A method, system, electronic device and storage medium for bump prediction correction

Abstract

This invention relates to the field of aviation safety prediction technology, and discloses a turbulence prediction and correction method, system, electronic device, and storage medium. The method includes acquiring multi-dimensional flight state data and performing classification preprocessing, including anomaly detection and missing value recovery for turbulence intensity values ​​and vertical overload values; converting aircraft type identifiers into low-dimensional embedding vectors; and standardizing the distribution of vertical flight path angles and horizontal stabilizer distances. Based on the preprocessed data, a time window is constructed, input into a spatiotemporal feature extraction model for multi-dimensional feature fusion and temporal correlation analysis, and outputting a correction coefficient sequence corresponding to the time window. The turbulence intensity value at each time point within the time window is multiplied by the corresponding correction coefficient sequence point by point to generate a turbulence intensity prediction value sequence for non-level flight conditions. The spatiotemporal feature extraction model is trained based on a composite loss function including a key scenario enhancement mechanism and a non-negativity protection mechanism. This invention can improve the accuracy and reliability of turbulence prediction.

Description

Technical Field

[0001] This invention belongs to the field of aviation safety prediction technology, and in particular relates to a turbulence prediction and correction method, system, electronic device and storage medium. Background Technology

[0002] Currently, aviation turbulence prediction is mainly based on the equivalent turbulence intensity (EDR) value under level flight conditions. However, under non-level flight attitudes such as climb / descent, changes in the aircraft's vertical trajectory angle will introduce attitude disturbance components, causing significant deviations when the level flight EDR is directly applied to non-level flight conditions.

[0003] In existing technologies, traditional correction methods use linear compensation models. However, in reality, the relationship between attitude components and actual turbulence components is affected by the nonlinear coupling of multiple factors such as aircraft aerodynamic characteristics, flight path angle, and turbulence intensity. Linear models are difficult to accurately represent complex correlations.

[0004] Therefore, this application provides a turbulence prediction correction method to solve the above-mentioned technical problems. Summary of the Invention

[0005] The purpose of this invention is to provide a turbulence prediction correction method, system, electronic device and storage medium to solve the technical problem that the prediction of existing technologies is characterized by relatively simple factors, which makes it impossible to make multi-dimensional predictions and thus results in low prediction accuracy.

[0006] To address the aforementioned technical problems, this invention provides a turbulence prediction and correction method, comprising:

[0007] Acquire multi-dimensional flight status data, which includes turbulence intensity value in level flight, vertical overload value in non-level flight, vertical flight path angle, aircraft type identification, and horizontal stabilizer distance.

[0008] The multi-dimensional flight status data is classified and preprocessed, including anomaly detection and missing value recovery for the turbulence intensity value and the vertical overload value, converting the aircraft type identifier into a low-dimensional embedding vector, and performing distribution standardization processing on the vertical flight path angle and the horizontal stabilizer distance.

[0009] A time window is constructed based on the preprocessed multi-dimensional flight status data. The spatiotemporal feature extraction model is input to perform multi-dimensional feature fusion and time correlation analysis, and the correction coefficient sequence corresponding to the time window is output.

[0010] Multiply the turbulence intensity value at each time point within the time window with the corresponding correction coefficient sequence point by point to generate a turbulence intensity prediction value sequence for non-level flight conditions.

[0011] The spatiotemporal feature extraction model is trained based on a composite loss function that includes a key scenario enhancement mechanism and a non-negativity guarantee mechanism to optimize the prediction accuracy of the correction coefficient sequence.

[0012] In some specific embodiments, the multi-dimensional flight status data is classified and preprocessed, including anomaly detection and missing value recovery for the turbulence intensity value and the vertical overload value, converting the aircraft type identifier into a low-dimensional embedding vector, and performing distribution standardization processing on the vertical flight path angle and the horizontal stabilizer distance, further including:

[0013] Filter out bump intensity values ​​that exceed the preset intensity range and vertical overload values ​​that exceed the preset load range;

[0014] Based on the time series continuous interpolation method, the missing turbulence intensity value and vertical overload value are recovered by the correlation of data at adjacent time points;

[0015] The model identifier is mapped to a continuous vector space of fixed dimensions to generate an embedded vector;

[0016] Perform a zero-mean, unit-variance linear transformation on the vertical flight path angle and the horizontal stabilizer distance.

[0017] In some specific embodiments, constructing a time window based on the preprocessed multi-dimensional flight status data further includes:

[0018] Set the length of the continuous time window that covers the entire lifecycle of the turbulence event;

[0019] A sliding window mechanism with a preset overlap ratio is used to segment the time series, maintaining the temporal continuity of adjacent windows;

[0020] The preprocessed multi-dimensional flight status data is concatenated along the feature dimensions to generate a combined feature vector.

[0021] The time window for constructing the three-dimensional tensor sample includes dimensions such as time step dimension, combined feature dimension, and single channel dimension.

[0022] In some specific embodiments, the spatiotemporal feature extraction model is input for multidimensional feature fusion and temporal correlation analysis, and the output is a sequence of correction coefficients corresponding to the time window, further including:

[0023] The feature compression path includes extracting long-term trend features through multi-level time-dimensional downsampling operations. Each level contains convolution operations at different time scales, gradually expanding the temporal receptive field.

[0024] The feature fusion layer includes aggregating the long-term trend features and eliminating information redundancy to form global features;

[0025] The feature recovery path includes reconstructing the temporal resolution through temporal upsampling operations and fusing global features and local transient features through a cross-level feature connection mechanism to generate spatiotemporal features;

[0026] The coefficient mapping layer includes converting spatiotemporal features into a complete time-dimensional sequence of corrected coefficients.

[0027] In some specific embodiments, the feature compression path includes extracting long-term trend features through multi-level time-dimensional downsampling operations, each level containing convolution operations at different time scales to gradually expand the temporal receptive field, and further includes:

[0028] The temporal convolution operation sequence with increasing dilation rate is executed to gradually expand the temporal receptive field by adjusting the sampling interval of the convolution kernel;

[0029] A non-linear activation function and batch normalization are introduced after each convolution operation to enhance feature representation capabilities.

[0030] Insert a temporal compression operation after the specified convolutional layer to reduce temporal resolution while increasing feature depth;

[0031] The output includes a composite feature tensor that includes the long-term aerodynamic effects of the aircraft type, the evolution trend of the vertical flight path angle, and the baseline level of the turbulence intensity value.

[0032] In some specific embodiments, the turbulence intensity value at each time point within the time window is multiplied by the corresponding correction coefficient sequence point by point to generate a turbulence intensity prediction value sequence for non-level flight conditions, further including:

[0033] Get the turbulence intensity values ​​at each time point within the current time window;

[0034] Call the trained spatiotemporal feature extraction model and output the correction coefficient sequence for the corresponding time window;

[0035] The turbulence intensity value at the same time point is multiplied by the corresponding correction coefficient to generate the turbulence intensity prediction value.

[0036] Output a real-time updated sequence of predicted turbulence intensity values.

