model to obtain a trained pilot control mode prediction model.

[0014] Optionally, the historical operation control data for any pilot in any flight mission profile includes a historical control stick operation sequence and a historical throttle lever operation sequence, wherein the historical control stick operation sequence includes a plurality of historical control stick operation data, and the historical throttle lever operation sequence includes a plurality of historical throttle lever operation data;

[0015] The specific process of determining the control mode type includes:

[0016] determining a first sequence position of optimal historical stick operation data based on the historical stick operation sequence and a first threshold, wherein the first sequence position is used to evaluate the timeliness of the pilot's stick control;

[0017] calculating, based on the historical sequence of stick operations, a degree of fluctuation in stick operations over a period of time, wherein the degree of fluctuation is used to evaluate the smoothness of the pilot's stick control;

[0018] determining a second sequence position of optimal historical throttle stick operation data based on the historical throttle stick operation sequence and a second threshold, wherein the second sequence position is used to evaluate the timeliness of the pilot's throttle stick control;

[0019] calculating a first deviation degree according to the first sequence position, the second sequence position, and a third threshold, wherein the first deviation degree is used to evaluate the coordination of pilot control;

[0020] Combining the corresponding first sequence position, the fluctuation degree and the first deviation degree into a data point;

[0021] A clustering algorithm is used to perform clustering processing on all the data points to determine the control mode type.

[0022] Optionally, the specific training process of the real-time flight risk warning model includes:

[0023] For any pilot's historical flight data corresponding to a known control mode type under any flight mission profile, different slice reference units are determined based on the sampling frequency of warning variables and the actual number of warnings required, wherein the warning variables include warning time and / or warning altitude, and the historical flight data includes historical aircraft performance data and historical flight environment data;

[0024] Slicing the historical flight data according to different slicing reference units to obtain different second training data sets, wherein each second training data set includes data slices corresponding to a plurality of warning points, and the warning points are determined according to the slicing positions of the slicing reference units;

[0025] For different second training data sets, the historical flight data in each data slice is used as feature data, and the risk result of each warning point is used as label data to train the first flight risk real-time warning model;

[0026] Calculating the average recall rate of different first flight risk real-time warning models;

[0027] Selecting an optimal slice reference unit corresponding to the first flight risk real-time warning model having an average recall rate greater than a first recall rate threshold, and using a warning point having a first recall rate greater than a second recall rate threshold as the target warning point, wherein the first recall rate is the recall rate of the warning point corresponding to the optimal slice reference unit;

[0028] Using historical flight data in the data slice corresponding to each target warning point under the same control mode type as feature data, and using the risk result of each target warning point as label data, training a second flight risk real-time warning model corresponding to different control mode types, wherein each of the second flight risk real-time warning models includes a decision tree model, a regression model, an XGBOOST model, and a LightGBM model;

[0029] Calculate the second recall rate of the second flight risk real-time warning model under different types;

[0030] The optimal second flight risk real-time warning model whose second recall rate is greater than the third recall rate threshold is selected as the trained flight risk real-time warning model corresponding to the final different control mode types.

[0031] Optionally, the specific calculation process of the first sequence position includes:

[0032] sequentially calculating the absolute values ​​of differences between adjacent historical joystick operation data;

[0033] The sequence position of the historical joystick operation data whose absolute value is greater than the first threshold is used as the first sequence position.

[0034] Optionally, the calculation expression for the fluctuation degree is:

[0035]

[0036] Wherein, HC_S represents the degree of fluctuation; n represents the number of the historical joystick operation data; MCS j represents the j-th historical joystick operation data; represents the mean value of all the historical joystick operation data.

[0037] Optionally, the calculation expression of the first deviation degree is:

[0038] HC_C=||HC_T M -HC_T A |-HC_C SOP |;

[0039] Wherein, HC_C represents the first deviation degree; HC_T M Indicates the first sequence position; HC_T A Indicates the second sequence position; HC_C SOP represents the third threshold.

[0040] In a second aspect, the present application provides a computer device comprising: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement a real-time warning method for aircraft flight risks as described in the first aspect above.

[0041] In a third aspect, the present application provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the real-time warning method for aircraft flight risks described in the first aspect above.

[0042] In a fourth aspect, the present application provides a computer program product, comprising a computer program, which, when executed by a processor, implements the real-time warning method for aircraft flight risks described in the first aspect above.

