A flight route static temperature and perturbation wind prediction precision improvement method and system based on flight data and reanalysis data assimilation

Abstract

This invention discloses a method and system for improving the prediction accuracy of static temperature and disturbance wind along a flight route based on the assimilation of flight data and reanalysis data. The method includes: filtering and imputing missing values ​​in the target route's QAR flight data to obtain static temperature and disturbance wind meteorological elements; adapting the reanalysis data to the scale of the meteorological elements derived from the QAR data using three-dimensional linear interpolation; constructing a variational assimilation system by combining a dynamic background field error matrix, an observation error matrix, and atmospheric physical constraints; fusing the QAR and reanalysis data through variational assimilation to generate high-resolution optimized data; comparing the errors between the assimilated data and the original reanalysis data, iteratively adjusting parameters, and outputting a high-resolution route dataset; optimizing the prediction model based on a Long Short-Term Memory (LSTM) network, comparing the prediction accuracy of the assimilated and unassimilated data, achieving improved prediction accuracy of meteorological elements along the route for the next hour, and outputting a route atmospheric static temperature and disturbance wind predictor.

Description

Technical Field

[0001] This invention belongs to the field of meteorological data processing and aviation safety technology, specifically relating to a method and system for improving the accuracy of route static temperature and disturbance wind prediction based on the assimilation of flight data and reanalysis data. Background Technology

[0002] Meteorological reanalysis data, as the core data for weather forecasting and route planning, relies on global observation data such as satellites and ground stations. It is used to generate historical or real-time datasets (such as ERA5, NCEP / NCAR, CRA) through fixed models and assimilation techniques. Although it has the advantages of wide coverage and continuous time series, and can provide basic meteorological elements such as static temperature and wind field required for routes, it is limited by the global uniform grid design. Its resolution is far lower than the needs of small and medium-scale route analysis. It cannot distinguish local turbulence, takeoff and landing wind shear and other small and medium-scale dangerous weather phenomena. Moreover, the error is large when meteorological elements are interpolated to the route range, which makes it difficult to meet the requirements of high-precision data for civil aviation safety scenarios.

[0003] Flight data collected by airborne sensors during aircraft flight, especially QAR (Quick Access Recorder) data, has unique advantages such as second-level sampling rate and parameter density. It can directly invert the meteorological elements of the route, such as static temperature and disturbance wind, through meteorological parameters such as total atmospheric temperature, static pressure, Mach number, ground speed, and aircraft attitude (roll angle, pitch angle, yaw angle). It can become a key data source to fill the gaps in the route area coverage in reanalysis data and provide important support for improving the accuracy of route meteorological data. Summary of the Invention

[0004] The purpose of this invention is to address the problem of insufficient accuracy in predicting static temperature and disturbance wind data for the next hour using traditional route forecasting methods that rely on original reanalysis data. This invention proposes a method and system for improving the accuracy of static temperature and disturbance wind forecasts based on the assimilation of flight data and reanalysis data. This invention aims to improve the resolution of current route meteorological data and enhance the accuracy of route meteorological forecasts for the next hour. It also solves the technical problem that traditional reanalysis data has low grid resolution and low meteorological element prediction accuracy, making it difficult to support aviation safety requirements such as route turbulence warnings and wind shear monitoring for different aircraft types and routes.

[0005] To achieve the above objectives, the present invention provides the following solution: A method for improving the accuracy of route static temperature and disturbance wind prediction based on the assimilation of flight data and reanalysis data, the method comprising: Step 1: Filter and fill in missing values ​​in the QAR data of the target route, and invert the static temperature and disturbance wind to obtain the QAR observation dataset; Step 2: Select reanalysis data covering the target route as the background field. Based on the QAR observation dataset, adapt the reanalysis data to the scale of meteorological elements derived from the QAR data through three-dimensional linear interpolation. Combine the dynamic background field error matrix B, the observation error matrix R and atmospheric physical constraints to construct a variational assimilation system. Step 3: Divide the data into a 1-hour main window and sub-window, solve the variational assimilation system using the conjugate gradient method, and generate preliminary optimized meteorological reanalysis data after 0.1°×0.1° gridding and noise removal; Step 4: Compare the errors between the preliminary optimized meteorological reanalysis data and the original reanalysis data, iteratively adjust the variational assimilation system parameters according to priority, and output a high-resolution route dataset, i.e., assimilation data. Step 5: Based on the LSTM prediction model optimization, compare the prediction accuracy of assimilated and unassimilated data to improve the prediction accuracy of airway meteorological elements for the next hour, and finally output the airway atmospheric calm temperature and disturbance wind predictor.

