CN-122015907-A - Method for calibrating sensing device in non-ideal environment

CN122015907ACN 122015907 ACN122015907 ACN 122015907ACN-122015907-A

Abstract

The invention discloses an interactive multi-model-based adaptive kernel bandwidth maximum entropy Kalman filtering DVL calibration method, and belongs to the technical field of navigation sensor device calibration in natural environments. The method comprises the steps of S1, initializing a calibration system, S2, collecting data of a real sensor at the q moment, calculating a proportional error at the current moment by using the collected data, S3, realizing bandwidth self-adaption by using a multi-model interaction algorithm frame, S4, filtering an installation angle by using maximum entropy Kalman filtering, S5, calculating model likelihood probability, updating model probability, S6, interactively outputting, calculating an output estimation state and an estimation covariance matrix of a normalized model, and S7, correcting the system state by using the estimation state error. The method can accurately calibrate the installation error and the structural parameter of the DVL, and improves the navigation precision.

Inventors

QIAN MIN
SHI YAN
ZHANG FANSHENG
CHEN CAN
YANG ZI

Assignees

南京理工大学
南京理工大学淮安研究院

Dates

Publication Date: 20260512
Application Date: 20251231

Claims (9)

1. The adaptive kernel bandwidth maximum entropy Kalman filter DVL calibration method based on the interactive multi-model is characterized by comprising the following steps of: s1, initializing a calibration system, setting Markov transition matrixes and initial model probabilities among different sub-filters of a multi-model system according to prior information, and finishing initialization of each sub-filter; s2, acquiring data of a real sensor at the moment q, and calculating a proportional error at the current moment by using the acquired data according to the working principle of DVL; S3, the installation angle estimation system uses a multi-model interaction algorithm framework to realize bandwidth self-adaption; s4, filtering the installation angle by using maximum entropy Kalman filtering; s5, calculating model likelihood probability and updating the model probability; s6, interactively outputting, and calculating the output estimation state of the normalized model And estimating covariance matrix ; S7, correcting the system state by using the estimated state error.
2. The adaptive kernel bandwidth maximum entropy Kalman filtering DVL calibration method based on the interactive multi-model of claim 1, wherein in S1, the specific implementation contents are: The acquired data comprise the original output speed of the DVL, the output of the SINS/GPS fusion navigation system and the output of the IMU gyroscope, and the proportion error at the current moment is calculated by using the acquired data according to the DVL speed measurement model; According to the operating principle of DVL, the velocity error model is expressed as: , Wherein, the Representing the output speed of the current DVL; Representing the proportional error of DVL; representing a transformation matrix of DVL and SINS; Representing a coordinate transformation matrix of the carrier system to the navigation system; A speed truth value at the current time DVL; Representing a projection of an angular velocity of the carrier coordinate system relative to the navigational coordinate system in the carrier coordinate system; Lever arms representing DVL and SINS; representing a velocity error vector; when calculating the proportional error, time integration is carried out on the speed, random noise is effectively smoothed, and the proportional error at the current moment is obtained, specifically: , , Wherein, the Representing the proportional error of DVL; the lever arms representing the DVL and SINS, Representing a projection of the angular velocity of the carrier coordinate system relative to the navigational coordinate system in the carrier coordinate system, Representing zero bias of the gyroscope; projection of the earth rotation speed on the navigation system; Representing a projection of an angular velocity of the navigation system relative to the earth system under the navigation system; representing the coordinate transformation matrix estimation value from the carrier system to the navigation system; Representing a sampling period; Representing the projection of the angular velocity of the carrier relative to the inertial frame in the carrier frame.
3. The adaptive kernel bandwidth maximum entropy Kalman filtering DVL calibration method based on the interactive multi-model of claim 1, wherein in S3, the specific implementation contents are: s31, initializing a multi-model system model, and setting Markov transfer matrixes among different sub-filters according to prior information; S32, inputting interaction, namely inputting posterior states and covariance of each sub-filter at the q-1 moment, a Markov transition matrix and prior probability of each model, and calculating transition probability of the mixed normalization model from the model j to the model k; S33, inputting the estimated states of the models And corresponding hybrid model normalized transition probabilities Calculating a mixed estimation state of the model k; s34, inputting an estimated covariance matrix of each model and a corresponding mixed model normalized transition probability The hybrid estimation covariance of model k is calculated.
4. The method for calibrating the adaptive kernel bandwidth maximum entropy Kalman filter DVL based on the interactive multi-model according to claim 3, wherein in S32, the specific implementation contents are as follows: , , Wherein, the Representing the normalization constant(s), Representing the transition probability of the mixed normalization model from model j to model k; representing the transition probability of model j to model k; expressed in time Probability of model k; expressed in time Probability of model j.
5. The method for calibrating the adaptive kernel bandwidth maximum entropy Kalman filter DVL based on the interactive multi-model according to claim 3, wherein in S33, the specific implementation contents are as follows: , Wherein, the Expressed in time After mixing, estimating a mixed initial state prepared for the model k; expressed in time Estimating the state of a model j; Representing the transition probability of the hybrid normalized model from model j to model k.
6. The method for calibrating the adaptive kernel bandwidth maximum entropy Kalman filter DVL based on the interactive multi-model according to claim 3, wherein in S34, the specific implementation contents are as follows: ; Wherein, the Expressed in time A mixed initial covariance matrix prepared for model k; Representing the transition probability of the mixed normalization model from model j to model k; expressed in time A posterior estimation error covariance matrix of the model j; expressed in time Estimating the state of a model j; expressed in time After mixing, a mixed initial state estimate is prepared for model k, T representing the transpose operation.
7. The adaptive kernel bandwidth maximum entropy Kalman filtering DVL calibration method based on the interactive multi-model of claim 1, wherein in S4, the specific implementation process is as follows: S41, initializing a filtering system; s42, defining the installation angle error of the system Is the difference between the theoretical installation angle and the actual error angle; s43, constructing an observation error filtering model of the system; S44, carrying out prediction updating on the system; S45, calculating a related entropy; S46, calculating the information New information covariance 。
8. The adaptive kernel bandwidth maximum entropy Kalman filtering DVL calibration method based on the interactive multi-model of claim 1, wherein in S5, the specific implementation process is as follows: calculating model likelihood probabilities The method specifically comprises the following steps: ; Wherein, the Representing an innovation covariance matrix; Representing an innovation vector; representing a transpose of the innovation vector; M represents the dimension of the multi-element normal distribution; Updating model probabilities The calculation formula of (2) is as follows: , Wherein, the Representing model likelihood probabilities; representing the normalization constant, and l representing the total number of hybrid models.
9. The adaptive kernel bandwidth maximum entropy Kalman filtering DVL calibration method based on the interactive multi-model of claim 1, wherein in S7, the specific implementation process is as follows: , Wherein, the Representing an estimated installation angle rotation matrix at the moment q; representing an estimated installation angle rotation matrix at the moment q-1; representing the identity matrix; Indicating the mounting angle error at time q.

