A data anomaly detection method based on Mahalanobis distance chi-square detection

Abstract

This application provides a data anomaly detection method based on Mahalanobis distance and chi-square detection, relating to the field of autonomous navigation and positioning detection technology. The method includes the following steps: In the extended Kalman filter correction stage, the residual vector of the sensor observation data is calculated, then the Mahalanobis distance is calculated based on the residual vector, and chi-square detection is performed using the Mahalanobis distance to determine if abnormal data exists. When abnormal data is detected, a correction suppression mechanism is triggered, and an abnormal data processing mode is entered. In the extended Kalman filter correction stage, this application first accurately calculates the residual vector of the sensor observation data, then scientifically calculates the Mahalanobis distance based on it, and effectively identifies abnormal data through chi-square detection. Once an anomaly is detected, the correction suppression mechanism is immediately triggered, entering the abnormal data processing mode to avoid erroneous correction of the system state by abnormal data, thereby significantly improving the system's positioning accuracy and operational stability, and providing a more reliable guarantee for autonomous navigation and positioning systems.

Description

Technical Field

[0001] This application relates to the field of autonomous navigation and positioning detection technology, and in particular to a data anomaly detection method based on Mahalanobis distance chi-square detection. Background Technology

[0002] In applications of ground-penetrating radar robots and similar autonomous navigation and positioning systems, the accuracy and reliability of sensor data are crucial to ensuring the effective operation of the system. Such systems typically require the fusion of sensor data from multiple sources and heterogeneous types, such as the global navigation satellite system receiver, inertial measurement unit, and chassis odometer in a real-time dynamic positioning system. Only by fusing data from these sensors can the system achieve high-precision environmental modeling and autonomous navigation functions.

[0003] However, in real-world operating environments, sensor data is often affected by various adverse factors, such as environmental noise, signal obstruction, and sensor malfunctions. These factors can lead to outliers in the observed data. Traditionally, extended Kalman filtering, as a commonly used state estimation method, has been widely applied in the field of multi-sensor fusion positioning. However, this method has significant shortcomings when dealing with outliers. Specifically, during the correction phase, extended Kalman filtering directly uses the observation residuals to update the system state, without fully considering the potential negative impact of outliers on the system state. Once outliers appear in the sensor data, these outliers will directly interfere with the system state estimation results, leading to a significant reduction in positioning accuracy and potentially causing system instability. Summary of the Invention

[0004] This application provides a data anomaly detection method based on Mahalanobis distance chi-square detection, which aims to solve the problem that sensor data in ground-penetrating radar robots and similar autonomous navigation and positioning systems are affected by environmental noise, signal obstruction, sensor malfunctions and other factors, resulting in abnormal values ​​that affect the positioning accuracy and stability of the system.

[0005] To address the aforementioned technical problems, this application provides a data anomaly detection method based on Mahalanobis distance chi-square detection, comprising the following steps: calculating the residual vector of sensor observation data during the correction stage of extended Kalman filtering; calculating the Mahalanobis distance based on the residual vector; performing chi-square detection based on the Mahalanobis distance to determine whether abnormal data exists; and triggering a correction suppression mechanism and entering anomaly data processing mode when abnormal data is detected.

[0006] In some implementations, the observation model in extended Kalman filtering is used to calculate the residual vector between the observed and predicted values, expressed as: ,in, For a moment The observed residual vector, For the observed values, For the observation matrix, This is for predicting the state.

[0007] In some implementations, the formulas used to calculate the Mahalanobis distance include: ,in, Indicates at time The calculated Mahalanobis distance, For a moment The prediction covariance matrix of the current state based on information from the previous time step. Let be the transpose of the observation matrix. To observe the noise covariance matrix.

[0008] In some implementations, the chi-square detection anomaly detection step includes setting a chi-square distribution threshold. ,in The degrees of freedom are related to the dimension of the residual vector. If the threshold is exceeded, the current observed data is determined to be abnormal data.

[0009] In some implementation schemes, the specific operation of entering the abnormal data processing mode after triggering the correction suppression mechanism is as follows: When abnormal data is detected, the system triggers a two-level processing mechanism; in the first level of processing, the Kalman gain update is frozen so that the subsequent positioning process relies only on the predicted state and does not use the currently abnormal observation data to correct the state; in the second level of processing, multi-sensor compensation is performed, using the inertial measurement unit acceleration integral to compensate for the linear velocity drift of the global navigation satellite system receiver, and the compensation formula is: ,in, For the compensated speed, For the speed of the Global Navigation Satellite System receiver, This is the acceleration of the inertial measurement unit.

[0010] Some implementation schemes also include using a short-time kinematics model to constrain the rate of change of the carrier's attitude angle, determining a reasonable range of values ​​for the rate of change of attitude angle based on the short-time kinematics model, and using a limiting output strategy to correct when the calculated rate of change of attitude angle exceeds the range.

