Steady-state flight data adaptive extraction method based on KNN-DBSCAN
The adaptive adjustment of clustered hyperparameters through the KNN-DBSCAN algorithm solves the problem of poor steady-state feature extraction effect in aircraft QAR data, improves the flexibility and accuracy of data processing, and reduces the impact of data fluctuations on system performance.
Patent Information
- Application Number
- CN202411304851.5
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-09-19
- Publication Date
- 2024-12-13
AI Technical Summary
The prior art has problems in aircraft QAR data processing with poor steady-state feature extraction effect, high requirements for hyperparameter setting of clustering algorithms, and poor robustness, resulting in data fluctuations affecting system performance evaluation.
Adaptive extraction method of steady-state flight data based on KNN-DBSCAN is adopted, and the hyperparameters of the clustering algorithm are adaptively adjusted through standardized processing, sliding average filtering, clustering feature calculation, feature coordinate system construction, neighborhood radius and density estimation, and the clustering algorithm is adaptively adjusted. Realize steady-state data extraction.
It improves the flexibility and adaptability of data processing, reduces the impact of data fluctuations on system performance, improves clustering effect and efficiency, and reduces the need for manual modification of hyperparameters.
Smart Images
Figure CN119128557A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to a method for adaptively extracting steady-state flight data, and more particularly to a method for adaptively extracting steady-state flight data based on KNN-DBSCAN. Background Art
[0002] As the most comprehensive and accurate flight data recorded during the operation of an aircraft, the Quick Access Recorder (QAR) data has great research value. It records various position parameters, motion parameters, operating parameters, control parameters, and alarm information of the aircraft during the entire flight process. It is widely used in flight quality monitoring, flight process visualization simulation reproduction, maintenance, accident investigation, etc., and has a great role in promoting the formation of active and predictive safety management models for data supervision departments or aircraft operating companies. However, in the process of using QAR data, the original QAR data has problems such as missing, zero jumping, heterogeneity, and uneven sampling, which need to be solved through data preprocessing steps.
[0003] For a civil aircraft to complete a complete flight segment, many onboard systems need to work together. Each onboard system has a complex internal structure, with multiple subsystems interconnected and data coupled. Due to the transition of flight phases, changes in environmental conditions, and the influence of the actuation of cross-linked components, the recorded QAR data will fluctuate and become unstable. In the prior art, a threshold method or a clustering method is usually used to select data within a fixed numerical range or with similar characteristics, and an expert experience method is used to delete QAR data within fixed conditions or time periods to reduce the impact of data fluctuations. Due to the differences in aircraft flight routes, flight environments, flight areas, and aircraft configurations, traditional threshold methods, clustering methods, and expert experience methods have problems such as loss of useful data, high requirements for hyperparameter settings, and poor robustness. In order to stably reflect the actual state of the onboard system under normal operating conditions, an adaptive extraction method for steady-state flight data is urgently needed. Summary of the invention
[0004] The technical problem to be solved by the present invention is to provide a KNN-DBSCAN-based steady-state flight data adaptive extraction method to overcome the shortcomings of the prior art, which solves the problems of poor steady-state feature extraction effect, high requirements for hyperparameter setting of clustering algorithms, and poor robustness.
[0005] The technical solution adopted by the present invention is: a method for adaptively extracting steady-state flight data based on KNN-DBSCAN, including being able to adaptively adjust the hyperparameters of the clustering algorithm and extract data under the stable working state of the corresponding components according to the actual conditions of the aircraft's flight route, flight environment, flight area and aircraft configuration, and utilizing the characteristics of the flight data to be tested. The method for adaptively extracting steady-state flight data includes the following steps:
[0006] 1) Standardize the original flight data;
[0007] 2) Perform sliding average filtering on the flight data that needs to extract steady state;
[0008] 3) Calculate the clustering features of the flight sequence required for steady-state extraction, including gradient values and mean offsets, and perform data normalization;
[0009] 4) Using clustering features to construct clustering feature coordinate system and distance matrix;
[0010] 5) Use the k nearest neighbor algorithm to obtain the neighborhood radius ρ of each feature point in the cluster feature coordinate system k , the neighborhood radius ρ k Contains k feature points;
[0011] 6) Calculate the standard deviation of the neighborhood radius;
[0012] 7) Calculate the density estimation function of the neighborhood radius using the Gaussian kernel density estimation method;
[0013] 8) Calculate the maximum search neighborhood radius of clustering hyperparameters;
[0014] 9) Count the number of data values in each column of the distance matrix that are less than the maximum search neighborhood radius of the clustering hyperparameter, and calculate the average number of data values in all columns of the distance matrix that are less than the maximum search neighborhood radius Eps of the clustering hyperparameter, and obtain the minimum number of cluster points of the clustering hyperparameter;
[0015] 10) The maximum search neighborhood radius and the minimum number of cluster points of the clustering hyperparameters are used as hyperparameters of the density-based clustering algorithm to cluster the flight sequence m required for steady-state extraction and realize adaptive extraction of steady-state flight data.
