Train satellite positioning fault detection method and system based on data density driving
By employing a data density-driven fault detection method, DBSCAN clustering and local outlier factor algorithms are used to identify and eliminate faulty satellite observations, thus solving the problem of detection accuracy in complex environments for satellite navigation and achieving highly reliable fault detection.
Patent Information
- Application Number
- CN202610844751.4
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2026-06-11
- Publication Date
- 2026-08-25
AI Technical Summary
Existing satellite navigation fault detection methods have low accuracy in complex non-Gaussian environments, and are prone to missed detections and false detections, affecting the reliability and accuracy of train positioning results.
A data density-driven fault detection method is adopted. The DBSCAN clustering algorithm is used to identify and remove noise points, the local outlier factor algorithm and kernel density estimation are used to determine the fault detection threshold, and the local subset monitoring strategy is combined to identify and exclude faulty satellite observations.
Significantly improves the accuracy and reliability of fault detection in complex non-Gaussian environments, reduces the rate of missed detections and false detections, and ensures the safety and continuity of the train operation control system.
Smart Images
Figure CN122632281A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of train satellite positioning technology, and in particular to a train satellite positioning fault detection method and system based on data density. Background Technology
[0002] With the continuous advancement of intelligent railway construction, the Global Navigation Satellite System (GNSS), with its unique advantages such as all-weather, all-time, high coverage, and high-precision positioning, is gradually becoming an important technological support for the next generation of train operation control systems. Satellite navigation-based train operation control systems can utilize satellite positioning information to achieve key functions such as train position awareness, track occupancy detection, and train integrity checks, thereby significantly reducing reliance on trackside equipment, lowering the construction and operation maintenance costs of the railway system, and effectively improving the autonomy and intelligence level of the train operation control system. Therefore, promoting the deep integration of satellite navigation technology and train operation control systems is of significant strategic importance for realizing the intelligent upgrade of the railway transportation system.
[0003] However, in practical engineering applications, GNSS signals are susceptible to various adverse factors during propagation and reception. Specifically, clock drift, orbital parameter errors, ionospheric delay, multipath reflection, and interference from buildings or terrain can all lead to abnormal deviations in satellite observations, resulting in step faults or gradual slope faults. For train operation control systems with extremely high safety requirements, these faults can directly cause unacceptable errors in train positioning results, and even affect train operation safety. Therefore, effective fault detection and identification of satellite observation data is essential to ensure the safety and reliability of train operation.
[0004] To address the aforementioned issues, academia and engineering have conducted extensive research, resulting in various satellite navigation fault detection methods. These methods are primarily based on statistical hypothesis testing theory, identifying anomalous satellite observations by analyzing the consistency relationships between redundant observation data. Based on the different ways of utilizing epoch observation data, existing methods can be broadly categorized into two types: snapshot-based fault detection methods, mainly including pseudorange comparison, least squares residual, and deconvolution methods, etc., utilize only the observation information at the current moment for fault detection, exhibiting good detection capability for step faults but insufficient sensitivity to slowly changing slope faults; and sequential fault detection methods, which combine temporal correlation information between multiple epochs, typically constructing fault detection statistics based on state estimation methods such as Kalman filtering, demonstrating good detection performance for slowly changing slope faults. To further improve the accuracy of fault detection, existing research has introduced techniques such as signal weight adjustment, robust estimation, and sliding windows, optimizing the observation noise covariance matrix or filter gain to improve the reliability of filtering information and fault detection effectiveness.
[0005] However, in-depth analysis reveals that all existing methods are based on a common statistical assumption: that the observation residuals or filtered information follow a Gaussian distribution. The actual operating environment of trains is extremely complex and variable. Satellite observation data is easily affected by multipath effects, obstruction interference, and signal reflection, leading to significant non-Gaussian characteristics in observation errors. This non-Gaussian characteristic is particularly pronounced in typical railway operating scenarios such as urban canyons, mountain tunnels, or densely forested areas. Under these conditions, traditional fault detection methods based on fixed probability distribution models struggle to accurately describe the inherent statistical characteristics of actual observation data, easily resulting in missed faults, misclassifying real faults as normal, or false faults, misclassifying normal observations as faults, thus severely impacting the reliability and accuracy of train positioning results. Therefore, a new satellite positioning fault detection method that can break free from the constraints of the Gaussian distribution assumption and adapt to complex non-Gaussian observation environments is urgently needed. Summary of the Invention
[0006] In view of this, embodiments of the present invention provide a train satellite positioning fault detection method and system based on data density to solve the problems of existing technologies, such as reliance on the prior assumption that the observation data follows a Gaussian distribution, low fault detection accuracy in complex non-Gaussian environments, and easy occurrence of missed detections and false detections.
