Air route turbulence intensity prediction method and system based on multi-source data assimilation and time-varying weighted set

Abstract

The invention belongs to the technical field of aviation safety and weather prediction crossing, and discloses an air route turbulence intensity prediction method and system based on multi-source data assimilation and time-varying weighted collection.The method comprises the steps that multi-source data are directionally collected, inversion and turbulence diagnosis index calculation are conducted on the multi-source data based on different stages of flight, and the multi-source data are obtained; a turbulence dissipation rate EDR index is calculated; the method comprises the following steps: by taking a temperature-pressure wind field inverted by multi-source data as background data and a temperature-pressure wind field inverted by QAR data as observation data, realizing multi-source data assimilation through Kalman filtering, generating a high-temporal-spatial-resolution meteorological data set on an air route, and outputting an error covariance matrix representing the accuracy of an assimilation result; high-resolution assimilation data is used as input, an EDR index obtained by inverting a vertical wind frequency through QAR data is used as a reference label, a weight coefficient of a turbulence diagnosis index is optimized in real time through a machine learning method, a time-varying weighted set turbulence intensity index is constructed, and turbulence intensity prediction confidence is output.

Description

Technical Field

[0001] This invention belongs to the interdisciplinary field of aviation safety and meteorological forecasting, specifically involving a method and system for predicting the intensity of turbulence along a flight path based on multi-source data assimilation and time-varying weighted sets. Background Technology

[0002] Turbulence along flight paths can cause sudden aircraft vibrations, which can range from minor issues like passenger comfort and increased crew workload to serious damage to onboard equipment and threats to flight structure safety. Therefore, turbulence prediction is an indispensable core component of the civil aviation safety system. Turbulence formation is related to multiple factors, including atmospheric thermal instability and dynamic disturbances. Thermal instability often stems from temperature differences between air masses at different altitudes; the convective motion of rising warm air and sinking cold air easily disrupts airflow stratification, generating turbulence. Dynamic disturbances are often caused by external forces such as topographical obstruction and jet shear, forcing airflow to change its motion and forming irregular vortices, thus inducing turbulence. Mixed causes involve the combined effects of thermal and dynamic forces; for example, the interplay between thermal disturbances caused by convection and dynamic disturbances caused by jet shear can make the vortex motion of airflow more intense and disordered. These factors combine to make the generation and evolution of turbulence more complex.

[0003] Existing technologies often integrate multi-source meteorological data using simple weighted averaging or direct splicing. While this can make preliminary use of information from different data sources, it still has limitations: First, the characteristics and data reliability of wind speed, temperature, and air pressure data from different sources vary greatly at specific flight stages. If data fusion is performed simply, the results are often difficult to guarantee reliability. Second, the turbulence intensity index is fixed. Traditional methods use fixed weights to combine causal parameters, which cannot adapt to the differences in the contribution of turbulence causes under different scenarios, and cannot provide a confidence assessment of the prediction results, resulting in uncertainty in prediction accuracy and reliability.

[0004] Multi-source meteorological data includes sounding data, wind profiler radar data, satellite cloud wind guidance data, satellite occultation data, AMDAR (Aircraft Weather Data Relay System) data, and reanalysis data, which can provide large-scale, multi-dimensional information on temperature, pressure, and wind fields. Radiosonde data provides horizontal wind speed and temperature / pressure data at the end of the flight path and in the low and mid-altitude regions, with a spatial resolution of 10 km and a temporal resolution of 6-12 hours. Wind profiler radar data provides horizontal and vertical wind speeds at the end of the flight path and in the low-altitude regions, with a spatial resolution of 0.5-2 km and a temporal resolution of 6 minutes. Satellite cloud-guided wind data provides horizontal wind speeds during the cruise phase and in the upper atmosphere, with a spatial resolution of 10 km and a temporal resolution of 3-6 hours. Satellite occultation data provides temperature / pressure data during the cruise phase and in the upper atmosphere, with a spatial resolution of 20 km and a temporal resolution of 1 hour. AMDAR data provides horizontal wind speed and temperature / pressure data for the entire flight path (end of the flight path + cruise phase), with a spatial resolution of approximately 1 km and a temporal resolution in the minute range. Reanalysis data provides horizontal and vertical wind speed and temperature / pressure data for the entire airspace. It integrates basic observation data with numerical model assimilation to ultimately obtain meteorological data with a spatial resolution of 25 km and a temporal resolution of 1 hour. QAR data can invert local temperature, pressure, wind and EDR index (turbulent dissipation rate) along the flight path in real time by using meteorological parameters such as total atmospheric temperature, static pressure, Mach number, ground speed and aircraft attitude (roll angle, pitch angle, yaw angle). The synergy of various data has the advantage of temporal and spatial complementarity.