[0043] According to the specific embodiments provided in this application, this application discloses the following technical effects:

[0044] The present application provides a method, device, medium and product for real-time early warning of aircraft flight risks. The method first obtains flight data at the current moment, uses part of the data in the flight data at the current moment to characterize the degree of influence of pilot human factors, then uses a trained pilot control mode prediction model to process the flight data at the current moment, outputs the control mode type at the current moment, and then inputs the aircraft performance data and flight environment data in the flight data at the current moment into the corresponding trained real-time early warning model for flight risks, outputs the risk result of the target warning point under the current control mode type, where the target warning point is the warning location point. Finally, when the risk result indicates that the target warning point is at risk, a real-time early warning is issued, thereby achieving real-time early warning of aircraft flight risks from two perspectives: pilot human factors and the selection of real-time early warning timing, improving the accuracy of the early warning results, and providing technical support for further ensuring aviation flight safety. BRIEF DESCRIPTION OF THE DRAWINGS

[0045] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0046] Figure 1 This is a diagram of the application environment of a real-time early warning method for aircraft flight risks in Example 1 of the present application;

[0047] Figure 2 A flowchart of a real-time early warning method for aircraft flight risks provided in Example 1 of the present application;

[0048] Figure 3 Schematic diagram of the pilot control mode calculation process in Example 1 of the present application;

[0049] Figure 4 Schematic diagram of the pilot control mode prediction model in Example 1 of the present application;

[0050] Figure 5 Schematic diagram of the pilot control mode prediction model training process in Example 1 of the present application;

[0051] Figure 6Schematic diagram of the segmentation process of the HCM dataset in Example 1 of the present application;

[0052] Figure 7 Schematic diagram of the training process of the real-time warning model for flight risk based on the pilot control mode in Example 1 of the present application;

[0053] Figure 8 Schematic diagram of the pilot control mode clustering results in Example 1 of the present application;

[0054] Figure 9 This is a process diagram of the pilot control mode prediction method based on flight data in Example 1 of the present application;

[0055] Figure 10 This is a schematic diagram of the aircraft operation risk warning result in Example 1 of the present application;

[0056] Figure 11 A schematic diagram of the structure of a computer device provided in Example 2 of the present application. DETAILED DESCRIPTION

[0057] The following will be combined with the drawings in the embodiments of this application to clearly and completely describe the technical solutions in the embodiments of this application. Obviously, the embodiments described are only part of the embodiments of this application, not all of the embodiments. Based on the embodiments in this application, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of this application.

[0058] In order to make the above-mentioned purposes, features and advantages of the present application more obvious and easy to understand, the present application is further described in detail below with reference to the accompanying drawings and specific implementation methods.

[0059] Example 1

[0060] The embodiment of the present application provides a real-time warning method for aircraft flight risks, which can be applied to Figure 1In the application environment shown, the terminal 102 communicates with the server 104 via a network. The data storage system can store data that the server 104 needs to process. The data storage system can be set up separately, integrated on the server 104, or placed on the cloud or other servers. The terminal 102 can send the current flight data to the server 104. After receiving the current flight data, the server 104 first inputs the current flight data into a trained pilot control mode prediction model and outputs the current control mode type. The aircraft performance data and flight environment data in the current flight data are then input into a corresponding trained real-time flight risk warning model, which outputs the risk result of the target warning point under the current control mode type. Finally, when the risk result indicates that the target warning point is at risk, an early warning is issued. The server 104 can feedback the obtained risk result to the terminal 102. In addition, in some embodiments, the real-time warning method for aircraft flight risks can also be implemented independently by the server 104 or the terminal 102. For example, the terminal 102 can directly use the real-time warning method for aircraft flight risks to process the flight data at the current moment, or the server 104 can obtain the flight data at the current moment from the data storage system and then use the real-time warning method for aircraft flight risks to process it.

[0061] Terminal 102 may include, but is not limited to, various desktop computers, laptops, smartphones, tablet computers, IoT devices, and portable wearable devices. IoT devices may include smart speakers, smart TVs, smart air conditioners, and smart car devices. Portable wearable devices may include smart watches, smart bracelets, and head-mounted devices. Server 104 may be implemented as a standalone server or a server cluster consisting of multiple servers, or may be a cloud server.

[0062] In an exemplary embodiment, Figure 2 As shown, a real-time warning method for aircraft flight risk is provided. The method is executed by a computer device, specifically a computer device such as a terminal or a server, or a terminal and a server. In the embodiment of the present application, the method is applied to Figure 1 The server 104 in the example is used as an example to illustrate the process, including the following steps 201 to 204.

[0063] Step 201: Acquire flight data at the current moment, wherein the flight data includes speed deviation, trajectory deviation, flight altitude, aircraft performance data, and flight environment data. The speed deviation represents the pilot's situational awareness of the aircraft, the trajectory deviation represents the pilot's control decision-making ability for the aircraft, and the flight altitude represents the difficulty of the pilot controlling the aircraft.

[0064] Step 202: Input the current flight data into a trained pilot control mode prediction model, and output the current control mode type. The current control mode type remains unchanged in the current flight mission profile. The flight mission profile refers to a specific flight operation of the aircraft, including takeoff, climb, cruise, descent, and landing.