[0007] On one hand, the present invention provides a data density-driven train satellite positioning fault detection method, the method comprising: Raw observation data is collected using a satellite receiver. The carrier-to-noise ratio, satellite elevation angle, and pseudorange residual are extracted as satellite signal feature quantities that reflect the quality of satellite observation data. A raw observation feature dataset containing multiple samples is constructed, with each sample containing the three feature dimensions of the carrier-to-noise ratio, the satellite elevation angle, and the pseudorange residual. The original observation feature dataset is processed using a density-based clustering algorithm to identify and remove noise points with a sample density lower than a preset threshold in the feature space, thereby obtaining a normal observation dataset. The filtered innovation squared is extracted from the normal observation dataset as a sample feature quantity to construct a normal sample feature set; the local outlier factor of each sample in the normal sample feature set is calculated based on the local outlier factor algorithm to obtain the normal sample local outlier factor set; and the probability density distribution of the normal sample local outlier factor set is fitted by the kernel density estimation algorithm, and the fault detection threshold is determined according to the preset false alarm rate. The real-time observations at the current moment are used as the samples to be detected. The local outlier factor of the sample to be detected relative to the feature set of normal samples is calculated. If the local outlier factor of the sample to be detected is less than or equal to the fault detection threshold, the current observation data is determined to be normal. Otherwise, the current observation data is determined to contain a fault measurement. After determining that a faulty measurement exists, a local subset monitoring strategy is adopted to eliminate the observations corresponding to each satellite one by one to construct multiple observation subsets. The local outlier factor of each observation subset is calculated, and the satellite observations corresponding to the minimum local outlier factor are identified as faulty measurements and excluded.
[0008] In some embodiments, the density-based clustering algorithm is the DBSCAN algorithm; The specific method for identifying and removing noise points with a sample density below a preset threshold in the neighborhood is as follows: Set a neighborhood radius and a minimum number of neighborhood samples. Define sample points whose number of samples within the neighborhood radius is greater than or equal to the minimum number of neighborhood samples as core points. Define sample points whose number of samples within the neighborhood radius is less than the minimum number of neighborhood samples and do not belong to the neighborhood of core points as noise points and remove them.
[0009] In some embodiments, the local outlier factor of each sample in the normal sample feature set is calculated based on the local outlier factor algorithm, including: For a sample point in the normal sample feature set, calculate the Kth distance of the sample point, and determine the Kth distance neighborhood of the sample point based on the Kth distance; Calculate the reachable distance from the sample point in the Kth distance neighborhood to the sample point: ; in, Represents sample points to sample point The reachable distance; Represents sample points The Kth distance; Represents sample points With sample points The Euclidean distance between them; This indicates taking the maximum value; Calculate the local reachability density of the sample point based on the reachability distances from all sample points in the Kth distance neighborhood to the sample point: ; in, Represents sample points Locally achievable density; This represents the number of sample points contained in the Kth distance neighborhood; Represents the Kth distance neighborhood; Calculate the local outlier factor of the sample point based on its local reachability density and the local reachability density of each sample point in its K-th distance neighborhood: ; in, Represents sample points Local outlier factors.
[0010] In some embodiments, a kernel density estimation algorithm is used to fit the probability density distribution of the local outlier set of the normal samples, and a fault detection threshold is determined based on a preset false alarm rate, including: The Gaussian kernel function is used as the kernel function for kernel density estimation, and a smoothing parameter is set. The probability density function of the random variable representing local outliers in normal samples is calculated using the kernel density estimation method: ; in, Indicates the probability density at The value at; This represents the total number of normal samples; Indicates the smoothing parameter; Represents the Gaussian kernel function; This represents a random variable representing a local outlier in a normal sample. Represents the first in the set of local outliers in normal samples One sample; For a pre-set false alarm rate, calculate the fault detection threshold, which satisfies: ; in, Represents random variables Less than the fault detection threshold The probability of; This indicates the preset false alarm rate.
[0011] In some embodiments, calculating the local outlier factor of the sample to be detected relative to the feature set of normal samples includes: Calculate the Euclidean distance between the sample to be detected and each sample in the feature set of normal samples, and determine the nearest neighbor set of the sample to be detected based on the preset neighborhood parameters; Calculate the reachability distance from the sample points within the nearest neighbor set to the sample to be detected: ; in, Represents sample points to the sample to be tested The reachable distance; Represents sample points The Kth distance; Represents sample points With the sample to be tested The Euclidean distance between them; This indicates taking the maximum value; Calculate the local reachability density of the sample to be detected based on the reachability distances from all sample points in the nearest neighbor set to the sample to be detected: ; in, Indicates the sample to be tested Locally achievable density; This indicates the number of sample points contained in the nearest neighbor set; Represents the nearest neighbor set; Based on the local reachability density of the sample to be detected and the local reachability density of each sample point in its nearest neighbor set, calculate the local outlier factor of the sample to be detected: ; in, Indicates the sample to be tested Local outlier factors.
[0012] In some embodiments, a local subset monitoring strategy is adopted to construct multiple observation subsets by successively eliminating the observations corresponding to each satellite, calculating the local outlier factor for each observation subset, and identifying the satellite observations corresponding to the minimum local outlier factor as fault measurements, including: The observations corresponding to each satellite are removed from all observation data at the current moment to construct the corresponding observation subset, and the sample to be detected and its local outlier factor corresponding to the observation subset are calculated. Compare the magnitudes of the local outliers corresponding to each subset of observations, determine the subset of observations corresponding to the minimum local outlier, and identify the satellite observations removed from this subset as fault measurements: ; in, Indicates a satellite that has been identified as faulty; Indicates taking the first Local outlier factors corresponding to each subset of observations The value that yields the minimum value.
