Airborne navigation data cleaning method based on multi-source sensing data

Abstract

The invention relates to the technical field of electrical digital data processing, in particular to an airborne navigation data cleaning method based on multi-source sensing data, which comprises the following steps: acquiring parameter data of airborne navigation at a plurality of historical time nodes, constructing a parameter sequence, and constructing a multi-scale window by taking any parameter data in the parameter sequence as a center; and calculating an abnormal score of any parameter data under any scale window by using an improved local outlier factor algorithm, judging the parameter data of which the mean value of the abnormal scores of any parameter data under all scale windows is greater than a set threshold value as abnormal data, and cleaning the abnormal data. The problem of misjudgment or missed judgment in the data cleaning process is solved.

Description

Technical Field

[0001] The present invention relates to the technical field of electrical digital data processing. More specifically, the present invention relates to an airborne navigation data cleaning method based on multi-source sensing data. Background Art

[0002] In the fields of modern aerospace, unmanned aerial vehicle systems, and intelligent aircraft navigation control, the airborne navigation system has become a core component for ensuring flight safety and performing high-precision tasks. With the continuous improvement of the intelligence level of flight platforms, airborne navigation systems usually integrate a variety of sensors, including but not limited to the Global Positioning System (GPS), barometric altimeter, Inertial Measurement Unit (IMU), magnetometer, etc. These sensors collect multi-dimensional navigation parameters such as longitude, latitude, altitude, three-axis acceleration, and three-axis angular velocity at high frequencies and continuously output a large amount of time-series data during flight. Such multi-source heterogeneous, high-dimensional, and high-frequency navigation data provides rich information for flight control and path planning, but at the same time, there are problems such as frequent noise interference, frequent local mutation anomalies, and complex spatio-temporal characteristics, which are extremely likely to have a negative impact on downstream data fusion, state estimation, and control strategies, and may even cause safety hazards such as flight yaw and attitude estimation distortion. Therefore, how to efficiently and accurately clean and identify abnormal airborne navigation data during flight has become an important issue that urgently needs to be solved in navigation data preprocessing technology.

[0003] Currently, academia and the engineering community have proposed a variety of anomaly detection methods based on statistical analysis, time-series modeling, clustering analysis, and machine learning for navigation data cleaning. Among them, the Local Outlier Factor (LOF) algorithm has become an important tool for detecting anomalies in navigation data due to its excellent performance in processing high-dimensional data and local density differences. The LOF algorithm identifies data with a significantly lower local density than its neighbor samples by comparing the local density relationship between a sample and its neighbor samples, thereby determining it as an outlier.

[0004] However, when the existing LOF algorithm is applied to navigation data processing, there are several key technical bottlenecks. Especially in terms of parameter selection, its performance highly depends on the setting of the neighborhood size value. A too small value will cause the algorithm to overly focus on local minor perturbations, easily misjudge noise or normal local fluctuations as anomalies, and reduce the detection accuracy; while a too large value will smooth out the true local changes, resulting in a slow response to sudden anomalies and even missed detection. In a dynamic flight environment, the fluctuation characteristics of navigation parameters vary greatly in different time periods. A fixed value cannot balance the recognition effects in different scenarios, thereby leading to problems of misjudgment or missed judgment during the data cleaning process. Summary of the Invention

[0005] To solve the problem of misjudgment or missed judgment in the data cleaning process proposed in the above background technology, the present invention provides the following solutions.

[0006] The present invention provides an airborne navigation data cleaning method based on multi-source sensing data, including: collecting parameter data of airborne navigation at multiple historical time nodes and constructing it into a parameter sequence, and constructing a multi-scale window centered on any parameter data in the parameter sequence; using an improved local outlier factor algorithm to calculate the anomaly score of any parameter data under any scale window, and determining the parameter data whose average value of the anomaly scores of any parameter data under all scale windows is greater than a set threshold as abnormal data, and cleaning the abnormal data; wherein, the improved local outlier factor algorithm includes value, the value is positively correlated with the initial value and negatively correlated with the degree of fluctuation within the scale window where any parameter data is located; the degree of fluctuation is the difference between the anomaly degree of the parameter data under the scale window and the anomaly degree of the parameter sequence.