[0065] Before step 202, the pilot control mode prediction model needs to be trained. The specific training process includes:

[0066] (1) Constructing a first training data set, wherein the first training data set includes historical flight data and control mode types of different pilots flying a specific aircraft model under a flight mission profile, and the control mode types are obtained by processing the pilots' historical operation control data using a clustering algorithm.

[0067] like Figure 3 As shown in the figure, the specific process of determining the control mode type is as follows:

[0068] 1) Obtain the pilot's historical operational control data.

[0069] Obtain the historical pilot control sequence HCS=[MCS,ACS] of a certain pilot under a specific flight mission profile, that is, historical operation control data, where MCS=[MCS1,MCS2,...,MCS n ], which represents a historical stick operation sequence consisting of several historical stick operation data (i.e., the pilot's stick control data); ACS = [ACS1, ACS2, ..., ACS n ] represents a historical sequence of throttle stick operations (i.e., the pilot's throttle stick control data); n is determined by the length of the specific flight mission profile. Given that flight control is primarily based on the joystick, both timeliness and smoothness are evaluated based on the MCS.

[0070] 2) Timeliness assessment.

[0071] A first sequence position of optimal historical stick operation data is determined based on the historical stick operation sequence and a first threshold, wherein the first sequence position is used to evaluate the timeliness of the pilot's stick control.

[0072] The specific calculation process of the first sequence position includes:

[0073] The absolute values ​​of the differences between adjacent historical joystick operation data are calculated in sequence.

[0074] The sequence position of the historical joystick operation data whose absolute value is greater than the first threshold is used as the first sequence position.

[0075] In the selected mission profile, the discrete pilot stick operation sequence is assumed to be MCS = {MCS1, MCS2, ..., MCS n}, n represents the length of the historical pilot stick operation data. MCS j Represents the historical pilot's stick operation data at time j, and calculates ΔMCS in sequence j =|MCS j -MCS j-1 |, j = 2, 3, ..., n, if ΔMCS j >k M , then HC_T M = j, and stop calculating ΔMCS j Among them, k M Indicates the first threshold

[0076] The first threshold is determined by the pilot's comprehensive mission profile and historical MCS change experience. It is used to judge the slope of the pilot's stick operation sequence and define the corresponding sequence position as timeliness. The stick operation sequence formed by different independent variables has different meanings of the corresponding sequence position, for example:

[0077] A time-arranged joystick operation sequence is formed with time as the independent variable. If the joystick position changes by 4° at the 10th second, when the first threshold is 3°, the timeliness of the joystick is 10 seconds.

[0078] A control stick operation sequence arranged according to height is formed with height as the independent variable. If the control stick position changes by 4° at 5 meters above the ground, when the first threshold is 3°, the timeliness of the control stick is 5 meters.

[0079] 3) Stability assessment.

[0080] The degree of fluctuation of the control stick operation over a period of time is calculated based on the historical control stick operation sequence, wherein the degree of fluctuation is used to evaluate the smoothness of the pilot's control of the control stick.

[0081] The calculation expression of the fluctuation degree of any pilot in any flight mission profile is:

[0082]

[0083] Wherein, HC_S represents the degree of fluctuation; n represents the number of the historical joystick operation data; MCS j represents the j-th historical joystick operation data; represents the mean value of all the historical joystick operation data.

[0084] 4) Coordination assessment.

[0085] 4-1) Determining a second sequence position of optimal historical throttle stick operation data based on the historical throttle stick operation sequence and a second threshold, wherein the second sequence position is used to evaluate the timeliness of the pilot's throttle stick control, and the second threshold is determined based on the pilot's comprehensive mission profile and historical changes in the MCS.

[0086] In the selected mission profile, the discrete pilot throttle stick operation sequence is assumed to be ACS = {ACS1, ACS2, ..., ACS n}, n represents the length of the historical pilot throttle stick operation data. ACS j Represents the historical pilot's throttle stick operation data at time j, and calculates ΔACS in sequence j =|ACS j -ACS j-1 |,j=2,3,…,n. If ΔACS j >k A , then HC_T A = j, and stop calculating ΔACS j Among them, k A represents the second threshold.

[0087] 4-2) Calculating a first degree of deviation based on the first sequence position, the second sequence position, and a third threshold, wherein the first degree of deviation is used to evaluate the coordination of pilot control.

[0088] The calculation expression of the first deviation degree is:

[0089] HC_C=||HC_T M -HC_T A |-HC_C SOP | (2);

[0090] Wherein, HC_C represents the first deviation degree; HC_T M Indicates the first sequence position; HC_T A Indicates the second sequence position; HC_C SOP represents the third threshold.

[0091] HC_C SOP Depends on technical or regulatory requirements of pilot control.

[0092] 5) Using a clustering algorithm to perform clustering processing on all data points to determine the control mode type, wherein a data point consists of the corresponding first deviation degree, the fluctuation degree and the first sequence position.