[0006] Preferred methods for filtering and imputing missing values ​​in the QAR data of the target route, and retrieving static temperature and disturbance wind to obtain the QAR observation dataset include: QAR data that fall within the latitude and longitude boundaries of the target route, are within the corresponding altitude range, and are within the set time range are selected from the airline database to form the target route QAR raw dataset; Extracting the parameter total atmospheric temperature from the original QAR dataset ,Mach number , ground speed Eastward component of ground velocity Northward component Vertical component Aircraft roll angle Yaw angle Pitch angle Angle of attack Outlier removal is performed, and then missing data is filled in; Based on the parameters in the processed QAR original dataset, meteorological elements are inverted to obtain the QAR observation dataset; Among them, inversion static temperature include: ; in, The specific heat ratio of air; Inverting the disturbed wind includes: calculating the speed of sound. Calculate vacuum velocity Inverted three-axis winds: Eastward winds North wind Vertical wind ; ; ; ; in, is the gas constant for dry air.

[0007] Preferably, based on QAR observation datasets, a method for constructing a variational assimilation system by adapting reanalysis data to the scale of meteorological elements derived from QAR data through three-dimensional linear interpolation, and combining the dynamic background field error matrix B, observation error matrix R, and atmospheric physical constraints, includes: Select reanalysis data covering the target route area and extract static temperature from the reanalysis data. East wind North wind Vertical wind Data serves as the initial background field for the assimilation system; A three-dimensional linear interpolation method was used to interpolate the background field resolution to the spatiotemporal scale of the QAR observation dataset; Based on the spatiotemporal scale of the QAR observation dataset, a variational assimilation objective function, i.e., a variational assimilation system, is constructed. Among them, methods for interpolating the background field resolution to the spatiotemporal scale of the QAR observation dataset using three-dimensional linear interpolation include: Latitude and longitude interpolation based on the longitude of the QAR observation point along the target route. ,latitude Centered on the background field, select four adjacent grid points. Bilinear interpolation was used for calculation. , Background field parameter values ​​at that location; Height interpolation based on the height of the QAR observation point Centered on the background field, select two adjacent height layers. Linear interpolation is used for calculation. Background field parameter values ​​at that location; Time interpolation uses the timestamps of QAR observation points. Centered on the background field, select two adjacent time points. Linear interpolation is used for calculation. Background field parameter values ​​at that location; The variational assimilation objective function is: ; in, This is the error term; For physical constraints; in, ;in, For the optimization analysis field to be determined, As background scene, The QAR observational meteorological data retrieved in step one, All are in matrix form: , This represents the index of the lattice point in the field that participated in assimilation. The background field error covariance matrix; Here are the covariance matrices of the QAR observation error and the background field error. form: QAR observation error covariance matrix form: ;in, This represents the standard deviation of the error for each parameter; obs For observational data; in, ; ;in, This represents the specific value of the i-th item in the physical constraint term matrix. For the first The easterly wind in the comparison item, For the earth to turn the wind, ; Coriolis parameters, ; This is the Earth's rotational angular velocity; The latitude of the target route is represented in rad. For height, The vertical pressure gradient is determined by the background pressure field. calculate; Given atmospheric density, from the ideal gas law calculate; It is the acceleration due to gravity; To constrain the weights, the actual wind variance is statistically adjusted based on the percentage of the squared mean of the atmospheric geostrophic balance deviation in mid- and low-latitude regions.

[0008] Preferred methods for generating preliminary optimized meteorological reanalysis data include: dividing data into a 1-hour main window and sub-windows, solving the variational assimilation system using the conjugate gradient method, and then performing 0.1°×0.1° gridding and noise removal. Let the analysis field be set initial value Set a convergence threshold The initial value is 10 -3 ; Calculate the gradient initial gradient Search direction initial value Set as Update step size Set to 0.1; gradient calculation is as follows: ; By step size Updated analysis field: ,in, This represents the solution required in the variational assimilation system. The k One iteration term, This represents the solution required in the variational assimilation system. The k Each iteration term synchronously calculates the new gradient. ;like Stop calculation and output the current analysis field. ; The arithmetic mean of the sub-window analysis fields is taken to obtain the preliminary optimized analysis field for the 1-hour assimilation window; The target route grid is divided into 0.1° × 0.1°. The nearest neighbor interpolation method is used to preliminarily optimize the parameter values ​​of the analysis field. The value is assigned to the corresponding grid, meaning that the value of each grid is equal to the optimized value of the nearest QAR observation point; A 3-point moving average was used to take the mean of itself and the two adjacent grids for each grid, eliminating isolated noise and generating preliminary optimized meteorological reanalysis data.

[0009] Preferred methods for comparing the errors between the preliminarily optimized meteorological reanalysis data and the original reanalysis data, iteratively adjusting the variational assimilation system parameters according to priority, and outputting a high-resolution route dataset include: Set the assimilated data as ,in It uses a 0.1° × 0.1° high-resolution grid index. For height indexing, For time indexing; The original reanalysis data are set as follows ,in A low-resolution grid index of 0.25° × 0.25° is used. For height indexing, For time indexing; Includes elements and Consistent, all are ,Will Matched to four dimensions of time and space The grid points are obtained. ; based on , and Set the error based on grid points and time error and corresponding indicators ; in, ; ;in, , The maximum value of the latitude index. The maximum value of the longitude index. The maximum value of the height index. The maximum value of the time index, the error threshold. Set them respectively to, , , , ; If any grid point or time-related error index fails to meet the standard, parameters are adjusted according to priority. Different failure scenarios are then investigated, and finally, the convergence threshold is applied. Error matrix Parameter optimization is performed in the order of assimilation sub-window step size. During optimization, both the convergence threshold and the sub-window step size are reduced, and the error matrix is ​​adjusted according to the location of substandard grid points. If the error does not meet the standard in the route area, it is reduced. Increase If the standard is not met in non-airway areas, the risk will increase. Shrink After completing the parameter iterative optimization, a high-resolution, high-precision dataset that meets the error index is output.

