Air pressure altitude dynamic correction method based on airborne data
By constructing a sliding window and LSTM model, barometric altitude is corrected in real time based on airborne data, solving the problems of incomplete barometric altitude correction and insufficient accuracy in existing technologies. This achieves full-range dynamic correction and high-precision correction, thereby improving flight safety.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- CIVIL AVIATION UNIV OF CHINA
- Filing Date
- 2026-01-29
- Publication Date
- 2026-06-23
AI Technical Summary
Existing methods can only correct barometric altitude for a small number of positioning points during the aircraft approach process, and the correction accuracy is low. They cannot achieve dynamic correction throughout the entire process, which increases the risk of the aircraft approaching obstacles under low temperature conditions.
A dynamic pressure altitude correction method based on airborne data is adopted. By constructing a sliding window, data cleaning and LSTM model, parameters are screened using recursive feature elimination method, outliers are detected and corrected, and the sliding window size is optimized to achieve real-time dynamic correction of pressure altitude.
It enables dynamic pressure-altitude correction throughout the aircraft approach, improving correction accuracy, reducing obstacle crossing risks under low-temperature conditions, and providing correction capabilities under high-temperature conditions, thereby enhancing flight safety.
Smart Images
Figure CN122258947A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to airport flight safety technology, and in particular to a method for dynamic correction of barometric altitude based on airborne data. Background Technology
[0002] In the final stages of flight operations, the main guidance methods for aircraft approach and landing include precision approach, near-precision approach, and non-precision approach. Due to terrain or cost reasons, special airports often cannot achieve two-way precision approach. To achieve continuous descent approach, some airports choose Baro-VNAV to implement near-precision approach procedures. This type of procedure uses barometric altitude as the primary reference for vertical control; therefore, the accuracy of the barometric altimeter is crucial to operational safety.
[0003] The barometric altimeter is currently the primary altitude indication device for aircraft. Its basic working principle is to provide current altitude information based on a standard barometric deceleration rate. When atmospheric conditions deviate from standard conditions, the barometric altimeter reading will also deviate, especially during approach and landing in low-temperature conditions. In such cases, the barometric altimeter reading may be higher than the actual value, meaning the actual flight altitude is lower than expected, posing a risk of the aircraft approaching obstacles. With the increasing use of Baro-VNAV (Baro-Vertical Navigation) procedures at some airports, the issue of barometric altimeter correction is becoming increasingly important.
[0004] Currently, there are two main methods for correcting barometric altitude: 1. Use correction tables. ICAO document 8168, "Air Navigation Services Procedures for Aircraft Operations—Visual and Instrument Flight Procedure Design," provides three barometric altitude correction tables, each corresponding to a different runway threshold altitude. When using these tables, the corrected barometric altitude value can be obtained by directly looking up the table or interpolating based on the airport temperature and location point altitude. Additionally, aircraft manufacturers typically provide correction tables in their crew operating manuals.
[0005] 2. Use the correction formula. ICAO document 8168 also provides a barometric altitude correction formula. This formula has higher accuracy than the correction table, but it is more complex to calculate and is often used to correct the altitude of the positioning point during flight approach procedure design.
[0006] The above methods currently have certain shortcomings in flight operations, mainly including: 1. Difficulty in achieving dynamic correction throughout the entire approach. During approach and landing, the crew's operational load is relatively large, and it is not feasible to manually make real-time corrections using correction tables or formulas. Therefore, in practice, only the barometric altitude of the approach program positioning points is usually corrected, and there is a lack of automated dynamic correction methods.
[0007] 2. The correction accuracy is too low. The correction table or correction formula relies on airport temperature as the main parameter input. However, the temperature of each positioning point to be corrected is different from that of the airport temperature because it is in the air and far away. This leads to errors in the existing method. In addition, barometric altitude is not only affected by temperature, but may also be affected by other factors in the air. The existing method does not consider other factors, resulting in low accuracy. Summary of the Invention
[0008] The purpose of this invention is to provide a method for dynamic barometric altitude correction based on airborne data. This addresses the problem that existing methods, when using barometric vertical navigation procedures, can only correct a limited number of positioning points within the approach procedure, cannot perform real-time dynamic correction, and have low accuracy. This method enables dynamic barometric altitude correction throughout the entire approach process, improves the accuracy of barometric altitude correction, and ensures flight safety when using barometric vertical navigation procedures.