[0093] Calculate the pilot control mode based on HC_T (first sequence position), HC_S, and HC_C: HC_T, HC_S, and HC_C of the control sequence HC form a coordinate point HC = [HC_T, HC_S, HC_C] in three-dimensional space. A clustering method is used to mine the pilot control mode HCM. This embodiment uses K-means clustering as an example to analyze several control modes HCM existing in HC based on the K-means clustering algorithm. The specific steps are as follows:

[0094] 5-1) Define the number of clusters k. The number of clusters is generally determined by flight practice requirements and the amount of HCS data;

[0095] 5-2) Distance metric. Assigning the object points to the cluster closest to the cluster center requires a nearest neighbor metric. Since HC = [HC_T, HC_S, HC_C] is a point in three-dimensional Euclidean space, the Euclidean distance is used to measure the distance between each point. The calculation formula is:

[0096]

[0097] 5-3) Calculation of new cluster centers [HC_Tc, HC_Sc, HC_Cc]k. For the k clusters generated after classification, the point with the smallest distance to the mean of the other points in the cluster is used as the cluster center [HC_Tc, HC_Sc, HC_Cc], and the cluster centers are calculated repeatedly;

[0098] 5-4) Whether to stop K-means. When the cluster center no longer changes or the number of cycles reaches a given value, the set of categories representing the HCM and the corresponding cluster centers [HC_Tc, HC_Sc, HC_Cc]k are obtained;

[0099] 5-5) Obtain the HCM. Define the pilot control mode based on the data features of the cluster centers [HC_Tc, HC_Sc, HC_Cc]k. The SOP mode indicates timely stick control, relatively stable stick position control, and moderate stick and throttle coordination. The conservative mode indicates early stick control, very stable stick position control, and poor stick and throttle coordination. The adventurous mode indicates late stick control, relatively stable stick position control, and good stick and throttle coordination.

[0100] (2) Using the historical flight data as feature data and the control mode type as label data, the pilot control mode prediction model is trained to obtain a trained pilot control mode prediction model.

[0101] Based on the HCM labels obtained in step (1), the prediction features of HCM are extracted from the perspective of “human-machine-environment”, and the corresponding HCM machine learning prediction model is constructed, such as Figure 4 shown.

[0102] like Figure 4 The features used to build the HCM prediction model for human-in-the-loop control systems shown in the figure are mainly divided into three categories, including human factors, machine factors, and environmental factors. Human factors include situational awareness, decision performance, and task difficulty:

[0103] Situational awareness represents a pilot's ability to perceive the current situation in their mission environment. Its magnitude is influenced by a person's attention, physiological, and psychological states, and directly determines their ability to gather information about their surroundings. Therefore, it can impact their ultimate HCM. Aircraft speed deviation is used to characterize situational awareness. Greater speed deviation indicates less accurate situational awareness; smaller speed deviation indicates more accurate situational awareness.

[0104] Decision-making performance represents the performance of the pilot in making decisions on the current human-computer interaction state based on the information input of situational awareness under the current control task, using his attention resources, control knowledge and control experience. It is ultimately reflected in the performance of the human-computer interaction task and is represented by aircraft track deviation. The larger the track deviation, the less accurate the pilot's decision on the current aircraft control; the smaller the track deviation, the more accurate the pilot's decision on the current aircraft control.

[0105] Mission difficulty represents the current level of control difficulty in human-machine interaction and reflects the system's risk. Higher mission difficulty demands greater control decisions and may also increase the pilot's physiological and psychological state, such as increased stress and fatigue. For typical aviation flights, mission difficulty is measured by altitude: the lower the altitude, the more difficult the mission.

[0106] The above three variables describe the pilot's situational awareness and decision-making state during information processing. Task difficulty is a key variable influencing decision-making and should be selected in conjunction with the pilot's control mission profile. Furthermore, performance factors (i.e., historical aircraft performance data) and environmental factors (i.e., historical flight environment data) are input into the machine learning model as HCM features, ultimately outputting the HCM.

[0107] Assuming that the pilot's HCM remains unchanged within a certain mission profile, that is, the pilot tends to choose a fixed HCM over a period of time, the above-mentioned pilot information processing process based on HCM characteristics and corresponding HCM data is simulated through machine learning, and finally the HCM prediction is achieved. The specific steps are as follows: Figure 5As shown:

[0108] (a) With reference to the typical flight mission process and according to the mission profile determined by the warning requirements, the prediction sequence dataset DHCM for HCM is collected.

[0109] (b) Calculate HCM based on step (1) as the label of the dataset DHCM.

[0110] (c) Based on the warning requirements, the warning variables WV, such as time and altitude, are confirmed. Based on the assumption that HCM is constant within the mission profile, the DHCM is segmented using the WV acquisition frequency to increase the sample size of the DHCM dataset and improve the accuracy of DHCM analysis based on machine learning. The flight data segmentation process is as follows: Figure 6 As shown, the sequence data is segmented according to the warning variables to form the DHCM for HCM analysis.