[0010] Preferred methods for comparing the prediction accuracy of assimilated and unassimilated data based on LSTM prediction model optimization include: The assimilated data used for prediction was selected from... Extract a certain length of route segment data. Unassimilated data are raw reanalysis data. The result obtained by three-dimensional linear interpolation is... Same dimension ,in For waypoint indexing, one point is taken every 0.1° along the way. For height indexing, For time indexing; The prediction model uses an LSTM time series model, based on... and Predicted static temperature along the flight path for the next hour East wind North wind and vertical wind The model formula is: ; in , These are the meteorological element values ​​for the corresponding sequence. It is the length of the time series input to the model. Model parameters, including the weight matrix and bias terms .

[0011] The present invention also provides a system for improving the accuracy of route static temperature and disturbance wind prediction based on flight data and reanalysis data assimilation. The system is used to implement the aforementioned method and includes: a data acquisition module, a construction module, an optimization module, a comparison module, and a prediction module. The acquisition module is used to filter and fill in missing values ​​in the QAR data of the target route, and to retrieve the static temperature and disturbance wind to obtain the QAR observation dataset. The construction module is used to select reanalysis data covering the target route as the background field, and based on the QAR observation dataset, adapt the reanalysis data to the scale of meteorological elements derived from the QAR data through three-dimensional linear interpolation. Combined with the dynamic background field error matrix B, the observation error matrix R and atmospheric physical constraints, a variational assimilation system is constructed. The optimization module is used to divide the data into a 1-hour main window and a sub-window, solve the variational assimilation system using the conjugate gradient method, and generate preliminary optimized meteorological reanalysis data after 0.1°×0.1° gridding and noise removal. The comparison module is used to compare the errors between the preliminary optimized meteorological reanalysis data and the original reanalysis data, iteratively adjust the variational assimilation system parameters according to priority, and output a high-resolution route dataset, i.e., assimilation data. The prediction module is used to optimize the LSTM prediction model, compare the prediction accuracy of assimilated and unassimilated data, improve the prediction accuracy of airway meteorological elements for the next hour, and finally output the airway atmospheric calm temperature and disturbance wind predictor.

[0012] Compared with the prior art, the beneficial effects of the present invention are as follows: (1) Relying on two types of data sources, QAR data and reanalysis data, meteorological elements are inverted through physical formulas and three-dimensional linear interpolation is adapted to the route scale. No additional observation equipment and data are required, and the system can be applied to the prediction of calm temperature and disturbance winds of any route without reconstruction.

[0013] (2) Using the method of forward assimilation fusion + backward accuracy verification, the optimized data is first generated through variational assimilation, and then the assimilated data is compared with the original reanalysis data to calculate the error. The parameters are then adjusted iteratively based on priority. In the prediction stage, the model output accuracy of assimilated and unassimilated data is compared. The dual verification and optimization ensure the accuracy of the data and the reliability of the prediction. Attached Figure Description

[0014] To more clearly illustrate the technical solution of the present invention, the drawings used in the embodiments are briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0015] Figure 1 This is a flowchart of a method for improving the prediction accuracy of route static temperature and disturbance wind based on the assimilation of flight data and reanalysis data in an embodiment of the present invention. Figure 2 This is a three-dimensional interpolation structure diagram of the background field of the reanalysis data in an embodiment of the present invention; Figure 3 This is a schematic diagram illustrating the assimilation window setting and data matching in an embodiment of the present invention; Figure 4 This is a schematic diagram illustrating the parameter optimization priority for assimilation data error accuracy verification in an embodiment of the present invention; Figure 5 This is a flowchart illustrating the iterative optimization process of LSTM prediction model parameters in an embodiment of the present invention. Detailed Implementation

[0016] 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.

[0017] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0018] Example 1 This invention provides a method for improving the accuracy of route-based static temperature and disturbance wind predictions based on the assimilation of flight data and reanalysis data. This method addresses the issues of coarse grids and low accuracy in route scenarios by mining route meteorological information from QAR data and combining it with variational assimilation techniques. Simultaneously, it further improves the accuracy of the assimilated data's future one-hour route meteorological predictions through LSTM time series prediction model optimization. This provides directly applicable high-quality meteorological data support for turbulence warnings and risk avoidance during takeoff and landing in the civil aviation field, and is suitable for meteorological data processing and prediction applications for civil aviation routes. The specific process is as follows... Figure 1 As shown, it includes five main steps: Step 1: Using the target route QAR data, after directional filtering and missing value completion, invert the static temperature and disturbance wind to obtain the QAR observation dataset.