[0037] In some specific embodiments, the spatiotemporal feature extraction model is trained based on a composite loss function including a key scene enhancement mechanism and a non-negativity guarantee mechanism to optimize the prediction accuracy of the correction coefficient sequence, further including:

[0038] The scenario weighting mechanism includes dynamically calculating the error weighting coefficients at each time point based on the turbulence intensity value and the absolute value of the vertical flight path angle, thereby strengthening the error contribution of key scenarios;

[0039] The error term includes the mean squared error between the predicted turbulence intensity value and the actual turbulence intensity value after the execution scenario weighting mechanism.

[0040] Constraints include applying a penalty function to negative values ​​in the series of correction coefficients;

[0041] The comprehensive optimization term includes combining the error term and the constraint term in a preset ratio to generate a composite loss function.

[0042] Based on the same concept, the present invention also provides a turbulence prediction and correction system, comprising:

[0043] The multi-dimensional data acquisition module is configured to acquire multi-dimensional flight status data, which includes turbulence intensity value in level flight, vertical overload value in non-level flight, vertical flight path angle, aircraft type identification, and horizontal stabilizer distance.

[0044] The multi-dimensional data processing module is configured to perform classification preprocessing on the multi-dimensional flight status data, including anomaly detection and missing value recovery of the turbulence intensity value and the vertical overload value, converting the aircraft type identifier into a low-dimensional embedding vector, and performing distribution standardization processing on the vertical flight path angle and the horizontal stabilizer distance.

[0045] The correction coefficient sequence generation module is configured to construct a time window based on the preprocessed multi-dimensional flight state data, input the spatiotemporal feature extraction model to perform multi-dimensional feature fusion and time correlation analysis, and output the correction coefficient sequence corresponding to the time window.

[0046] The turbulence intensity prediction value sequence generation module is configured to multiply the turbulence intensity value at each time point within the time window with the corresponding correction coefficient sequence point by point to generate a turbulence intensity prediction value sequence for non-level flight conditions.

[0047] The spatiotemporal feature extraction model optimization module is configured to train the spatiotemporal feature extraction model based on a composite loss function including a key scene enhancement mechanism and a non-negativity guarantee mechanism, so as to optimize the prediction accuracy of the correction coefficient sequence.

[0048] Based on the same concept, the present invention also provides an electronic device, including: a processor, a communication interface, a memory, and a communication bus, wherein the processor, the communication interface, and the memory communicate with each other through the communication bus; the memory stores a computer program, and when the computer program is executed by the processor, the processor performs the steps of a turbulence prediction and correction method.

[0049] Based on the same concept, the present invention also provides a computer-readable storage medium storing a computer program executable by an electronic device, which, when run on the electronic device, causes the electronic device to perform the steps of a turbulence prediction and correction method.

[0050] Compared with existing technologies, its advantages are as follows:

[0051] This invention discloses a turbulence prediction correction method, system, electronic device, and storage medium, which uses multi-dimensional features for expression, optimizing computational efficiency while ensuring physical rationality, and improving prediction accuracy and reliability. Attached Figure Description

[0052] Other features, objects, and advantages of this application will become more apparent from the following detailed description of non-limiting embodiments with reference to the accompanying drawings:

[0053] Figure 1 This is a flowchart illustrating some specific embodiments of the turbulence prediction and correction method of the present invention;

[0054] Figure 2 This is a schematic diagram of the structure of a turbulence prediction and correction system according to some specific embodiments of the present invention;

[0055] Figure 3 This is a schematic diagram of the structure of an electronic device according to some specific embodiments of the present invention;

[0056] In the diagram, 710 is the processor; 720 is the memory; 730 is the input device; and 740 is the output device. Detailed Implementation

[0057] To make the objectives, technical solutions, and advantages of this application clearer, the application will be further described in detail below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments in this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0058] The terminology used in the embodiments of this application is for the purpose of describing particular embodiments only and is not intended to limit the application. The singular forms “a,” “said,” and “the” used in the embodiments of this application and the appended claims are also intended to include the plural forms, and “multiple” generally includes at least two unless the context clearly indicates otherwise.

[0059] It should be understood that the term "and / or" used in this article 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, and B existing alone. Additionally, the character " / " in this article generally indicates that the preceding and following related objects have an "or" relationship.

[0060] It should be understood that although the terms first, second, third, etc., may be used in the embodiments of this application, these descriptions should not be limited to these terms. These terms are only used to distinguish the descriptions. For example, first may also be referred to as second without departing from the scope of the embodiments of this application, and similarly, second may also be referred to as first.

[0061] Depending on the context, the words “if” or “suppose” as used here can be interpreted as “when” or “in response to determination” or “in response to detection.” Similarly, depending on the context, the phrases “if determination” or “if detection (of the stated condition or event)” can be interpreted as “when determination” or “in response to determination” or “when detection (of the stated condition or event)” or “in response to detection (of the stated condition or event).”

[0062] It should also be noted that the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that an article or device that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such an article or device. Without further limitation, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the article or device that includes said element.

[0063] It should be noted that any symbols and / or numbers present in the specification that are not marked in the accompanying drawings are not reference numerals.

[0064] Reference Figure 1 A turbulence prediction correction method, comprising:

[0065] S101, acquire multi-dimensional flight status data, wherein the multi-dimensional flight status data includes turbulence intensity value in level flight, vertical overload value in non-level flight, vertical flight path angle, aircraft type identification, and horizontal stabilizer distance.

[0066] S102, perform classification preprocessing on the multi-dimensional flight status data, including anomaly detection and missing value recovery of the turbulence intensity value and the vertical overload value, convert the aircraft type identifier into a low-dimensional embedding vector, and perform distribution standardization processing on the vertical flight path angle and the horizontal stabilizer distance.

[0067] S103, construct a time window based on the preprocessed multi-dimensional flight status data, input the spatiotemporal feature extraction model to perform multi-dimensional feature fusion and time correlation analysis, and output the correction coefficient sequence corresponding to the time window;

[0068] S104, multiply the turbulence intensity value at each time point within the time window with the corresponding correction coefficient sequence point by point to generate a turbulence intensity prediction value sequence for non-level flight conditions;

[0069] S105, the spatiotemporal feature extraction model is trained based on a composite loss function including a key scenario enhancement mechanism and a non-negativity guarantee mechanism to optimize the prediction accuracy of the correction coefficient sequence.

[0070] Specifically, in this embodiment of the invention, multi-dimensional flight state data including level flight turbulence intensity value, non-level flight vertical overload value, vertical flight path angle, aircraft type identification, and horizontal stabilizer distance are acquired; outlier points in the turbulence intensity value and vertical overload value are removed based on a physical threshold range, and missing values ​​are filled in using temporal adjacent interpolation; the aircraft type identification is transformed into a fixed-dimensional dense embedded vector through vector space mapping to preserve aerodynamic similarity; the vertical flight path angle and horizontal stabilizer distance are standardized by performing a zero-mean, unit-variance linear transformation; based on covering the complete turbulence cycle... The time window length and preset overlap rate are used to construct a three-dimensional sample tensor. The samples are input into the spatiotemporal feature extraction model, which extracts global features containing long-term aircraft characteristics and flight path trends through compressed paths. After fusion by the bottleneck layer, the temporal resolution is reconstructed by the recovery path combined with jump connections, and finally the correction coefficient sequence is output. The level flight turbulence intensity value at each time point is multiplied by the corresponding correction coefficient to generate the non-level flight prediction value. The model is trained through a composite loss function, which includes dynamic weight terms determined by turbulence intensity and flight path angle to ensure the accuracy of key scenarios, while non-negative constraint terms ensure the physical rationality of the prediction value.