[0111] (d) Calculate the HCM prediction data features and extract the HCM analysis features and HCM labels from DHCM to form the dataset DHCM-ML for machine learning.

[0112] (e) Use DHCM-ML to run different types of machine learning models, such as DL, Random Forest XGBoost, LGBoost, etc., compare model performance and select the optimal model as the prediction model of HCM. The evaluation indicators include accuracy A, precision P, recall R and comprehensive evaluation index F1 value to verify the usability of the model. A represents the probability that all sample risks are correctly classified; P represents the proportion of samples classified as risky that are actually risky samples. The higher the P value, the better the model accuracy; R represents the classification accuracy of risk samples. The higher the R value, the better the model's risk warning effect; F1 is a comprehensive evaluation index of the model. The higher the F1, the more effective the experimental method. The calculation formulas for P, R and F1 values ​​are shown in formula (4):

[0113]

[0114] In the formula, TP means the number of positive classes predicted as positive classes; FP means the number of negative classes predicted as positive classes.

[0115] It should be noted that this embodiment focuses on extracting the pilot control mode from the pilot's stick operation data and throttle lever operation data. In addition, more operation data, such as rudder, instrument panel, etc., can be integrated to enrich the flight operation data, and attempts can be made to extract data features from more feature dimensions such as operation accuracy, standardization of operation sequence, etc.; the clustering algorithm can also adopt hierarchical clustering, density clustering and model clustering methods, etc.

[0116] Step 203: Input the aircraft performance data and flight environment data in the flight data at the current moment into the corresponding trained real-time flight risk warning model, and output the risk result of the target warning point under the control mode type at the current moment, wherein the risk result indicates whether the target warning point is at risk, and the control mode type of the corresponding trained real-time flight risk warning model is the same as the control mode type at the current moment.

[0117] Before step 203, it is necessary to train a real-time flight risk warning model, specifically including:

[0118] Method 1:

[0119] (aa) For any pilot's historical flight data corresponding to a known control mode type under any flight mission profile, different slice reference units are determined based on a sampling frequency of warning variables and actual warning number requirements, wherein the warning variables include warning time and / or warning altitude, and the historical flight data includes historical aircraft performance data and historical flight environment data.

[0120] (bb) Slicing the historical flight data according to different slicing reference units to obtain different second training data sets, wherein each second training data set includes data slices corresponding to a number of warning points, and the warning points are determined according to the slicing positions of the slicing reference units.

[0121] (cc) For different second training data sets, the historical flight data in each of the data slices is used as feature data, and the risk result of each warning point is used as label data to train the first flight risk real-time warning model.

[0122] (dd) Calculating the average recall rate of different first flight risk real-time warning models.

[0123] (ee) Selecting the optimal slice reference unit corresponding to the first flight risk real-time warning model whose average recall rate is greater than a first recall rate threshold, and using the warning point whose first recall rate is greater than a second recall rate threshold as the target warning point, wherein the first recall rate is the recall rate of the warning point corresponding to the optimal slice reference unit.

[0124] (ff) Using the historical flight data in the data slice corresponding to each of the target warning points under the same control mode type as feature data, and the risk result of each of the target warning points as label data, the second flight risk real-time warning model corresponding to different control mode types is trained, wherein the types of each of the second flight risk real-time warning models include decision tree model (DT), regression model (RM), XGBOOST model and LightGBM model.

[0125] (gg) Calculate the second recall rate of the second flight risk real-time warning model under different types.

[0126] (hh) Selecting the optimal second flight risk real-time warning model whose second recall rate is greater than the third recall rate threshold as the trained flight risk real-time warning model corresponding to the final different control mode types.

[0127] Method 2:

[0128] The steps in method 2 are only different from step (cc) in method 1. The improved step (cc') in method 2 is:

[0129] (cc') For different second training data sets, the historical flight data in the data slice corresponding to each warning point under the same control mode type is used as feature data, and the risk result of each warning point is used as label data to train the first flight risk real-time warning model corresponding to different control mode types.

[0130] The following describes the implementation process of Method 2 using a specific example.

[0131] The changes in flight risks during the warning cycle are uncertain, that is, the accident risk prediction models and prediction accuracies of different warning positions are different. The warning position has an important impact on the changes in human-machine system risks over time. In the field of civil aviation safety, radio altitude is usually used to provide risk warnings for takeoff and landing processes. Furthermore, risk warnings are also limited by the warning mechanism and real-time data transmission. The warning mechanism includes actual mission requirements (such as the number of warnings) and the pilot's emergency response time, and the real-time data transmission limitations include data collection frequency and data delay. Therefore, the training process of the accident risk warning model of the "human-in-the-loop" control system based on HCM is as follows: Figure 7 shown.