[0009] The technical solution adopted in this invention is: a method for dynamic correction of barometric altitude based on airborne data, the steps of which are as follows: I. Constructing a sliding window First, available airborne parameters are obtained from historical airborne data. Position parameters, motion parameters, and environmental parameters are selected as candidate parameters for the sliding window. The final sliding window parameters are obtained by using the recursive feature elimination method. The data corresponding to the final sliding window parameters are extracted from the dynamically received airborne data, and the sliding window is initialized and updated.
[0010] II. Cleaning sliding window data After initializing and updating the sliding window, first determine whether the sliding window data is empty. For empty values, use the previous value filling method to correct them. For the data that has completed the empty value judgment and correction, the median absolute deviation method, interquartile range method, and standard score method are fused to detect outliers. The detected outliers are corrected using the linear extrapolation method to obtain the cleaned sliding window data.
[0011] III. Constructing an LSTM model and optimizing the sliding window size An LSTM model is constructed based on the cleaned sliding window data. Based on aircraft performance and approach procedure constraints, candidate intervals for sliding window size are determined. Data with different sliding window sizes are loaded into the LSTM model. The root mean square error (RMSE) index is used to evaluate the performance of the LSTM model under different sliding window sizes. The window size with the best performance index is selected as the final result of sliding window optimization.
[0012] IV. Dynamic Correction of Barometric Altitude Using LSTM Model Based on the final result of the sliding window optimization, the dynamically received airborne data is loaded into the LSTM model, and the LSTM model outputs the corrected barometric altitude, thus achieving dynamic correction of barometric altitude based on airborne data.
[0013] The design principle of this invention: Barometric vertical navigation is an important guidance method for aircraft approach and landing. Its vertical control relies primarily on barometric altitude, which is significantly affected by temperature, especially low temperatures causing the barometer reading to be higher than the actual altitude, posing a risk of approaching obstacles. This invention addresses the problem of barometric altitude correction for aircraft under high / low temperature conditions by utilizing onboard data and a correction model to achieve dynamic correction of barometric altitude.
[0014] The key steps of this method include sliding window construction, data cleaning, and LSTM model construction. First, the sliding window is constructed. Since the LSTM model used subsequently often employs a sliding window as its input structure, the dynamically received airborne data needs to be organized into a sliding window format to adapt to the characteristics of the LSTM model. First, the parameters used in the sliding window are selected using a recursive feature elimination method. Then, the data corresponding to these parameters is extracted from the received airborne data to initialize and update the sliding window.
[0015] Next, the sliding window is cleaned, which involves detecting and correcting outliers and null values. Null values can generally be identified directly, and correction can be achieved by filling in the gaps with values from the previous time step. Detecting and correcting outliers is more complex. Outliers are abnormal values that significantly deviate from the overall characteristics of the data. Based on experience using airborne data, when many parameters are used, it is generally difficult to avoid the presence of outliers in the data. Failure to detect and correct them will seriously affect subsequent model training and prediction. Considering the scenario of dynamically correcting barometric altitude, the data that can be used for detection and correction only includes data from the time step before the current moment. Therefore, common detection and correction methods need to be modified according to the characteristics of the sliding window.
[0016] This invention integrates multiple detection methods to detect outliers within a sliding window. After detecting outliers, a linear extrapolation method is used for correction to ensure that the corrected data conforms to the original trend as much as possible.
[0017] Finally, a Long Short-Term Memory (LSTM) neural network model was constructed and optimized for barometric altitude correction. The LSTM model is an improved neural network model based on the Recurrent Neural Network (RNN), commonly used in natural language processing and time series forecasting. Based on the characteristics of the cleaned sliding window data, the LSTM model parameters were set, with the sliding window size having a significant impact on model performance and requiring focused optimization. Candidate intervals for sliding window sizes were constructed based on aircraft performance and approach procedure constraints. Data with different sliding window sizes were loaded into the LSTM model, and the Root Mean Square Error (RMSE) metric was used to evaluate the LSTM model's performance. The optimized model was then used for barometric altitude correction.
[0018] The beneficial effects of this invention are: 1. It can perform dynamic corrections throughout the entire approach. Based on dynamically received airborne data and LSTM models, it can continuously obtain corrected barometric altitudes, thereby achieving corrections throughout the entire approach and landing process and overcoming the limitation of existing methods that can only correct some positioning points.
