A reactive wind shear alarm reliability evaluation method based on flight parameter data

Abstract

This invention belongs to the field of aircraft avionics airworthiness technology, specifically disclosing a reactive wind shear warning reliability assessment method based on flight parameter data. The method includes the following steps: extraction of flight wind shear operation data; wind shear warning event labeling; calculation of wind shear warning reliability assessment indicators; construction and training of a VAE-LSTM model; and using the trained VAE-LSTM model to assess the reliability of reactive wind shear warnings. This invention assesses and labels the reliability of wind shear warning events by analyzing sample data. Then, based on the VAE-ISTM model, it transforms wind shear identification reliability into an anomaly detection problem of time-series indicators, selecting the cumulative sum of energy variables as the judgment indicator. It designs and proposes an assessment method for the reliability of reactive wind shear warnings during the approach phase based on the cumulative sum of energy variables, providing technical support for avionics equipment manufacturers to optimize the design of reactive wind shear warning systems and for regulatory authorities to better conduct airworthiness certification of related equipment.

Description

Technical Field

[0001] This invention relates to the field of aircraft avionics airworthiness technology, specifically to a reactive wind shear warning reliability assessment method based on flight parameter data. Background Technology

[0002] Low-altitude wind shear is a weather phenomenon extremely detrimental to flight safety. It occurs rapidly, leaving pilots little time for reflection, decision-making, and action; failure to respond correctly and promptly can result in severe personnel and economic losses. To help pilots identify wind shear in a timely manner, onboard avionics provides a reactive wind shear warning system (RWS). This system uses aircraft inertial parameters and aerodynamic data to solve flight dynamics equations, calculating the rate of change of horizontal and vertical wind speed components in the aircraft's current location, thereby assessing the degree of continuous energy loss and attenuation. Analysis of crew responses and QAR data from flights triggering RWS alarms suggests that existing reactive wind shear warning systems may have some false alarms and missed alarms, adversely affecting normal flight operations and unnecessarily increasing the crew's workload.

[0003] In the prior art, there is an airborne wind shear meter and wind shear detection method (201710514160.1). This invention provides an airborne wind shear meter, including a radar antenna, a signal transmission system, a wind shear data processing system, and a wind shear warning output display system. It also provides a wind shear detection method based on this wind shear meter. There is also an airport low-altitude wind shear detection method and system (201610005901.9). This invention uses boundary layer wind profiler radar to detect the wind field over the airport in real time. The obtained actual meteorological observation data is used to issue warnings for wind shear in both the horizontal and vertical directions when the wind shear information exceeds the system threshold. Finally, there is a method for evaluating the operational quality of approach phase wind shear based on QAR data (202310231815.X). This invention uses QAR data to evaluate the operational quality of the aircraft crew after wind shear occurs during the approach phase. It is mainly used to evaluate the operational quality of the aircraft crew and does not involve judging the effectiveness of the warning.

[0004] Currently, existing research on wind shear mainly focuses on alarm identification methods and optimization, primarily addressing the principles and processes of alarms and how to improve them. However, there is a gap in the assessment and judgment of the accuracy and reliability of reactive wind shear alarm identification. This poses significant challenges to the airworthiness certification and improved design of airborne reactive wind shear alarm systems. Current research on reactive wind shear alarms concentrates on the design and improvement of alarm detection methods; research on alarm reliability assessment has not yet been reported.

[0005] Given that QAR data can accurately reflect the pilot's maneuvers, aircraft status, and the characteristics of the external environment, a reactive wind shear warning reliability assessment method based on flight parameter data can be established by selecting relevant parameters that can objectively reflect the aircraft's status in the wind shear region. Summary of the Invention

[0006] To address the problems existing in the prior art, this invention provides a reliability assessment method for reactive wind shear warnings based on flight parameter data. First, sample data is collected, and the reliability of wind shear warning events is assessed and calibrated through analysis of the sample data. Then, based on the VAE-ISTM model, the reliability of wind shear identification is transformed into an anomaly detection problem of time-series indicators. The cumulative sum of energy variables is selected as the judgment index. A reliability assessment method for reactive wind shear warnings during the approach phase based on the cumulative sum of energy variables is designed and proposed. This provides technical support for avionics equipment manufacturers to optimize the design of reactive wind shear warning systems and for regulatory authorities to better conduct airworthiness certification of related equipment, thus solving the problems mentioned in the background art.

