CN-121980203-A - Offshore moving target track prediction method and system for intention guidance and environment constraint

CN121980203ACN 121980203 ACN121980203 ACN 121980203ACN-121980203-A

Abstract

The invention provides a method and a system for predicting a track of a moving target on the sea, which aims at guiding and restraining the environment, and belongs to the technical field of offshore traffic control and track prediction. The method comprises the steps of extracting an offshore traffic channel based on AIS track data, counting environmental characteristics of channel nodes and edges, representing a historical track as a collection of channel nodes and edges, constructing a behavior simulation learning model, solving through a neural network to obtain a global rewarding function, training an intention distinguishing model, outputting behavior probabilities of short-time channel nodes and long-time ports or anchors, decomposing a speed sequence by means of wavelet transformation under the guidance of intention based on observation tracks and intention distinguishing results, predicting speed change, and interpolating to generate future track coordinates. According to the invention, the channel network is constructed in a data driving mode, so that subjective deviation of artificial rules is avoided, an intention guiding mechanism is introduced to inhibit long time sequence prediction error accumulation, and the adaptability to complex ocean scenes is enhanced by combining environment characteristic statistics.

Inventors

YANG WENTAO
WANG FENGJIE
LIAO MENGGUANG

Assignees

湖南科技大学三亚研究院

Dates

Publication Date: 20260505
Application Date: 20260408

Claims (8)