[0011] In some implementation schemes, the short-time kinematic model can be expressed as the following equation:

[0012] in, These are the rates of change of roll angle, pitch angle, and yaw angle, respectively. These are the angular velocities of the carrier around the x, y, and z axes, respectively. These are the linear accelerations of the carrier along the x, y, and z axes, respectively. Let the calculated rate of change of heading angle be a function related to various motion parameters. > Then the corrected rate of change of heading angle is set to If the calculated rate of change of heading angle satisfies Then the corrected rate of change of heading angle is set to .

[0013] By adopting the above technical solution, this application has the following beneficial effects compared with the prior art: In the extended Kalman filter correction stage, this application first accurately calculates the residual vector of sensor observation data, then scientifically calculates the Mahalanobis distance based on it, and effectively identifies abnormal data through chi-square detection. Once an anomaly is detected, the correction suppression mechanism is immediately triggered to enter the abnormal data processing mode, avoiding the erroneous correction of the system state by abnormal data, thereby significantly improving the system's positioning accuracy and operational stability, and providing a more reliable guarantee for autonomous navigation and positioning systems. Attached Figure Description

[0014] To more clearly illustrate the related technologies or the technical solutions in the embodiments of this application, the accompanying drawings used in the description of the related technologies or the embodiments of this application will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of this application, and not all embodiments. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort. Figure 1 This is a flowchart illustrating a data anomaly detection method based on Mahalanobis distance chi-square detection, provided in an embodiment of this application. Detailed Implementation

[0015] like Figure 1 As shown, a data anomaly detection method based on Mahalanobis distance chi-square detection is applied to ground penetrating radar robots and similar autonomous navigation and positioning systems.

[0016] In the correction phase of the extended Kalman filter, this method first uses the observation model in the extended Kalman filter to calculate the residual vector between the sensor observations and the predicted values, expressed as: ,in, For a moment The observed residual vector, For the observed values, For the observation matrix, To predict the state, this residual vector can accurately reflect the deviation between the observed data and the expected state, laying the foundation for subsequent accurate judgment of whether the data is abnormal. It effectively avoids the interference of abnormal data caused by environmental noise, signal blockage and sensor failure in the system state estimation, and ensures the positioning accuracy and stable operation of the system.

[0017] Then, the Mahalanobis distance is calculated based on the residual vector. The formula used to calculate the Mahalanobis distance includes: ,in, Indicates at time The calculated Mahalanobis distance, For a moment The prediction covariance matrix of the current state based on information from the previous time step. Let be the transpose of the observation matrix. To observe the noise covariance matrix, this distance comprehensively considers information such as the residual vector, the system's prediction covariance, and observation noise. It can more scientifically measure the degree of deviation between the observed data and the system's prediction, providing a key basis for accurately judging whether the data is abnormal, effectively avoiding the adverse effects of abnormal data on system state estimation, and ensuring the positioning accuracy and operational stability of autonomous navigation and positioning systems such as ground-penetrating radar robots.

[0018] After calculating the Mahalanobis distance, chi-square detection is performed based on the Mahalanobis distance to determine if there is any outlier data. Specifically, this is done by setting a chi-square distribution threshold related to the dimension of the residual vector. ,in For degrees of freedom, the calculated Mahalanobis distance is compared with a threshold. If If the threshold is exceeded, the current observation data is determined to be abnormal data. This detection method can scientifically quantify the degree of data abnormality, effectively identify abnormal values ​​caused by environmental interference or sensor failure, avoid the traditional method of directly using abnormal data, which leads to a decrease in system positioning accuracy and instability, and ensure the reliability of autonomous navigation and positioning systems.

[0019] In autonomous navigation and positioning systems such as ground-penetrating radar robots, once abnormal data is detected, the system will trigger a correction and suppression mechanism, then enter the abnormal data processing mode, and simultaneously start a two-level processing mechanism.

[0020] In the first stage of processing, the system freezes the Kalman gain update. As a result, the subsequent positioning process will rely solely on the predicted state and will no longer use abnormal observation data to correct the system state. This is because if abnormal observation data is used for correction, it is very likely to cause the system state estimation to deviate. Moreover, this deviation will further increase over time, seriously affecting the positioning accuracy of the system. By freezing the Kalman gain update, this risk is effectively avoided, ensuring the relative stability of the system state estimation during the abnormal data processing stage. In the second stage of processing, the system performs multi-sensor compensation. Specifically, it uses the acceleration integral of the inertial measurement unit to compensate for the linear velocity drift of the global navigation satellite system receiver. The compensation formula is as follows: ,in, For the compensated speed, For the speed of the Global Navigation Satellite System receiver, The acceleration of the inertial measurement unit is compensated in this way to correct velocity drift caused by various factors, improve the accuracy of velocity data, and thus provide more reliable data support for the system's positioning.

[0021] During the operation of the system, the change in the attitude angle of the carrier directly affects the state estimation and positioning accuracy of the system. If the change in attitude angle, especially the change in heading angle, is unreasonable, it is very likely to cause the heading angle to diverge, resulting in serious positioning deviation of the system or even failure to work normally. Therefore, based on the above technical solution, a short-time kinematic model is also introduced to constrain the rate of change of the carrier's attitude angle.