[0005] Based on this characteristic, this invention uses Kalman filtering to assimilate multi-source data and output the error covariance matrix. It dynamically optimizes the weights of six parameters through machine learning, constructs a set index that matches the EDR index retrieved from QAR data, and generates prediction confidence, thus overcoming the limitations of traditional methods. Summary of the Invention

[0006] To address the problems of insufficient utilization of traditional route turbulence prediction data, fixed ensemble index weights, and lack of assimilation accuracy characterization and prediction confidence assessment, this invention provides a method and system for predicting route turbulence intensity based on multi-source data assimilation and time-varying weighted ensembles. It generates high-resolution assimilated data by fusing multi-source temperature, pressure, and wind data using Kalman filtering and outputs an accuracy covariance matrix. Combined with machine learning, it dynamically optimizes the weights of six turbulence diagnostic indices, achieving accurate prediction and confidence assessment of turbulence intensity. This method is suitable for route turbulence early warning and flight safety management in transport aviation and general aviation.

[0007] To achieve the above objectives, the present invention provides the following solution: A method for predicting the intensity of turbulence along a flight path based on multi-source data assimilation and time-varying weighted sets, the method comprising: Multi-source wind speed data, multi-source temperature and air pressure data, and QAR data are collected in a targeted manner. Based on different stages of flight, the multi-source data are inverted and turbulence diagnostic index is calculated, and the turbulence dissipation rate (EDR) index is estimated. Using the temperature, pressure and wind fields retrieved from multi-source data as background data and the temperature, pressure and wind fields retrieved from QAR data as observation data, multi-source data assimilation is achieved through Kalman filtering to generate a high spatiotemporal resolution meteorological dataset along the flight route, and an error covariance matrix characterizing the accuracy of the assimilation results is output. Using high-resolution assimilation data as input and the EDR index obtained by inverting vertical wind frequency from QAR data as the benchmark label, the weight coefficients of the turbulence diagnostic index are optimized in real time through machine learning methods to construct a time-varying weighted ensemble turbulence intensity index, and output the turbulence intensity value and the turbulence intensity prediction confidence based on the assimilation accuracy covariance matrix and weight coefficients.

[0008] Preferred methods for inverting multi-source data based on different stages of flight include: ; in, For static temperature, Total atmospheric temperature Mach number, The specific heat ratio of air; ; in, For static pressure, For air pressure, altitude The international standard atmospheric temperature at that location It is the acceleration due to gravity. The molar mass of dry air, Gas constant of dry air This refers to the tropospheric temperature lapse rate. ; in, For the speed of sound, The pitch angle, Yaw angle For roll angle, For the angle of attack, Vacuum speed, The eastward component of the ground velocity. The northward component of the ground velocity. This represents the vertical component of the ground velocity. The wind is from the east. The wind is from the north. It is a vertical wind.

[0009] Preferably, the method for calculating the turbulence diagnostic index includes: Thermogeneous index level temperature gradient calculate: ; Dynamics Factors Index Dutton Index calculate: ; Ellrod1 index (dynamic causation index) calculate: ; Richardson Index of Mixed Causes calculate: ; in, Potential temperature, calculated from temperature and air pressure. , , , These represent latitude, longitude, and altitude, respectively. Mixed origin index MOSCAT2 index calculate: ; Mixture origin index level temperature gradient absolute value / Richardson number calculate: ; in, East-west winds, The wind is from the north and south. For vertical wind, For temperature, This refers to air pressure.

[0010] Preferred methods for calculating the turbulent dissipation rate (EDR) index include: ; in, For the first The EDR index mapped from the turbulence diagnostic index For the first Turbulence diagnostic index , Indicates standard deviation, As a benchmark for turbulence intensity, b This is an empirical coefficient.