[0013] In some embodiments, the satellite elevation angle is calculated using the satellite position and the receiver position, as shown in the formula: ; in, Indicates the first The satellite elevation angle of each satellite; Represents the arcsine function; This represents the receiver's coordinates in the Z direction within the northeast-central coordinate system; Indicates the first The coordinates of the satellite in the Z direction of the northeast celestial coordinate system; Indicates the receiver and the first The spatial distance between satellites; The pseudorange residual is calculated by the difference between the original observed pseudorange and the calculated pseudorange, using the following formula: ; in, Represents pseudorange residuals; Indicates the original observation pseudorange; , , This represents the satellite's position coordinates on the three axes in a Cartesian coordinate system. , , This represents the three-axis position coordinates of the receiver in a spatial rectangular coordinate system, estimated by the positioning algorithm.
[0014] In some embodiments, the squared filtered information is obtained in the following manner: The observation data in the normal observation dataset is processed by Kalman filtering using the positioning algorithm of the satellite receiver to obtain the filtered innovation vector. The elements of the filtered innovation vector are squared and then summed to obtain the filtered innovation square. The filtered innovation square is used to characterize the degree of deviation between the observed data and the predicted data.
[0015] On the other hand, the present invention also provides a data density-driven train satellite positioning fault detection system, including a processor, a memory, and a computer program or instructions stored in the memory. The processor is used to execute the computer program or instructions, and when the computer program / instructions are executed, the system implements the steps of any of the methods mentioned above.
[0016] On the other hand, the present invention also provides a computer-readable storage medium having a computer program or instructions stored thereon, characterized in that the computer program or instructions, when executed by a processor, implement the steps of any of the methods mentioned above.
[0017] This invention introduces a data density-driven detection approach, overcoming the constraints of traditional satellite navigation fault detection methods that rely on the statistical assumption that observation residuals or filtered information follow a Gaussian distribution. This invention requires no prior assumptions about the data's distribution; it relies solely on the data's density distribution characteristics in the feature space for anomaly detection. Therefore, it can adaptively adapt to the complex and ever-changing railway operating environment, fundamentally avoiding performance degradation caused by distribution assumption mismatch.
[0018] Furthermore, by preprocessing the original observation data using the DBSCAN density clustering algorithm, non-line-of-sight signals with significant pseudorange errors can be effectively identified and removed in the feature space, thereby obtaining a high-quality normal observation dataset for offline modeling. This ensures that the training data used in the subsequent establishment of the local outlier factor model is relatively normal, avoiding the contamination of model parameters by abnormal data, and thus improving the reliability of the fault detection model.
[0019] Furthermore, an offline detection model is constructed using the local outlier factor algorithm, and the detection threshold is adaptively determined using the kernel density estimation method. Compared to traditional methods that rely on fixed thresholds or specific quantile tables, this invention can calculate the decision threshold based on the actual distribution of normal data, accurately controlling the false alarm rate even under non-Gaussian distribution conditions while maintaining a high fault detection rate. This mechanism enables this invention to have good detection sensitivity for both step faults and gradually changing ramp faults, overcoming the inherent deficiency of snapshot methods in being insensitive to ramp faults.
[0020] Furthermore, a local subset monitoring strategy is introduced during the online detection phase. By systematically eliminating satellite observations and comparing the local outliers of each subset, the faulty satellite can be accurately located. This strategy does not require pre-assumed fault modes or complex fault identification logic; it can quickly and accurately identify and eliminate fault sources simply through density comparison, thereby ensuring the continuity and availability of the train positioning system.
[0021] In summary, this invention can significantly improve the accuracy and reliability of satellite positioning fault detection in complex non-Gaussian observation environments, and reduce the missed detection rate and false detection rate, which has important engineering application value for ensuring the safe operation of train operation control systems.
[0022] Additional advantages, objects, and features of the invention will be set forth in part in the description which follows, and will also become apparent in part to those skilled in the art upon studying the description, or may be learned by practice of the invention. The objects and other advantages of the invention can be realized and obtained by means of the structures specifically pointed out in the description and drawings.
[0023] Those skilled in the art will understand that the objectives and advantages achievable with the present invention are not limited to those specifically described above, and that the above and other objectives achievable with the present invention will become clearer from the following detailed description. Attached Figure Description
[0024] The accompanying drawings, which are included to provide a further understanding of the invention and form part of this application, are not intended to limit the invention. In the drawings: Figure 1 This is a schematic diagram of the steps of a train satellite positioning fault detection method based on data density in one embodiment of the present invention.
[0025] Figure 2 This is a flowchart illustrating a data density-driven train satellite positioning fault detection method according to an embodiment of the present invention. Detailed Implementation
[0026] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the embodiments and accompanying drawings. Here, the illustrative embodiments and descriptions of this invention are used to explain the invention, but are not intended to limit the invention.
[0027] It should also be noted that, in order to avoid obscuring the invention with unnecessary details, only the structures and / or processing steps closely related to the solution according to the invention are shown in the accompanying drawings, while other details that are not closely related to the invention are omitted.
[0028] It should be emphasized that the term "including / comprises" as used herein refers to the presence of a feature, element, step, or component, but does not exclude the presence or addition of one or more other features, elements, steps, or components.
[0029] It should also be noted that, unless otherwise specified, the term "connection" in this article can refer not only to a direct connection, but also to an indirect connection involving an intermediary.