[0007] The above technical solution effectively improves the ability to identify local anomalies by introducing a multi-scale window mechanism, which fully considers the change characteristics of navigation parameter data at different time scales. By constructing a multi-scale window and calculating the anomaly score at each scale, it can comprehensively capture the outlier behavior of data at different time granularities and avoid the problems of missed detection or misjudgment that may occur under a single scale.

[0008] Furthermore, the value is: , is the initial value, is the degree of fluctuation within the scale window where the parameter data is located, is the floor function.

[0009] The above technical solution realizes the adaptive adjustment of the neighborhood scale by dynamically combining the value of the clustering parameter with the local fluctuation degree, thereby significantly improving the adaptability and robustness of the anomaly detection algorithm under different data states. Specifically, when the data is in a stable state with small fluctuations, the algorithm tends to use a larger neighborhood range to enhance the ability to identify subtle outliers; while in an unstable state with large fluctuations, the neighborhood range will automatically shrink, thus reducing misjudgments caused by local drastic changes.

[0010] Furthermore, the anomaly degree under the scale window where the parameter data is located is: , is the th data under the scale window, is the mean of all data under the said scale window, is the number of data under the said scale window, is the mean of all data in the difference sequence corresponding to the said parameter sequence.

[0011] The above technical solution effectively enhances the sensitivity and discriminability to local abnormal fluctuations by introducing an abnormal degree calculation method that combines the data deviation degree and the overall change level within the sliding window. Specifically, by measuring the deviation degree between each data in the window and the mean value, and combining the overall change level after the difference of the entire parameter sequence as an adjustment factor, the evaluation of the abnormal degree not only reflects the local data distribution characteristics but also takes into account the global change trend. This design that integrates local deviation and global dynamics can more accurately identify the atypical fluctuations occurring at a specific moment. Especially when dealing with the minute mutations caused by external interference or sensing errors during flight, it can significantly improve the sensitivity and stability of anomaly detection, providing a more reliable basis for subsequent data cleaning and state recognition.

[0012] Furthermore, it also includes calculating the distance between any two parameter data of the local outlier factor algorithm , , is the th parameter data and the th parameter data's Euclidean distance, is the th parameter data and the th parameter data's acquisition time interval serial number.

[0013] The above technical solution constructs a more spatiotemporally correlated distance metric method by introducing both spatial differences and time interval serial numbers when calculating the distance between any two parameter data, thereby enhancing the anomaly detection's perception ability of time series features; traditional methods often ignore the possible differences in behavior patterns caused by acquisition time differences, while the above technical solution incorporates the time factor into the distance calculation, which can not only more accurately distinguish short-term mutations from long-term trend changes but also effectively suppress the misjudgment problem caused by too large a time span.

[0014] Furthermore, the said parameter data are the airborne longitude, latitude, altitude, three-axis acceleration or three-axis angular velocity.

[0015] Furthermore, the said difference sequence is obtained by performing a first-order difference on the parameter sequence.

[0016] Furthermore, the airborne altitude is collected by a barometric altimeter, the airborne longitude and latitude data are collected by a GPS, and the airborne three-axis acceleration and three-axis angular velocity are collected by an IMU.

[0017] Further, the construction of the multi-scale window includes: centering on any parameter data in the parameter sequence, respectively selecting different window lengths, and symmetrically expanding corresponding numbers of adjacent data points on both sides of the any parameter data to form multiple sliding windows centered on the any parameter data and having different scales, thereby obtaining the constructed multi-scale window.

[0018] The above technical solution forms a multi-scale observation perspective by constructing multiple symmetric sliding windows with different lengths around any data point in the parameter sequence, effectively enhancing the ability to jointly perceive local features and overall trends of the data; through sliding windows of different scales, sudden changes in the short term and slow changes in the medium and long term can be captured simultaneously, making the detection of abnormal points more comprehensive and accurate, and reducing the deviation risk caused by a fixed window scale.

[0019] Further, the cleaning of abnormal data includes: when the parameter data is determined to be abnormal data, the three-point smoothing method is used to correct the parameter data.

[0020] Further, the lengths of the multi-scale windows include 3, 5, and 7.