[0011] Preferably, the method of using temperature, pressure, and wind fields retrieved from multi-source data as background data and temperature, pressure, and wind fields retrieved from QAR data as observation data, dynamically correcting background data bias through Kalman filtering, generating a high spatiotemporal resolution assimilated meteorological dataset along the flight route, and outputting a covariance matrix characterizing the accuracy of the assimilation results includes: Based on the analysis field at the previous moment, predict the background field and prediction error covariance at the current moment; Calculate the optimal fusion weights based on the prediction error and the observation error; Based on the optimal fusion weights and combined with QAR data, the predicted state is corrected to obtain the optimal analysis field; Based on the optimal analysis field, update the error estimate of the analysis field at the current moment; Among them, the methods for predicting the background field and prediction error covariance at the current moment based on the analysis field at the previous moment include: ; ; in, for The time-predicted state vector has a dimension of 5×1, which is the estimated value of the background field. for The time-mapping analysis field vector has a dimension of 5×1. for The time-prediction error covariance matrix has a dimension of 5×5. for The error covariance matrix is ​​analyzed at each moment, with a dimension of 5×5. State transition matrix The transpose of the matrix, 5×5; initial state Let the background field be defined, and the initial error covariance matrix be defined. Let the mean error covariance of the historical data be used. The system noise matrix; The background error covariance matrix; Methods for calculating the optimal fusion weights based on prediction and observation errors include: ; in, for The time-major Kalman gain, with a dimension of 5×5, represents the correction weight of the observed data to the background field. Observation matrix The transpose of the matrix, because It is the identity matrix. , The observation error covariance matrix; Methods for obtaining the optimal analysis field by correcting the predicted state based on the optimal fusion weights and combining QAR data include: ; in, for The final optimized analysis field vector at any given time has a dimension of 5×1, which is the output data after assimilation. This represents the observation increment, with a dimension of 5×1, reflecting the difference between observed and predicted data. For observation vectors; Methods for updating the error estimate of the analysis field at the current moment based on the optimal analysis field include: ; in, for The error covariance matrix is ​​analyzed at each moment, with a dimension of 5×5. It is a 5×5 identity matrix.

[0012] Preferably, the method for constructing a time-varying weighted ensemble turbulence intensity index includes: ; in, It is a time-varying weighted ensemble turbulence intensity index. The optimal dynamic weighting coefficients are... For the first The EDR index is mapped from the turbulence diagnostic index.

[0013] Preferred methods for calculating the confidence level of turbulence intensity prediction include: Combining the accuracy covariance matrix of the assimilation results with the dynamic weighting coefficients, the confidence level for turbulence intensity prediction in the 0-1 interval is calculated using the following formula: ; in, The total error variance of the assimilation results. Let be the analysis error covariance matrix of the assimilation results. For matrix trace operations, The first of the error covariance matrix One diagonal value; The confidence level is calculated using the "1-normalized error" method. Ensure the result is in the interval [0, 1]. The formula for the historical maximum possible total error variance is as follows: .

[0014] The present invention also provides a route turbulence intensity prediction system based on multi-source data assimilation and time-varying weighted set. The system is used to implement the aforementioned method and includes: an inversion module, an assimilation module, and an optimization module. The inversion module is used to collect multi-source wind speed data, multi-source temperature, air pressure data and QAR data in a directional manner, perform inversion and turbulence diagnostic index calculation on the multi-source data based on different stages of flight, and deduce the turbulence dissipation rate (EDR) index. The assimilation module is used to use the temperature, pressure and wind fields retrieved from multi-source data as background data and the temperature, pressure and wind fields retrieved from QAR data as observation data. It then uses Kalman filtering to assimilate the multi-source data, generate a high spatiotemporal resolution meteorological dataset along the flight route, and output an error covariance matrix that characterizes the accuracy of the assimilation results. The optimization module is used to take high-resolution assimilation data as input, use the EDR index obtained by inverting vertical wind frequency from QAR data as the benchmark label, optimize the weight coefficients of the turbulence diagnostic index in real time through machine learning methods, construct a time-varying weighted ensemble turbulence intensity index, and output the turbulence intensity value and the turbulence intensity prediction confidence based on the assimilation accuracy covariance matrix and weight coefficients.

[0015] Compared with the prior art, the beneficial effects of the present invention are as follows: (1) Efficient and quantifiable multi-source data fusion: By fusing the wide coverage advantage of multi-source temperature, pressure and wind data with the high-resolution route of QAR data through Kalman filtering, high spatiotemporal resolution assimilation data of the route is generated, and the error covariance matrix of the assimilation result is output, providing high-quality input and accuracy reference for turbulence prediction. (2) Weight optimization and dynamic adaptation to multiple parameters: Machine learning adjusts the weights of the six causal indices according to the time series, breaking through the limitations of traditional fixed weights, so that the set of indices matches the differences in the contribution of turbulence causes under different scenarios; (3) Enhanced data adaptability by region: Based on the characteristics of data sources near the terminal area and during the high-altitude cruise phase, the interpolation resolution and state transition matrix were set differently to adapt the meteorological data of different regions. This avoided data redundancy or distortion while improving the utilization efficiency of the entire route data. Attached Figure Description

[0016] 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.

[0017] Figure 1 This is an overall flowchart of a method for predicting the intensity of turbulence along a flight path based on multi-source data assimilation and time-varying weighted sets, according to the present invention. Detailed Implementation

[0018] 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.