[0019] 2. It can achieve higher correction accuracy and improve flight safety when using barometric vertical navigation procedures. This invention uses a recursive feature elimination method to filter and obtain the final sliding window parameters, which can capture the influence of multiple factors in the aircraft state and external environment, and the correction accuracy is higher than that of existing methods that only consider airport temperature.
[0020] This method features high accuracy and real-time performance, overcoming the shortcomings of existing methods that can only correct barometric altitude at specific locations. It achieves dynamic barometric altitude correction throughout the entire approach process, improving obstacle crossing safety under low-temperature conditions. At the same time, it has the ability to correct barometric altitude under high-temperature conditions, and can be widely applied to flight operations using barometric vertical navigation procedures. Attached Figure Description
[0021] Figure 1 This is a flowchart illustrating the overall process of the barometric altitude dynamic correction method based on airborne data according to the present invention. Figure 2 for Figure 1 Flowchart of the sliding window parameter selection method using the recursive feature elimination method; Figure 3 This is a schematic diagram showing the parameter importance ranking in an embodiment of the present invention; Figure 4 This is a data flow diagram of the cleaning sliding window according to an embodiment of the present invention; Figure 5 This is an example diagram illustrating the outlier correction effect in an embodiment of the present invention; Figure 6 This is a diagram illustrating the effect of air pressure altitude correction in an embodiment of the present invention. Detailed Implementation
[0022] The present invention will be further described below with reference to the accompanying drawings and embodiments.
[0023] like Figure 1 As shown, a method for dynamic correction of barometric altitude based on airborne data is as follows: 1. Constructing a sliding window: First, obtain available airborne parameters from historical airborne data, select position parameters, motion parameters, and environmental parameters as candidate parameters for the sliding window, and use the recursive feature elimination method to filter and obtain the final sliding window parameters. Extract the data corresponding to the final sliding window parameters from the dynamically received airborne data, and initialize and update the sliding window.
[0024] This step organizes the received airborne data into a sliding window format to facilitate subsequent input into the LSTM model.
[0025] 1. Due to the large number of airborne data parameters, in order to save computational resources and improve the accuracy of subsequent model corrections, it is necessary to select the parameters used in the sliding window. The final sliding window parameters are obtained using the recursive feature elimination method. The selection process involves the following steps (e.g., ...). Figure 2 (as shown) A1. Select the random forest regression model as the estimator for the recursive feature elimination method, extract the data corresponding to the sliding window candidate parameters, load them into the random forest regression model for calculation, and output the weights of each candidate parameter. A2. Sort the parameters by importance according to their weights; A3. Based on the ranking of the importance of each parameter, eliminate the least important parameter from the candidate parameters to obtain new candidate parameters, and then return to step A1. A4. Repeat the operation until the number of candidate parameters reaches the specified value. The candidate parameters at this point are the final sliding window parameters.
[0026] Taking a certain aircraft model as an example, over 3300 usable airborne parameters were obtained from historical airborne data. Position parameters, motion parameters, and environmental parameters were selected as candidate parameters for a sliding window. During the initial screening, the data corresponding to these parameters were extracted and loaded into a random forest regression model. The model outputs the weights of each parameter, and the candidate parameters are ranked according to their importance. Figure 3 As shown in the figure, the horizontal axis represents the parameter weights. It can be seen from the figure that parameters 10 and 11 have a weight of 0, making them the least important parameters in this selection process, and therefore they should be excluded. The excluded candidate parameters are then selected again, and after multiple selections, the final sliding window parameters are obtained.
[0027] 2. Initialize and update sliding window data: After determining the final sliding window parameters, initialize and update the sliding window data according to the following procedure: a. Initialization of sliding window data: Initially, the data corresponding to the sliding window parameters is extracted from the received airborne data to form the sliding window.
[0028] b. Updating sliding window data: After dynamically receiving new airborne data, extract the data corresponding to the sliding window parameters from the new airborne data and add it to the sliding window. At the same time, delete the data at the first moment in the sliding window to ensure that the window size remains unchanged.