[0019] Step 2: Using reanalysis data covering the target route as the background field, the data is adapted to the scale of meteorological elements derived from QAR data through three-dimensional linear interpolation. A variational assimilation system is then constructed by combining the dynamic background field error matrix B, the QAR observation error matrix R, and atmospheric physical constraints.

[0020] Step 3: Divide the data into a 1-hour main window and sub-window, solve the objective function using the conjugate gradient method, and generate preliminary optimized meteorological reanalysis data by 0.1°×0.1° gridding and noise removal.

[0021] Step 4: Compare the assimilated data with the original reanalysis data to calculate the error, iteratively adjust the assimilation model parameters according to priority, and output a high-resolution route dataset.

[0022] Step 5: Based on the LSTM prediction model optimization, compare the prediction accuracy of assimilated data and unassimilated data, and achieve a prediction accuracy of ≥20% for assimilated data of airway meteorological elements in the next hour compared with unassimilated data, and finally generate an airway atmospheric calm temperature and disturbance wind prediction system.

[0023] Specifically: Step 1: Using the target route QAR data, after directional filtering and missing value completion, invert the static temperature and disturbance wind to obtain the QAR observation dataset.

[0024] Sub-step 1-A: QAR data acquisition and preprocessing.

[0025] Define the latitude and longitude boundaries, key altitude layers, and time range of the target route. Select QAR data from the airline database that fall within the latitude and longitude boundaries of the target route, are at the corresponding altitude range, and are within the set time range to form the original QAR dataset of the target route.

[0026] Extracting the parameter total atmospheric temperature from the original QAR dataset ,Mach number Aircraft roll angle Yaw angle Pitch angle Angle of attack , ground speed Eastward component of ground velocity Northward component Vertical component Outlier removal is performed, and missing data is then added to ensure the continuity of data in the spatiotemporal dimensions.

[0027] Sub-step 1-B: Meteorological elements (static temperature, disturbing wind) are inverted to obtain the QAR observation dataset.

[0028] Based on the parameters in step 1-A, the static temperature is inverted using the following formula. East wind North wind Vertical wind : (1); in, The specific heat ratio of air can be adjusted according to the target flight path altitude: 1.40 for the troposphere and 1.39 for the stratosphere. The inversion of disturbance wind involves three steps: calculating the speed of sound... Calculate vacuum velocity Inverting the three-axis wind .

[0029] (2); (3); (4); in, The value of is the gas constant for dry air, which is 287 J / (kg·K).

[0030] Step 2: Using reanalysis data covering the target route as the background field, the data is adapted to the scale of meteorological elements derived from QAR data through three-dimensional linear interpolation. A variational assimilation system is then constructed by combining the dynamic background field error matrix B, the QAR observation error matrix R, and atmospheric physical constraints.

[0031] Sub-step 2-A: Background field acquisition and preprocessing.

[0032] The background data source is selected from reanalysis data covering the target flight path area. The acquisition range is 50km outside the latitude and longitude boundaries of the target flight path, the altitude is from ground level to maximum flight altitude, and the time is consistent with the range set in step 1-A. Static temperature is extracted from the original reanalysis data. East wind North wind Vertical wind Data serves as the initial background field for the variational assimilation system.

[0033] Then adopt Figure 2 The illustrated three-dimensional linear interpolation method interpolates the background field resolution to the spatiotemporal scale of the QAR observation dataset: latitude and longitude interpolation is based on the longitude of the QAR observation points along the target route. ,latitude Centered on the background field, select four adjacent grid points. Bilinear interpolation was used for calculation. , Background field parameter values ​​at the location; height interpolation based on the height of the QAR observation point. Centered on the background field, select two adjacent height layers. Linear interpolation is used for calculation. Background field parameter values ​​at the location; time interpolation using QAR observation timestamps. Centered on the background field, select two adjacent time points. Linear interpolation is used for calculation. Background field parameter values ​​at that location.

[0034] Sub-step 2-B: Construction of variational assimilation objective function (variational assimilation system).

[0035] A three-dimensional variational assimilation (3D-Var) framework is adopted, and the variational assimilation objective function includes error terms and physical constraint terms: (5); in, Let be the overall objective function.

[0036] For the error term: (6); in, For the optimization analysis field to be determined, As background scene, The QAR observational meteorological data retrieved in step one, All are in matrix form: , This represents the index of the lattice point in the field that participated in assimilation. The background field error covariance matrix; Here is the QAR observation error covariance matrix. Background field error covariance matrix. form: (7); QAR observation error covariance matrix form: (8); in, The standard deviation of the error for each parameter is... b , obs For observational data, meteorological reanalysis data of the target route over the past year were collected and compared with QAR observation datasets for calculation. Taking static temperature in the background field error as an example: (9); (10); in, This refers to the QAR observation data interpolated onto the corresponding grid points of the background field. The background field has a total of A number of time-space matching comparison items are available. This refers to the reanalysis data interpolated to the corresponding grid points in the QAR observation dataset. The observation set contains a total of A number of time-space matching comparison items are available. Refers to the first Comparison items.