[0071] For example, taking the climb phase as an example: acquire 60 consecutive seconds of data including the level flight EDR sequence 0.3g, 0.4g, 0.5g... overload values ​​1.2g, 1.3g, 1.4g... flight path angles 4°, 5°, 6°... aircraft type identifier B737 horizontal stabilizer distance 12 meters; filter out outliers with EDR greater than 2.0g or overload greater than 3.0g and interpolate missing values ​​using the mean of the preceding and following times; map B737 into an 8-dimensional vector [-0. 2,0.5,1.1,-0.3,0.7,0.4,-1.2,0.9]; The original mean of the flight path angle is 5.3° with a standard deviation of 1.8. After standardization, the value of 6° in the 3rd second is converted to (6-5.3) / 1.8=0.39; A 60-second window sample tensor is constructed with a time step of 60×features×12×channels, where the features include an 8-dimensional aircraft vector + standardized flight path angle + standardized stabilizer distance + EDR + overload value; After inputting into the spatiotemporal model, the compression path performs three-level processing. The first level uses a 3×1 convolution kernel to sample at an interval of 1 unit, outputting 30×12×2 features. The second level has an interval of 2 units, outputting 15×12×4 features. The third level has an interval of 4 units, outputting 15×12×8 features. The recovery path upsamples the third-level output to 30×12×4 and concatenates it with the first-level activation features. After 3×1 convolution, the output is 30×12×4. It continues to upsample and concatenate the original features, finally outputting a 60×1 correction coefficient sequence, such as [1.1,1]. [.2,1.3...]; Multiply the 3rd second level flight EDR of 0.5g by the coefficient 1.3 to get the predicted value of 0.65g; In the loss function calculation, the weight of the 3rd second = 0.5 × 0.5g + 0.3 × |6°| + 0.2 = 0.25 + 1.8 + 0.2 = 2.25. The error term is 2.25 × (0.65g - true value 0.62g)² = 0.002. The non-negative constraint term applies a penalty to the negative coefficient. The total loss is the sum of the weighted error and the constraint term.

[0072] In some applications, the multi-dimensional flight status data undergoes classification preprocessing, including anomaly detection and missing value recovery for the turbulence intensity values ​​and vertical overload values; converting the aircraft type identifier into a low-dimensional embedding vector; and performing distribution standardization processing on the vertical flight path angle and the horizontal stabilizer distance, including filtering out turbulence intensity values ​​greater than a preset intensity range and vertical overload values ​​greater than a preset load range; recovering missing turbulence intensity values ​​and vertical overload values ​​based on a time series continuous interpolation method through data correlation between adjacent time points; mapping the aircraft type identifier to a fixed-dimensional continuous vector space to generate an embedding vector; and performing a zero-mean, unit-variance linear transformation on the vertical flight path angle and the horizontal stabilizer distance.

[0073] Understandably, reasonable ranges for turbulence intensity and vertical overload are set, and abnormal data points exceeding these ranges are filtered out. An interpolation method based on time series continuity is used to recover missing turbulence intensity and vertical overload values ​​by leveraging the numerical correlation between adjacent time points. Discrete aircraft type identifiers are transformed into fixed-dimensional continuous vector representations using vector space mapping technology, generating low-dimensional embedding vectors that preserve similarity between aircraft types. Standardized linear transformations are performed on the vertical flight path angle and horizontal stabilizer distance, achieving a zero-mean, unit-variance distribution transformation by subtracting the mean and dividing by the standard deviation.

[0074] For example, taking B737 aircraft data as an example: the turbulence intensity value of 2.5g at the 15th second was detected as abnormal and removed because it exceeded the preset threshold of 2.0g; the turbulence intensity value at the 20th second was missing, so the arithmetic mean of the adjacent values ​​of 1.2g at the 19th second and 1.3g at the 21st second, 1.25g, was used for interpolation; the aircraft type identifier B737 was mapped to an 8-dimensional embedding vector [0.2,-0.5,1.1,0.7,-0.3,0.9,0.4,-0.1]; the original data sequence of vertical flight path angles was 3.5°, 4.2°, 5.1°... The mean of the 60-second window was calculated to be 4.3° with a standard deviation of 0.8, so the value of 5.1° at the 3rd second was standardized to (5.1-4.3) / 0.8=1.0; the horizontal stabilizer distance was a fixed value of 12 meters, the mean of the entire dataset was 10.5 meters with a standard deviation of 2.5 meters, and the standardization yielded (12-10.5) / 2.5=0.6.

[0075] In some applications, a time window is constructed based on the preprocessed multi-dimensional flight status data, including setting the length of a continuous time window that covers the entire life cycle of a turbulence event; using a sliding window mechanism with a preset overlap ratio to segment the time series and maintain the temporal continuity of adjacent windows; splicing the preprocessed multi-dimensional flight status data along the feature dimensions to generate a combined feature vector; and constructing a time window for a three-dimensional tensor sample, with dimensions including the time step dimension, the combined feature dimension, and the single-channel dimension.

[0076] Understandably, a continuous time window length covering the entire lifecycle of a turbulence event is set to ensure the capture of the entire process of turbulence initiation, development, and dissipation; a sliding window mechanism with a preset overlap ratio is used to segment the time series data, and the temporal continuity of adjacent windows is maintained by controlling the window movement step size to avoid information fragmentation; the standardized vertical flight path angle and standardized horizontal stabilizer distance are embedded in low-dimensional vectors along the feature dimensions to form multi-dimensional feature vectors by splicing and combining them; a three-dimensional tensor sample structure containing the time step dimension, combined feature dimension, and single-channel dimension is constructed to preserve the time series characteristics and feature correlation of the original data.

[0077] For example, taking continuous flight data as an example: a 60-second time window is set to cover the typical turbulence event cycle; a 50% overlap sliding window is used, i.e., sliding once every 30 seconds, with the first window containing data from the 1st to the 60th second and the second window containing data from the 31st to the 90th second; after splicing and preprocessing, the features include an 8-dimensional aircraft type embedding vector [0.2, -0.5, 1.1, 0.7, -0.3, 0.9, 0.4, -0.1], a standardized flight path angle of 1.0, a standardized stabilizer distance of 0.6, a level flight turbulence intensity of 1.25g, and a vertical overload of 1.3g, forming a 12-dimensional combined feature vector; a 3-dimensional tensor sample is constructed with a time step dimension of 60 corresponding to 60-second sampling points, a combined feature dimension of 12 corresponding to 12 feature parameters, and a single channel dimension of 1 representing a single data stream. The first window sample is represented as a 60×12×1 tensor structure.

[0078] In some applications, the spatiotemporal feature extraction model is input for multidimensional feature fusion and temporal correlation analysis, and the output is a sequence of corrected coefficients corresponding to the time window. This includes a feature compression path, which extracts long-term trend features through multi-level time dimension downsampling operations, with each level containing convolution operations at different time scales to gradually expand the temporal receptive field; a feature fusion layer, which aggregates the long-term trend features and eliminates information redundancy to form global features; a feature recovery path, which reconstructs the temporal resolution through time dimension upsampling operations and combines global features with local transient features through cross-level feature connection mechanisms to generate spatiotemporal features; and a coefficient mapping layer, which converts the spatiotemporal features into a complete sequence of corrected coefficients in the time dimension.