[0029] Taking a specific flight data as an example, after the final sliding window parameters are determined, the sliding window data is initialized and updated. The relevant parameters of the sliding window are shown in Table 1 (in which "NaN" represents a null value): Table 1. Data on relevant parameters of the sliding window
[0030] II. Cleaning sliding window data After initializing and updating the sliding window, first determine whether the sliding window data is empty. For empty values, use the previous value filling method to correct them. For the data that has completed the empty value judgment and correction, the median absolute deviation method, interquartile range method, and standard score method are fused to detect outliers. The detected outliers are corrected using the linear extrapolation method to obtain the cleaned sliding window data.
[0031] The sliding window already constructed in step one needs further cleaning. The process for cleaning the sliding window data is detailed below. Figure 4 First, null values are identified and corrected. If the sliding window data is null, the previous value filling method is used for correction. Then, outliers are detected and corrected. When detecting outliers, this invention uses a fusion of multiple detection algorithms for processing, and then uses linear extrapolation to correct the outliers.
[0032] 1. Null value detection and correction Upon receiving airborne data, the system first checks if the sliding window data is null. If it is null, it fills the data with the value from the previous moment. Taking the data in Table 1 as an example, the relevant parameter "Total Temperature" contains null values ("NaN" represents null values in Table 1). The relevant parameter data of the sliding window after correcting for null values is shown in Table 2. Table 2. Sliding window parameters after correcting for null values.
[0033] 2. Outlier detection After null value detection and correction, the sliding window data still needs to be subjected to outlier detection. To improve the detection rate, the fusion of median absolute deviation, interquartile range, and standard score methods is used to detect outliers.
[0034] The method of detecting outliers by combining the median absolute deviation method, interquartile range method, and standard score method is as follows: B1. First, use the median absolute deviation method for detection: For each parameter in the sliding window, first calculate the median of the parameter data, and then calculate the median absolute deviation (MAD) using the following formula: MAD =Med ( |x i - x m | ) (1) In the formula, x i For parameter data, x m The median of the parameter data. Med To obtain the median; then calculate the median absolute deviation score (MAD_score) for each parameter's data: (2) In the formula, k The scaling factor (usually 1.4826) is used to treat parameter data whose median absolute deviation score (MAD_score) exceeds the set threshold as outliers.
[0035] Taking the sliding window parameters in Table 2 as an example, the median absolute deviation (MAD) of the relevant parameter "total temperature" is calculated to be 0.1 according to formula (1), and the median absolute deviation (MAD) of the relevant parameter "vacuum speed" is 0.5. Then, the median absolute deviation score (MAD_score) of the sliding window parameters is calculated according to formula (2), as shown in Table 3: Table 3. Median absolute deviation fractional data of sliding window related parameters.
[0036] The threshold for the median absolute deviation score (MAD_score) of the relevant parameters was set to 2.5. As can be seen from Table 3, the median absolute deviation score (MAD_score) of the total temperature and vacuum velocity parameters at each time point within the sliding window did not exceed the limit, meaning that no outliers were detected.
[0037] B2. For sliding window parameter data that has been tested using the median absolute deviation method, use the interquartile range method: Sort each parameter data in the sliding window from smallest to largest. The first quartile, Q1, is the value at the 25th percentile of the sorted value, and the third quartile, Q3, is the value at the 75th percentile of the sorted value. The interquartile range (IQR) is then calculated using the following formula: IQR = Q3 - Q1 (3) If parameter data x i Meet the conditions x i <Q1- m × IQR or x i >Q3 + m × IQR In the formula m If the interquartile range coefficient (usually taken as 1.5) is used, then this parameter data... x i It is considered an outlier.
[0038] Taking the sliding window parameters in Table 2 as an example, the data of each parameter in the sliding window are sorted to obtain the quartiles Q1 and Q3 of each parameter, and the interquartile range IQR is calculated according to formula (3). The specific calculation results are shown in Table 4: Table 4. Calculation results of relevant parameters for the interquartile range method.
[0039] From Table 4, we can conclude that: Q1-m × IQR =-4.0-1.5×0.2=-4.3 Q3 + m × IQR= -3.8 + 1.5 × 0.2 = -3.5 Taking the "total temperature" parameter data in Table 2 as an example, the total temperature data at each time point was neither less than -4.3 nor greater than -3.5. Therefore, no outliers were found when using the interquartile range method.