[0019] 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.

[0020] Example 1 like Figure 1 As shown, this invention provides a method for predicting the intensity of turbulence along a flight path based on multi-source data assimilation and time-varying weighted sets. The method includes: Step 1: Multi-source data collection, preprocessing, and calculation of turbulence diagnostic indices Multi-source wind speed data, multi-source temperature and pressure data, and QAR data are collected in a targeted manner. After basic preprocessing, based on the temperature, pressure, and wind parameters retrieved from various data sources, six core indices characterizing different causes of turbulence (thermal, dynamic, and mixing) are calculated, and all indices are mapped to EDR indices using specified formulas.

[0021] Step 2: Kalman Filtering Multi-Source Data Assimilation Using temperature, pressure and wind fields retrieved from multi-source data as background data and temperature, pressure and wind fields retrieved from QAR data as observation data, the background data bias is dynamically corrected by Kalman filtering to generate a high spatiotemporal resolution assimilation meteorological dataset along the flight route, and outputs a covariance matrix characterizing the accuracy of the assimilation results.

[0022] Step 3: Construction of time-varying weighted ensemble exponent, output of turbulence intensity and confidence level Using the high-resolution assimilation data generated in step two as input, and the EDR index obtained by inverting vertical wind speed from QAR data as the benchmark label, the weight coefficients of the six turbulence diagnostic indices are optimized in real time through machine learning methods to construct a time-varying weighted set of turbulence intensity indices. The turbulence intensity value and the turbulence intensity prediction confidence in the [0,1] interval based on the assimilation accuracy covariance matrix and weight coefficients are output.

[0023] The specific implementation process is as follows: Step 1: Multi-source data collection, preprocessing, and calculation of turbulence diagnostic indices.

[0024] Sub-step 1-A: Multi-source data preprocessing and temperature, pressure and wind inversion.

[0025] Define the target flight path, altitude, and time range, and collect five types of meteorological parameters: east-west winds. North-south wind Vertical wind ,temperature air pressure .

[0026] Based on the coverage area attributes of each data source (airway terminal / low and medium altitude, cruise segment / high altitude) and the differences in the spatiotemporal resolution of the original data, a targeted differentiated interpolation strategy is adopted to balance data accuracy and availability: For the airway terminal and low and medium altitude areas, thanks to the dense distribution of data sources such as radiosonde data, wind profiler radar data, and AMDAR data, and the relatively high spatial and temporal resolution of the original data, in order to meet the refined requirements of turbulence prediction for flight in the terminal area, all multi-source data in this area are uniformly interpolated to a grid with a high spatial resolution of 0.01°×0.01° and a high temporal resolution of 1 minute; while for the cruise segment and high altitude areas, due to the sparse distribution of core data sources such as satellite cloud wind data and satellite occultation data, and the low spatial and temporal resolution of the original data, forcibly pursuing high resolution would easily lead to data distortion. Therefore, the data in this area are interpolated to a low-resolution grid adapted to its data characteristics, such as a spatial resolution of 0.1°×0.1° and a temporal resolution of 10 minutes. Through regionally adapted interpolation processing, a multi-resolution background data source covering the entire airway with a resolution that accurately matches the regional data conditions is finally formed.

[0027] Flight data collection targets the route and collects QAR data from multiple aircraft types, specifically the total atmospheric temperature. static pressure ,Mach number Altitude Pitch angle Yaw angle Roll angle Angle of attack , ground speed Eastward component of ground velocity Northward component of ground speed Vertical component of ground velocity Vertical acceleration Then invert the static temperature. air pressure East wind North wind and 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. (2) in, altitude The international standard atmospheric temperature at that location It is the acceleration due to gravity. The molar mass of dry air, Gas constant of dry air This refers to the tropospheric temperature lapse rate. (3) in, For the speed of sound, Vacuum speed, The wind is from the east. The wind is from the north. For vertical wind. QAR data is sampled at the time resolution step of the background data source and matched to the corresponding grid of the background data source by latitude and longitude to ensure spatiotemporal synchronization for easy assimilation.

[0028] Sub-step 1-B: Calculation of turbulence diagnostic index and mapping of EDR index.

[0029] East-west winds obtained from background data North-south wind Vertical wind ,temperature air pressure Five parameters are used to calculate six turbulence diagnostic indices corresponding to three types of turbulence causes: thermal, dynamic, and mixing.

[0030] Thermogeneous index level temperature gradient The calculation is as follows: (4) Dynamics Factors Index Dutton Index The calculation is as follows: (5) Ellrod1 index (dynamic causation index) The calculation is as follows: (6) Richardson Index of Mixed Causes The calculation is as follows: (7) in, Potential temperature, calculated from temperature and air pressure. .