[0021] The beneficial effects of the present invention are as follows: The present invention constructs a parameter sequence by fusing multi-source sensing data, and combines a multi-scale sliding window and an improved local outlier factor algorithm, which can comprehensively evaluate the abnormal characteristics of data from multiple time scales, improving the robustness and accuracy of abnormal detection. Introducing a mechanism for adaptively adjusting the neighborhood range according to the degree of fluctuation enables flexible identification of abnormal data under different noise levels or dynamic change environments, enhancing the adaptability of the algorithm to sudden perturbations and complex flight states. At the same time, by introducing a joint distance calculation method for time and space, the description effect of the evolution law of time series data is improved, avoiding misjudgment caused by simple numerical anomalies. For the identified abnormal data, the three-point smoothing method is used for correction, effectively removing the anomalies while retaining the trend characteristics of the original data, significantly improving the continuity and stability of the data, and providing a more reliable input basis for subsequent navigation calculations and state evaluations. Overall, the present invention has high adaptability, high precision, and good engineering feasibility, and can meet the strict requirements of the airborne navigation system for high-quality data input in a changing environment. Description of the Drawings

[0022] Figure 1 It is a flowchart schematically showing an airborne navigation data cleaning method based on multi-source sensing data according to an embodiment of the present invention. Detailed Embodiments

[0023] An embodiment of an airborne navigation data cleaning method based on multi-source sensing data.

[0024] As shown Figure 1 in the figure, the flowchart of the airborne navigation data cleaning method based on multi-source sensing data according to the embodiment of the present invention includes the following steps: S1: Collect the parameter data of the airborne navigation at multiple historical time nodes and construct them into a parameter sequence, and construct a multi-scale window centered on any parameter data in the parameter sequence.

[0025] In one embodiment, in order to achieve refined analysis and multi-dimensional dynamic perception of the aircraft state, the collected parameter data includes but is not limited to multi-dimensional time series information such as the longitude, latitude, altitude, three-axis acceleration, and three-axis angular velocity of the aircraft. These parameter data are continuously sampled based on multiple historical time nodes, so as to form a multi-dimensional parameter sequence with time correlation and spatial continuity; Specifically, the airborne altitude data is collected in real time by a high-precision barometric altimeter, which can effectively capture the minute altitude changes of the aircraft in the vertical direction and has high sensitivity and resolution; the longitude and latitude data are obtained by GPS to ensure high-precision positioning of the aircraft position within the range, especially in dynamic flight trajectory reconstruction and track matching; the three-axis acceleration and three-axis angular velocity data are collected by an inertial measurement unit (IMU), and then reflect the linear motion state and rotational attitude changes of the aircraft in the three spatial axes in real time; The construction of the multi-scale window includes: centering on any parameter data in the parameter sequence, respectively selecting different window lengths, and symmetrically expanding the corresponding number of adjacent data points on the left and right of the any parameter data to form multiple sliding windows centered on the any parameter data and having different scales, so as to obtain the constructed multi-scale window. The lengths of the multi-scale windows include 3, 5, and 7. Through the multi-scale window construction mechanism, feature information can be extracted in parallel on multiple time scales, and the double sensitivity to short-term local changes and long-term overall trends can be improved.

[0026] S2: Use the improved local outlier factor algorithm to calculate the anomaly score of the any parameter data under any scale window.

[0027] In one embodiment, the improved local outlier factor algorithm includes a value, and the value is: where is the initial value, is the floor function.

[0028] By introducing a neighbor number adjustment mechanism that is adaptively associated with the degree of local fluctuation, the neighborhood range is dynamically adjusted in the improved local outlier factor algorithm, thereby enhancing the flexibility and accuracy of outlier recognition in different fluctuating environments. When the local fluctuation is large, the algorithm automatically reduces the neighborhood range to enhance the sensitivity to local mutations; while in the region with small fluctuations, a larger neighborhood is maintained to enhance the stability and robustness of judgment. The above mechanism effectively balances local sensitivity and global consistency, improves the adaptability of outlier detection, and is particularly suitable for complex time-series data scenarios with significant dynamic changes.

[0029] The degree of fluctuation is the difference between the degree of abnormality of the parameter data in the scale window where it is located and the degree of abnormality of the parameter sequence; by calculating the difference between the degree of abnormality of the parameter data in the local scale window and the overall abnormality degree of the entire parameter sequence, the relative fluctuation of the current data point in the local environment is effectively measured.