[0040] B3. Finally, the sliding window parameter data, after being tested using the interquartile range method, are further tested using the standard score method: For each parameter data point in the sliding window, the standard score is calculated according to the following formula. z : (4) In the formula, u The mean of the parameter data. σ The standard deviation of the parameter data; the calculated standard score z The value is compared with the set threshold, and parameter data that exceeds the threshold is regarded as outliers.
[0041] Taking the sliding window parameters in Table 2 as an example, the mean and standard deviation of the sliding window parameters are first calculated as shown in Table 5: Table 5. Mean and Standard Deviation of Relevant Parameters for Sliding Window
[0042] Note: The average value of the sliding window parameter "Total Temperature" u and standard deviation σ The calculation process is a conventional calculation known in the field, and will not be described in detail here.
[0043] Mean u and standard deviation σ Substituting into formula (4), we obtain the standard scores of the relevant parameter data within the sliding window. z As shown in Table 6: Table 6 Standard Scores of Sliding Window Related Parameters
[0044] The threshold for the relevant parameters in the standard score test was set to 3.0. Taking the "total temperature" parameter data as an example, the standard score of total temperature at each time point in Table 6 was compared with the threshold of 3.0. None of the scores exceeded the threshold, so no outliers appeared in the total temperature parameter data.
[0045] 3. Correction of outliers After outliers are detected, they need to be corrected to ensure the proper functioning of subsequent prediction algorithms. In scenarios involving dynamic correction of barometric altitude, only data up to the current moment is available, excluding data from subsequent moments. Therefore, this invention uses linear extrapolation to correct the detected outliers, i.e., calculating the correction value: (5) In the formula, x* t The value is the adjustment value of the sliding window parameters at time t; x t-1 For t - Sliding window parameter data at time 1 x t-2 For t - The sliding window parameter data at time 2. The correction value calculated using formula (5). x* t Replace the original outlier.
[0046] Taking the pitch angle parameter data of a sliding window during the approach of a certain flight as an example, an outlier was detected at t=368s. t- Pitch angle at time 1 x 367 = 1.8° and t- Pitch angle at time 2 x 366 = 1.8°, substituting the pitch angle parameter data at the two moments into formula (5), we obtain the pitch angle correction value at t=368s: .
[0047] See the example of the effect after correcting outliers using pitch angle parameters. Figure 5 In the figure, the blue dashed line represents the original data, and the red solid line represents the corrected data. As can be seen from the figure, the overall trend of the data remains unchanged after correcting outliers using this invention.
[0048] III. Constructing an LSTM model and optimizing the sliding window size An LSTM model is constructed based on the cleaned sliding window data. Based on aircraft performance and approach procedure constraints, candidate intervals for sliding window size are determined. Data with different sliding window sizes are loaded into the LSTM model. The root mean square error (RMSE) index is used to evaluate the performance of the LSTM model under different sliding window sizes. The window size with the best performance index is selected as the final result of sliding window optimization.
[0049] 1. Constructing an LSTM model Long Short-Term Memory (LSTM) neural networks show good prediction performance for time series data. Based on the characteristics of the cleaned sliding window, this invention constructs an LSTM barometric altitude correction model, setting the model's hyperparameters according to the following rules: (1) Use the number of filtered sliding window parameters as the input size; (2) Based on the characteristics of the data, a shallower LSTM layer (1-2 layers) is selected, and a bidirectional recurrent network is not used; (3) The learning rate should be selected in the range of approximately 0.01-0.001.
[0050] 2. Optimize the size of the sliding window After constructing the LSTM model, the hyperparameters need to be adjusted to optimize the model's performance. The size of the sliding window significantly impacts the accuracy of the prediction results; a larger window better captures long-term dependencies, while a smaller window helps detect short-term fluctuations. To achieve the best prediction results, this method constructs time windows of different sizes based on aircraft performance and approach procedure constraints, trains LSTM models on each window, evaluates the differences in prediction performance, and finally determines the optimal time window size.
[0051] The following method is used to construct the candidate range for the sliding window size: First, determine the reference length for the sliding window size based on aircraft performance and approach procedures. L : (6) In the formula, k This is the scaling factor. D IF The distance from the intermediate approach positioning point to the runway nose during the approach procedure, in nautical miles; V app The approach speed is given in knots, and 3600 is the conversion constant from hours to seconds.