[0079] Understandably, the feature compression path extracts long-term trend features through multi-level time-dimensional downsampling operations. Each level uses convolution operations at different time scales to gradually expand the temporal receptive field to capture the evolution trend of aircraft aerodynamic characteristics and flight path angle. The feature fusion layer aggregates the long-term trend features output by the compression path and eliminates redundancy through an information condensation mechanism to form fused features that characterize the global flight state. The feature recovery path reconstructs the temporal resolution step by step through time-dimensional upsampling operations and combines the fused features with the local transient features of the compression path through a cross-level feature connection mechanism to generate spatiotemporal features containing global constraints and local details. The coefficient mapping layer converts the spatiotemporal features into a sequence of correction coefficients of the same length as the original time window through one-dimensional convolution operations, with each coefficient corresponding to a nonlinear correction factor at a specific time point.

[0080] For example, taking a 12-dimensional feature tensor with a 60-second time window as an example: the first stage of the feature compression path uses a 3-time-point scale convolutional kernel with a sampling interval of 1 unit, inputting a 60×12×1 tensor and outputting a 30×12×2 feature tensor; the second stage uses a 3-time-point scale convolutional kernel with a sampling interval of 2 units, inputting a 30×12×2 tensor and outputting a 15×12×4 feature tensor; the third stage uses a 3-time-point scale convolutional kernel with a sampling interval of 4 units, outputting a 15×12×8 feature tensor; the feature fusion layer performs a 4-interval convolution on the third stage output to generate a 15×12×8 global feature; The feature recovery path first upsamples the global features to a 30×12×4 tensor and concatenates them with the 30×12×2 features output from the first stage to form a 30×12×6 tensor. After a 2-interval convolution, it outputs 30×12×4 features. It then upsamples the features to 60×12×2 and concatenates them with the original input to form a 60×12×3 tensor. After a 1-interval convolution, it outputs 60×12×2 spatiotemporal features. The coefficient mapping layer converts the spatiotemporal features into a 60×1 corrected coefficient sequence through a 1×1 convolution. For example, time point 3 corresponds to a coefficient of 1.32, and time point 58 corresponds to a coefficient of 0.98.

[0081] In some applications, the feature compression path includes extracting long-term trend features through multi-level time-dimensional downsampling operations. Each level contains convolutional operations at different time scales, gradually expanding the temporal receptive field. This includes performing a sequence of temporal convolutional operations with increasing expansion rates and gradually expanding the temporal receptive field by adjusting the sampling interval of the convolutional kernels. Non-linear activation functions and batch normalization are introduced after each level of convolutional operation to enhance feature representation. Time-dimensional compression operations are inserted after a specified convolutional level to reduce temporal resolution while increasing feature depth. The output is a composite feature tensor containing the long-term aerodynamic effects of the aircraft type, the evolution trend of the vertical flight path angle, and the baseline level of turbulence intensity values.

[0082] Understandably, a sequence of temporal convolution operations with increasing expansion rates is implemented to achieve exponential expansion of the receptive field by adjusting the sampling interval of the convolution kernel; a nonlinear activation function is introduced after each convolution operation to perform feature transformation, and batch standardization is used to stabilize the data distribution and enhance the model's expressive power; a temporal compression operation is inserted after a specific convolution level to simultaneously reduce temporal resolution and increase feature depth; the final output is a multidimensional composite feature tensor that includes the long-term aerodynamic characteristics of the aircraft, the continuous evolution trend of the vertical flight path angle, and the baseline level of turbulence intensity.

[0083] For example, taking a 60×12×1 sample tensor as input: the first-level convolution uses a 3×1 kernel with a sampling interval of 1 unit, covering 3 time points, and outputs 30×12×2 dimensions; the second-level convolution uses a 3×1 kernel with a sampling interval of 2 units, expanding the receptive field to 7 time points, and outputs 15×12×4 features; the third-level convolution uses a 3×1 kernel with a sampling interval of 4 units, reaching a receptive field of 15 time points, and outputs 15×12×8 features; ReLU activation is performed after each convolution level. Function processing and batch standardization are performed. For example, the feature value at time point 5 in the second-level output is adjusted from 1.25 to 0.82 after BN. After the third level, a max pooling operation is inserted to fix the time resolution to 15 points. The final output composite feature tensor includes aircraft type influence components, such as the value of 0.73 in the third channel, which represents the aerodynamic characteristics of the B737. The flight path angle trend component, such as the value of -0.15 in the seventh channel, indicates the continuous climb state. The turbulence baseline component, such as the value of 1.28 in the 12th channel, reflects the strong turbulence background of the current flight segment.

[0084] In some applications, the turbulence intensity value at each time point within the time window is multiplied by the corresponding correction coefficient sequence point by point to generate a turbulence intensity prediction value sequence for non-level flight conditions. This includes obtaining the turbulence intensity value at each time point within the current time window, calling the trained spatiotemporal feature extraction model, and outputting the correction coefficient sequence for the corresponding time window; performing a dot product operation on the turbulence intensity value at the same time point and the corresponding correction coefficient to generate turbulence intensity prediction values; and outputting a real-time updated turbulence intensity prediction value sequence.

[0085] Understandably, the process involves obtaining a sequence of level flight turbulence intensity values ​​for all sampling moments within the current time window; calling a pre-trained spatiotemporal feature extraction model to output a sequence of correction coefficients perfectly aligned with the time window; performing element-wise multiplication on the level flight turbulence intensity value and correction coefficient for each identical sampling moment to generate an instantaneous prediction value; and outputting a continuous time domain non-level flight turbulence intensity prediction sequence through a sliding window update mechanism.

[0086] For example, taking the time window of 31-90 seconds as an example: obtain the sequence of level flight turbulence intensity values ​​[0.3g, 0.4g, 0.5g...] (60 values ​​in total); call the trained model to output the correction coefficient sequence [1.1, 1.2, 1.3...] (60 dimensions); perform calculations at the 3rd second position (33rd second in the window): 0.5g × 1.3 = 0.65g; calculate at the 58th second position (88th second in the window): 0.8g × 0.98 = 0.784g; output the complete prediction sequence [0.33g, 0.48g, 0.65g...]; when the window slides to 61-120 seconds, update the output of the new prediction sequence in real time.

[0087] In some applications, the spatiotemporal feature extraction model is trained based on a composite loss function, including a key scenario enhancement mechanism and a non-negativity guarantee mechanism, to optimize the prediction accuracy of the correction coefficient sequence. This includes a scenario weighting mechanism, which dynamically calculates the error weight coefficients at each time point based on the turbulence intensity value and the absolute value of the vertical flight path angle, thereby enhancing the error contribution of key scenarios; an error term, which includes the mean squared error between the predicted turbulence intensity value and the actual turbulence intensity value after implementing the scenario weighting mechanism; a constraint term, which includes applying a penalty function to negative values ​​in the correction coefficient series; and a comprehensive optimization term, which includes combining the error term and the constraint term in a preset ratio to generate a composite loss function.