[0007] To achieve the above objectives, the present invention provides the following technical solution: a reactive wind shear alarm reliability assessment method based on flight parameter data, comprising the following steps:

[0008] S1. Extraction of flight wind shear operation data;

[0009] S2, Wind shear warning event labeling;

[0010] S3, Calculation of reliability assessment indicators for wind shear alarm;

[0011] S4. Construction and training of VAE-LSTM model;

[0012] S5. Use the trained VAE-LSTM model to evaluate the reliability of reactive wind shear alarms.

[0013] Preferably, in step S1, flight data is exported, and events that trigger reactive wind shear alarms during the approach phase are filtered out. Finally, the flight wind shear operation data of the approach phase that triggered the reactive wind shear alarm is extracted.

[0014] Preferably, in step S2, all wind shear alarm events are divided into two categories: valid wind shear alarms and invalid wind shear alarms. A valid wind shear alarm is one in which wind shear is significant, specifically referring to: significant changes in wind speed and direction, with airspeed changing by more than 15 kt in a short period of time or airspeed / climax changing by more than 500 ft / min in a short period of time. An invalid wind shear alarm is one in which there are no significant wind shear characteristics, specifically referring to: no significant changes in wind speed and direction or airspeed / climax rapidly recovering after the change.

[0015] Preferably, in step S3, the cumulative sum of energy variability during the final approach phase is calculated based on the extracted sample events that triggered the reactive wind shear alarm; specifically, this includes: calculating the aircraft's energy variability E from the parameters extracted in step S1 from the flight wind shear operation data. t The calculation formula is as follows:

[0016]

[0017] The cumulative sum of energy variation rates during the final approach phase is calculated using the following formula:

[0018]

[0019] Where IVV is the rate of increase / decrease, TAS is the vacuum rate, and t start The time point at which the final approach phase begins, t end This is the point in time when the final approach phase ends.

[0020] Preferably, step S4 specifically includes the following steps:

[0021] S41. Based on the calculated cumulative sum of energy variation rates during the final approach phase, select data from normal approaches before triggering the alarm as the training and test sets; the training set data format is X. train = [x1,x2,…,x L [], where L represents the time span of the training data; the test set data is defined as X. test =[x1,x2,...,x L' ], L' is the length of the test set;

[0022] S42. Perform data preprocessing on the training and test sets, including: standardizing the training set by calculating its mean and variance as standardization parameters; then, using these standardization parameters, performing the same standardization on the test set to ensure consistent data distribution between the training and test sets. The standardization process is expressed by the following formula:

[0023]

[0024] Where x is the original data, mean is the mean of the training set, and std is the variance of the training set;

[0025] S43. The VAE-LSTM model is trained using preprocessed training set data and then validated using test set data to obtain the trained VAE-LSTM model.

[0026] Preferably, in step S43, during the VAE-LSTM model training phase, the original data is first divided into multiple subsequences using a sliding window, and the window data w is defined. t Based on a given time t and window length l, the original data X = [x1,...,x] can be... L-1 ,x L The data is divided into multiple sliding window data W = [w1,...,w...] L-l ,w L-l+1 ], containing L-l+1 windows of data, where each window of data is represented as w t =[x t-l+1 ,...,x t-1 ,x t ];

[0027] The VAE model is pre-trained using the pre-divided window data W, and the weights of the encoder and decoder in the VAE model are saved. Then, the window data is combined and concatenated to further divide it into n sliding windows. The encoder of the pre-trained VAE model extracts features, and the extracted feature vectors are input into the LSTM model for prediction. Then, the decoder with fixed weights restores the data to the initial data dimension, and finally, the reconstructed data is calculated. and raw input data The arithmetic square root is used as the reconstruction error loss. recon The formula is expressed as follows;

[0028]

[0029] Preferably, in step S5, the detection results of the VAE-LSTM model are compared with the wind shear alarm event rating to obtain a wind shear alarm reliability assessment.