[0016] The KNN-DBSCAN-based adaptive extraction method for steady-state flight data of the present invention can realize adaptive clustering hyperparameter adjustment for QAR data generated by different flights, thereby effectively extracting the stable working state data of relevant components of the airborne system, reducing the impact of data fluctuations on system performance evaluation, and having high robustness. The beneficial effects of the present invention are:
[0017] 1. The method of the present invention can adaptively extract the steady-state characteristics of flight data according to the actual conditions of the aircraft's flight route, flight environment, flight area, and aircraft configuration, utilizing the characteristics of the flight data to be measured, thereby improving the flexibility and adaptability of data processing;
[0018] 2. The method of the present invention adaptively calculates the hyperparameters of the DBSCAN clustering algorithm, including the maximum search neighborhood radius and the minimum number of cluster points, according to the different mathematical distributions of the required clustering feature data. A better clustering effect can be achieved without manual modification of the hyperparameters, thereby improving the efficiency and accuracy of data processing. BRIEF DESCRIPTION OF THE DRAWINGS
[0019] Figure 1 It is a flow chart of the method for adaptively extracting steady-state flight data based on KNN-DBSCAN of the present invention;
[0020] Figure 2 It is a clustering feature coordinate system diagram of the present invention;
[0021] Figure 3 is a k-neighborhood radius distribution diagram of the present invention;
[0022] Figure 4 is a distribution diagram of the derivative of the kernel density estimation function of the present invention;
[0023] Figure 5 It is the adaptive clustering result diagram of the present invention;
[0024] Figure 6 This is a diagram showing the effect of the adaptive extraction of steady-state features of the present invention. DETAILED DESCRIPTION
[0025] The following is a detailed description of the method for adaptively extracting steady-state flight data based on KNN-DBSCAN of the present invention in conjunction with the embodiments and drawings.
[0026] like Figure 1 As shown, the method for adaptively extracting steady-state flight data based on KNN-DBSCAN of the present invention includes being able to adaptively adjust the hyperparameters of the clustering algorithm and extract data under the stable working state of the corresponding components according to the actual conditions of the flight route, flight environment, flight area and aircraft configuration of the aircraft and the characteristics of the flight data to be measured. The method for adaptively extracting steady-state flight data includes the following steps:
[0027] 1) Standardize the original flight data;
[0028] The original flight data refers to Quick Access Recorder (QAR) data, and the standardized processing includes:
[0029] (1) Convert the string format data in the quick access recorder data into digital format;
[0030] (2) Perform linear interpolation on data with missing values.
[0031] 2) Perform sliding average filtering on the flight data that needs to extract steady state. The specific calculation process is as follows:
[0032]
[0033] Where o is the sliding time window created, o = (1,1,···,1) 1×L , o is convolved with the flight data x from which steady-state extraction is required, and the convolution operation result is divided by the sliding time window size L to obtain the flight sequence m for steady-state extraction required after sliding average filtering.
[0034] 3) Calculate the clustering features of the flight sequence required for steady-state extraction, including gradient values and mean offsets, and perform data normalization;
[0035] The calculation requires the gradient value g of the i-th data sample in the flight sequence m extracted in a steady state i and the mean shift d of the i-th data sample i , the specific calculation process is as follows:
[0036] g i =m i -m i-1 ,i=2,3...n (2)
[0037] g1=0 (3)
[0038]
[0039] In the formula, m i represents the i-th data sample in flight sequence m for the required steady-state extraction, represents the mean of the flight sequence m required for steady-state extraction, and n is the total number of data samples in the flight sequence m required for steady-state extraction.
[0040] The data normalization process is to normalize the gradient value g of the data sample i , the mean shift d of the data sample i Perform Z-score data normalization, the specific calculation is as follows:
[0041]
[0042] In the formula, is the gradient value of the ith data sample after data normalization, is the mean shift of the ith data sample after data normalization, The average of the gradient values of all data samples in the flight sequence m extracted for the desired steady state, is the average value of the mean shift of all data samples in the flight sequence m required for steady-state extraction, and n is the total number of data samples in the flight sequence m required for steady-state extraction.
[0043] 4) Use clustering features to construct clustering feature coordinate system and distance matrix; where:
[0044] Figure 2 It is a cluster feature coordinate system diagram of the present invention. As shown in the figure, the abscissa of the cluster feature coordinate system is the gradient value of the data sample in the flight sequence m required for steady-state extraction, and the ordinate is the mean offset of the data sample in the flight sequence m required for steady-state extraction;
[0045] The distance matrix DIST N×N It is expressed as:
[0046]
[0047] Among them, N is the total number of feature points in the cluster feature coordinate system, which is equal to the total number n of data samples in the flight sequence m required for steady-state extraction. The data value in the i′th row and j′th column of the distance matrix is the feature point of the i′th data sample in the flight sequence m required for steady-state extraction in the cluster feature coordinate system. and the feature point of the j′th data sample in the cluster feature coordinate system The Euclidean distance between .