[0088] Understandably, the scenario weighting mechanism dynamically calculates the error weighting coefficients at each time point through the turbulence intensity value and the absolute value of the vertical flight path angle, thereby strengthening the error contribution weight of key flight scenarios; the error term calculates the mean square error between the predicted turbulence intensity value and the actual value after scenario weighting; the constraint term applies a penalty function to the negative values ​​in the correction coefficient sequence to ensure physical rationality; and the comprehensive optimization term linearly combines the error term and the constraint term according to a preset ratio to generate a differentiable composite loss function.

[0089] For example, taking the 3-second time point as an example: the level flight turbulence intensity value is 0.5g, the absolute value of the vertical flight path angle is 6°, and the dynamic weighting coefficient is calculated as 0.5×0.5g+0.3×6°+0.2×1=0.25+1.8+0.2=2.25; the predicted value at this point is 0.65g, the actual value is 0.62g, the squared error is (0.65-0.62)²=0.0009, and the weighted error is 2.25×0.0009=0.002025; if the correction coefficient at this point is -0.1, then the constraint term is 50×max(0,-(-0.1))=5.0; assuming the total error term is 0.15 and the total constraint term is 8.0, the composite loss is 0.15 + 8.0=8.15. The loss for the entire time window is the average of all points.

[0090] Another embodiment of the present invention is described below:

[0091] Input features: Level flight EDR ( ): Turbulence signal from the event data logger during level flight (sampling frequency 1Hz, unit) Non-level flight auxiliary data: overload data ( ): Vertical overload sensor data (reflecting instantaneous force, sampling frequency 1Hz); Vertical flight path angle ( Vertical velocity With ground speed The ratio ( Dimensionless, sampling frequency 1Hz); Model: such as B737, A320, C919, etc. (categorical variable); Distance from the horizontal stabilizer to the center of lift ( ): Fixed structural parameters of the aircraft model (unit: meters). Tag data: Actual non-level flight EDR ( : Turbulence signal recorded by EDR in non-level flight conditions (climb / descent) (sampling frequency 1Hz, aligned with level flight EDR time).

[0092] Outlier and Missing Value Handling: Outliers: Remove values ​​exceeding the physically reasonable range from EDR and overload data (e.g., EDR > 2.0g, overload > 3.0g); Missing Values: Complete using linear interpolation (because the data is a time series, adjacent time points are highly correlated). Model Coding Optimization: Embedded Coding Scheme: Map the model to an integer index (e.g., B737→1, A320→2, C919→3…); Convert the index into an 8-dimensional dense vector through the embedding layer (…). Vector parameters are optimized through model training to bring similar models (such as those from the same manufacturer) closer in vector distance; feature standardization: for numerical features ( Standardize:

[0093]

[0094] in The mean, The standard deviation is calculated based on the training set to avoid data leakage.

[0095] Time window partitioning (sample construction): window size (60 sampling points), sliding step (30 sampling points), sample tensor construction: each sample is a three-dimensional tensor.

[0096] ;

[0097] in (Time step) (Feature dimensions: 8-dimensional model embedding vector + 4 numerical features) The label represents the actual non-level flight EDR for the corresponding time window. .

[0098] The U-shaped dilated convolution module takes samples within a time window as input and outputs correction coefficients for each time step. Ultimately through " "Predicting non-level flight EDR."

[0099] Shrinking the path (downsampling + dilated convolution) compresses the time dimension, expands the receptive field, and captures global features (such as the long-term impact of device type). The trend changes provide a basis for distinguishing between "attitude-added components" and "real turbulence components".

[0100] Global characteristics specifically include: the continuous influence of the aircraft's aerodynamic characteristics (e.g., the attitude-added components of larger aircraft are gentler than those of smaller aircraft); and the trend of the vertical trajectory angle (e.g., during continuous climb). For a positive and stable condition, the correction factor needs to be shifted overall; the baseline level of the level flight EDR (if the overall turbulence is strong in a certain flight segment, the correction factor needs to be adapted to the baseline).

[0101] Structural and dimensional design (with) (For example, a layer)

[0102] Level 1:

[0103] enter: ;

[0104] It is the input tensor, and its dimension is Where 60 represents the time step, i.e., data from 60 consecutive time points; 12 represents the number of features, including different features such as EDR in level flight, overload, and vertical flight path angle; and 1 represents the number of channels, which is a single channel at the initial input. Dilated convolution: kernel size 3×1 (3 points in the time direction, 1 point in the feature direction), dilation rate... Input channel 1 → Output channel 2; kernel size 3×1 considers 3 adjacent time points in the time direction, and only 1 feature point in the feature direction. Expansion rate. This indicates that the spacing between kernel elements is 1 during convolution. The input channel count is 1, and the output channel count becomes 2 after convolution. Activation and batch normalization:

[0105]

[0106] First, batch normalization (BN) is performed on the result of the dilated convolution. Then, the ReLU activation function is applied. Max pooling (stride 2): Output (Time step halved, number of channels doubled).

[0107] Level 2:

[0108] Dilated convolution: kernel size 3×1, dilation rate Input channel 2 → Output channel 4; the number of input channels is 2, and the number of output channels becomes 4, further increasing the diversity of features. Output after activation, batch normalization, and pooling: .

[0109] Level 3:

[0110] Dilated convolution: kernel size 3×1, dilation rate Input channel 4 → Output channel 8; the number of input channels is 4, and the number of output channels becomes 8, further increasing the expressive power of the features. Output after activation and batch normalization (without pooling, used as the bottleneck layer input): .

[0111] The bottleneck layer integrates global features from the contraction path to further compress redundant information. Calculation:

[0112] ;

[0113] This is the output feature map of the third layer of the contraction path. It contains feature information extracted from the previous layers. : Here This represents the expansion rate, with a value of 4. When... At this time, the spacing between convolution kernel elements is 4. : Indicates a dilated convolution operation. Uses the dilation rate. Dilated convolution, for the input Convolution processing is performed to extract a wider range of features. : Indicates batch normalization operation. : Represents the activation function. It is the output feature map of the bottleneck layer. : Indicates the dimension of the output feature map. Where 15 represents the time step, 12 represents the number of features, and 8 represents the number of channels.

[0114] The expansion path (upsampling + skip connections) restores the global features of the bottleneck layer to the original time scale through upsampling, and compares them with the local features of the corresponding layer of the contraction path (such as instantaneous overload peaks, etc.). Mutations (spontaneous splicing) are used to achieve the fusion of "global constraints + local details".

[0115] Structural and dimensional design (corresponding to the reverse of the contraction path):

[0116] Level 2:

[0117] Transposed convolution (upsampling): kernel size 3×1, stride 2, input channels 8 → output channels 4. Upsampling to The kernel size of 3×1 indicates that three adjacent time points are considered in the time direction, and one feature point is considered in the feature direction. The step size of 2 is the upsampling step size. By setting the step size to 2, the input... (dimension is) Upsampling The time step changed from 15 to 30, and the number of channels changed from 8 to 4. Skip connections: activation features of layer 1 of the contraction path. Concatenation (channels 4+2=6); Dilated convolution: kernel size 3×1, dilation rate Input channel 6 → Output channel 4, Output A 3×1 kernel size is used to capture local correlations in the temporal direction. Expansion rate. This increases the spacing between elements in the convolution kernel by 2, thereby expanding the receptive field. The input channels are 6, and after convolution, the output channels become 4, resulting in the output... Its dimensions are .