[0052] Let the candidate range for the sliding window size be [ W min , W max The lower limit of the candidate range for the sliding window size. W min Select using the following formula: (7) Upper limit of the candidate range for sliding window size W max Select using the following formula: (8) In the formula, w 1 、w 2 is an empirical constant for the sliding window configuration.
[0053] Taking a certain aircraft type and a certain approach flight as an example, the distance between the intermediate approach positioning point and the runway threshold... D IF = 6 nautical miles, approach speed 180 knots, scaling factor k Taking 2, the length of the sliding window size reference sequence is calculated according to formula (6):
[0054] Note: The sliding window size is referenced to the sequence length. L It's just numbers, no units.
[0055] Then, calculate the lower limit of the candidate interval for the sliding window size according to formula (7), and take the empirical constant. w 1=5, then W min =max{5,240 / 100}=5, calculate the upper limit of the candidate interval for the sliding window size according to formula (8), and take the empirical constant. w 2=45, then W max =min{45, 240 / 10}=24, so the candidate range for the sliding window size is [5, 24]; taking the interval of each selected sliding window size as 5, the selected sliding window sizes are 5, 10, 15, and 20.
[0056] Data with different sliding window sizes were loaded into the LSTM model, and the root mean square error (RMSE) was used to evaluate the performance of the LSTM model under different sliding window sizes. The window size with the best performance index was selected as the final result of sliding window optimization. Taking flight data as an example, the RMSE of the LSTM model with different window sizes is shown in Table 7: Table 7 shows the root mean square error (RMSE) for different window sizes.
[0057] As can be seen from Table 7, the root mean square error (RMSE) is minimized when the sliding window size is 10. Therefore, 10 should be selected as the optimal sliding window size as the final result of sliding window optimization.
[0058] Note: The calculation of root mean square error (RMSE) is a conventional calculation known in the field and will not be described in detail here.
[0059] IV. Dynamic Correction of Barometric Altitude Using LSTM Model Based on the final result of the sliding window optimization, the dynamically received airborne data is loaded into the LSTM model, and the LSTM model outputs the corrected barometric altitude, thus achieving dynamic correction of barometric altitude based on airborne data.
[0060] After completing the LSTM model construction and optimization, the dynamically received airborne data is loaded into the LSTM model according to the optimal sliding window size (10) for dynamic pressure altitude correction. The LSTM model output is the corrected pressure altitude. For airborne data within a certain time period, the pressure altitude correction effect is as follows: Figure 6 As shown in the figure, the solid blue line represents the actual air pressure altitude data, and the dashed red line represents the air pressure altitude after model correction. The corrected air pressure altitude is closer to the actual air pressure altitude.
[0061] To verify the effectiveness of this invention, the following comparison of its effects with existing pressure correction methods is presented: Taking a specific flight as an example, the airport temperature is -6℃, and the approach procedure used includes the initial approach positioning point, intermediate approach positioning point, and final approach positioning point.
[0062] Using the existing correction table method to correct the barometric altitude at each positioning point, the root mean square error (RMSE) of this method was calculated to be 105.1.
[0063] The barometric altitude at each positioning point was corrected using the existing correction formula, and the root mean square error (RMSE) of this method was calculated to be 93.7.
[0064] Using this invention to correct barometric altitude, the effect is as follows: Figure 6 As shown in the figure, the present invention achieves continuous dynamic correction. The root mean square error (RMSE) of the method is calculated to be 39.9420, which is less than that of the correction table method and the correction formula method. Therefore, it can be concluded that the present invention is superior to the existing correction table method and correction formula method in terms of pressure altitude correction accuracy, and can better meet the operational requirements.
Claims
1. A method for dynamic correction of barometric altitude based on airborne data, characterized in that, The method is as follows: I. Constructing a sliding window First, available airborne parameters are obtained from historical airborne data. Position parameters, motion parameters, and environmental parameters are selected as candidate parameters for the sliding window. The final sliding window parameters are obtained by using the recursive feature elimination method. The data corresponding to the final sliding window parameters are extracted from the dynamically received airborne data, and the sliding window is initialized and updated. II. Cleaning sliding window data After initializing and updating the sliding window, first determine whether the sliding window data is empty. For empty values, use the previous value filling method to correct them. For the data that has completed the empty value judgment and correction, the median absolute deviation method, interquartile range method, and standard score method are fused to detect outliers. The detected outliers are corrected using the linear extrapolation method to obtain the cleaned sliding window data. III. Constructing an LSTM model and optimizing the sliding window size An LSTM model is constructed based on the cleaned sliding window data. Based on aircraft performance and approach procedure constraints, candidate intervals for sliding window size are determined. Data with different sliding window sizes are loaded into the LSTM model. The root mean square error (RMSE) index is used to evaluate the performance of the LSTM model under different sliding window sizes. The window size with the best performance index is selected as the final result of sliding window optimization. IV. Dynamic Correction of Barometric Altitude Using LSTM Model Based on the final result of the sliding window optimization, the dynamically received airborne data is loaded into the LSTM model, and the LSTM model outputs the corrected barometric altitude, thus achieving dynamic correction of barometric altitude based on airborne data.