[0030] In the following description, embodiments of the invention will be illustrated with reference to the accompanying drawings. In the drawings, the same reference numerals represent the same or similar parts, or the same or similar steps.
[0031] It should be emphasized here that the step markers mentioned below are not a limitation on the order of the steps, but should be understood as meaning that the steps can be executed in the order mentioned in the embodiments, or in a different order than in the embodiments, or several steps can be executed simultaneously.
[0032] To address the problems of existing technologies, such as reliance on the prior assumption that observed data follows a Gaussian distribution, low accuracy in fault detection under complex non-Gaussian environments, and susceptibility to missed and false detections, this invention provides a data density-driven train satellite positioning fault detection method, such as... Figure 1 As shown, the method includes the following steps S101~S105: Step S101: Collect raw observation data using a satellite receiver, extract the carrier-to-noise ratio, satellite elevation angle, and pseudorange residual as satellite signal feature quantities that reflect the quality of satellite observation data, and construct a raw observation feature dataset containing multiple samples. Each sample contains three feature dimensions: carrier-to-noise ratio, satellite elevation angle, and pseudorange residual.
[0033] Step S102: The original observation feature dataset is processed using a density-based clustering algorithm. Noise points with sample density below a preset threshold in the feature space are identified and removed to obtain a normal observation dataset.
[0034] Step S103: Extract the squared filtered innovation from the normal observation dataset as a sample feature quantity to construct a normal sample feature set; calculate the local outlier factor of each sample in the normal sample feature set based on the local outlier factor algorithm to obtain the local outlier factor set of normal samples; and use the kernel density estimation algorithm to fit the probability density distribution of the local outlier factor set of normal samples, and determine the fault detection threshold according to the preset false alarm rate.
[0035] Step S104: Take the real-time observation at the current moment as the sample to be detected, and calculate the local outlier factor of the sample to be detected relative to the feature set of normal samples; if the local outlier factor of the sample to be detected is less than or equal to the fault detection threshold, then the current observation data is determined to be normal; otherwise, the current observation data is determined to contain fault measurement.
[0036] Step S105: After determining that there is a faulty measurement, a local subset monitoring strategy is adopted to eliminate the observations corresponding to each satellite one by one to construct multiple observation subsets. The local outlier factor of each observation subset is calculated, and the satellite observations corresponding to the minimum local outlier factor are identified as faulty measurements and excluded.
[0037] like Figure 2The diagram shown is a flowchart of a data density-driven train satellite positioning fault detection method.
[0038] In step S101, feature quantities that can effectively reflect the quality of satellite observation data are extracted from the raw observation data collected by the satellite receiver. The raw pseudorange observation data is mapped to a multidimensional feature space, providing a feature basis for subsequent density-based clustering analysis and outlier detection.
[0039] First, raw observation data is collected using a satellite receiver. During the propagation of GNSS signals, direct signals and non-line-of-sight signals exhibit different physical characteristics. Therefore, this invention extracts three complementary feature dimensions based on this characteristic: carrier-to-noise ratio, satellite elevation angle, and pseudorange residual.
[0040] Carrier-to-noise ratio (CNR) is the ratio of the carrier power of a satellite navigation signal to the noise power per unit bandwidth. Direct signals typically have a relatively high CNR due to less propagation path loss; conversely, non-line-of-sight (NLS) signals experience significant energy attenuation due to obstructions from trees, buildings, etc., resulting in a significantly lower CNR. Therefore, CNR serves as an effective indicator for distinguishing between direct and NLS signals.
[0041] Satellite elevation angle can reflect the reliability of observation data from the perspective of spatial geometry. Satellites with higher elevation angles are less likely to have their signal propagation paths blocked by obstacles such as buildings and trees, and are more likely to be regarded as direct signals; while low elevation angle signals, because their propagation paths are closer to the ground plane, are more susceptible to obstruction and reflection from the surrounding environment, and are usually non-line-of-sight signals.
[0042] In some embodiments, the satellite elevation angle is calculated using the satellite position and the receiver position, as shown in formula (1): ; (1) in, Indicates the first The satellite elevation angle of each satellite; Represents the arcsine function; This represents the receiver's coordinates in the Z direction within the northeast-central coordinate system; Indicates the first The coordinates of the satellite in the Z direction of the northeast celestial coordinate system; Indicates the receiver and the first The spatial distance between satellites.
[0043] The pseudorange residual is the difference between the original observed pseudorange and the calculated pseudorange. It is calculated by back-calculating the pseudorange from the current position to the satellite using the estimated position from the satellite receiver and the satellite's position. Since pseudorange is calculated based on signal propagation time, reflected signals have longer propagation paths and longer propagation times, resulting in larger pseudorange values. Therefore, a larger pseudorange residual indicates that the signal is more likely to be a non-line-of-sight signal.
[0044] In some embodiments, the pseudorange residual is calculated by the difference between the original observed pseudorange and the calculated pseudorange, as shown in formula (2): ; (2) in, Represents pseudorange residuals; Indicates the original observation pseudorange; , , This represents the satellite's position coordinates on the three axes in a Cartesian coordinate system. , , This represents the three-axis position coordinates of the receiver in a spatial rectangular coordinate system, estimated by the positioning algorithm.