[0037] For physical constraint terms: (11); (12); in, This represents the specific value of the i-th item in the physical constraint term matrix. For the first The easterly wind in the comparison item, For the earth to turn the wind, ; Coriolis parameters, ; The Earth's rotational angular velocity is 7.292 × 10⁻⁶. -5 rad / s; The latitude of the target route is represented in rad. For height, The vertical pressure gradient is determined by the background pressure field. calculate; Given atmospheric density, from the ideal gas law calculate; The acceleration due to gravity is 9.8 m / s². 2 ; To constrain weights, the actual wind variance is statistically adjusted based on the squared mean of the atmospheric geostrophic balance deviation in mid- and low-latitude regions as a percentage of the total deviation. (Mid-latitude flight routes) Low-latitude air routes .

[0038] Step 3: Divide the data into a 1-hour main window and sub-window, solve the variational assimilation objective function using the conjugate gradient method, and generate preliminary optimized meteorological reanalysis data after 0.1°×0.1° gridding and noise removal.

[0039] Sub-step 3-A: Solve for the minimum value of the objective function using the conjugate gradient method.

[0040] use Figure 3 The assimilation window setting shown allows for the filtering of spatiotemporally overlapping QAR observation data and background field data within each sub-window, resulting in the acquisition of QAR and background field matching data for each sub-window.

[0041] Taking a single sub-window as an example, to solve for the minimum of the objective function: Let the analysis field be... initial value Set a convergence threshold The initial value is 10 -3 ; Calculate the gradient initial gradient Search direction initial value Set as Update step size Set to 0.1; gradient calculation is as follows: (13); By step size Updated analysis field: ,in, This represents the solution required in the variational assimilation system. The k One iteration term, This represents the solution required in the variational assimilation system. The k Each iteration term synchronously calculates the new gradient. ;like Stop calculation and output the current analysis field. ; Finally, the arithmetic mean of the sub-window analysis fields is taken to obtain the preliminary optimized analysis field for the 1-hour assimilation window.

[0042] Sub-step 3-B: Post-processing of the analysis field, simplifying meshing and noise removal.

[0043] First, spatial gridding is performed, dividing the target flight path into a 0.1° × 0.1° grid. Then, the nearest neighbor interpolation method is used to preliminarily optimize the parameter values ​​of the analysis field. The values ​​are assigned to the corresponding grids, meaning each grid value is equal to the optimized value of the nearest QAR observation point. Finally, a 3-point moving average is used to take the mean of the value of each grid and the two adjacent grids to eliminate isolated noise.

[0044] Step 4: Compare the preliminary optimized meteorological reanalysis data with the original reanalysis data to calculate the error, iteratively adjust the variational assimilation system parameters according to priority, and output a high-resolution route dataset.

[0045] Sub-step 4-A: Define the error index between the assimilated data and the original reanalysis data.

[0046] The assimilated data is a high-resolution dataset generated after the assimilation and fusion steps in the previous steps. ,in It uses a 0.1° × 0.1° high-resolution grid index. For height indexing, Indexed by time; the original reanalysis data are the reanalysis data before assimilation. ,in A low-resolution grid index of 0.25° × 0.25° is used. For height indexing, For time indexing, Includes elements and Consistent, all are .Will Matched to four dimensions of time and space The grid points are obtained. ,make sure and The grid and time are all aligned.

[0047] Define the error based on grid points respectively and time error and corresponding indicators : (14); (15); in, , The maximum value of the latitude index. The maximum value of the longitude index. The maximum value of the height index. The maximum value of the time index, the error threshold. Set them respectively to, , , , .

[0048] Sub-step 4-B: Iterative parameter adjustment, outputting a high-resolution route dataset.

[0049] If the error index at any grid point or in time does not meet the standard, then... Figure 4 The parameter priority diagram shows the adjustment of parameters in steps 2-B and 3-A of the preceding steps. Detection is performed according to different non-compliance conditions, and then based on the convergence threshold. Error matrix Parameter optimization is performed in the order of assimilation of the sub-window step size. During optimization, both the convergence threshold and the sub-window step size are reduced, and the error matrix is ​​adjusted according to the location of substandard grid points. If the error does not meet the standard in the route area, it is reduced. Increase If the standard is not met in non-airway areas, the risk will increase. Shrink To reduce Increase For example, the formula is as follows: (16); (17); (18); in, This represents the sub-window time step. After completing iterative parameter optimization, the output is a high-resolution, high-precision dataset that meets the error metrics. A 0.1°×0.1° route-specific grid preserves small- to medium-scale features along the route, which is used for higher-precision route data prediction in subsequent steps.

[0050] Step 5: Based on the LSTM prediction model optimization, compare the prediction accuracy of assimilated and unassimilated data to improve the prediction accuracy of airway meteorological elements for the next hour, and finally output the airway atmospheric calm temperature and disturbance wind predictor.

[0051] Sub-step 5-A: Establish a prediction model and define the benchmarks and indicators for comparing prediction accuracy.