[0132] (A) Based on the data between 1300ft and 50ft, the warning points {WP1, WP2,…, WP n}, n is determined by the sampling frequency of the warning variable WP and the recall rate of the final model.

[0133] (B) Data slice range of the slice reference unit ΔWP=(WP max -WP min ) / n, construct the risk warning data set DW for machine learning training = {DW-S i}, i=1, 2, ..., n, in this embodiment, the data slicing ranges of the selected slicing reference units are Δ50ft, Δ100ft, Δ200ft, Δ300ft, Δ400ft and Δ500ft respectively.

[0134] (C) In each data slice DW-S i The prediction results based on HCM are i Group them to form the DW-S-HCM data set.

[0135] (D) Referring to the flight dynamics equation, machine state parameters and environmental change parameters are selected as the features X for risk prediction.

[0136] (E) Set risk label Y based on the typical risks of the “human-in-the-loop” control system and the corresponding risk management threshold.

[0137] (F) Train the machine learning model based on DW-S-HCM and calculate DW-S i The average risk recall rate R of each model in the group on the test data DW-S .

[0138] (G) Select ΔEW so that the R of DW-Si DW-S If the value is higher than 70%, the embodiment finally selects Δ400 ft and performs three landing risk warnings at the warning points of 100 ft, 500 ft and 900 ft.

[0139] (H) Training different types of machine learning models at the target warning point corresponding to ΔEW.

[0140] (I) The machine learning model with the highest risk recall rate is selected as the optimal HCM-based real-time early warning of accident risks, as shown in Table 1.

[0141] Table 1 Recall rates of target warning points of different types of machine learning models under different control modes

[0142]

[0143] Step 204: When the risk result indicates that the target warning point is at risk, a real-time warning is issued.

[0144] This embodiment, with aviation flight safety as the broader context, addresses the issue of flight risk prevention and control in the big data era. It proposes a method for improving real-time flight risk warnings based on flight data from two perspectives: pilot human factors and the timing of real-time warnings. This provides technical support for further ensuring aviation flight safety.

[0145] The real-time early warning method for aircraft flight risks provided by this embodiment has the following advantages:

[0146] 1. Effectively identify pilot control modes: This embodiment mines typical pilot control modes based on flight control data from the perspective of pilot control behavior. Specifically, three types of flight control data features are calculated from the pilot's control data on the joystick and throttle: timeliness, stability, and coordination. Based on these three types of data features, typical control modes contained in different pilot control data are clustered, such as Figure 8 shown.

[0147] Figure 8 The clustering results indicate that the pilots represented in the data exhibit three types of HCM. Based on practical experience in civil aviation safety, such as in the Flight Procedures Manual, HCM is defined to facilitate understanding by safety managers. The cluster centers and their corresponding coordinates are used to label each HCM as SOP-type, conservative, and adventurous. The corresponding interpretation table is shown below:

[0148] Table 2 Pilot control mode explanation

[0149]

[0150] 2. Accurately predict the real-time pilot control mode: The pilot's control mode during the real-time operation of the aircraft will have a direct impact on the aircraft's operating results. It is necessary to give the current pilot's control mode based on real-time flight data in actual operation to better evaluate its possible impact on flight safety. The present invention summarizes the typical pilot control mode prediction features from the "human-machine environment" factors formed by the pilot's control mode, and especially proposes three types of features from the pilot's cognitive perspective, namely situational awareness, decision-making performance, and task difficulty, and realizes accurate prediction of the pilot's control mode based on machine learning methods. The pilot control mode prediction sample results are as follows: Figure 9 shown.

[0151] Assuming that the pilot's selected flight control mode remains unchanged during the landing phase (50 ft to 0 ft, which lasts approximately 5 seconds) between 1300 ft and 50 ft, the historical high-dimensional sequence data can be sliced ​​by altitude to obtain sample data for building the pilot's HCM prediction model. Based on the samples constructed from the flight slice data and the HCM warning features, a machine learning-based HCM analysis model was constructed. Its performance on the test dataset is shown below:

[0152] Table 3 Comparison of prediction performance of HCM machine learning models

[0153]

[0154] As shown in Table 3, the decision tree has the best comprehensive performance for HCM, with all performance indicators exceeding 90%. Therefore, the decision tree is selected as the HCM analysis model for real-time warning of aircraft landing risks, so as to make more accurate warnings on the landing accident risks brought by different HCMs.

[0155] 3. Effective early warning of aircraft operation risks: Existing aircraft operation risk warnings often ignore the impact of real-time pilot control modes on flight safety. This embodiment provides a real-time early warning method for aircraft risks under typical flight profiles based on accurate prediction of real-time pilot control modes, and can provide accurate risk warning results at the optimal warning location. The sample results of aircraft operation risk warning are as follows: Figure 10 As shown. Figure 10 It can be seen that compared with not considering HCM, the HCM-based real-time flight risk warning model proposed in the invention can recall more risk samples. For risk warning, a difference of 1% may mean a serious flight accident. Therefore, accident risk prediction based on HCM has important practical significance.