[0031] Mixed origin index MOSCAT2 index The calculation is as follows: (8) Mixture origin index level temperature gradient absolute value / Richardson number The calculation is as follows: (9) in, These represent latitude, longitude, and altitude, respectively, and the partial derivatives are calculated using the differences between adjacent grids.

[0032] The above six indices are mapped to the EDR index using a specified formula: (10) in, For the first The EDR index mapped from the turbulence diagnostic index b This is an empirical coefficient. For the first Turbulence diagnostic index , Indicates standard deviation, The turbulence intensity benchmark is calculated in sub-step 3-A.

[0033] Step 2: Kalman filtering for multi-source data assimilation.

[0034] Sub-step 2-A: Definition of assimilation variables and correlation matrix.

[0035] State vector : The time-assimilation analysis field vector, with a dimension of 5×1, contains 5 core meteorological variables, defined as follows: (11) in, for The temperature of the atmosphere after assimilation at any given time. The atmospheric pressure after assimilation at any given time. , , for The east-west, north-south, and vertical components of the wind field after time-space assimilation; Background Field The temperature and pressure wind field is obtained by interpolation of multi-source data and serves as the initial estimate for assimilation. Its dimension is consistent with the state vector, and each component corresponds to a temperature. air pressure East-west wind North-south wind Vertical wind ; Observation vector The wind field, composed of temperature and pressure data inverted from QAR data, has a dimension of 5×1 and corresponds one-to-one with the state vector variables, as defined below: (12) in, , Temperature and calibrated air pressure retrieved from QAR data. , , The three-axis wind components retrieved from QAR data; State transition matrix The format is as follows: (13) in, This represents the constructor of a diagonal matrix. Due to the high time resolution, the terminal region exhibits minimal state changes, and the diagonal elements... to Based on the historical time period state change rate setting, values ​​of 0.9-0.99 or 1.01-1.1 reflect weak state changes. State changes are greater during the cruise phase and high-altitude regions due to low time resolution. (Diagonal elements...) to Based on the historical period state change rate setting, a value of 0.8-0.9 or 1.1-1.2 is used to reflect state changes more significantly.

[0036] Observation matrix The background error covariance matrix is ​​a 5×5 identity matrix. and observation error covariance matrix It is a 5×5 identity matrix, in the following form: (14) (15) (16) in, These represent the error variances of the five parameters in the background field. These represent the error variances of the observed values ​​for the five parameters; System noise matrix It is a 5×5 diagonal matrix, with the diagonal elements set as the average error between historical predicted data and historical observed values, representing the model uncertainty. This represents the average error of the corresponding parameter.

[0037] (17) Sub-step 2-B: Assimilation iteration and error covariance matrix output.

[0038] State prediction: Based on the analysis field at the previous moment, predict the background field and prediction error covariance at the current moment; (18) (19) in, for The time-predicted state vector has a dimension of 5×1, which is the estimated value of the background field. for The time-mapping analysis field vector has a dimension of 5×1. for The time-prediction error covariance matrix has a dimension of 5×5. for The error covariance matrix is ​​analyzed at each moment, with a dimension of 5×5. State transition matrix The transpose of the matrix, 5×5; initial state Let the background field be defined, and the initial error covariance matrix be defined. Let the mean error covariance of the historical data be used. Kalman gain calculation: Calculate the optimal fusion weights based on the prediction error and observation error; (20) in, for The time-major Kalman gain, with a dimension of 5×5, represents the correction weight of the observed data to the background field. Observation matrix The transpose of the matrix, because It is the identity matrix. ; Analysis field update: Combine QAR data to correct the prediction state and obtain the optimal analysis field; (twenty one) in, for The final optimized analysis field vector at any given time has a dimension of 5×1, which is the output data after assimilation. The observation increment, with a dimension of 5×1, reflects the difference between the observed data and the predicted data; Analysis error covariance update: Updates the error estimate of the analysis field at the current time. (twenty two) in, for The error covariance matrix is ​​analyzed at each moment, with a dimension of 5×5. It is a 5×5 identity matrix.

[0039] Through the above iterative process, regionally adapted assimilation datasets are finally generated: a 0.01°×0.01° spatial resolution and a 1-minute temporal resolution dataset for the route terminal and low-altitude region; and a 0.1°×0.1° spatial resolution and a 10-minute temporal resolution dataset for the cruise segment and high-altitude region. Both datasets contain complete meteorological variables such as temperature, air pressure, and triaxial wind field, as well as the time analysis error covariance matrix obtained from the final assimilation iteration process. The error covariance matrix, as a core indicator characterizing the accuracy of the assimilation results, provides a foundation for subsequent confidence calculations.