[0118] Level 1:

[0119] Transposed convolution: stride 2, input channel 4 → output channel 2, upsampled to With a step size of 2, the input features (dimension 1) are... Upsampling The time step changed from 30 to 60, and the number of channels changed from 4 to 2. Skip connections: input features of layer 0 of the contraction path. Concatenation (channel count 2+1=3); shrinking the input features of layer 0 of the path Dimension is This is combined with the upsampled features (dimension 1). The concatenation is performed along the channel dimension to obtain a new feature with 2+1=3 channels. Dilated convolution: kernel size 3×1, dilation rate... Input channel 3 → Output channel 2, Output A 3×1 kernel size is used to capture local temporal correlations. Expansion rate. This is a standard convolution operation. The input has 3 channels; after convolution, the output has 2 channels, resulting in the output... Its dimensions are .

[0120] Level 0:

[0121] Dilated convolution: kernel size 3×1, dilation rate Input channel 2 → Output channel 1, Output A 3×1 kernel size is used to capture local temporal correlations. Expansion rate. This is a standard convolution operation. The input has two channels; after convolution, the output has only one channel, resulting in the final output. Its dimensions are The dimension is the same as that of the original input data.

[0122] The correction coefficient output layer maps the fused features to the correction coefficients at each time step. .calculate:

[0123] ;

[0124] : is the output feature map of layer 0 of the expansion path, which is the fused feature obtained after all the previous processing steps. : Represents a 1×1 convolution operation. : This is the output of the correction coefficient output layer, which is of dimension . The tensor is denoted by 60, which represents the time step, the same as the time step of the original input data; and 1 represents the number of channels, with each time step corresponding to a correction coefficient. : Indicates the dimension of the output tensor, i.e., 60 time steps, with each time step corresponding to a correction coefficient.

[0125] The predicted non-level flight EDR is:

[0126] ;

[0127] : This indicates that the predicted non-level flight EDR is related to time. The function. : This indicates the EDR in level flight, i.e., the original level flight turbulence data. : Represents the correction factor, which is the output of the correction factor output layer relative to time. The corresponding scalar value.

[0128] Model training and optimization:

[0129] The dataset is divided by flight segments (to avoid the same flight data appearing in both the training and test sets): 70% for training, 15% for validation, and 15% for test.

[0130] Loss function:

[0131] ;

[0132] : Represents the loss function, used to measure the difference between the model's predictions and the actual results. : Indicates the number of samples, that is, the total number of samples contained in the training set. : Indicates the number of time steps for each sample. Each sample has Each time step. : indicates the first The first sample The weights of each time step are used to differentiate error penalties for different samples and time steps. : indicates the model's prediction of the first The sample at the th The non-level flight EDR value at each time step. : indicates the first The sample at the th The true non-level flight EDR value at each time step. : Represents the weight coefficient of the non-negative constraint term, used to control the strength of the non-negative constraint. : Represents the correction factor, which is the output of the correction factor output layer relative to time. The corresponding scalar value. : Represents the non-negativity constraint function, when At that time, the value of this item is Otherwise, it is 0.

[0133] Weighted error term: weights ,in (Strong turbulence weight) (Weight of large flight path angle) (Basic weights) strengthen error penalties in key scenarios; : is the first The first sample Weights for each time step. This is the weight for strong turbulence, with a value of 0.5. When the EDR (Electronic Rumble Rate) is at level flight... When the value is large, it indicates that the scene is in a strong turbulence scenario. At this time, the weight will be increased accordingly, and the penalty for error will be heavier, so as to improve the prediction accuracy of the model in the strong turbulence scenario. This is the weight of the maximum flight path angle, with a value of 0.3. It represents the absolute value of the flight path angle. When the value is larger, it indicates a large flight path angle scenario, and the weight will increase to strengthen the penalty for errors in that scenario. : This is the base weight, with a value of 0.2. It provides a basic weight for each time step, ensuring that there is a certain error penalty even in non-critical scenarios. : is the first The first sample The level flight EDR value at each time step. : is the first The first sample The absolute value of the trajectory angle at each time step. Non-negative constraint terms: ,avoid (Ensure that the predicted non-level flight EDR is non-negative and conforms to the physical meaning of turbulence intensity). : This is the weighting coefficient for the non-negative constraint term, with a value of 50, used to control the strength of the non-negative constraint. : This is a correction factor; the purpose of the non-negative constraint term is to ensure .

[0134] Optimizers and Regularization, Optimizer: Adam (learning rate) Regularization: Dropout (probability 0.2, to prevent overfitting) + weight decay ( (Constrained parameter size). Training process: Iterate 50 times, calculating the loss of the training set and validation set in each round; Early stopping strategy: Stop training when the validation set loss increases for 5 consecutive rounds, and save the optimal model (the model with the smallest validation set loss).

[0135] Model validation and evaluation:

[0136] EDR prediction error:

[0137] Root Mean Square Error (RMSE): ;

[0138] : Indicates root mean square error. : Indicates the number of samples. : Indicates the number of time steps for each sample. Each sample has Each time step. : Represents the EDR value predicted by the model. : Represents the actual EDR value. : indicates a summation operation, which sums the differences between the predicted and actual values ​​for all samples at all time steps. : Represents the square of the difference between the predicted value and the actual value. This indicates that the average of the summation results is obtained, yielding the average squared error. : This indicates that the square root of the mean square error is taken to obtain the root mean square error. Its unit is the same as that of EDR, which makes it easier to intuitively understand the magnitude of the error.

[0139] Mean Absolute Error (MAE): ;

[0140] : Indicates the mean absolute error. : Sample size. Number of time steps for each sample. : The EDR value predicted by the model. : The actual EDR value. The summation operation sums the absolute differences between the predicted and actual values ​​for all samples at all time steps. : Represents the absolute difference between the predicted value and the true value. Absolute operation makes the sign of the error not affect the result, and only the magnitude of the error matters. The average of the summations yields the mean absolute error, which reflects the average deviation between the predicted and actual values. Coefficient of determination (COP) ):

[0141] .

[0142] : indicates the coefficient of determination. : Represents the sum of squared residuals, which is the sum of the squared differences between the model's predicted values ​​and the actual values, reflecting the magnitude of the error in the model's prediction. : represents the total sum of squares, where It is the average of the true values, and the total sum of squares reflects the dispersion of the true values. This indicates the proportion of total variance that the model can explain. If the model's predictions are perfectly accurate, then the sum of squared residuals is 0. If the model's predictions are the same as the average of the actual values, then If the model's predictions are worse than the average of the actual values, then It may be a negative number. Requirements: .

[0143] Key scenario performance: Strong turbulence sample: , requiring 10% higher than the overall level; : indicates that the level flight EDR is greater than The sample is defined as a strongly turbulent sample. It is gravitational acceleration. It is the EDR value in level flight.

[0144] Large diameter angle sample: , requiring It is 10% higher than the overall level.