[0022] The short-time kinematic model can be expressed as the following equation:

[0023] in, These are the rates of change of roll angle, pitch angle, and yaw angle, respectively. These are the angular velocities of the carrier around the x, y, and z axes, respectively. These are the linear accelerations of the carrier along the x, y, and z axes, respectively. It is a function related to each motion parameter.

[0024] This model establishes precise equations based on motion parameters such as angular velocity and linear acceleration of the carrier around each axis, accurately describing the attitude angle changes of the carrier under different motion states. Based on this model, a reasonable range for the attitude angle change rate can be determined. When the calculated attitude angle change rate exceeds this range, the system will employ an output limitation strategy for correction. If the calculated heading angle change rate satisfies... > Then the corrected rate of change of heading angle is set to If the calculated rate of change of heading angle satisfies Then the corrected rate of change of heading angle is set to In this way, the short-time kinematic model effectively ensures that the carrier's attitude angle changes conform to the real kinematic laws, fundamentally avoiding the problem of heading angle divergence, further enhancing the system's stability and positioning reliability during abnormal data processing, and providing a solid guarantee for the accurate positioning of systems such as ground-penetrating radar robots in complex environments.

[0025] It should be noted that the several embodiments shown above in this application are described in a progressive manner, with each embodiment focusing on the differences from other embodiments. Similar or identical parts between embodiments can be referred to interchangeably. It should also be noted that in the textual description of this application, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply such an actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements may include not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus; and, without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0026] Furthermore, those skilled in the art can implement or use this application by practicing the several embodiments shown above. Various modifications to the embodiments shown above will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments not shown without departing from the spirit or scope of this application. Therefore, this application is not to be limited to the several embodiments shown above, but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A data anomaly detection method based on Mahalanobis distance chi-square detection, characterized in that, Includes the following steps: During the correction phase of the extended Kalman filter, the residual vector of the sensor observation data is calculated; Calculate Mahalanobis distance based on residual vector; Chi-square detection is performed based on Mahalanobis distance to determine if there is any abnormal data; When abnormal data is detected, the correction and suppression mechanism is triggered, and the abnormal data processing mode is entered.

2. The data anomaly detection method based on Mahalanobis distance chi-square detection according to claim 1, characterized in that, Using the observation model in the extended Kalman filter, the residual vector between the observed and predicted values ​​is calculated as follows: in, For a moment The observed residual vector; For the observed values, The observation matrix; This is for predicting the state.

3. The data anomaly detection method based on Mahalanobis distance chi-square detection according to claim 2, characterized in that, The formulas used to calculate Mahalanobis distance include: in, Indicates at time The calculated Mahalanobis distance; For a moment The prediction covariance matrix of the current state based on information from the previous time step; This is the transpose of the observation matrix; To observe the noise covariance matrix.

4. The data anomaly detection method based on Mahalanobis distance chi-square detection according to claim 3, characterized in that, The steps for chi-square testing to determine abnormalities include: Set chi-square distribution threshold ,in The degrees of freedom are related to the dimension of the residual vector; like If the threshold is exceeded, the current observed data is determined to be abnormal data.

5. The data anomaly detection method based on Mahalanobis distance chi-square detection according to claim 4, characterized in that, The specific steps for entering the abnormal data processing mode after triggering the correction and suppression mechanism are as follows: When abnormal data is detected, the system triggers a two-level processing mechanism; In the first stage of processing, the Kalman gain update is frozen so that the subsequent localization process relies solely on the predicted state and does not use the currently present anomaly observation data to correct the state. In the second-stage processing, multi-sensor compensation is performed, using the inertial measurement unit's acceleration integral to compensate for the linear velocity drift of the global navigation satellite system receiver. The compensation formula is as follows: in, The compensated speed; For the speed of the Global Navigation Satellite System receiver; This is the acceleration of the inertial measurement unit.

6. The data anomaly detection method based on Mahalanobis distance chi-square detection according to claim 5, characterized in that, It also includes using a short-time kinematics model to constrain the rate of change of the carrier's attitude angle, determining a reasonable range of values ​​for the rate of change of attitude angle based on the short-time kinematics model, and using a limiting output strategy to correct when the calculated rate of change of attitude angle exceeds the range.

7. The data anomaly detection method based on Mahalanobis distance chi-square detection according to claim 6, characterized in that, The short-time kinematic model can be expressed as the following equation: in, These are the rates of change of roll angle, pitch angle, and yaw angle, respectively. These are the angular velocities of the carrier around the x, y, and z axes, respectively. These are the linear accelerations of the carrier along the x, y, and z axes, respectively. It is a function related to each motion parameter; If the calculated rate of change of heading angle satisfies > Then the corrected rate of change of heading angle is set to ; If the calculated rate of change of heading angle satisfies Then the corrected rate of change of heading angle is set to .

Images