[0030] The degree of abnormality of the parameter data in the scale window where it is located is: , is the th data in the scale window, is the mean value of all data in the scale window, is the number of data in the scale window, is the mean value of all data in the difference sequence corresponding to the parameter sequence. The difference sequence is obtained by performing a first-order difference on the parameter sequence.

[0031] By evaluating the deviation degree of the parameter data from the mean value within the scale window at a specific scale window and combining the overall average change level of the parameter sequence difference, a measure of abnormality degree that takes into account both local volatility and global dynamic changes is constructed. It can not only accurately identify the degree of abnormality of local data in its neighborhood, but also effectively introduce the adjustment effect of the overall change trend of the parameter sequence, thereby avoiding misjudgment caused by short-term noise or minor perturbations, and improving the robustness and sensitivity of outlier detection. In addition, the above measure has good generality and scalability, can be applied to different types of multi-dimensional time-series data, and helps to construct a more stable and efficient flight state monitoring and early warning.

[0032] Furthermore, it also includes calculating the distance between any two parameter data in the local outlier factor algorithm , , is the th parameter data and the th parameter data, which is the Euclidean distance, is the th parameter data and the The serial number of the time interval for collecting parameter data.

[0033] When calculating the distance between any two parameter data in the above technical solution, the spatial difference of the data itself and the interval of the collection time are comprehensively considered, and a metric mechanism of spatio-temporal coupling is effectively introduced. Compared with the traditional method based only on the difference of data values, it can more accurately reflect the evolution law and correlation characteristics of parameter data in time series, and avoid misjudgment caused by ignoring time factors.

[0034] S3: Determine the parameter data whose mean value of the anomaly scores of any parameter data under all scale windows is greater than the set threshold as anomaly data, and clean the anomaly data.

[0035] In one embodiment, in order to improve the comprehensive determination ability of multi-scale abnormal behaviors and further improve the overall quality stability of parameter data and the reliability of subsequent processing, a discrimination and cleaning mechanism based on multi-scale anomaly scores is proposed. The above mechanism first statistically analyzes the anomaly scores of any parameter data under all preset scale windows, and calculates the mean value of its anomaly scores at each scale; if the mean value exceeds the anomaly determination threshold set by the system, it can be preliminarily determined that the parameter data shows significant deviation behaviors at multiple time scales and has a high anomaly credibility. By this method, misjudgment caused by local fluctuations or occasional anomalies at a single scale can be avoided, and it is ensured that the identified anomaly data points have stronger global consistency and time series reliability.

[0036] Based on the above determination results, in order to further eliminate the influence of anomaly data on the overall sequence smoothness and subsequent modeling effect, an anomaly data cleaning mechanism is introduced. During the cleaning process, for the data points determined to be abnormal, the three-point smoothing method is used for numerical correction, that is, a new value that more conforms to the local trend is generated by weighted averaging the values of its adjacent front and back two data points with itself. The three-point smoothing method is a classic and efficient sliding mean correction strategy, and its main advantage is that it can not only effectively suppress short-term noise and sudden anomalies in the data, but also largely retain the local trend characteristics of the original data, avoiding information loss caused by over-smoothing.

[0037] The introduction of the above cleaning strategy significantly enhances the robustness of the data processing chain. On the one hand, while ensuring the reliability of the anomaly detection results, it further improves the continuity and stability of the data; on the other hand, by retaining the original structural characteristics of the data, it provides higher-quality input data for subsequent state analysis, prediction modeling and behavior recognition, and improves the accuracy, robustness and generalization ability of the overall system.