[0030] Preferably, to detect whether the wind shear alarm of the VAE-LSTM model is abnormal, a threshold is set using the F1 score. When the reconstruction error exceeds this threshold, it indicates that the reconstructed data of the current window differs significantly from the actual data, and the alarm is reasonable. The F1 score calculation formula is expressed as follows:

[0031]

[0032] In the formula: P represents detection accuracy; R represents recall; F1 represents F1 score; n TP A true positive result represents the number of data points that are actually abnormal even if the test result is incorrect; n FP A false positive represents the number of data points that tested abnormal but were actually normal; n FN A false negative indicates the number of data points that tested normal but were actually abnormal.

[0033] The beneficial effects of this invention are as follows: Based on QAR data and the cumulative sum of energy variability during the final approach phase, this invention designs a method for evaluating the reliability of wind shear warnings through a VAE-LSTM model and expert verification. 1) This method, through the analysis of QAR data, can serve as a way to judge the optimization effect during subsequent warning system optimization; 2) By analyzing the cumulative sum of energy variability through VAE-LSTM and combining it with warnings from airborne avionics equipment, the pilot's wind shear recovery maneuvers can be evaluated more effectively, providing a more reasonable explanation for some maneuvers occurring before the system warning time; 3) Through statistical analysis of a large amount of real wind shear data, the algorithm structure can be continuously optimized to obtain an index that better reflects the intensity of wind shear. Attached Figure Description

[0034] Figure 1 This is a schematic diagram of the reliability assessment process for the wind shear alarm system in an embodiment of the present invention;

[0035] Figure 2 This is a schematic diagram of the architecture of the approach stage reactive wind shear alarm reliability assessment method in an embodiment of the present invention;

[0036] Figure 3 This is a flowchart illustrating the calculation of energy variability accumulation in an embodiment of the present invention;

[0037] Figure 4 This is a flowchart of the evaluation process based on VAE-LSTM in an embodiment of the present invention;

[0038] Figure 5 This is a diagram of the internal structure of the LSTM in an embodiment of the present invention;

[0039] Figure 6 This is a schematic diagram of IAS and IVV data before and after a certain RWS at Lhasa Airport in Embodiment 3 of the present invention;

[0040] Figure 7 This is a schematic diagram of IAS and IVV data before and after two RWS events at Kunming Airport in Embodiment 3 of the present invention;

[0041] Figure 8 This is a schematic diagram of the wind shear segment reconstruction error at Lhasa Airport in Embodiment 3 of the present invention;

[0042] Figure 9 This is a schematic diagram of the wind shear segment reconstruction error at Kunming Airport in Embodiment 3 of the present invention;

[0043] Figure 10 This is a schematic diagram of wind shear detection at Lhasa Airport based on VAE-LSTM in Embodiment 3 of the present invention;

[0044] Figure 11 This is a schematic diagram of wind shear detection at Kunming Airport based on VAE-LSTM in Embodiment 3 of the present invention;

[0045] Figure 12 This is a schematic diagram of IAS and IVV data for a certain wind shear event at Kunming Airport. Detailed Implementation

[0046] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0047] Example 1

[0048] This invention designs a reactive wind shear warning reliability assessment method based on flight parameter data, based on the cumulative sum of energy variability during the final approach phase of an aircraft. This method can be used as a benchmark to optimize wind shear warning systems. The specific process is as follows: Figure 1 As shown, it includes the following steps:

[0049] Step 1: Extraction of flight wind shear operation data

[0050] Export flight data using decoding software such as AirFase or AGS, filter out events that trigger reactive wind shear alarms during the approach phase, and extract the final approach phase flight wind shear operation data that triggered the reactive wind shear alarm based on the required QAR data fields and flight phase.

[0051] Step 2: Wind Shear Alarm Event Labeling

[0052] The airborne system records all wind shear warning events, but not all warnings indicate that the aircraft has actually encountered wind shear. Therefore, flight experts need to analyze the QAR data to classify all wind shear warning events into two categories: valid wind shear warnings (significant wind shear) and invalid wind shear warnings (no significant wind shear characteristics).