[0048] 5) Use the k nearest neighbor algorithm to obtain the neighborhood radius ρ of each feature point in the cluster feature coordinate system k , the neighborhood radius ρ k It contains k feature points, and the value of k is 5-20.
[0049] Figure 3 It is a k-neighborhood radius distribution diagram of the present invention, wherein the horizontal axis is the size of the neighborhood radius, the vertical axis is the number of feature points, and the meaning of each point in the diagram is the number of feature points in the feature coordinate system that is less than the neighborhood radius;
[0050] 6) Calculate the standard deviation of the neighborhood radius;
[0051] 7) Calculate the density estimation function of the neighborhood radius using the Gaussian kernel density estimation method; including:
[0052] First, the window width h of the Gaussian kernel density function is calculated using the Silverman bandwidth estimation method and the standard deviation of the neighborhood radius:
[0053]
[0054] Where r is the neighborhood radius ρ k The total number of , σ is the standard deviation of the neighborhood radius;
[0055] Then use Gaussian kernel density estimation to get the density estimation function of the neighborhood radius It is expressed as:
[0056]
[0057] In the formula, is the neighborhood radius of the t-th feature point, h is the window width of the Gaussian kernel density function, and K(u) is the Gaussian kernel function with u as the independent variable, defined as:
[0058]
[0059] 8) Calculate the maximum search neighborhood radius of clustering hyperparameters;
[0060] Includes: Density estimation function for neighborhood radius Take the derivative and calculate the neighborhood radius value corresponding to the minimum value of the derivative, which is the maximum search neighborhood radius Eps of the clustering hyperparameter.
[0061] The specific calculation of Eps is as follows:
[0062]
[0063] In the formula, it is assumed that the radius of the t1th neighborhood is The derivative of the density estimation function is the minimum value of the derivative function, expressed as:
[0064]
[0065] Figure 4 It is a distribution diagram of the derivative of the kernel density estimation function of the present invention, wherein the horizontal axis is the size of the neighborhood radius, and the vertical axis is the derivative value of the density estimation function. The dotted line in the figure represents the minimum value of the derivative of the density estimation function, and its corresponding neighborhood radius value is the part of the density estimation function with the most drastic value change, that is, the maximum search neighborhood radius of the clustering hyperparameter.
[0066] 9) Count the number of data values in each column of the distance matrix that are less than the maximum search neighborhood radius of the clustering hyperparameter, and calculate the average number of data values in all columns of the distance matrix that are less than the maximum search neighborhood radius of the clustering hyperparameter Eps, and obtain the minimum number of cluster points MinPts of the clustering hyperparameter, expressed as:
[0067]
[0068] In the formula, p i′Indicates the number of data values in the i′th column of the distance matrix that are less than the clustering hyperparameter maximum search neighborhood radius Eps.
[0069] 10) The maximum search neighborhood radius and the minimum number of cluster points of the clustering hyperparameters are used as hyperparameters of the density-based clustering algorithm (DBSCAN algorithm) to cluster the flight sequence m required for steady-state extraction and realize adaptive extraction of steady-state flight data.
[0070] Figure 5 This is the adaptive clustering result diagram of the present invention, in which the horizontal axis is the gradient value and the vertical axis is the mean offset. The "x" symbol represents the feature point with steady-state characteristics extracted by the method of the present invention, and the "." symbol represents the feature point with non-steady-state characteristics extracted by the method of the present invention.
[0071] Figure 6 It is a diagram showing the effect of the adaptive extraction of steady-state features of the present invention. In the diagram, the symbol "x" represents the steady-state working sample point selected by the method of the present invention, and the symbol "." represents the non-steady-state working sample point identified by the method of the present invention.