[0045] Finally, the three features of each sample—carrier-to-noise ratio, satellite elevation angle, and pseudorange residual—are combined to construct an original observation feature dataset containing multiple samples. Each sample is represented as a feature vector containing the three feature dimensions mentioned above.
[0046] In step S102, an unsupervised density clustering method is used to identify and remove abnormal observation data with significant pseudorange errors in the feature space, thereby obtaining a clean dataset composed of normal observation data with small pseudorange errors, which provides a high-quality data foundation for the training of the subsequent offline detection model.
[0047] In the feature space, normal observation data, due to their similar quality characteristics, such as high carrier-to-noise ratio, moderate elevation angle, and small pseudorange residuals, will exhibit a continuous high-density distribution; while abnormal observations with significant pseudorange errors, such as non-line-of-sight signals and multipath interference signals, deviate from the normal data distribution, and have a lower sample density in their neighborhood, and can be identified as outliers.
[0048] In some embodiments, the density-based clustering algorithm described above is the DBSCAN (Density-Based Spatial Clustering of Applications with Noise) density clustering algorithm.
[0049] First, the features in the original observation feature dataset are subjected to min-max normalization to eliminate the influence of different units on the distance calculation. Then, Euclidean distance is used to measure the similarity between samples, as shown in formula (3): ; (3) in, Represents sample points and The Euclidean distance between them; This represents the feature dimension index of the sample, with a value range of 100. These correspond to three feature dimensions: carrier-to-noise ratio, satellite elevation angle, and pseudorange residual, respectively. Indicates the first The sample at the th Values can be taken in each feature dimension; Indicates the first The sample at the th The values can be taken in each feature dimension.
[0050] Next, the neighborhood radius parameter is set to construct the neighborhood. For example, the neighborhood radius is denoted as... Neighborhood is denoted as .
[0051] Then, a minimum neighborhood sample number parameter is set to determine the core point. For example, the minimum neighborhood sample number is denoted as... MinPts Specifically, for sample points its neighborhood Defined as all and Distance not exceeding The set of sample points. If the number of sample points within the neighborhood radius is greater than or equal to the minimum neighborhood sample number ( ), then the sample points It was determined to be the core point.
[0052] Subsequently, clusters are constructed using density reachability relationships: if sample points Located at the core point Within its neighborhood, it is called from Density reaches directly.
[0053] If a sample sequence exists ,satisfy Then it is called from Density reachable. Choose any unvisited core point as the starting point, recursively search all sample points that are density reachable from it, and continuously expand to form clusters.
[0054] For non-core points, if they do not belong to the neighborhood of any core point, they are defined as noise points (i.e., anomalous samples). Normal GNSS observation data exhibits a continuous high-density distribution in the feature space, forming major clusters; while anomalous observations with severe multipath errors, non-line-of-sight propagation, or signal obstruction have discrete feature distributions, making it difficult to form high-density regions, and are therefore identified as noise points. After removing the observation data corresponding to noise points, the normal observation dataset is obtained.
[0055] The DBSCAN algorithm used in this invention does not require pre-specifying the number of clusters and can discover clusters of arbitrary shapes, and has a good ability to identify outliers in the feature space.
[0056] In step S103, using the normal observation dataset obtained in step S102, an offline detection model describing the distribution characteristics of normal data is established based on the Local Outlier Factor (LOF) algorithm, and the kernel density estimation (KDE) method is used to adaptively determine the fault detection threshold, providing a decision benchmark for subsequent online fault detection.
[0057] First, the normal observation dataset obtained in step S102 is filtered using the positioning algorithm of the satellite receiver.
[0058] In some embodiments, the present invention employs Kalman filtering to process the observed data and extracts a innovation vector from the filter. The innovation vector reflects the degree of deviation between the observed data and the predicted data. The squared elements of the innovation vector are then summed to obtain the filtered innovation square, which serves as a sample feature quantity characterizing the quality of the observed data. For example, the normal sample feature set is denoted as... ,in, This represents the total number of normal samples. Indicates the first The squared filtered innovation of each sample. This feature is used to characterize the degree of deviation between the observed data and the predicted data.
[0059] The local outlier factor algorithm determines the outlier status of a sample point by comparing its local density with that of other sample points in its neighborhood. Its calculation process includes: First, for a sample point in the normal sample feature set, calculate its K-th distance, which is the Euclidean distance between the sample point and its (K+1)-th nearest neighbor after sorting the data in the distance space in ascending order. Then, determine the K-th distance neighborhood of the sample point based on the K-th distance.
[0060] Secondly, the reachable distance from the sample point in the Kth distance neighborhood to the sample point in the normal sample feature set is calculated, as shown in formula (4): ; (4) in, Represents sample points to sample point The reachable distance; Represents sample points The Kth distance; Represents sample points With sample points The Euclidean distance between them; This indicates taking the maximum value.
[0061] The reachability design makes distance calculation more stable and avoids misjudgments caused by distance fluctuations in sparse data areas.
[0062] Next, based on the reachability distances from all sample points in the Kth distance neighborhood to the sample point, the local reachability density of the sample point is calculated, as shown in formula (5): ; (5) in, Represents sample points Locally achievable density; This represents the number of sample points contained in the Kth distance neighborhood; This represents the Kth distance neighborhood.
