CN-121977539-A - Unmanned ship cluster co-location method based on SOM network self-adaptive screening

CN121977539ACN 121977539 ACN121977539 ACN 121977539ACN-121977539-A

Abstract

The invention discloses an unmanned ship cluster co-location method based on SOM network self-adaptive screening, and belongs to the field of navigation and location. The method comprises the steps of uniformly modeling outputs of GNSS, INS, GMNS and other heterogeneous navigation sources into Gaussian probability models, converting positioning information of a cooperative ship into indirect estimation of a target ship by utilizing inter-ship ranging direction-finding information to form a complete positioning information set, inputting position data into an SOM neural network for clustering, adopting a maximum inter-class variance algorithm to adaptively determine a threshold value based on node density, screening out high-confidence positioning information, regarding the screened information as a point on a Riemann manifold based on an information geometric theory, calculating a weighted geometric center of the information to obtain a final fusion result, and feeding back the fusion result to optimize cooperative observation parameters of the next iteration period. The method can adaptively remove wild values and uniformly process heterogeneous data, and remarkably improves the robustness, precision and instantaneity of the large-scale unmanned ship cluster in the cooperative positioning in the complex dynamic sea area.

Inventors

TANG CHENGKAI
Zhou Xunbin
CHENG ZHUO
ZHANG LINGLING
LIU YANGYANG
Danzesho
Chen Aomi
ZHU YANBING
WANG CHENYU
ZHANG SONGNIAN

Assignees

西北工业大学

Dates

Publication Date: 20260505
Application Date: 20260409

Claims (8)

1. An unmanned ship cluster co-location method based on SOM network self-adaptive screening is characterized by comprising the following steps: Step 1, modeling heterogeneous navigation sources by normalization information probability, namely respectively constructing unified information probability models for GNSS, GMNS and INS carried by unmanned vessels, calculating position correction quantity by using a least square method for the GNSS navigation sources, calculating positioning error covariance by using an error propagation law, and adopting an online self-adaptive statistical fitting method for GMNS navigation sources and INS navigation sources; Step 2, position conversion of cooperative observation information, namely converting navigation source positioning information of the cooperative ship into indirect position estimation and corresponding covariance matrix of the ship to be positioned by utilizing distance measurement and direction measurement information of the cooperative ship relative to the ship to be positioned in the unmanned ship cluster, and combining the positioning information of each navigation source of the ship to be positioned to generate a complete positioning information pair set; Step 3, self-adaptive data screening based on SOM neural network, specifically comprising: step 3.1, preprocessing data, namely, carrying out standardization processing on the position estimation vectors in the set by using the positioning information; step 3.2, SOM network mapping, namely inputting the standardized position vector into a SOM neural network for training, and establishing a mapping relation from each input position vector to two-dimensional grid nodes; step 3.3, counting the node density, namely counting the quantity of the input position vectors gathered by each grid node after the SOM network is converged to obtain the density of each node; Step 3.4, self-adaptive threshold screening, namely calculating an optimal density screening threshold value by utilizing a maximum inter-class variance algorithm based on the node density set obtained in the step 3.3, and screening out original positioning information pairs corresponding to nodes with the node density higher than the threshold value according to the mapping relation established in the step 3.2 to form an optimal positioning information pair set; Step 4, information geometric multisource fusion, namely based on the information geometric theory, regarding each position estimation in the optimized positioning information pair set obtained by screening in the step 3 as a point on a Riemann manifold, and calculating a weighted geometric center of a Gaussian probability density function corresponding to the point estimation to obtain a final fusion position estimation and a fusion covariance matrix of the ship to be positioned; and step 5, feedback closed-loop optimization, namely correcting the initial value and variance of the cooperative ranging direction-finding information in the next iteration period by utilizing the final fusion position estimation obtained in the iteration to form a closed-loop optimization flow.
2. The unmanned ship cluster co-location method of claim 1, wherein the information probability model is constructed for GMNS and INS navigation sources by: Calculating a position estimate of a navigation source output at a previous time Final fusion position with last moment The difference between them to obtain an observation error sample ; Based on the deviation vector of the previous moment by using an exponentially weighted moving average rule Sum covariance matrix Recursively updating the bias vector at the current time Covariance matrix The update formula is as follows: In the formula, And The corresponding bias term forgetting factors and covariance matrix forgetting factors are respectively; Taking the position estimation output by the navigation source at the current moment as the mean value Will be updated As a covariance matrix Building a normalized probability model 。
3. The unmanned ship cluster co-location method according to claim 1 or 2, wherein in step 2, the conversion formula of the position of the ship to be located by the cooperating ship is: In the formula, Is the first Ship to be positioned for multiple iterations Is the first of (2) Personal cooperating ship Is the first of (2) Navigation source positioning results of individual positioning sources, Is the first Sub-iteration cooperative ship Is a result of the positioning of (a) Converted vessel to be positioned Is provided with a plurality of location information, Is the first Sub-iteration cooperative ship Relative to the vessel to be positioned Ranging and direction finding information of (1), wherein In order to measure the value of the distance, And Polar angle and azimuth angle in the direction-finding information respectively, Is the first Sub-iteration cooperative ship Is a result of the positioning of (a) Is a covariance matrix of the error of (a), Is the first The vessel to be positioned is obtained through iterative conversion Positioning information of (a) Is a covariance matrix of the error of (a), Is the ranging variance.
4. The unmanned ship cluster co-location method according to claim 1 or 2, wherein in step 3.1, the normalization process is specifically: Computing all position estimate vectors In each dimension Sample mean value And standard deviation of Wherein , For the number of samples in the raw navigational positioning data: Pressing the button Calculating to obtain a standardized vector Wherein Representation of In the first place Normalized eigenvalues in the individual dimensions.
5. The method for co-locating unmanned ship clusters according to claim 1 or 2, wherein in step 3.2, the training process of the SOM neural network comprises the steps of, for an input standardized vector By calculating a weight vector for each of its and every neuron node in the SOM network Euclidean distance between to determine the best matching unit, index of said best matching unit From the following components And iteratively updating the weight vectors of the best matching unit and the nodes in the neighborhood thereof according to the following formula: In the formula, For the number of training iterations, In order to decay the learning rate over time, Is indexed as Best matching unit node and node of (a) A neighborhood function between.
6. The unmanned ship cluster co-location method of claim 5, wherein the neighborhood function is a gaussian neighborhood function and is defined as: In the formula, And Respectively represent the index as Best matching unit node and node of (a) Geometric locations in a two-dimensional grid of the SOM network, Is the neighborhood radius decaying over time.
7. The unmanned ship cluster co-location method according to claim 1 or 2, wherein in step 3.4, the optimal density screening threshold is calculated by the following formula: In the formula, The threshold value is selected for the optimal density, For the set of node densities, For candidate thresholds traversed between node density minima and maxima, And Respectively by The number of background class nodes and foreground class nodes divided for the threshold value is the proportion of the total number of the nodes, And And the average densities of the background class node and the foreground class node are respectively.
8. The unmanned ship cluster co-location method according to claim 1 or 2, wherein in step 4, the final fusion position estimate and fusion covariance matrix are calculated by the following formula: In the formula, For the final fusion of the position estimation results, In order to finally fuse the covariance matrix, To optimize the number of pairs in the set of positioning pairs, the first The information pair is , Is the corresponding weight factor.