[0156] The present application also provides an application scenario, which applies the above-mentioned real-time early warning method for aircraft flight risks. Specifically: the real-time early warning method for aircraft flight risks provided by this embodiment can be applied to the scenario of aviation safety management. The aviation safety management scenario includes flight operation links, maintenance and emergency response links. The real-time early warning method for aircraft flight risks provided by this embodiment belongs to the safety monitoring link in flight operations. By adopting the method of this embodiment during the safety monitoring link, a flight risk warning can be issued in a timely manner to remind pilots or ground operators to take corresponding countermeasures to ensure flight safety. The real-time early warning method for aircraft flight risks provided by this embodiment helps to improve the efficiency and accuracy of aviation safety management and ensure the safety and reliability of flight operations.

[0157] Example 2

[0158] This embodiment provides a computer device, which can be a server or a terminal. Its internal structure diagram can be as follows: Figure 11As shown. The computer device includes a processor, a memory, an input / output interface (Input / Output, abbreviated as I / O) and a communication interface. The processor, memory and input / output interface are connected through a system bus, and the communication interface is connected to the system bus through the input / output interface. The processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program and a database. The internal memory provides an environment for the operation of the operating system and computer program in the non-volatile storage medium. The input / output interface of the computer device is used to exchange information between the processor and an external device. The communication interface of the computer device is used to communicate with an external terminal through a network connection. When the computer program is executed by the processor, it implements a real-time early warning method for aircraft flight risks in Example 1.

[0159] Those skilled in the art will understand that Figure 11 The structure shown in the figure is only a block diagram of a part of the structure related to the solution of the present application, and does not constitute a limitation on the computer device to which the solution of the present application is applied. The specific computer device may include more or fewer components than shown in the figure, or combine certain components, or have a different component arrangement.

[0160] Example 3

[0161] This embodiment provides a computer device including a memory and a processor. The memory stores a computer program. When the processor executes the computer program, it implements the real-time warning method for aircraft flight risks in Example 1.

[0162] Example 4

[0163] This embodiment provides a computer-readable storage medium storing a computer program. When the computer program is executed by a processor, the method for real-time early warning of aircraft flight risks in embodiment 1 is implemented.

[0164] Example 5

[0165] This embodiment provides a computer program product, including a computer program. When the computer program is executed by a processor, it implements the real-time warning method for aircraft flight risks in Example 1.

[0166] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, stored data, displayed data, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties, and the collection, use and processing of relevant data must comply with relevant regulations.

[0167] Those skilled in the art will appreciate that all or part of the processes in the above-mentioned embodiment methods can be implemented by instructing the relevant hardware through a computer program, and the computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to memory, database or other media used in the embodiments provided in this application may include at least one of non-volatile and volatile memory. Non-volatile memory may include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetic random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory may include random access memory (RAM) or external cache memory, etc. By way of illustration and not limitation, RAM may be in various forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM).

[0168] The databases involved in the various embodiments provided herein may include at least one of a relational database and a non-relational database. Non-relational databases may include, but are not limited to, distributed databases based on blockchains. The processors involved in the various embodiments provided herein may include, but are not limited to, general-purpose processors, central processing units, graphics processing units, digital signal processors, programmable logic units, data processing logic units based on quantum computing, and the like.

[0169] The technical features of the above embodiments can be combined arbitrarily. To make the description concise, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0170] This document uses specific examples to illustrate the principles and implementation methods of this application. The description of the above examples is only intended to help understand the method and core concept of this application. At the same time, for those skilled in the art, based on the concept of this application, there may be changes in the specific implementation methods and application scope. In summary, the content of this specification should not be understood as limiting this application.