2. The method for dynamic correction of barometric altitude based on airborne data according to claim 1, characterized in that, In step one, the recursive feature elimination method is used to filter and obtain the final sliding window parameters. The filtering process performs the following operations: A1. Select the random forest regression model as the estimator for the recursive feature elimination method, extract the data corresponding to the sliding window candidate parameters, load them into the random forest regression model for calculation, and output the weights of each candidate parameter. A2. Sort the parameters by importance according to their weights; A3. Based on the ranking of the importance of each parameter, eliminate the least important parameter from the candidate parameters to obtain new candidate parameters, and then return to step A1. A4. Repeat the operation until the number of candidate parameters reaches the specified value. The candidate parameters at this point are the final sliding window parameters.
3. The method for dynamic correction of barometric altitude based on airborne data according to claim 1, characterized in that, In step two, outliers are detected by combining the median absolute deviation method, interquartile range method, and standard score method, following these steps: B1. First, use the median absolute deviation method for detection: For each parameter in the sliding window, first calculate the median of the parameter data, and then calculate the median absolute deviation (MAD) using the following formula: MAD =Med ( |x i - x m | )（1） In the formula, x i For parameter data, x m The median of the parameter data. Med To take the median; Then calculate the median absolute deviation score (MAD_score) for each parameter data: （2） In the formula, k The scaling factor is used to treat parameter data whose median absolute deviation score (MAD_score) exceeds a set threshold as outliers. B2. For sliding window parameter data that has been tested using the median absolute deviation method, use the interquartile range method: Sort each parameter data in the sliding window from smallest to largest. The first quartile, Q1, is the value at the 25th percentile of the sorted value, and the third quartile, Q3, is the value at the 75th percentile of the sorted value. The interquartile range (IQR) is then calculated using the following formula: （3） If parameter data x i Meet the conditions x i < Q1- m × IQR or x i > Q3+ m × IQR In the formula m If the interquartile range coefficient is used, then this parameter data... x i Consider it an outlier; B3. Finally, the sliding window parameter data, after being tested using the interquartile range method, are further tested using the standard score method: For each parameter data point in the sliding window, the standard score is calculated according to the following formula. z : （4） In the formula, u The mean of the parameter data. σ The standard deviation of the parameter data; Calculate the standard score z The value is compared with the set threshold, and parameter data that exceeds the threshold is regarded as outliers.
4. The method for dynamic correction of barometric altitude based on airborne data according to claim 1, characterized in that, In step two, the detected outliers are corrected using linear extrapolation, i.e., the correction value is calculated: （5） In the formula, x* t The value is the adjustment value of the sliding window parameters at time t; x t-1 For t - Sliding window parameter data at time 1 x t-2 For t - Sliding window parameter data at time 2.
5. The method for dynamic correction of barometric altitude based on airborne data according to claim 1, characterized in that, In step three, based on the constraints of aircraft performance and approach procedures, the candidate range for the sliding window size is constructed using the following method: First, determine the reference length for the sliding window size based on aircraft performance and approach procedures. L : (6) In the formula, k This is the scaling factor. D IF The distance from the intermediate approach positioning point to the runway nose during the approach procedure, in nautical miles; V app Approach speed of the aircraft type, in knots, 3600 is the conversion constant from hours to seconds; Let the candidate range for the sliding window size be [ W min , W max The lower limit of the candidate range for the sliding window size. W min Select using the following formula: (7) Upper limit of the candidate range for sliding window size W max Select using the following formula: (8) In the formula, w 1 、w 2 is an empirical constant for the sliding window configuration.