[0040] Step 3: Construction of time-varying weighted set exponent and output of turbulence intensity.

[0041] Sub-step 3-A: Setting the baseline label and initial weights.

[0042] The EDR index, obtained from the vertical wind speed inversion based on QAR data (Equation 3), is used as the benchmark label for turbulence intensity. The calculation formula is as follows. After obtaining the vertical wind speed, the power spectrum of the vertical wind speed is estimated using the periodic spectrum estimation method: (twenty three) in, To represent a complex number, It is the vertical wind data sequence number. It is the first item , It is the sampling frequency of vertical wind, where This represents the vertical wind data acquired within 10 seconds. For the period of the spectrum, Then, the power spectrum of the actual vertical wind is divided by the theoretical vertical wind power spectrum within a certain frequency range to estimate the turbulence intensity estimated from the QAR data. : (twenty four) in, This represents the bias correction term; in EDR exponential estimation algorithms, 1.3 is often chosen as an empirical value. and Corresponding to frequency and upper and lower boundaries, and Corresponding to angular frequencies and And further correspondence in the spatial frequency domain and , This is the power spectrum of the theoretical turbulence model.

[0043] The initial weights are based on turbulence physics mechanisms, setting initial weights for the three types of turbulence diagnostic indices, satisfying the condition that the sum of the weights is 1, and the constraint conditions are as follows. , Indicates the first The initial weights of each diagnostic index.

[0044] Sub-step 3-B: Weight adjustment of the weighted set index based on GBDT.

[0045] Input five meteorological parameters from the high-resolution assimilation dataset , , , , And 6 turbulence diagnostic indices mapped to the EDR index , , , , , The input features are 11 dimensions; the output target is the dynamic weight coefficients of 6 turbulence diagnostic indices. The constraints are , Indicates the first The final weight of each diagnostic index.

[0046] Using a 1-minute time interval of high-resolution assimilated data as the sample unit, each sample contains 11-dimensional input features and corresponding... The dataset is divided into training, validation, and test sets in an 8:1:1 ratio.

[0047] A gradient boosting decision tree (GBDT) model is adopted as the machine learning model. This model has a strong fitting ability for nonlinear features and flexible adaptability to weight constraints, and can efficiently perform dynamic optimization of turbulence parameter weights. The loss function is the mean squared error loss (MSE), and the objective is to minimize the time-varying weighted ensemble exponent. The difference is expressed by the following formula: (25) in, This is the mean squared error loss value. For the sample size, For the first Inverting the true label from QAR data of a sample, For the first The predicted turbulence intensity for each sample is calculated using the current weighting coefficients. .

[0048] During model training, a Lagrange multiplier constraint is added to ensure that the sum of the weights is 1. The constraint function is as follows: (26) in, For a constrained loss function, It is a Lagrange multiplier.

[0049] The training optimization process involves initializing the weight coefficients as follows: Input the training set into the machine learning model and calculate the constrained loss function. The weight coefficients are updated by finding the minimum value of the loss function using the backpropagation algorithm. After each training round, the loss value is evaluated using the validation set. If the loss on the validation set decreases by less than 0.5% for five consecutive rounds, training is stopped; the optimal dynamic weight coefficients are output. The weights sum to 1.

[0050] Sub-step 3-C: Output of time-varying weighted ensemble turbulence intensity index and prediction confidence.

[0051] Based on the optimized weighting coefficients, a time-varying weighted ensemble turbulence intensity index is constructed. : (27) The physical quantities are consistent with those in step 3-B. Then, combining the accuracy covariance matrix of the assimilation result output in step 2-B with the dynamic weighting coefficients obtained in step 3-B, the prediction confidence level of turbulence intensity in the 0-1 interval is calculated using the following formula. (28) in, The total error variance of the assimilation results. The analysis error covariance matrix of the assimilation results in step 2-B is shown below. For matrix trace operations, The first of the error covariance matrix One diagonal value; then the confidence level is calculated using the "1 - normalized error" form. Ensure the result is in the interval [0, 1]. This represents the historical maximum possible total error variance.

[0052] (29) Each time a new set of assimilated data and QAR data is acquired, incremental training of the model is triggered, and the weight coefficients are updated in real time. confidence level of turbulence intensity prediction .

[0053] In practical applications, this process follows a cycle of real-time data access, regional assimilation, model iterative updates, and continuous result output: After flight takeoff, the system continuously receives multi-source data and QAR data from the flight route, processes and assimilates it according to differentiated strategies for the vicinity of the terminal area and high-altitude cruising, automatically triggers incremental training to update weights and confidence levels, and refreshes weight coefficients every 15 to 30 minutes based on the update frequency of multi-source data until the flight lands, thus achieving dynamic turbulence prediction throughout the entire flight.