Claims

1. A real-time early warning method for aircraft flight risks, characterized in that: The aircraft flight risk real-time early warning method includes: Acquiring current flight data, wherein the flight data includes speed deviation, trajectory deviation, flight altitude, aircraft performance data, and flight environment data, wherein the speed deviation represents the pilot's situational awareness of the aircraft, the trajectory deviation represents the pilot's control decision-making ability for the aircraft, and the flight altitude represents the pilot's control difficulty for the aircraft; Inputting the current flight data into a trained pilot control mode prediction model to output the current control mode type, wherein the current control mode type remains unchanged in the current flight mission profile, wherein the flight mission profile refers to a specific flight operation of the aircraft, and the specific flight operation includes takeoff, climb, cruise, descent, and landing; Inputting aircraft performance data and flight environment data in the current flight data into a corresponding trained real-time flight risk warning model, and outputting a risk result of a target warning point under the current control mode type, wherein the risk result indicates whether the target warning point is at risk, and the control mode type of the corresponding trained real-time flight risk warning model is the same as the current control mode type; When the risk result indicates that the target warning point is at risk, a real-time warning is issued; The specific training process of the pilot control mode prediction model includes: Constructing a first training dataset, wherein the first training dataset includes historical flight data and control mode types of different pilots flying a specific aircraft model under a flight mission profile, the historical flight data including historical speed deviations, historical trajectory deviations, historical flight altitudes, historical aircraft performance data, and historical flight environment data, and the control mode types are obtained by processing the pilots' historical operational control data using a clustering algorithm; Using the historical flight data as feature data and the control mode type as label data, training the pilot control mode prediction model to obtain a trained pilot control mode prediction model; The historical operation control data for any pilot in any flight mission profile includes a historical control stick operation sequence and a historical throttle lever operation sequence, wherein the historical control stick operation sequence includes a plurality of historical control stick operation data, and the historical throttle lever operation sequence includes a plurality of historical throttle lever operation data; The specific process of determining the control mode type includes: determining a first sequence position of optimal historical stick operation data based on the historical stick operation sequence and a first threshold, wherein the first sequence position is used to evaluate the timeliness of the pilot's stick control; calculating, based on the historical sequence of stick operations, a degree of fluctuation in stick operations over a period of time, wherein the degree of fluctuation is used to evaluate the smoothness of the pilot's stick control; determining a second sequence position of optimal historical throttle stick operation data based on the historical throttle stick operation sequence and a second threshold, wherein the second sequence position is used to evaluate the timeliness of the pilot's throttle stick control; calculating a first deviation degree according to the first sequence position, the second sequence position, and a third threshold, wherein the first deviation degree is used to evaluate the coordination of pilot control; Combining the corresponding first sequence position, the fluctuation degree and the first deviation degree into a data point; A clustering algorithm is used to perform clustering processing on all the data points to determine the control mode type.

2. The method for real-time early warning of aircraft flight risk according to claim 1, characterized in that: The specific training process of the flight risk real-time warning model includes: For any pilot's historical flight data corresponding to a known control mode type under any flight mission profile, different slice reference units are determined based on the sampling frequency of warning variables and the actual number of warnings required, wherein the warning variables include warning time and / or warning altitude, and the historical flight data includes historical aircraft performance data and historical flight environment data; Slicing the historical flight data according to different slicing reference units to obtain different second training data sets, wherein each second training data set includes data slices corresponding to a plurality of warning points, and the warning points are determined according to the slicing positions of the slicing reference units; For different second training data sets, the historical flight data in each data slice is used as feature data, and the risk result of each warning point is used as label data to train the first flight risk real-time warning model; Calculating the average recall rate of different first flight risk real-time warning models; Selecting an optimal slice reference unit corresponding to the first flight risk real-time warning model having an average recall rate greater than a first recall rate threshold, and using a warning point having a first recall rate greater than a second recall rate threshold as the target warning point, wherein the first recall rate is the recall rate of the warning point corresponding to the optimal slice reference unit; Using historical flight data in the data slice corresponding to each target warning point under the same control mode type as feature data, and using the risk result of each target warning point as label data, training a second flight risk real-time warning model corresponding to different control mode types, wherein each of the second flight risk real-time warning models includes a decision tree model, a regression model, an XGBOOST model, and a LightGBM model; Calculate the second recall rate of the second flight risk real-time warning model under different types; The optimal second flight risk real-time warning model whose second recall rate is greater than the third recall rate threshold is selected as the trained flight risk real-time warning model corresponding to the final different control mode types.

3. The real-time early warning method for aircraft flight risk according to claim 1, characterized in that: The specific calculation process of the first sequence position includes: sequentially calculating the absolute values ​​of differences between adjacent historical joystick operation data; The sequence position of the historical joystick operation data whose absolute value is greater than the first threshold is used as the first sequence position.

4. The real-time early warning method for aircraft flight risk according to claim 1, characterized in that: The calculation expression of the fluctuation degree is: Wherein, HC_S represents the degree of fluctuation; n represents the number of the historical joystick operation data; MCS j represents the j-th historical joystick operation data; represents the mean value of all the historical joystick operation data.

5. The real-time early warning method for aircraft flight risk according to claim 1, characterized in that: The calculation expression of the first deviation degree is: HC_C=||HC_T M -HC_T A |-HC_C SOP |; Wherein, HC_C represents the first deviation degree; HC_T M Indicates the first sequence position; HC_T A Indicates the second sequence position; HC_C SOP represents the third threshold.

6. A computer device comprising: A memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement a real-time early warning method for aircraft flight risks according to any one of claims 1 to 5.

7. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the real-time early warning method for aircraft flight risks according to any one of claims 1 to 5 is implemented.

8. A computer program product comprising a computer program, characterized in that When the computer program is executed by a processor, the real-time early warning method for aircraft flight risks according to any one of claims 1 to 5 is implemented.

Images