[0052] The assimilated data used for prediction is selected from the output of step three. Extract a certain length of route segment data. Unassimilated data are raw reanalysis data. The result obtained by three-dimensional linear interpolation is... Same dimension .in For waypoint indexing, one point is taken every 0.1° along the way. For height indexing, For time indexing.

[0053] The prediction model uses an LSTM time series model, based on... and Predicted static temperature along the flight path for the next hour East wind North wind and vertical wind The model formula is: (19); in, , These are the meteorological element values ​​for the corresponding sequence. It is the length of the time series input to the model. Model parameters, including the weight matrix and bias terms .

[0054] The baseline for comparing the forecast results is the raw reanalysis data from the next hour. Interpolate the data to waypoints to obtain the actual flight path data for the next hour. ; using real data Define the root mean square error of assimilated data prediction as a benchmark. Root mean square error of prediction for unassimilated data Precision improvement rate Three indicators: (20); (twenty one); (twenty two); in, This represents the maximum value of the waypoint index. The maximum value of the height index. For data The corresponding prediction results For data The corresponding prediction results.

[0055] Sub-step 5-B: Model parameter optimization, outputting a route atmospheric calm temperature and disturbance wind predictor.

[0056] To minimize and To optimize the LSTM model parameters Learning rate and the length of the input time series ,use Figure 5 The following is the optimization process for the relevant parameters of the prediction model: (1) Initialize the parameters: weight matrix Normal initialization was used, with a mean of 0 and a standard deviation of 0.01; bias term Initialize to 0; initial learning rate set to The initial time series length is set to... .

[0057] (2) Parameter iterative optimization: Divide the historical dataset into training set and validation set in a 7:3 ratio; calculate the loss function of the training set, which is the mean square error between the predicted value and the true value; (twenty three); in, The time series length of the training set; calculate the loss function pair. gradient Then use Optimizer update parameters ; (twenty four); in, kThis represents the number of iterations. If the validation set Five consecutive iterations of improvement halved the learning rate. Generate a candidate set by taking values ​​at intervals of 1. Iterate through the candidate set value, These represent the minimum and maximum values, respectively.

[0058] (3) Set convergence conditions: validation set The value is higher than the target threshold to ensure generalization ability; five consecutive iterations. Decrease ≤ 0.01; Parameter change over 10 consecutive iterations Training set : Static temperature ≤ 0.8℃, horizontal wind ≤ 0.7m / s, vertical wind ≤ 0.5m / s.

[0059] After optimization to meet the convergence condition, the final output is an optimized atmospheric static temperature and disturbance wind predictor for the flight path, which satisfies the following conditions for any flight path point r and altitude h. .

[0060] Through the above steps, this invention can generate a high-resolution route dataset after acquiring QAR data and reanalysis data of the target route, and obtain a corresponding route atmospheric calm temperature and disturbance wind prediction system for predicting high-precision calm temperature and disturbance wind data for the route in the next hour. This supports aviation safety applications such as route turbulence warning and takeoff and landing wind shear monitoring.

[0061] This invention is compatible with QAR data from multiple aircraft models and mainstream reanalysis data. It has low engineering costs and can be directly applied to aviation safety scenarios such as en-route turbulence warning, flight plan optimization, and takeoff and landing wind shear monitoring.

[0062] Example 2 The present invention also provides a system for improving the accuracy of route static temperature and disturbance wind prediction based on flight data and reanalysis data assimilation. The system is used to implement the method described in Embodiment 1. The system includes: a data acquisition module, a construction module, an optimization module, a comparison module, and a prediction module. The acquisition module is used to filter and fill in missing values ​​in the QAR data of the target route, and to retrieve the static temperature and disturbance wind to obtain the QAR observation dataset. The module is used to select reanalysis data covering the target route as the background field. Based on the QAR observation dataset, the reanalysis data is adapted to the scale of meteorological elements derived from the QAR data through three-dimensional linear interpolation. Combining the dynamic background field error matrix B, the observation error matrix R and atmospheric physical constraints, a variational assimilation system is constructed. The optimization module is used to divide the data into a 1-hour main window and a sub-window, solve the variational assimilation system using the conjugate gradient method, and generate preliminary optimized meteorological reanalysis data after 0.1°×0.1° gridding and noise removal. The comparison module is used to compare the errors between the preliminary optimized meteorological reanalysis data and the original reanalysis data, iteratively adjust the variational assimilation system parameters according to priority, and output a high-resolution route dataset, i.e., assimilation data. The prediction module is used to optimize the LSTM prediction model, compare the prediction accuracy of assimilated and unassimilated data, improve the prediction accuracy of airway meteorological elements for the next hour, and finally output the airway atmospheric calm temperature and disturbance wind predictor.

[0063] The embodiments described above are merely preferred embodiments of the present invention and are not intended to limit the scope of the present invention. Various modifications and improvements made to the technical solutions of the present invention by those skilled in the art without departing from the spirit of the present invention should fall within the protection scope defined by the claims of the present invention.