[0145] : Samples with an absolute value of a flight path angle greater than 5° are defined as large flight path angle samples. It is the flight path angle.

[0146] This embodiment is based on level flight EDR and combines data such as overload, vertical flight path angle, aircraft type, and horizontal stabilizer distance in non-level flight states. First, the aircraft type features are optimized through embedding encoding, and samples are constructed through standardization and time window segmentation. Then, a U-shaped dilated convolution module is used to capture global features through shrinking paths, fuse information through bottleneck layers, and restore details through expanding paths combined with skip connections, outputting correction coefficients. Through a weighted loss function (strengthening the error penalty for strong turbulence and large flight path angle scenarios) and non-negative constraints, the model is trained, and finally, the non-level flight EDR is predicted by "level flight EDR × correction coefficient", realizing accurate correction of turbulence data under non-level flight attitude and improving the prediction accuracy of key scenarios.

[0147] For the purpose of simplicity, the method steps disclosed in the above embodiments are described as a series of actions. However, those skilled in the art should understand that the embodiments of the present invention are not limited to the described order of actions, because according to the embodiments of the present invention, some steps can be performed in other orders or simultaneously. Furthermore, those skilled in the art should also understand that the embodiments described in the specification are all preferred embodiments, and the actions involved are not necessarily essential to the embodiments of the present invention.

[0148] like Figure 2As shown, the present invention also provides a turbulence prediction and correction system, comprising:

[0149] The multi-dimensional data acquisition module 201 is configured to acquire multi-dimensional flight status data, wherein the multi-dimensional flight status data includes turbulence intensity value in level flight, vertical overload value in non-level flight, vertical flight path angle, aircraft type identification, and horizontal stabilizer distance.

[0150] The multi-dimensional data processing module 202 is configured to perform classification preprocessing on the multi-dimensional flight status data, including anomaly detection and missing value recovery of the turbulence intensity value and the vertical overload value, converting the aircraft type identifier into a low-dimensional embedding vector, and performing distribution standardization processing on the vertical flight path angle and the horizontal stabilizer distance.

[0151] The correction coefficient sequence generation module 203 is configured to construct a time window based on the preprocessed multi-dimensional flight state data, input the spatiotemporal feature extraction model to perform multi-dimensional feature fusion and time correlation analysis, and output the correction coefficient sequence corresponding to the time window.

[0152] The turbulence intensity prediction value sequence generation module 204 is configured to multiply the turbulence intensity value at each time point within the time window with the corresponding correction coefficient sequence point by point to generate a turbulence intensity prediction value sequence for non-level flight conditions.

[0153] The spatiotemporal feature extraction model optimization module 205 is configured to train the spatiotemporal feature extraction model based on a composite loss function including a key scene enhancement mechanism and a non-negativity guarantee mechanism, so as to optimize the prediction accuracy of the correction coefficient sequence.

[0154] It is worth noting that although only some basic functional modules are disclosed in the embodiments of this invention, it does not mean that the composition of this system is limited to the above-mentioned basic functional modules. On the contrary, what this embodiment intends to express is that, based on the above-mentioned basic functional modules, those skilled in the art can arbitrarily add one or more functional modules in combination with existing technology to form an infinite number of embodiments or technical solutions. That is to say, this system is open rather than closed. The fact that this embodiment only discloses a few basic functional modules should not be considered as the scope of protection of the claims of this invention being limited to the disclosed basic functional modules. At the same time, for the convenience of description, the above device is described separately according to its functions as various units and modules. Of course, in implementing this invention, the functions of each unit and module can be implemented in one or more software and / or hardware.

[0155] like Figure 3As shown, the present invention also provides an electronic device, including: a processor, a communication interface, a memory, and a communication bus, wherein the processor, the communication interface, and the memory communicate with each other through the communication bus; the memory stores a computer program, and when the computer program is executed by the processor, the processor performs the steps of a turbulence prediction and correction method.

[0156] Figure 3 This is a schematic diagram of the structure of an electronic device provided in an embodiment of the present invention. For example... Figure 3 The structure shown in this embodiment of the invention includes an electronic device comprising one or more processors 710 and a memory 720; the processors 710 in this electronic device may be one or more. Figure 3 Taking a processor 710 as an example; a memory 720 is used to store one or more programs; the one or more programs are executed by the one or more processors 710, causing the one or more processors 710 to implement a turbulence prediction correction method as described in any one of the embodiments of the present invention.

[0157] The electronic device may also include an input device 730 and an output device 740.

[0158] The processor 710, memory 720, input device 730, and output device 740 in this electronic device can be connected via a bus or other means. Figure 3 Taking the example of a connection between China and Israel via a bus.

[0159] The memory 720 in this electronic device serves as a computer-readable storage medium, capable of storing one or more programs. These programs can be software programs, computer-executable programs, or modules, such as the program instructions / modules corresponding to the turbulence prediction and correction method provided in this embodiment of the invention. The processor 710 executes various functional applications and data processing of the electronic device by running the software programs, instructions, and modules stored in the memory 720, thereby implementing the turbulence prediction and correction method described in the above embodiment.

[0160] The memory 720 may include a program storage area and a data storage area. The program storage area may store the operating system and applications required for at least one function; the data storage area may store data created based on the use of the electronic device. Furthermore, the memory 720 may include high-speed random access memory and may also include non-volatile memory, such as at least one disk storage device, flash memory device, or other non-volatile solid-state storage device. In some instances, the memory 720 may further include memory remotely located relative to the processor 710, which can be connected to the device via a network. Examples of such networks include, but are not limited to, the Internet, intranets, local area networks, mobile communication networks, and combinations thereof.

[0161] Input device 730 can be used to receive input digital or character information, and to generate key signal inputs related to user settings and function control of the electronic device. Output device 740 may include display devices such as a display screen.

[0162] The present invention also provides a computer-readable storage medium storing a computer program executable by an electronic device, which, when run on the electronic device, causes the electronic device to perform the steps of a turbulence prediction and correction method.

[0163] Specifically, the computer storage medium in this embodiment of the invention can be any combination of one or more computer-readable media. The computer-readable medium can be a computer-readable signal medium or a computer-readable storage medium. For example, a computer-readable storage medium can be—but is not limited to—an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any combination thereof. More specific examples of computer-readable storage media (a non-exhaustive list) include: an electrical connection having one or more wires, a portable computer disk, a hard disk, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fiber, portable compact disk read-only memory (CD-ROM), optical storage device, magnetic storage device, or any suitable combination thereof. In this embodiment, the computer-readable storage medium can be any tangible medium containing or storing a program that can be used by or in conjunction with an instruction execution system, apparatus, or device.