Description

Unmanned ship cluster co-location method based on SOM network self-adaptive screening Technical Field The invention belongs to the technical field of navigation positioning, in particular to a co-positioning method suitable for a large-scale unmanned ship cluster, and more particularly relates to a co-positioning method based on an information geometric theory and performing optimization processing on unmanned ship cluster navigation information through a Self-Organizing Map (SOM) network Self-adaptive screening mechanism. Background The unmanned ship (Unmanned Surface Vehicle, USV) cluster has important application value in the fields of marine mapping, environmental monitoring and the like, and the robustness and instantaneity of the collaborative navigation system are key for realizing intelligent cluster control. The existing system usually integrates multi-source information such as a global satellite navigation system (Global Navigation SATELLITE SYSTEM, GNSS), an inertial navigation system (Inertial Navigation System, INS), a geomagnetic matching navigation system (Geomagnetic Matching Navigation System, GMNS) and the like for positioning. However, in complex water areas such as ports and coasts, GNSS signals are easily blocked and interfered by multipath effects, sensor data are severely disturbed by stormy waves, and positioning errors are accumulated. The existing method has obvious defects when dealing with the scenes, namely firstly, the data screening mostly adopts fixed threshold values set according to experience, so that 'wild values' under dynamic environments are difficult to adaptively reject, the effective information is wrongly deleted or abnormal noise is missed to judge, the robustness of a positioning system is seriously reduced, secondly, the heterogeneous navigation source data format, the resolving principle and the error characteristics are different, and a unified probability modeling and uncertainty characterization framework is lacking, so that the complementary advantages of navigation sources are difficult to fully utilize in a fusion layer, the system precision is greatly reduced when a single information source fails, thirdly, the data quantity is rapidly increased along with the enlargement of the cluster scale, the calculation burden of the traditional centralized processing or complex detection algorithm is heavy, and the millisecond-level real-time requirement is difficult to meet. Disclosure of Invention The invention aims to solve the problems of poor co-location precision and low system stability caused by non-uniform multi-source heterogeneous navigation data format, serious outlier interference and difficulty in considering real-time performance and robustness in the traditional screening method in the complex sea area environment of the existing large-scale USV cluster, and provides a large-scale USV cluster co-location method fused with SOM network self-adaptive screening technology and information geometry theory. According to the method, a normalized information probability model of the heterogeneous navigation source is constructed, the self-adaptive screening of abnormal data is realized by utilizing the topological mapping characteristic of the SOM neural network and the maximum inter-class variance principle, and high-precision fusion is completed by combining with an information geometric theory, so that high-reliability and high-real-time co-location of the USV cluster in a dynamic complex environment is realized. In order to achieve the above purpose, the technical solution provided by the present invention is: The invention provides an unmanned ship cluster co-location method based on SOM network self-adaptive screening, which comprises the following steps: Step 1, modeling heterogeneous navigation sources by normalization information probability, namely respectively constructing unified information probability models for GNSS, GMNS and INS carried by unmanned vessels, calculating position correction quantity by using a least square method for the GNSS navigation sources, calculating positioning error covariance by using an error propagation law, and adopting an online self-adaptive statistical fitting method for GMNS navigation sources and INS navigation sources; Step 2, position conversion of cooperative observation information, namely converting navigation source positioning information of the cooperative ship into indirect position estimation and corresponding covariance matrix of the ship to be positioned by utilizing distance measurement and direction measurement information of the cooperative ship relative to the ship to be positioned in the unmanned ship cluster, and combining the positioning information of each navigation source of the ship to be positioned to generate a complete positioning information pair set; Step 3, self-adaptive data screening based on SOM neural network, specifically comprising: step 3.1, preprocessing data, namely, car