[0038] The solution of the present invention constructs a parameter sequence by fusing multi-source sensing data, and introduces an adaptive anomaly detection and data cleaning mechanism under a multi-scale sliding window, realizing the accurate identification and effective repair of multi-source heterogeneous, fluctuating and locally mutated situations in airborne navigation data. By using an improved local outlier factor algorithm, the neighborhood range is dynamically adjusted according to the local fluctuation degree, improving the sensitivity and discrimination accuracy of the algorithm for abnormal points in complex flight states; at the same time, anomaly judgment is combined with the mean anomaly score under a multi-scale window, enhancing the robustness to sudden anomalies and periodic disturbances; further, through a comprehensive measurement method that fuses time interval and data distance, a similarity evaluation that is more in line with the time series characteristics of flight data is realized; the abnormal points are cleaned by the three-point smoothing method, taking into account both local trend preservation and mutation value suppression, effectively improving the continuity, reliability and engineering usability of the overall data. The above solution is particularly applicable to data preprocessing and anomaly detection scenarios in high-dynamic environments, providing a high-quality and stable data foundation for subsequent navigation, state estimation and flight control modeling.

[0039] In the description of this specification, "a plurality of", "several" means at least two, such as two, three or more, etc., unless otherwise specifically defined.

[0040] Although this specification has shown and described multiple embodiments of the present invention, it is obvious to those skilled in the art that such embodiments are provided only by way of example. Those skilled in the art will think of many changes, alterations and alternative ways without departing from the spirit and idea of the present invention. It should be understood that various alternative solutions to the embodiments of the present invention described herein can be adopted in the process of practicing the present invention.

Claims

1. An airborne navigation data cleaning method based on multi-source sensing data, characterized in that Including: Collecting the parameter data of the airborne navigation at multiple historical time nodes and constructing them into a parameter sequence, and constructing multi-scale windows centered on any parameter data in the parameter sequence; Calculating the anomaly score of any parameter data under any scale window by using an improved local outlier factor algorithm, determining the parameter data whose average value of the anomaly scores of any parameter data under all scale windows is greater than a set threshold as anomaly data, and cleaning the anomaly data; Among them, the improved local outlier factor algorithm includes value, and the value is positively correlated with the initial value and negatively correlated with the degree of fluctuation within the scale window where any of the parameter data is located; The degree of fluctuation is the difference between the degree of anomaly in the scale window where the parameter data is located and the degree of anomaly of the parameter sequence.

2. The airborne navigation data cleaning method based on multi-source sensing data according to claim 1, wherein The said value is: , is the initial value, is the degree of fluctuation within the scale window where the parameter data is located, is the floor function.

3. The airborne navigation data cleaning method based on multi-source sensing data according to claim 1, wherein The degree of abnormality under the scale window where the parameter data is located is: , is the th data under the scale window, is the mean of all data under the scale window, is the number of data under the scale window, is the mean of all data in the difference sequence corresponding to the parameter sequence.

4. The airborne navigation data cleaning method based on multi-source sensing data according to claim 1, characterized in that It further includes calculating the distance between any two parameter data in the local outlier factor algorithm , , being the Euclidean distance between the th parameter data and the th parameter data, and being the serial number of the time interval between the data collection of the th parameter data and the 5. The airborne navigation data cleaning method based on multi-source sensing data according to claim 1, characterized in that, The parameter data is the longitude, latitude, altitude, three-axis acceleration or three-axis angular velocity of the aircraft.

6. The airborne navigation data cleaning method based on multi-source sensing data according to claim 3, characterized in that The difference sequence is obtained by performing a first-order difference on the parameter sequence.

7. The airborne navigation data cleaning method based on multi-source sensing data according to claim 5, wherein Collecting the altitude of the aircraft by using a barometric altimeter, collecting the longitude and latitude of the aircraft by using a GPS, and collecting the three-axis acceleration and three-axis angular velocity of the aircraft by using an IMU.

8. The airborne navigation data cleaning method based on multi-source sensing data according to claim 1, wherein, The constructing of the multi-scale windows includes: centering on any parameter data in the parameter sequence, respectively selecting different window lengths, and symmetrically expanding the corresponding number of adjacent data points on the left and right of any parameter data to form multiple sliding windows centered on any parameter data and having different scales, so as to obtain the constructed multi-scale windows.

9. The airborne navigation data cleaning method based on multi-source sensing data according to claim 1, characterized in that, The cleaning of the anomaly data includes: when the parameter data is determined to be anomaly data, correcting the parameter data by using a three-point smoothing method.

10. The airborne navigation data cleaning method based on multi-source sensing data according to claim 1, wherein, The lengths of the multi-scale windows include 3, 5, and 7.

Images