Claims

1. A method for improving the accuracy of route static temperature and disturbance wind prediction based on the assimilation of flight data and reanalysis data, characterized in that, The method includes: Step 1: Filter and fill in missing values ​​in the QAR data of the target route, and invert the static temperature and disturbance wind to obtain the QAR observation dataset; Step 2: Select reanalysis data covering the target route as the background field. Based on the QAR observation dataset, adapt the reanalysis data to the scale of meteorological elements derived from the QAR data through three-dimensional linear interpolation. Combine the dynamic background field error matrix B, the observation error matrix R and atmospheric physical constraints to construct a variational assimilation system. Step 3: Divide the data into a 1-hour main window and sub-window, solve the variational assimilation system using the conjugate gradient method, and generate preliminary optimized meteorological reanalysis data after 0.1°×0.1° gridding and noise removal; Step 4: Compare the errors of the preliminarily optimized meteorological reanalysis data with the original reanalysis data, iteratively adjust the variational assimilation system parameters according to priority, and output a high-resolution route dataset, i.e., the assimilated data; if the error index at any grid point or time does not meet the standard, adjust the parameters according to the parameter priority, and detect different non-compliance situations, and then adjust according to the convergence threshold. Background field error matrix Observation error matrix The parameters are optimized in the order of the assimilation sub-window step size. During optimization, both the convergence threshold and the assimilation sub-window step size are reduced, and the background field error matrix is... and observation error matrix Adjustments are made based on the location of the non-compliant grid points. If the error is not compliant in the flight path area, the background field error matrix is ​​reduced. Increase the observation error matrix If the standard is not met in non-flight areas, the background field error matrix should be increased. Reduce the observation error matrix After completing the parameter iterative optimization, a high-resolution, high-precision dataset that meets the error index is output. Step 5: Based on the LSTM prediction model optimization, compare the prediction accuracy of assimilated and unassimilated data to improve the prediction accuracy of airway meteorological elements for the next hour, and finally output the airway atmospheric calm temperature and disturbance wind predictor.

2. The method for improving the accuracy of route static temperature and disturbance wind prediction based on flight data and reanalysis data assimilation as described in claim 1, characterized in that, Methods for filtering and imputing missing values ​​in QAR data of the target route, and retrieving static temperature and disturbing wind to obtain the QAR observation dataset include: QAR data that fall within the latitude and longitude boundaries of the target route, are within the corresponding altitude range, and are within the set time range are selected from the airline database to form the target route QAR raw dataset; Extracting the parameter total atmospheric temperature from the original QAR dataset ,Mach number , ground speed Eastward component of ground velocity Northward component Vertical component Aircraft roll angle Yaw angle Pitch angle Angle of attack Outlier removal is performed, and then missing data is filled in; Based on the parameters in the processed QAR original dataset, meteorological elements are inverted to obtain the QAR observation dataset; Among them, inversion static temperature include: ； in, The specific heat ratio of air; Inverting the disturbed wind includes: calculating the speed of sound. Calculate vacuum velocity Inverted three-axis winds: Eastward winds North wind Vertical wind ; ； ； ； in, is the gas constant for dry air.

3. The method for improving the accuracy of route static temperature and disturbance wind prediction based on flight data and reanalysis data assimilation as described in claim 2, characterized in that, Based on QAR observation datasets, a method for constructing a variational assimilation system by adapting reanalysis data to the scale of meteorological elements derived from QAR data through three-dimensional linear interpolation, and combining the dynamic background field error matrix B, observation error matrix R, and atmospheric physical constraints, includes: Select reanalysis data covering the target route area and extract static temperature from the reanalysis data. East wind North wind Vertical wind Data serves as the initial background field for the assimilation system; A three-dimensional linear interpolation method was used to interpolate the background field resolution to the spatiotemporal scale of the QAR observation dataset; Based on the spatiotemporal scale of the QAR observation dataset, a variational assimilation objective function, i.e., a variational assimilation system, is constructed. Among them, methods for interpolating the background field resolution to the spatiotemporal scale of the QAR observation dataset using three-dimensional linear interpolation include: Latitude and longitude interpolation based on the longitude of the QAR observation point along the target route. ,latitude Centered on the background field, select four adjacent grid points. Bilinear interpolation was used for calculation. , Background field parameter values ​​at that location; Height interpolation based on the height of the QAR observation point Centered on the background field, select two adjacent height layers. Linear interpolation is used for calculation. Background field parameter values ​​at that location; Time interpolation uses the timestamps of QAR observation points. Centered on the background field, select two adjacent time points. Linear interpolation is used for calculation. Background field parameter values ​​at that location; The variational assimilation objective function is: ； in, This is the error term; For physical constraints; in, ;in, For the optimization analysis field to be determined, As background scene, The QAR observational meteorological data retrieved in step one, All are in matrix form: , This represents the index of the lattice point in the field that participated in assimilation. The background field error matrix; Here are the observation error matrix and the background field error matrix. form: ; Observation error matrix form: ;in, This represents the standard deviation of the error for each parameter; , , , These represent the static temperature, easterly wind, northerly wind, and vertical wind obtained from QAR data inversion, respectively. in, ; ;in, This represents the specific value of the i-th item in the physical constraint term matrix. For the first The easterly wind in the comparison item, For the earth to turn the wind, ; Coriolis parameters, ; This is the Earth's rotational angular velocity; The latitude of the target route is represented in rad. For height, The vertical pressure gradient is determined by the background pressure field. calculate; Given atmospheric density, from the ideal gas law calculate; It is the acceleration due to gravity; To constrain the weights, the actual wind variance is statistically adjusted based on the percentage of the squared mean of the atmospheric geostrophic balance deviation in mid- and low-latitude regions.