[0053] Step 3: Calculation of Reliability Assessment Indicators for Wind Shear Alarm

[0054] Energy variability represents the aircraft's own energy level under the influence of wind shear in the wind shear area. It can comprehensively reflect the state of the aircraft. Select the sample events that triggered the reactive wind shear alarm extracted in step 1 and calculate the cumulative sum of energy variability in the final approach phase.

[0055] Step 4: Evaluation method based on VAE-LSTM model

[0056] Variational autoencoder (VAE) models are generative deep learning models that can reduce data dimensionality while preserving data features. Long Short-Term Memory (LSTM) networks are a special type of recurrent neural network that can compensate for the shortcomings of VAE models in time series prediction. Based on the cumulative sum of energy variability in the final approach phase calculated in step 3, data from normal approaches before the alarm is triggered are selected as the training set. The data for the triggered reactive wind shear alarm are then tested, and the test results are compared with the event levels of each sample given by the flight expert in step 2 to verify the rationality of the method. A method for assessing alarm reliability based on the cumulative sum of energy variability is presented.

[0057] Example 2

[0058] A reactive wind shear alarm reliability assessment method based on flight parameter data includes the following steps:

[0059] Step 1: Extraction of flight wind shear operation data

[0060] Decoding software such as AIRFASE or AGS is used to read flight QAR data. Wind shear operation-related parameter extraction templates are established for different decoding libraries used by different aircraft. The flight QAR data is scanned using the templates to identify flights that trigger reactive wind shear alarms and extract wind shear operation-related data for those flights. The data table is shown in Table 1.

[0061] Table 1 Flight Wind Shear Operation Data

[0062]

[0063] Step 2: Wind Shear Alarm Event Labeling

[0064] like Figure 2 As shown, this method mainly learns the indicators through the model and compares them with the calibration of flight experts. Therefore, flight experts need to analyze the flight wind shear operation data that triggers the reactive wind shear alarm by combining the judgment parameters in the flight wind shear operation data extracted in step 1, and classify the events that trigger the reactive wind shear alarm as described in Table 2:

[0065] Table 2 Classification of Wind Shear Warning Events

[0066]

[0067] Step 3: Calculation of Reliability Assessment Indicators for Wind Shear Alarm

[0068] Because aircraft encounter strong airflow changes in a short period of time within the wind shear area, their airspeed and rate of climb / deceleration are significantly affected. However, since the units and numerical values ​​of airspeed and rate of climb / deceleration are not standardized, the energy variability index, which can comprehensively reflect the changes in both, is chosen as the evaluation index for the reliability of wind shear warnings. Figure 3 The calculation process of the energy variability index has been demonstrated. The specific calculation method of the index will be explained in detail below.

[0069] (1) Energy variability

[0070] The energy variability index, starting from the changes in aircraft energy, characterizes the impact of wind shear on aircraft energy attenuation. The aircraft's energy variability E is calculated from the parameters extracted from the flight wind shear operation data in step 1. t The calculation formula is shown in equation (1).

[0071]

[0072] (2) Cumulative sum of energy variability

[0073] In theory, during the final approach phase, the aircraft should maintain a stable approach along the glide path. Therefore, calculating the cumulative sum of energy variability and presenting the energy variability trend per second can better analyze the intensity of wind shear and thus assess the reliability of reactive wind shear warnings. The calculation formula is shown in (2).

[0074]

[0075] Among them, t start The time point at which the final approach phase begins, t end This is the point in time when the final approach phase ends.

[0076] Step 4: Evaluation based on VAE-LSTM model

[0077] This method is based on univariate time series anomaly detection analysis to assess the reliability of reactive wind shear warnings. It mainly consists of two parts: model training and anomaly detection. The training set data format is X. train =[x1,x2,...,x L [], where L represents the time span of the training data. The test set data is defined as X. test =[x1,x2,...,x L' L' is the length of the test set. Since QAR data is time series data, a fixed-length input and output are generated using a sliding window approach, and the anomaly score between the reconstructed sample and the normal sample is used to determine whether it is defined as an anomaly. Figure 4 The overall architecture of the model is presented.