Description

Method for calibrating sensing device in non-ideal environment Technical Field The invention belongs to the technical field of navigation sensor device calibration in natural environment, and particularly relates to an adaptive kernel bandwidth maximum entropy Kalman filtering DVL calibration method based on an interactive multi-model. Background The amphibious unmanned platform is used as an underwater special task execution instrument, and accurate position information of the amphibious unmanned platform is obtained in real time, so that the amphibious unmanned platform is a precondition for completing other autonomous tasks. At present, one of the most widely used navigation modes in underwater vehicles is a strapdown inertial navigation/Doppler velocimeter (SINS/DVL) integrated navigation system. The DVL measures the speed of the carrier relative to the seabed or the water layer based on the Doppler effect, and can realize short-time autonomous navigation with higher precision by combining an inertial navigation system. In a typical underwater task that lacks GPS signal and external base station support, DVL provides indispensable motion constraint information. However, in practical application, the installation error of the DVL can significantly influence the navigation precision, so that the method has important significance in accurately modeling the error and on-line correction in long-endurance autonomous operation. Especially for an amphibious unmanned platform, due to the limitation of miniaturized design, the DVL installation space is extremely limited, and the sensor is difficult to be aligned with an SINS coordinate system strictly due to special working conditions such as bottoming crawling, the installation deviation is often large, and the traditional calibration method based on small-angle approximation is limited in applicability under such a scene. In addition, the observed signal of DVL in the underwater environment is easily influenced by multipath reflection, suspended matter shielding and flow field disturbance, and the measurement noise of the DVL often presents non-Gaussian and time-varying statistical characteristics, so that the calibration difficulty is further increased. The existing calibration method mainly comprises an off-line calibration technology based on a static or specific maneuvering track and an on-line estimation method based on Extended Kalman Filtering (EKF) or Unscented Kalman Filtering (UKF). The online filtering method is mostly based on least square or Gaussian noise assumption, and estimation deviation and even divergence are easy to occur in a non-Gaussian noise environment. In recent years, there are also researches on the introduction of a robust estimation theory (such as Huber kernel) to improve noise adaptability, but these methods generally rely on manually adjusting kernel parameters, and are difficult to cope with the actual scene of time-varying noise statistics. In addition, the single kernel bandwidth is easy to over fit or under fit when facing different noise distributions, and the self-adaptive adjustment capability for complex noise environments is lacking. Aiming at the problems, the application provides an interactive multi-model-based adaptive kernel bandwidth maximum entropy Kalman filter DVL calibration method, which aims to realize high-precision and adaptive online calibration of large installation angle errors by combining a multi-model frame with maximum entropy robust estimation and improve the navigation reliability of an amphibious unmanned platform in a complex underwater environment. Disclosure of Invention Aiming at the problems mentioned in the background art, the invention provides an adaptive kernel bandwidth maximum entropy Kalman filtering DVL calibration method based on an interactive multi-model, which solves the problems that Doppler Velocimeter (DVL) measurement errors in a non-ideal underwater environment are not subjected to Gaussian distribution, outlier interference exists and the like, and remarkably improves the robustness and adaptability of a calibration algorithm. The technical scheme adopted by the invention is as follows in order to solve the technical problems: An adaptive kernel bandwidth maximum entropy Kalman filter DVL calibration method based on an interactive multi-model comprises the following steps: s1, initializing a calibration system, setting Markov transition matrixes and initial model probabilities among different sub-filters of a multi-model system according to prior information, and finishing initialization of each sub-filter; s2, acquiring data of a real sensor at the moment q, and calculating a proportional error at the current moment by using the acquired data according to the working principle of DVL; S3, the installation angle estimation system uses a multi-model interaction algorithm framework to realize bandwidth self-adaption; s4, filtering the installation angle by using maximum entr