[0063] Local reachability density is the reciprocal of the average reachability distance from each point in the neighborhood to the sample point, reflecting the density of data around the sample point.
[0064] Finally, based on the local reachability density of the sample point and the local reachability density of each sample point in its K-th distance neighborhood, the local outlier factor of the sample point is calculated, as shown in Equation (6): ; (6) in, Represents sample points Local outlier factors.
[0065] The Local Outlier Factor (LOF) value characterizes the degree of anomaly of a sample point relative to its neighborhood data. If the LOF value is approximately 1, it indicates that the sample point has a similar density to its neighborhood points and is considered a normal point; if it is greater than 1, it indicates that the sample point has a significantly lower density than its neighborhood points and exhibits outlier characteristics.
[0066] By traversing all samples in the normal sample feature set, we obtain the set of local outliers among the normal samples, which can be denoted as... .
[0067] Since real-world data often does not follow a Gaussian distribution, this invention uses kernel density estimation, a nonparametric method, to fit the probability density distribution of local outliers in normal samples, thereby adaptively determining the detection threshold.
[0068] In some embodiments, a Gaussian kernel function is used as the kernel function for kernel density estimation, and a smoothing parameter is set. The probability density function of the random variable of the local outlier factor in the normal sample is calculated by the kernel density estimation method, as shown in formula (7): ; (7) in, Indicates the probability density at The value at; This represents the total number of normal samples; Indicates the smoothing parameter; Represents the Gaussian kernel function; This represents a random variable representing a local outlier in a normal sample. Represents the first in the set of local outliers in normal samples One sample.
[0069] For the preset false alarm rate, calculate the fault detection threshold to satisfy formula (8): ; (8) in, Represents random variables Less than the fault detection threshold The probability of; This indicates the preset false alarm rate, which can be adjusted according to system security requirements.
[0070] This fault detection threshold serves as the decision threshold for determining whether the observed data is normal during the subsequent online detection phase.
[0071] In step S104, the real-time observation at the current moment is taken as the sample to be detected, and its local outlier factor in the offline detection model is calculated. By comparing it with the detection threshold determined in step S103, it is determined whether there is a fault measurement in the current observation data.
[0072] First, the real-time observations at the current moment are preprocessed in the same way as in the offline phase: the carrier-to-noise ratio, satellite elevation angle and pseudorange residual features are extracted, and the filtered innovation square is calculated through the positioning algorithm to obtain the sample to be detected.
[0073] Then, following the same method as calculating the local outlier factor in step S103, the local outlier factor of the sample to be detected relative to the normal sample feature set is calculated. The specific process includes: Calculate the Euclidean distance between the sample to be detected and each sample in the feature set of normal samples, and determine the nearest neighbor set of the sample to be detected based on the preset neighborhood parameter (the same K value as in the offline stage).
[0074] The reachable distance from the sample points in the nearest neighbor set to the sample to be detected is calculated as shown in formula (9): ; (9) in, Represents sample points to the sample to be tested The reachable distance; Represents sample points The Kth distance; Represents sample points With the sample to be tested The Euclidean distance between them; This indicates taking the maximum value.
[0075] Based on the reachability distances from all sample points in the nearest neighbor set to the sample to be detected, the local reachability density of the sample to be detected is calculated, as shown in formula (10): ; (10) in, Indicates the sample to be tested Locally achievable density; This indicates the number of sample points contained in the nearest neighbor set; This represents the nearest neighbor set.
[0076] Based on the local reachability density of the sample to be detected and the local reachability density of each sample point in its nearest neighbor set, the local outlier factor of the sample to be detected is calculated as shown in formula (11): ; (11) in, Indicates the sample to be tested Local outlier factors.
[0077] Finally, fault determination is made by comparing the local outlier factor of the sample to be tested with the fault detection threshold: If the local outlier factor of the sample to be detected is less than or equal to the fault detection threshold ( This indicates that the current observed sample is embedded within the normal data distribution area, and its local density is comparable to that of normal samples, thus determining that the current observed data is normal.
[0078] If the local outlier factor of the sample to be detected is greater than the fault detection threshold ( This indicates that the current observation sample is located outside the normal data distribution area, and its local density is significantly lower than that of the normal sample, indicating that there is a fault measurement in the current observation data.
[0079] This decision mechanism does not rely on the assumption that the observed data follows a Gaussian distribution. It makes judgments based entirely on the density distribution characteristics of the data itself, and has a natural adaptability to anomalies under non-Gaussian distributions.
[0080] In step S105, after a fault is detected in step S104, a local subset monitoring strategy is used to accurately locate the specific satellite that caused the fault, and its observations are excluded from the navigation solution to ensure the continuity and reliability of train positioning.
[0081] Since single-satellite failure is the most common scenario, this invention employs a step-by-step elimination strategy to locate the source of the failure.
[0082] In some embodiments, the observations corresponding to each satellite are removed from all observation data at the current moment to construct a corresponding observation subset, and the sample to be detected and its local outlier factor corresponding to the observation subset are calculated according to the method in step S104.
[0083] The local outlier factor is compared for each subset of observations. The identification logic is as follows: if the satellites removed are normal satellites, the remaining subset of observations still contains faulty satellites, and the local outlier factor of this subset is still relatively high; if the satellites removed are faulty satellites, the remaining subset of observations only contains normal observations, and the local outlier factor of this subset will be significantly reduced, becoming the minimum value among all subsets.