[0078] (1) Data preprocessing

[0079] Because the time between the wind shear warning being triggered and the pilot taking over control is short, interpolation is performed on the calculated cumulative energy rate data to increase the amount of data for model training.

[0080] Given that QAR parameters are time-series data, standardization is typically performed during data preprocessing to further improve the model's robustness. Standardization aims to eliminate dimensional differences between features, making them more similar in scale and helping the model learn data patterns more effectively. This data preprocessing method is applied to both the training and test sets. First, the training set is standardized, and its mean and variance are calculated as standardization parameters. Then, using these parameters, the test set is subjected to the same standardization process to ensure consistent data distribution between the training and test sets. Standardization can be expressed by the following formula:

[0081]

[0082] Where x is the original data, mean is the mean of the training set, and std is the variance of the training set.

[0083] (2) VAE model

[0084] The VAE model, short for Variational Autoencoder, aims to reconstruct the original data using a set of encoders and decoders, making the generated data x' as similar as possible to the original data x. Compared to traditional AE models that seek mappings between single values, VAE models seek mappings between distributions, further enhancing the data generation capability. By maximizing the lower bound of evidence function ELBO, it ensures that x' most similar to the original data x can be generated from the approximate posterior distribution p(z|x) that is closest to the prior distribution q(z). The lower bound of evidence function is...

[0085] ELBO = E q [logp(x|z)]-KL(q(z)||p(z|x)) (4)

[0086] Among them, E q [logp(x|z)] is the reconstruction error, which measures whether the decoder can accurately reconstruct the original data when the latent variable z is sampled from the approximate posterior distribution p(z|x). KL(q(z)||p(z|x)) is the KL divergence, which measures the difference between the approximate posterior distribution p(z|x) and the prior distribution q(z) of the latent variable z.

[0087] This invention selects a CNN (Convolutional Neural Network) as the fully connected layer in the VAE (Visual Algorithm for Enhancing the Model) model. To enable the model to learn the features of different parts of the QAR data, the original data is divided into multiple subsequences using a sliding window approach. The window data w is defined as follows: t Based on a given time t and window length l, the original data X = [x1, ..., x...] can be... L-1 ,x L The data is divided into multiple sliding window data W = [w1, ..., w...] L-l ,w L-l+1 ], containing L-l+1 windows of data, where each window of data can be represented as w t =[x t-l+1 ,...,x t-1 ,x t ].

[0088] To enable the model to learn more features contained in the data, convolutional layers with strides are used instead of pooling layers. This allows the model to flexibly choose appropriate downsampling methods based on the characteristics of the data, rather than using a fixed preset pooling operation. The partitioned window data W is input to the encoder. Two one-dimensional convolutional layers process the input data to extract features from the temporal data. Then, two fully connected layers are used to calculate the mean and log-variance of the latent variable z. Finally, a reparameterization technique is used to combine the mean, log-variance, and random noise of the latent variable z to generate the latent variable z. The decoder reconstructs the latent features z back to the original input window size through convolution. The convolution operation formula is as follows:

[0089]

[0090] In the formula, Let f(·) be the i-th feature obtained from the k-th layer through convolution, and f(·) be the ReLU activation function. The weights used in the i-th convolution kernel of the k-th layer are the weights used in the operation, * represents the convolution operator, X k-1 The features output by the (k-1)th layer (the first layer is the input layer, W) This represents the bias of the i-th convolutional kernel in the k-th layer.

[0091] (3) LSTM model

[0092] LSTM stands for Long Short Term Memory network. By adding special gating structures (forget gate, input gate, and output gate), the LSTM model can selectively remember or forget specific information from previous events, thus solving the gradient vanishing and gradient exploding problems of traditional RNNs (Recurrent Neural Networks). The LSTM structure is as follows: Figure 5 As shown.