4. The method for improving the accuracy of route static temperature and disturbance wind prediction based on flight data and reanalysis data assimilation as described in claim 3, characterized in that, The method for generating preliminary optimized meteorological reanalysis data by dividing the data into a 1-hour main window and sub-windows, solving the variational assimilation system using the conjugate gradient method, and then performing 0.1°×0.1° gridding and noise removal includes: Let the analysis field be set initial value Set a convergence threshold The initial value is 10 -3 ; Calculate the gradient initial gradient Search direction initial value Set as Update step size Set to 0.1; gradient calculation is as follows: ； By step size Updated analysis field: ,in, This represents the solution required in the variational assimilation system. The k One iteration term, This represents the solution required in the variational assimilation system. The k Each iteration term synchronously calculates the new gradient. ;like Stop calculation and output the current analysis field. ; The arithmetic mean of the sub-window analysis fields is taken to obtain the preliminary optimized analysis field for the 1-hour assimilation window; The target route grid is divided into 0.1° × 0.1°. The nearest neighbor interpolation method is used to preliminarily optimize the parameter values ​​of the analysis field. The value is assigned to the corresponding grid, meaning that the value of each grid is equal to the optimized value of the nearest QAR observation point; A 3-point moving average was used to take the mean of itself and the two adjacent grids for each grid, eliminating isolated noise and generating preliminary optimized meteorological reanalysis data.

5. The method for improving the accuracy of route static temperature and disturbance wind prediction based on flight data and reanalysis data assimilation as described in claim 4, characterized in that, Methods for comparing the errors of the initially optimized meteorological reanalysis data with the original reanalysis data, iteratively adjusting the variational assimilation system parameters according to priority, and outputting high-resolution route datasets include: Set the assimilated data as ,in It uses a 0.1° × 0.1° high-resolution grid index. For height indexing, For time indexing; The original reanalysis data are set as follows ,in A low-resolution grid index of 0.25° × 0.25° is used. For height indexing, For time indexing; Includes elements and Consistent, all are ,Will Matched to four dimensions of time and space The grid points are obtained. ; based on , and Set the error based on grid points and time error and corresponding indicators ; in, ; ;in, , The maximum value of the latitude index. The maximum value of the longitude index. The maximum value of the height index. The maximum value of the time index, the error threshold. Set them respectively to, , , , .

6. The method for improving the accuracy of route static temperature and disturbance wind prediction based on flight data and reanalysis data assimilation as described in claim 5, characterized in that, Methods for comparing the prediction accuracy of assimilated and unassimilated data based on LSTM prediction model optimization include: from Extract a certain length of route segment data. Assimilated data is used for prediction, while unassimilated data serves as the raw reanalysis data. The result obtained by three-dimensional linear interpolation is... Same dimension ,in For waypoint indexing, one point is taken every 0.1° along the way. For height indexing, For time indexing; The prediction model uses an LSTM time series model, based on... and Predicted static temperature along the flight path for the next hour East wind North wind and vertical wind The model formula is: ； in , These are the meteorological element values ​​for the corresponding sequence. It is the length of the time series input to the model. Model parameters, including the weight matrix and bias terms .

7. A system for improving the accuracy of route static temperature and disturbance wind prediction based on the assimilation of flight data and reanalysis data, the system being used to implement the method described in any one of claims 1-6, characterized in that, The system includes: a data acquisition module, a data construction module, an optimization module, a comparison module, and a prediction module; The acquisition module is used to filter and fill in missing values ​​in the QAR data of the target route, and to retrieve the static temperature and disturbance wind to obtain the QAR observation dataset. The construction module is used to select reanalysis data covering the target route as the background field, and based on the QAR observation dataset, adapt the reanalysis data to the scale of meteorological elements derived from the QAR data through three-dimensional linear interpolation. Combined with the dynamic background field error matrix B, the observation error matrix R and atmospheric physical constraints, a variational assimilation system is constructed. The optimization module is used to divide the data into a 1-hour main window and a sub-window, solve the variational assimilation system using the conjugate gradient method, and generate preliminary optimized meteorological reanalysis data after 0.1°×0.1° gridding and noise removal. The comparison module is used to compare the errors between the preliminary optimized meteorological reanalysis data and the original reanalysis data, iteratively adjust the variational assimilation system parameters according to priority, and output a high-resolution route dataset, i.e., assimilation data. The prediction module is used to optimize the LSTM prediction model, compare the prediction accuracy of assimilated and unassimilated data, improve the prediction accuracy of airway meteorological elements for the next hour, and finally output the airway atmospheric calm temperature and disturbance wind predictor.

Images