1. The method for predicting the track of the moving target at sea for the purpose of guiding and environment constraint is characterized by comprising the following steps: step S1, based on an AIS track data set with unified space-time coordinates, an island data exclusion layer is combined, and the density of track sampling points accessed by a sea moving target history is considered to construct an offshore traffic channel network; Step S2, according to the sea traffic channel network, counting speed distribution, course distribution and sea depth distribution characteristics in the key nodes of the sea traffic channel and the neighborhood range of the sea traffic channel side; s3, representing the historical track of which each destination is a port or an anchor as a set consisting of key nodes and edges of each offshore traffic channel, and dividing the historical track into a training set and a verification set; s4, constructing a behavior simulation learning model, solving an objective function through a neural network model, outputting rewarding values of each directed node and edge of the offshore traffic channel, and obtaining a global rewarding function through weighted fusion; S5, training an intention judging model, inputting a track data set, a port or anchor ground set, marine traffic channel statistical characteristics and a global rewarding function, and outputting a key node of a marine moving target short-time marine traffic channel and long-time port or anchor ground behavior probability; And S6, judging short-time and long-time behavior intentions of the moving target based on the moving target observation track, the port or anchor set, the marine traffic channel statistical characteristics, the global rewarding function and the intent judging model, and predicting the moving target speed change by utilizing wavelet transformation decomposition speed sequence under the guidance of the intentions, and generating future track coordinates by interpolation.
2. The method for predicting a moving target trajectory at sea with intention to guide and environmental constraint according to claim 1, wherein the step S1 comprises the steps of: step S11, the unified coordinates, the abnormal points are removed, the points are sampled through the resampled track, and the local density of each point is calculated: step S12, calculating the minimum distance from each track sampling point to any higher-density point; Step S13, calculating decision values of sampling points of each track, sorting according to the decision values, determining the number of key nodes through inflection point analysis, and selecting the key nodes before The track sampling points are used as key node candidates of the sea traffic channel; S14, eliminating candidate nodes in the island data exclusion layer polygon to obtain an offshore traffic channel key node set; S15, constructing Voronoi units for each key node of the offshore transportation channel, and calculating unit boundaries through perpendicular bisectors; And S16, restoring the topological relation among the nodes based on the Voronoi diagram to form an offshore traffic channel network.
3. The method for predicting a moving target trajectory at sea with intention to guide and environmental constraint according to claim 1, wherein the step S2 comprises the steps of: S21, counting the speed distribution characteristics of all track sampling points in the neighborhood range of each key node of the offshore traffic channel, wherein the speed distribution characteristics comprise a maximum value, a minimum value, a median and an average value; s22, counting the distribution characteristics of the ground heading, wherein the distribution characteristics comprise a maximum value, a minimum value, a median value and an average value; s23, counting the ship head course distribution characteristics, including maximum value, minimum value, median and average value; s24, counting sea depth distribution characteristics including maximum value, minimum value, median and average value; And S25, combining the space position of each key node of the offshore traffic channel with the corresponding speed, the ground heading, the head heading and the sea depth distribution characteristics, and taking the space position as an environment characteristic vector of the node.
4. The method for predicting a moving target trajectory at sea with intention to guide and environmental constraint according to claim 1, wherein the step S3 comprises the steps of: step S31, matching the preprocessed track sampling points to key nodes of the sea traffic channel based on a proximity principle, and describing the track as a process of reaching a port or an anchor from a starting point through a node sequence; s32, interpolation is completed on AIS track data by utilizing a physical dynamics model, and the change relation of the position of the moving target along with time is described through a differential equation; Step S33, in the physical dynamics model, the change rate of the speed and the direction is further considered; step S34, discretizing a differential equation, and updating the speed and the direction according to the time interval; Step S35, updating the position coordinates by utilizing the updated speed and direction and combining the time interval; step S36, the processed track is expressed as a structured data sequence containing time stamp, position coordinates, ground heading, warhead heading, instantaneous speed and water depth information; step S37, screening the track with the destination point located in the predefined port or anchor area as an expert track, and representing the expert track as a structured form of a node access sequence and a destination label.
5. The method for predicting a moving target trajectory at sea with intention to guide and environmental constraint according to claim 1, wherein the step S4 comprises the steps of: Step S41, modeling the moving target behavior as a Markov decision process, constructing a reward function parameterized by a neural network, and evaluating instant rewards obtained under specific state-action pairs; Step S42, designing a multi-layer perceptron as a reward function approximator, inputting a state-action joint feature vector, and outputting a scalar reward value; Step S43, defining the accumulated rewards of any candidate path as the sum of rewards of each side in the path; S44, using a maximum entropy inverse reinforcement learning framework, and maximizing the probability of the expert trajectory in all feasible paths by taking the maximum log likelihood of the expert trajectory as an objective function; Step S45, training a neural network through an optimization algorithm until the log likelihood of a verification set is not improved any more, and obtaining a final rewarding function; And step S46, calculating expected rewards for each directed edge in the offshore traffic channel network, and obtaining a global rewards function by weighting and fusing according to the historical track access frequency.
6. The method for predicting a moving target trajectory at sea with intention to guide and environmental constraint according to claim 1, wherein the step S5 comprises the steps of: Step S51, extracting time sequence characteristics representing motion continuity and environment interaction characteristics representing the interaction relation between the target and the environment from the processed structured track data; Step S52, calculating the maximum accumulated rewards from the key nodes of the current offshore traffic channel to each candidate node as rewarding features based on the global rewarding function; Step S53, adding a long-short-time intention label to the input track segment, taking the actual next node after the last observation node as a short-time label, and taking the finally arrived port or anchor as a long-time label; step S54, carrying out balance treatment on the sample unbalance problem of the long-short time intention label by adopting an oversampling technology to generate a synthetic sample; And step S55, training a long-short intention judgment model, and outputting short intention and long intention of the moving target, wherein the short intention is the probability of selecting the key nodes of the next sea traffic channel, and the long intention is the probability distribution of the destination port/anchor.
7. The method for predicting a moving target trajectory at sea with intention to guide and environmental constraint according to claim 1, wherein the step S6 comprises the steps of: step S61, inputting a current observation track segment of a moving target, obtaining short-time intention probability distribution through a trained intention judgment model, and selecting a short-time channel node with highest probability and a long-time port/anchor as intention targets; Step S62, on the sea traffic channel network fused with the rewarding value, planning an optimal path from the current position to the short-time intended target and then from the short-time intended target to the long-time intended target by taking the maximum accumulated rewarding as a criterion, and connecting to obtain a complete planned path; s63, extracting a speed sequence from the historical observation sequence, performing multi-layer discrete wavelet transformation on the speed sequence, and decomposing the speed sequence into an approximate coefficient representing a low-frequency trend and a detail coefficient representing different frequency fluctuation; S64, constructing a path constraint prediction network, inputting a feature vector formed by splicing a wavelet coefficient with a current speed and a path reference point after standardization and flattening, and outputting a speed increment and a course angle of multiple steps in the future; step S65, according to the predicted speed increment and the course angle, the predicted coordinates at the future moment are reconstructed by combining with the kinematic model, and the multi-mode predicted track which accords with the navigation preference of the moving target is output.
8. An offshore moving object trajectory prediction system intended for guidance and environmental constraints, characterized in that it is used for implementing an offshore moving object trajectory prediction method intended for guidance and environmental constraints according to any one of claims 1 to 7, comprising: The channel extraction module is used for extracting the offshore traffic channel by combining the AIS track data set with the island data to exclude the layer and considering the density of track sampling points accessed by the history of the offshore moving target; The environment characteristic statistics module is used for counting speed distribution, course distribution and sea depth distribution characteristics in the neighborhood range of key nodes of the offshore traffic channel and the edges of the offshore traffic channel according to the offshore traffic channel; the track representation module is used for representing the historical track of which each destination is a port or an anchor as a set formed by key nodes and edges of each offshore traffic channel, and dividing the historical track into a training set and a verification set; The behavior simulation learning module is used for constructing a behavior simulation learning model, solving an objective function through a neural network model, outputting rewarding values of each directed node and edge of the offshore traffic channel, and obtaining a global rewarding function through weighting and fusion; The intention judging module is used for training an intention judging model, inputting a track data set, a port or anchor ground set, marine traffic channel statistical characteristics and a global rewarding function, and outputting the key nodes of the marine moving target short-time marine traffic channel and the long-time port or anchor ground behavior probability; The track prediction module is used for judging the short-time and long-time behavior intention of the moving target based on the moving target observation track, the port or anchor set, the marine traffic channel statistical characteristics, the global rewarding function and the intention judgment model, and predicting the moving target speed change by utilizing wavelet transformation decomposition speed sequence under the guidance of the intention, and generating future track coordinates by interpolation.