[0084] Therefore, the satellite observations corresponding to the local outlier factor minimum are identified as fault measurements, as shown in formula (12): ;(12) in, Indicates a satellite that has been identified as faulty; Indicates taking the first Local outlier factors corresponding to each subset of observations The value that yields the minimum value.
[0085] After identifying the faulty satellite, its corresponding observations are excluded from the navigation calculation, and train positioning calculations are performed using observation data from the remaining normal satellites. This strategy does not require prior knowledge of the fault mode (step fault or gradual ramp fault) and can effectively identify and exclude fault measurements in single-satellite fault scenarios.
[0086] Corresponding to the above method, the present invention also provides a data density-driven train satellite positioning fault detection system. The system includes a computer device, which includes a processor and a memory. The memory stores computer instructions, and the processor is used to execute the computer instructions stored in the memory. When the computer instructions are executed by the processor, the electronic device performs the steps of the method described above.
[0087] This invention also provides a computer-readable storage medium storing a computer program thereon, which, when executed by a processor, implements the steps of the aforementioned method. The computer-readable storage medium may be a tangible storage medium, such as random access memory (RAM), main memory, read-only memory (ROM), electrically programmable ROM, electrically erasable programmable ROM, register, floppy disk, hard disk, removable storage disk, CD-ROM, or any other form of storage medium known in the art.
[0088] Those skilled in the art will understand that the exemplary components, systems, and methods described in conjunction with the embodiments disclosed herein can be implemented in hardware, software, or a combination of both. Whether implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this invention. When implemented in hardware, it can be, for example, electronic circuits, application-specific integrated circuits (ASICs), appropriate firmware, plug-ins, function cards, etc. When implemented in software, the elements of this invention are programs or code segments used to perform the desired tasks. The programs or code segments can be stored in a machine-readable medium or transmitted over a transmission medium or communication link via data signals carried in a carrier wave.
[0089] It should be clarified that the present invention is not limited to the specific configurations and processes described above and shown in the figures. For the sake of brevity, detailed descriptions of known methods are omitted here. In the above embodiments, several specific steps are described and shown as examples. However, the method process of the present invention is not limited to the specific steps described and shown. Those skilled in the art can make various changes, modifications, and additions, or change the order of steps, after understanding the spirit of the present invention.
[0090] In this invention, features described and / or illustrated for one embodiment may be used in the same or similar manner in one or more other embodiments, and / or combined with or in place of features of other embodiments.
[0091] The above description is merely a preferred embodiment of the present invention and is not intended to limit the invention. For those skilled in the art, various modifications and variations can be made to the embodiments of the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.
Claims
1. A data density-driven train satellite positioning fault detection method, characterized in that, The method includes: Raw observation data is collected using a satellite receiver. The carrier-to-noise ratio, satellite elevation angle, and pseudorange residual are extracted as satellite signal feature quantities that reflect the quality of satellite observation data. A raw observation feature dataset containing multiple samples is constructed, with each sample containing the three feature dimensions of the carrier-to-noise ratio, the satellite elevation angle, and the pseudorange residual. The original observation feature dataset is processed using a density-based clustering algorithm to identify and remove noise points with a sample density lower than a preset threshold in the feature space, thereby obtaining a normal observation dataset. The filtered innovation squared is extracted from the normal observation dataset as a sample feature quantity to construct a normal sample feature set; the local outlier factor of each sample in the normal sample feature set is calculated based on the local outlier factor algorithm to obtain the normal sample local outlier factor set; and the probability density distribution of the normal sample local outlier factor set is fitted by the kernel density estimation algorithm, and the fault detection threshold is determined according to the preset false alarm rate. The real-time observations at the current moment are used as the samples to be detected. The local outlier factor of the sample to be detected relative to the feature set of normal samples is calculated. If the local outlier factor of the sample to be detected is less than or equal to the fault detection threshold, the current observation data is determined to be normal. Otherwise, the current observation data is determined to contain a fault measurement. After determining that a faulty measurement exists, a local subset monitoring strategy is adopted to eliminate the observations corresponding to each satellite one by one to construct multiple observation subsets. The local outlier factor of each observation subset is calculated, and the satellite observations corresponding to the minimum local outlier factor are identified as faulty measurements and excluded.
2. The train satellite positioning fault detection method based on data density driving according to claim 1, characterized in that, The density-based clustering algorithm is the DBSCAN algorithm; The specific method for identifying and removing noise points with a sample density below a preset threshold in the neighborhood is as follows: Set a neighborhood radius and a minimum number of neighborhood samples. Define sample points whose number of samples within the neighborhood radius is greater than or equal to the minimum number of neighborhood samples as core points. Define sample points whose number of samples within the neighborhood radius is less than the minimum number of neighborhood samples and do not belong to the neighborhood of core points as noise points and remove them.