[0093] Where, x t For the information input this time, c t-1 and c t These represent the cell states retained from the previous round and the current cell states, respectively. t-1 and h t These are the output information from the previous round and the current round, respectively. t This is a forget gate, where the sigmoid activation function controls the information retained from the previous cell state. t The input gate is used to determine the current input x through an activation function. t The information. Combining the results of the forgetting gate and the input gate, the current cell state c can be obtained. t o t As the output gate, the updated cell state c t The output h is determined by the activation function. t The information included is partial. The calculation formulas for each part are as follows:

[0094]

[0095] (4) VAE-LSTM model

[0096] The present invention employs the following process for detecting and verifying the effectiveness of wind shear warnings: first, the target data is preprocessed; then, the VAE-LSTM model is trained using training data; and finally, the trained model is used to process the test data to verify the model's effectiveness.

[0097] The preprocessing stage includes metric calculation, data standardization, and window partitioning. The model training stage first pre-trains the VAE model using the partitioned window data W, then saves the encoder and decoder weights in the VAE model, and then combines and concatenates the window data, further dividing it into n sliding windows. The encoder of the pre-trained VAE model extracts features, and the extracted feature vectors are input into the LSTM model for prediction. Then, the decoder with fixed weights restores the data to the initial data dimension, and finally, the reconstructed data is calculated. and raw input data The arithmetic square root is used as the reconstruction error loss. recon .

[0098]

[0099] To detect whether wind shear alarms are abnormal, a threshold needs to be set. If the reconstruction error exceeds this threshold, it indicates that the reconstructed data in the current window differs significantly from the actual data, and the alarm is reasonable; conversely, if the error is less than the threshold, the alarm is not valid. Based on the reconstruction error, 200 threshold reference values ​​are selected, and the F1 score for each threshold reference value is calculated. Since the F1 score comprehensively considers detection accuracy and recall, it can effectively cover all anomalies.

[0100]

[0101] In the formula: P represents detection accuracy, R represents recall, F1 represents F1 score, and n TP A true positive result represents the number of data points where the detected abnormality is also actually present. (n) FP A false positive represents the number of data points that tested abnormal but were actually normal, n. FN A false negative indicates the number of data points that appear normal but are actually abnormal. Considering that wind shear alarms and similar anomalies are typically sequential anomalies, all windows containing anomalies will be marked as abnormal.

[0102] Example 3

[0103] To ensure the applicability of the algorithm as much as possible, QAR data of landing flights at two airports where wind shear frequently occurs were selected as experimental subjects. Although wind shear warnings are mainly triggered during the final approach phase, the duration of the final approach phase may vary for different flights. To ensure training efficiency, 100 sets of normal approach and landing flight data were selected for each airport, including data from the first 200 seconds of the landing phase before the wind shear warning was triggered, as the training dataset. Considering that the triggering frequency of wind shear warnings is relatively low, 10 sets of flight data from Lhasa Airport and 12 sets from Kunming Airport that triggered reactive wind shear warnings during the final approach phase were selected. Furthermore, considering that after triggering a reactive wind shear warning, the pilot will perform a go-around maneuver, and after implementing a series of operations such as TOGA thrust, the aircraft will be affected not only by external wind shear but also by changes in response to the pilot's maneuvers. Therefore, data from 200 seconds prior to the reactive wind shear warning time was used as the test dataset. First, based on the advice of flight experts, the test dataset was analyzed and it was found that although the RWS was triggered in the second set of verification flight segment data at Lhasa Airport, the airspeed and rate of climb / descendancy did not change much. Figure 6 The data for this flight includes IAS and IVV, with the red vertical line representing the alarm time shown in the QAR data. Clearly, in the period prior to the alarm for the second flight segment, the IAS did not change by more than 15 kt, and the IVV did not change by more than 500 ft / min; therefore, it should not be considered that this flight segment encountered wind shear. Similarly, analysis of 12 sets of verification data from Kunming Airport revealed that two sets did not meet the loss criteria for airspeed and rate of climb / deceleration, as shown below. Figure 7As shown. After confirming that the remaining 19 sets of flight segment data were all genuine and valid RWS, the data for this flight segment was retained as invalid RWS for comparison and verification in the test set.