[0054] This invention ultimately outputs two types of turbulence intensity indices: those near the terminal area and those for high-altitude cruising. and turbulence intensity prediction confidence Among them, the turbulence intensity index near the terminal area has a spatial resolution of 0.01°×0.01° and a time resolution of 1 minute, while the high-altitude cruise turbulence intensity index has a spatial resolution of 0.1°×0.1° and a time resolution of 10 minutes, basically covering the full-time and spatial turbulence state of the target route; the prediction confidence level is a quantitative value in the interval [0, 1], reflecting the reliability of the prediction results. The three together provide a comprehensive, accurate and reliable quantitative early warning basis for air traffic control and flight planning.

[0055] Example 2 The present invention also provides a route turbulence intensity prediction system based on multi-source data assimilation and time-varying weighted set. The system is used to implement the method described in Embodiment 1. The system includes: an inversion module, an assimilation module, and an optimization module. The inversion module is used to collect multi-source wind speed data, multi-source temperature and air pressure data, and QAR data in a directional manner. Based on different stages of flight, it performs inversion on multi-source data and calculates turbulence diagnostic index, and calculates the turbulence dissipation rate (EDR) index. The assimilation module is used to use the temperature, pressure and wind fields retrieved from multi-source data as background data and the temperature, pressure and wind fields retrieved from QAR data as observation data. It achieves multi-source data assimilation through Kalman filtering, generates a high spatiotemporal resolution meteorological dataset along the flight route, and outputs an error covariance matrix that characterizes the accuracy of the assimilation results. The optimization module takes high-resolution assimilation data as input and uses the EDR index obtained by inverting vertical wind frequency from QAR data as the benchmark label. It optimizes the weight coefficients of the turbulence diagnostic index in real time through machine learning methods, constructs a time-varying weighted ensemble turbulence intensity index, and outputs the turbulence intensity value and the turbulence intensity prediction confidence based on the assimilation accuracy covariance matrix and weight coefficients.

[0056] 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 predicting the intensity of turbulence along a flight path based on multi-source data assimilation and time-varying weighted sets, characterized in that, The method includes: Multi-source wind speed data, multi-source temperature and air pressure data, and QAR data are collected in a targeted manner. Based on different stages of flight, the multi-source data are inverted and turbulence diagnostic index is calculated, and the turbulence dissipation rate (EDR) index is estimated. Using the temperature, pressure and wind fields retrieved from multi-source data as background data and the temperature, pressure and wind fields retrieved from QAR data as observation data, multi-source data assimilation is achieved through Kalman filtering to generate a high spatiotemporal resolution meteorological dataset along the flight route, and an error covariance matrix characterizing the accuracy of the assimilation results is output. Using high-resolution assimilation data as input and the EDR index obtained by inverting vertical wind frequency from QAR data as the benchmark label, the weight coefficients of the turbulence diagnostic index are optimized in real time through machine learning methods to construct a time-varying weighted ensemble turbulence intensity index, and output the turbulence intensity value and the turbulence intensity prediction confidence based on the assimilation accuracy covariance matrix and weight coefficients.

2. The method according to claim 1, characterized in that, Methods for inverting multi-source data based on different stages of flight include: ； in, For static temperature, The total atmospheric temperature, Mach number, The specific heat ratio of air; ； in, For static pressure, For air pressure, altitude The international standard atmospheric temperature at that location It is the acceleration due to gravity. The molar mass of dry air, Gas constant of dry air This refers to the tropospheric temperature lapse rate. ； in, For the speed of sound, The pitch angle, Yaw angle For roll angle, For the angle of attack, Vacuum speed, The eastward component of the ground velocity. The northward component of the ground velocity. This represents the vertical component of the ground velocity. The wind is from the east. The wind is from the north. It is a vertical wind.

3. The method according to claim 2, characterized in that, Methods for calculating turbulence diagnostic indices include: Thermogeneous index level temperature gradient calculate: ； Dynamics Factors Index Dutton Index calculate: ； Ellrod1 index (dynamic origin index) calculate: ； Richardson Index of Mixed Causes calculate: ； in, Potential temperature, calculated from temperature and air pressure. , , , These represent latitude, longitude, and altitude, respectively. Mixed origin index MOSCAT2 index calculate: ； Mixture origin index level temperature gradient absolute value / Richardson number calculate: ； in, East-west winds, The wind is from the north and south. For vertical wind, For temperature, This refers to air pressure.