[0164] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A turbulence prediction and correction method, characterized in that, include: Acquire multi-dimensional flight status data, which includes turbulence intensity value in level flight, vertical overload value in non-level flight, vertical flight path angle, aircraft type identification, and horizontal stabilizer distance. The multi-dimensional flight status data is classified and preprocessed, including anomaly detection and missing value recovery for the turbulence intensity value and the vertical overload value, converting the aircraft type identifier into a low-dimensional embedding vector, and performing distribution standardization processing on the vertical flight path angle and the horizontal stabilizer distance. A time window is constructed based on the preprocessed multi-dimensional flight status data. The spatiotemporal feature extraction model is input to perform multi-dimensional feature fusion and time correlation analysis, and the correction coefficient sequence corresponding to the time window is output. Multiply the turbulence intensity value at each time point within the time window with the corresponding correction coefficient sequence point by point to generate a turbulence intensity prediction value sequence for non-level flight conditions. The spatiotemporal feature extraction model is trained based on a composite loss function that includes a key scenario enhancement mechanism and a non-negativity guarantee mechanism in order to optimize the prediction accuracy of the correction coefficient sequence. The feature compression path includes extracting long-term trend features through multi-level time-dimensional downsampling operations, with each level containing convolution operations at different time scales to gradually expand the temporal receptive field. The feature fusion layer includes aggregating the long-term trend features and eliminating information redundancy to form global features; The feature recovery path includes reconstructing the temporal resolution through temporal upsampling operations and fusing global features and local transient features through a cross-level feature connection mechanism to generate spatiotemporal features; The coefficient mapping layer includes converting spatiotemporal features into a complete time-dimensional sequence of corrected coefficients; Among them, the scenario weighting mechanism includes dynamically calculating the error weighting coefficients at each time point based on the turbulence intensity value and the absolute value of the vertical flight path angle, thereby strengthening the error contribution of key scenarios; The error term includes the mean squared error between the predicted turbulence intensity value and the actual turbulence intensity value after the execution scenario weighting mechanism. Constraints include applying a penalty function to negative values ​​in the series of correction coefficients; The comprehensive optimization term includes combining the error term and the constraint term in a preset ratio to generate a composite loss function.

2. The turbulence prediction and correction method according to claim 1, characterized in that, The multi-dimensional flight status data undergoes classification preprocessing, including anomaly detection and missing value recovery for the turbulence intensity value and the vertical overload value, converting the aircraft type identifier into a low-dimensional embedding vector, and performing distribution standardization processing on the vertical flight path angle and the horizontal stabilizer distance, further including: Filter out bump intensity values ​​that exceed the preset intensity range and vertical overload values ​​that exceed the preset load range; Based on the time series continuous interpolation method, the missing turbulence intensity value and vertical overload value are recovered by the correlation of data at adjacent time points; The model identifier is mapped to a continuous vector space of fixed dimensions to generate an embedded vector; Perform a zero-mean, unit-variance linear transformation on the vertical flight path angle and the horizontal stabilizer distance.

3. The turbulence prediction and correction method according to claim 1, characterized in that, Constructing a time window based on the preprocessed multi-dimensional flight status data further includes: Set the length of the continuous time window that covers the entire lifecycle of the turbulence event; A sliding window mechanism with a preset overlap ratio is used to segment the time series, maintaining the temporal continuity of adjacent windows; The preprocessed multi-dimensional flight status data is concatenated along the feature dimensions to generate a combined feature vector. The time window for constructing the three-dimensional tensor sample includes dimensions such as time step dimension, combined feature dimension, and single channel dimension.

4. The turbulence prediction and correction method according to claim 1, characterized in that, The feature compression path includes extracting long-term trend features through multi-level time-dimensional downsampling operations. Each level contains convolution operations at different time scales, gradually expanding the temporal receptive field. Further components include: The temporal convolution operation sequence with increasing dilation rate is executed to gradually expand the temporal receptive field by adjusting the sampling interval of the convolution kernel; A non-linear activation function and batch normalization are introduced after each convolution operation to enhance feature representation capabilities. Insert a temporal compression operation after the specified convolutional layer to reduce temporal resolution while increasing feature depth; The output includes a composite feature tensor that includes the long-term aerodynamic effects of the aircraft type, the evolution trend of the vertical flight path angle, and the baseline level of the turbulence intensity value.

5. The turbulence prediction and correction method according to claim 1, characterized in that, The turbulence intensity value at each time point within the time window is multiplied by the corresponding correction coefficient sequence point by point to generate a turbulence intensity prediction sequence for non-level flight conditions, which further includes: Get the turbulence intensity values ​​at each time point within the current time window; Call the trained spatiotemporal feature extraction model and output the correction coefficient sequence for the corresponding time window; The turbulence intensity value at the same time point is multiplied by the corresponding correction coefficient to generate the turbulence intensity prediction value. Output a real-time updated sequence of predicted turbulence intensity values.

6. A turbulence prediction and correction system, characterized in that, include: The multi-dimensional data acquisition module is configured to acquire multi-dimensional flight status data, which includes turbulence intensity value in level flight, vertical overload value in non-level flight, vertical flight path angle, aircraft type identification, and horizontal stabilizer distance. The multi-dimensional data processing module is configured to perform classification preprocessing on the multi-dimensional flight status data, including anomaly detection and missing value recovery of the turbulence intensity value and the vertical overload value, converting the aircraft type identifier into a low-dimensional embedding vector, and performing distribution standardization processing on the vertical flight path angle and the horizontal stabilizer distance. The correction coefficient sequence generation module is configured to construct a time window based on the preprocessed multi-dimensional flight state data, input the spatiotemporal feature extraction model to perform multi-dimensional feature fusion and time correlation analysis, and output the correction coefficient sequence corresponding to the time window. The turbulence intensity prediction value sequence generation module is configured to multiply the turbulence intensity value at each time point within the time window with the corresponding correction coefficient sequence point by point to generate a turbulence intensity prediction value sequence for non-level flight conditions. The spatiotemporal feature extraction model optimization module is configured to train the spatiotemporal feature extraction model based on a composite loss function including a key scene enhancement mechanism and a non-negativity guarantee mechanism, so as to optimize the prediction accuracy of the correction coefficient sequence. The feature compression path includes extracting long-term trend features through multi-level time-dimensional downsampling operations, with each level containing convolution operations at different time scales to gradually expand the temporal receptive field. The feature fusion layer includes aggregating the long-term trend features and eliminating information redundancy to form global features; The feature recovery path includes reconstructing the temporal resolution through temporal upsampling operations and fusing global features and local transient features through a cross-level feature connection mechanism to generate spatiotemporal features; The coefficient mapping layer includes converting spatiotemporal features into a complete time-dimensional sequence of corrected coefficients; Among them, the scenario weighting mechanism includes dynamically calculating the error weighting coefficients at each time point based on the turbulence intensity value and the absolute value of the vertical flight path angle, thereby strengthening the error contribution of key scenarios; The error term includes the mean squared error between the predicted turbulence intensity value and the actual turbulence intensity value after the execution scenario weighting mechanism. Constraints include applying a penalty function to negative values ​​in the series of correction coefficients; The comprehensive optimization term includes combining the error term and the constraint term in a preset ratio to generate a composite loss function.

7. An electronic device, characterized in that, include: The system includes a processor, a communication interface, a memory, and a communication bus, wherein the processor, the communication interface, and the memory communicate with each other via the communication bus; the memory stores a computer program, which, when executed by the processor, causes the processor to perform the steps of the method according to any one of claims 1 to 5.

8. A computer-readable storage medium, characterized in that, It stores a computer program executable by an electronic device, which, when run on the electronic device, causes the electronic device to perform the steps of the method according to any one of claims 1 to 5.

Images