[0104] The VAE-LSTM model was used to detect flight data that triggered reactive wind shear warnings during the approach phase of 10 groups of flights to Lhasa Airport and 12 groups of flights to Kunming Airport. The reconstruction error was obtained as follows: Figure 8 , Figure 9 As shown, based on the optimal F1 principle, thresholds of 0.1961 and 0.2046 are selected for Lhasa Airport and Kunming Airport respectively in the preset thresholds. The red dotted line in the figure represents the selected threshold.

[0105] Based on the set threshold, the detection results are obtained as follows: Figure 10 , Figure 11 As shown in the figure, the blue curve represents the energy variability of the test flight. The alarm times of the airborne wind shear warning system are marked with red dashed lines, and the wind shear recognition windows calculated by the model are filled with green solid lines. As shown in the figure, 9 out of 10 reactive wind shear alarms at Lhasa Airport were successfully detected, all of which were valid RWS. The invalid RWS used for comparison verification in the second flight segment was not identified. All 10 sets of wind shear verification data at Kunming Airport were successfully detected, and the remaining 2 sets were identified as weak wind shear as previously determined. The accuracy results are shown in Table 3.

[0106] Table 3. Parameters for Detecting the Effectiveness of Wind Shear at Different Airports

[0107]

[0108] The main factors affecting accuracy and other metrics are the two sets identified multiple times. Therefore, the following analysis combines the original QAR data to explain the main reasons.

[0109] Case Analysis: By combining flight expert advice and raw QAR data, the main reasons for the model's early identification of wind shear in the example flight segments at Lhasa Airport and Kunming Airport are as follows: (1) In the model identification window data, the airspeed or rate of climb related to wind shear altitude showed large fluctuations; (2) The radio altitude of the model's early identification window position was above 2000ft, while the airborne wind shear warning system usually only activates when the radio altitude is below 1500ft. Therefore, it can be considered that the model's identification capability not only maintains a certain identification rate at low altitudes, but also has certain reference value for wind shear identification at high altitudes. The following is a specific analysis of whether its early identification is reliable, taking the second group of wind shear at Kunming Airport as an example. Figure 12As shown, the speed and rate of ascent / decline parameters for this flight segment are presented. It is easy to see that between 80 and 150, the IAS drops sharply multiple times (more than 15 kt), and even falls below VAPP for 8 consecutive seconds. Furthermore, the IVV drops significantly at 130 and 183 (more than 500 ft / min). Therefore, it can be concluded that the model detection results are accurate.

[0110] It should be noted that, in this invention, the terms "comprising," "including," or any other variations thereof are intended to cover a non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitation, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0111] The terminology used in the embodiments of this invention is for the purpose of describing particular embodiments only and is not intended to limit the invention. The singular forms “a,” “the,” and “the” as used in the embodiments of this invention and the appended claims are also intended to include the plural forms unless the context clearly indicates otherwise.

[0112] It should be understood that the term "and / or" used in this invention is merely a description of the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent: A existing alone, A and B existing simultaneously, or B existing alone. Furthermore, the character " / " in this invention generally indicates that the preceding and following related objects have an "or" relationship.

[0113] Depending on the context, the word "if" as used here can be interpreted as "when," "when," "in response to determination," or "in response to detection." Similarly, depending on the context, the phrase "if determination" or "if detection (of the stated condition or event)" can be interpreted as "when determination," "in response to determination," "when detection (of the stated condition or event)," or "in response to detection (of the stated condition or event)."

[0114] The terms "first" and "second" used in the embodiments are merely to distinguish similar objects and do not represent a specific ordering of objects. It is understood that "first" and "second" can be interchanged in a specific order or sequence where permitted. It should be understood that the objects distinguished by "first" and "second" can be interchanged where appropriate so that the embodiments described herein can be implemented in an order other than those illustrated or described herein.