3. The train satellite positioning fault detection method based on data density driving according to claim 1, characterized in that, The local outlier factor for each sample in the normal sample feature set is calculated based on the local outlier factor algorithm, including: For a sample point in the normal sample feature set, calculate the Kth distance of the sample point, and determine the Kth distance neighborhood of the sample point based on the Kth distance; Calculate the reachable distance from the sample point in the Kth distance neighborhood to the sample point: ； in, Represents sample points to sample point The reachable distance; Represents sample points The Kth distance; Represents sample points With sample points The Euclidean distance between them; This indicates taking the maximum value; Calculate the local reachability density of the sample point based on the reachability distances from all sample points in the Kth distance neighborhood to the sample point: ； in, Represents sample points Locally achievable density; This represents the number of sample points contained in the Kth distance neighborhood; Represents the Kth distance neighborhood; Calculate the local outlier factor of the sample point based on its local reachability density and the local reachability density of each sample point in its K-th distance neighborhood: ； in, Represents sample points Local outlier factors.
4. The train satellite positioning fault detection method based on data density driving according to claim 1, characterized in that, The probability density distribution of the local outlier set of the normal samples is fitted using a kernel density estimation algorithm, and a fault detection threshold is determined based on a preset false alarm rate, including: The Gaussian kernel function is used as the kernel function for kernel density estimation, and a smoothing parameter is set. The probability density function of the random variable representing local outliers in normal samples is calculated using the kernel density estimation method. ； in, Indicates the probability density at The value at; This represents the total number of normal samples; Indicates the smoothing parameter; Represents the Gaussian kernel function; This represents a random variable representing a local outlier in a normal sample. Represents the first in the set of local outliers in normal samples One sample; For a pre-set false alarm rate, calculate the fault detection threshold, which satisfies: ； in, Represents random variables Less than the fault detection threshold The probability of; This indicates the preset false alarm rate.
5. The train satellite positioning fault detection method based on data density driving according to claim 1, characterized in that, Calculating the local outlier factor of the sample to be detected relative to the feature set of normal samples includes: Calculate the Euclidean distance between the sample to be detected and each sample in the feature set of normal samples, and determine the nearest neighbor set of the sample to be detected based on the preset neighborhood parameters; Calculate the reachability distance from the sample points within the nearest neighbor set to the sample to be detected: ； in, Represents sample points to the sample to be tested The reachable distance; Represents sample points The Kth distance; Represents sample points With the sample to be tested The Euclidean distance between them; This indicates taking the maximum value; Calculate the local reachability density of the sample to be detected based on the reachability distances from all sample points in the nearest neighbor set to the sample to be detected: ； in, Indicates the sample to be tested Locally achievable density; This indicates the number of sample points contained in the nearest neighbor set; Represents the nearest neighbor set; Based on the local reachability density of the sample to be detected and the local reachability density of each sample point in its nearest neighbor set, calculate the local outlier factor of the sample to be detected: ； in, Indicates the sample to be tested Local outlier factors.
6. The train satellite positioning fault detection method based on data density driving according to claim 1, characterized in that, A local subset monitoring strategy is adopted, which involves systematically removing observations corresponding to each satellite to construct multiple observation subsets. The local outlier factor for each observation subset is calculated, and the satellite observation corresponding to the minimum local outlier factor is identified as a fault measurement, including: The observations corresponding to each satellite are removed from all observation data at the current moment to construct the corresponding observation subset, and the sample to be detected and its local outlier factor corresponding to the observation subset are calculated. By comparing the magnitudes of the local outliers corresponding to each subset of observations, the subset of observations corresponding to the minimum local outlier is determined, and the satellite observations removed from this subset are identified as fault measurements. ； in, Indicates a satellite that has been identified as faulty; Indicates taking the first Local outlier factors corresponding to each subset of observations The value that yields the minimum value.
7. The train satellite positioning fault detection method based on data density driving according to claim 1, characterized in that, The satellite elevation angle is calculated using the satellite position and receiver position, using the following formula: ； in, Indicates the first The satellite elevation angle of each satellite; Represents the arcsine function; This represents the receiver's coordinates in the Z direction within the northeast-central coordinate system; Indicates the first The coordinates of the satellite in the Z direction of the northeast celestial coordinate system; Indicates the receiver and the first The spatial distance between satellites; The pseudorange residual is calculated by the difference between the original observed pseudorange and the calculated pseudorange, using the following formula: ； in, Represents pseudorange residuals; Indicates the original observation pseudorange; , , This represents the satellite's position coordinates on the three axes in a Cartesian coordinate system. , , This represents the three-axis position coordinates of the receiver in a spatial rectangular coordinate system, estimated by the positioning algorithm.
8. The train satellite positioning fault detection method based on data density driving according to claim 1, characterized in that, The squared filtered information is obtained in the following way: The observation data in the normal observation dataset is processed by Kalman filtering using the positioning algorithm of the satellite receiver to obtain the filtered innovation vector. The filtered innovation vector is obtained by squaring each element and summing them. The squared filtering innovation is used to characterize the degree of deviation between the observed data and the predicted data.
9. A data density-driven train satellite positioning fault detection system, comprising a processor, a memory, and a computer program or instructions stored in the memory, characterized in that, The processor is used to execute the computer program or instructions, and when the computer program / instructions are executed, the system implements the steps of the method as described in any one of claims 1 to 8.
10. A computer-readable storage medium having a computer program or instructions stored thereon, characterized in that, When the computer program or instructions are executed by a processor, they implement the steps of the method as described in any one of claims 1 to 8.