Claims
1. A method for adaptively extracting steady-state flight data based on KNN-DBSCAN, characterized by: The method includes being able to adaptively adjust the hyperparameters of the clustering algorithm and extract data of the corresponding components in a stable working state according to the actual conditions of the aircraft's flight route, flight environment, flight area and aircraft configuration, and utilizing the characteristics of the flight data to be measured. The method includes the following steps: 1) Standardize the original flight data; 2) Perform sliding average filtering on the flight data that needs to extract steady state; 3) Calculate the clustering features of the flight sequence required for steady-state extraction, including gradient values and mean offsets, and perform data normalization; 4) Using clustering features to construct clustering feature coordinate system and distance matrix; 5) Use the k nearest neighbor algorithm to obtain the neighborhood radius ρ of each feature point in the cluster feature coordinate system k , the neighborhood radius ρ k Contains k feature points; 6) Calculate the standard deviation of the neighborhood radius; 7) Calculate the density estimation function of the neighborhood radius using the Gaussian kernel density estimation method; 8) Calculate the maximum search neighborhood radius of clustering hyperparameters; 9) Count the number of data values in each column of the distance matrix that are less than the maximum search neighborhood radius of the clustering hyperparameter, and calculate the average number of data values in all columns of the distance matrix that are less than the maximum search neighborhood radius Eps of the clustering hyperparameter, and obtain the minimum number of cluster points of the clustering hyperparameter; 10) The maximum search neighborhood radius and the minimum number of cluster points of the clustering hyperparameters are used as hyperparameters of the density-based clustering algorithm to cluster the flight sequence m required for steady-state extraction and realize adaptive extraction of steady-state flight data.
2. The method for adaptively extracting steady-state flight data based on KNN-DBSCAN according to claim 1, characterized in that: Step 1) The raw flight data refers to the quick access recorder data, and the standardization processing includes: (1) Convert the string format data in the quick access recorder data into digital format; (2) Perform linear interpolation on data with missing values.
3. The method for adaptively extracting steady-state flight data based on KNN-DBSCAN according to claim 1, characterized in that: In step 2), the flight data for which steady-state extraction is required is subjected to sliding average filtering, and the specific calculation process is as follows: Where o is the sliding time window created, o = (1,1,···,1) 1×L , o is convolved with the flight data x from which steady-state extraction is required, and the convolution operation result is divided by the sliding time window size L to obtain the flight sequence m for steady-state extraction required after sliding average filtering.
4. The method for adaptively extracting steady-state flight data based on KNN-DBSCAN according to claim 1, characterized in that: The calculation described in step 3) requires the gradient value g of the i-th data sample in the flight sequence m extracted in the steady state i and the mean shift d of the i-th data sample i , the specific calculation process is as follows: g i =m i -m i-1 ,i=2,3...n(2) g1＝0(3) In the formula, m i represents the i-th data sample in flight sequence m for the required steady-state extraction, represents the mean of the flight sequence m required for steady-state extraction, and n is the total number of data samples in the flight sequence m required for steady-state extraction.
5. The method for adaptively extracting steady-state flight data based on KNN-DBSCAN according to claim 1, characterized in that: The data normalization process described in step 3) is to normalize the gradient value g of the data sample i , the mean shift d of the data sample i Perform Z-score data normalization, the specific calculation is as follows: In the formula, is the gradient value of the ith data sample after data normalization, is the mean shift of the ith data sample after data normalization, The average of the gradient values of all data samples in the flight sequence m extracted for the desired steady state, is the average value of the mean shift of all data samples in the flight sequence m required for steady-state extraction, and n is the total number of data samples in the flight sequence m required for steady-state extraction.
6. The method for adaptively extracting steady-state flight data based on KNN-DBSCAN according to claim 1, characterized in that: The abscissa of the cluster feature coordinate system in step 4) is the gradient value of the data sample in the flight sequence m extracted in the desired steady state, and the ordinate is the mean offset of the data sample in the flight sequence m extracted in the desired steady state; the distance matrix DIST N×N It is expressed as: Among them, N is the total number of feature points in the cluster feature coordinate system, which is equal to the total number n of data samples in the flight sequence m required for steady-state extraction. The data value in the i′th row and j′th column of the distance matrix is the feature point of the i′th data sample in the flight sequence m required for steady-state extraction in the cluster feature coordinate system. and the feature point of the j′th data sample in the cluster feature coordinate system The Euclidean distance between .
7. The method for adaptively extracting steady-state flight data based on KNN-DBSCAN according to claim 1, characterized in that: The value of k in step 5) is 5-20.
8. The method for adaptively extracting steady-state flight data based on KNN-DBSCAN according to claim 1, characterized in that: Step 7) includes: First, the window width h of the Gaussian kernel density function is calculated using the Silverman bandwidth estimation method and the standard deviation of the neighborhood radius: Where r is the neighborhood radius ρ k The total number of , σ is the standard deviation of the neighborhood radius; Then use Gaussian kernel density estimation to get the density estimation function of the neighborhood radius It is expressed as: In the formula, is the neighborhood radius of the t-th feature point, h is the window width of the Gaussian kernel density function, and K(u) is the Gaussian kernel function with u as the independent variable, defined as:
9. The method for adaptively extracting steady-state flight data based on KNN-DBSCAN according to claim 1, characterized in that: Step 8) includes: density estimation function for neighborhood radius Take the derivative and calculate the neighborhood radius value corresponding to the minimum value of the derivative, which is the maximum search neighborhood radius Eps of the clustering hyperparameter.