[0115] Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions described in the foregoing embodiments or make equivalent substitutions for some of the technical features. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for assessing the reliability of reactive wind shear alarms based on flight parameter data, characterized in that, Includes the following steps: S1. Extraction of flight wind shear operation data; S2, Wind shear warning event labeling; S3. Calculation of wind shear alarm reliability assessment indicators; Based on the extracted sample events that triggered the reactive wind shear alarm, calculate the cumulative sum of energy variability during the final approach phase; Specifically, this includes: obtaining the aircraft's energy variability from the parameters calculated in the flight wind shear operation data extracted in step S1. The calculation formula is as follows: ； The cumulative sum of energy variation rates during the final approach phase is calculated using the following formula: ； in, IVV For the rate of rise and fall, TAS Vacuum rate, This marks the start of the final approach phase. This is the point in time when the final approach phase ends; S4. Construction and training of the VAE-LSTM model; specifically including the following steps: S41. Based on the calculated cumulative sum of energy variability during the final approach phase, select data from normal approaches before triggering the alarm as the training and test sets; the training set data format is as follows: , The training set data represents the time span of the training data; the test set data is defined as follows: , It is the length of the test set; S42. Perform data preprocessing on the training and test sets, including: standardizing the training set by calculating its mean and variance as standardization parameters; then, using these standardization parameters, performing the same standardization on the test set to ensure consistent data distribution between the training and test sets. The standardization process is expressed by the following formula: ； in It is the raw data. It is the mean of the training set. It is the variance of the training set; S43. Training the VAE-LSTM model: The VAE-LSTM model is trained using preprocessed training set data. During the VAE-LSTM model training phase, the original data is first divided into multiple subsequences using a sliding window method, defining the window data... Based on a given time and window length The raw data can be Divided into multiple sliding window data ,Include 1 window of data, where each window of data is represented as ; Use the predefined window data The VAE model is pre-trained, and the weights of the encoder and decoder in the VAE model are saved. Then, the window data is combined and concatenated, and further divided into... A sliding window The encoder of the pre-trained VAE model extracts features, and the extracted feature vectors are input into the LSTM model for prediction. Then, the decoder with fixed weights restores the data to the initial data dimension, and finally, the reconstructed data is calculated. and raw input data The arithmetic square root as the reconstruction error The formula is expressed as follows; ； Then, the test set data is used for validation to obtain the trained VAE-LSTM model; S5. Use the trained VAE-LSTM model to evaluate the reliability of reactive wind shear alarms.

2. The reactive wind shear alarm reliability assessment method based on flight parameter data according to claim 1, characterized in that: In step S1, flight data is exported, and events that trigger reactive wind shear alarms during the approach phase are filtered out. Finally, the flight wind shear operation data that triggered the reactive wind shear alarm during the final approach phase is extracted.

3. The reactive wind shear alarm reliability assessment method based on flight parameter data according to claim 1, characterized in that: In step S2, all wind shear alarm events are divided into two categories: valid wind shear alarms and invalid wind shear alarms. A valid wind shear alarm is one in which wind shear is significant, specifically referring to: significant changes in wind speed and direction, with airspeed changing by more than 15 kt or airspeed / climax changing by more than 500 ft / min in a short period of time. An invalid wind shear alarm is one in which there are no significant wind shear characteristics, specifically referring to: no significant changes in wind speed and direction or airspeed / climax recovering rapidly after the change.

4. The reactive wind shear alarm reliability assessment method based on flight parameter data according to claim 1, characterized in that: In step S5, the detection results of the VAE-LSTM model are compared with the wind shear alarm event rating to obtain a wind shear alarm reliability assessment.

5. The reactive wind shear alarm reliability assessment method based on flight parameter data according to claim 4, characterized in that: To detect whether wind shear alarms in the VAE-LSTM model are abnormal, an F1 score is used to set a threshold. When the reconstruction error exceeds this threshold, it indicates that the reconstructed data in the current window differs significantly from the actual data, and the alarm is reasonable. The F1 score calculation formula is expressed as follows: ； In the formula: This represents the accuracy of the test; Represents recall rate; Represents the F1 score; A true positive result represents the number of data points that are actually abnormal and also detected as abnormal. A false positive indicates the number of data points that tested abnormal but were actually normal. A false negative indicates the number of data points that tested normal but were actually abnormal.

Images