4. The method according to claim 3, characterized in that, Methods for calculating the turbulent dissipation rate (EDR) exponent include: ； in, For the first The EDR index mapped from the turbulence diagnostic index For the first Turbulence diagnostic index , Indicates standard deviation, As a benchmark for turbulence intensity, b This is an empirical coefficient.

5. The method according to claim 1, characterized in that, Using temperature, pressure, and wind fields retrieved from multi-source data as background data and temperature, pressure, and wind fields retrieved from QAR data as observation data, a high spatiotemporal resolution assimilated meteorological dataset along the flight route is generated by dynamically correcting background data bias using Kalman filtering, and a covariance matrix representing the accuracy of the assimilation results is output. The methods include: Based on the analysis field at the previous moment, predict the background field and prediction error covariance at the current moment; Calculate the optimal fusion weights based on the prediction error and the observation error; Based on the optimal fusion weights and combined with QAR data, the predicted state is corrected to obtain the optimal analysis field; Based on the optimal analysis field, update the error estimate of the analysis field at the current moment; Among them, the methods for predicting the background field and prediction error covariance at the current moment based on the analysis field at the previous moment include: ； ； in, for The time-predicted state vector has a dimension of 5×1, which is the estimated value of the background field. for The time-mapping analysis field vector has a dimension of 5×1. for The time-prediction error covariance matrix has a dimension of 5×5. for The error covariance matrix is ​​analyzed at each moment, with a dimension of 5×5. State transition matrix The transpose of the matrix, dimension 5×5; initial state Let the background field be defined, and the initial error covariance matrix be defined. Let it be the average error covariance of the historical data; The system noise matrix; The background error covariance matrix; Methods for calculating the optimal fusion weights based on prediction and observation errors include: ； in, for The time-major Kalman gain, with a dimension of 5×5, represents the correction weight of the observed data to the background field. Observation matrix The transpose of the matrix, because It is the identity matrix. , The observation error covariance matrix; Methods for obtaining the optimal analysis field by correcting the predicted state based on the optimal fusion weights and combining QAR data include: ； in, for The final optimized analysis field vector at any given time has a dimension of 5×1, which is the output data after assimilation. This represents the observation increment, with a dimension of 5×1, reflecting the difference between observed and predicted data. For observation vectors; Methods for updating the error estimate of the analysis field at the current moment based on the optimal analysis field include: ； in, for The error covariance matrix is ​​analyzed at each moment, with a dimension of 5×5. It is a 5×5 identity matrix.

6. The method according to claim 1, characterized in that, Methods for constructing time-varying weighted ensemble turbulence intensity indices include: ； in, It is a time-varying weighted ensemble turbulence intensity index. The optimal dynamic weighting coefficients are... For the first The EDR index is mapped from the turbulence diagnostic index.

7. The method according to claim 1, characterized in that, Methods for calculating the confidence level of turbulence intensity prediction include: Combining the accuracy covariance matrix of the assimilation results with the dynamic weighting coefficients, the confidence level for turbulence intensity prediction in the 0-1 interval is calculated using the following formula: ； in, The total error variance of the assimilation results. Let be the analysis error covariance matrix of the assimilation results. For matrix trace operations, The first of the error covariance matrix One diagonal value; The confidence level is calculated using the "1-normalized error" method. Ensure the result is in the interval [0, 1]. The formula for the historical maximum possible total error variance is as follows: 。 8. A route turbulence intensity prediction system based on multi-source data assimilation and time-varying weighted set, the system being used to implement the method described in any one of claims 1-7, characterized in that, The system includes: an inversion module, an assimilation module, and an optimization module; The inversion module is used to collect multi-source wind speed data, multi-source temperature, air pressure data and QAR data in a directional manner, perform inversion and turbulence diagnostic index calculation on the multi-source data based on different stages of flight, and deduce the turbulence dissipation rate (EDR) index. The assimilation module is used to use the temperature, pressure and wind fields retrieved from multi-source data as background data and the temperature, pressure and wind fields retrieved from QAR data as observation data. It then uses Kalman filtering to assimilate the multi-source data, generate a high spatiotemporal resolution meteorological dataset along the flight route, and output an error covariance matrix that characterizes the accuracy of the assimilation results. The optimization module is used to take high-resolution assimilation data as input, use the EDR index obtained by inverting vertical wind frequency from QAR data as the benchmark label, optimize the weight coefficients of the turbulence diagnostic index in real time through machine learning methods, construct a time-varying weighted ensemble turbulence intensity index, and output the turbulence intensity value and the turbulence intensity prediction confidence based on the assimilation accuracy covariance matrix and weight coefficients.

Images