Description

Offshore moving target track prediction method and system for intention guidance and environment constraint Technical Field The invention relates to the technical field of offshore traffic control and track prediction, in particular to a method and a system for predicting a track of an offshore moving target with intention guidance and environmental constraint. Background The marine moving target track prediction is one of the core technologies of ocean situation awareness and intelligent control decision, and is widely applied to the fields of ocean monitoring, navigation safety, port management and the like. The method can accurately predict the future motion trail of the offshore target, and has important significance for guaranteeing navigation safety, optimizing route planning and preventing offshore risks. However, the motion of moving targets at sea is comprehensively influenced by multiple factors such as sea condition changes, channel habits, port distribution, meteorological conditions and the like, and the lack of fixed structured network constraints has remarkable randomness and uncertainty in behavior, and especially long-time sequence track prediction faces the serious challenges of error accumulation. The existing offshore moving target track prediction methods are mainly divided into two types: One is a prediction method based on impersonation generation. This type of approach assumes that the motion behavior of a moving object follows certain potential rules or constraints, and simulates the future trajectory of the object by mining behavior patterns from expert experience or historical trajectory data, constructing rule models or generating models. Researchers typically abstract interactions between moving objects as attractive or repulsive forces, or model simulated object behavior decisions using cellular automata, probabilistic neural networks, and the like. With the development of deep learning, variations are introduced into track generation tasks from encoders, generation countermeasure networks, attention mechanisms, and the like, and future tracks with interpretability are generated by characterizing spatiotemporal features in learning extraction tracks. However, the method has strong subjectivity on the abstraction of the behavior rules, is difficult to comprehensively describe the real decision process under the complex ocean environment, and the generated track has the problems of authenticity and limited generalization capability. The other is a prediction method based on data learning. The method regards the future track of the moving target as the mapping result of the interaction of the historical behavior and the environmental characteristics, and builds an end-to-end prediction model by mining the motion mode, time sequence dependence and spatial association in the historical track. Common techniques include cyclic neural networks, long and short term memory networks, encoder-decoder structures, gaussian process regression, etc., and some studies introduce beam search, monte carlo sampling, etc. to address the need for multi-modal prediction. The method can model the behavior decision of the target in a high-dimensional space, has strong adaptability, but the prediction performance of the method is highly dependent on the quality and coverage of training data, and as the prediction duration is increased, the problem of error accumulation is increasingly prominent, so that the uncertainty change of future environment and target decision is difficult to effectively realize. In summary, the existing track prediction method has the following common problems in the offshore complex scene that (1) the method based on imitation learning is strong in subjectivity, track authenticity and generalization capability are insufficient, (2) the method based on data learning is high in data quality requirement, long-time sequence prediction error accumulation is serious, and interpretability is weak, (3) the offshore moving target is lack of definite ground object constraint, and is strong in behavior randomness, and the existing method is difficult to accurately capture long-time movement rules and intention changes. Therefore, a high-precision track prediction method capable of integrating environmental constraint and behavior intention and adapting to an offshore random scene is needed. Disclosure of Invention The invention aims to provide a method and a system for predicting a moving target track on the sea, which aim to solve the problems of weak authenticity, weak generalization capability, insufficient interpretability and long time sequence prediction error accumulation of the current track prediction method and effectively solve the problem of long time sequence track prediction error accumulation caused by the future scene randomness and behavior decision uncertainty of the moving target on the sea. In order to achieve the above object, the present invention provides a method fo