Semantic-based dynamic reasoning method and device for ship behavior

Through the dynamic inference method of ship behavior based on semantics, a semantic network is established and converted into a dynamic Bayesian network, which solves the problem of insufficient accuracy in identifying uncertain ship behavior in the existing technology, and achieves a deep understanding of massive data and More accurate behavioral reasoning supports more efficient water traffic management.

CN114548411BActive Publication Date: 2024-11-26WUHAN UNIV OF TECH
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202210131719.3
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2022-02-14
Publication Date
2024-11-26
Estimated Expiration
2042-02-14

AI Technical Summary

Technical Problem

The prior art is difficult to accurately identify and infer uncertain ship behavior from massive ship navigation data. Data-driven methods cannot reflect environmental interference, and methods based on knowledge modeling are insufficiently accurate.

Method used

Using semantic-based dynamic inference method of ship behavior, through the sorting of ship navigation data and factor extraction, a semantic network of ship behavior is established, transformed into a static Bayesian network, and time series information is given to obtain a dynamic Bayesian network to infer the ship's Hide behavior.

Benefits of technology

Effectively reasoning of potential ship behavior from a human cognitive perspective, improve decision support for water traffic management, and improve the accuracy of identification and reasoning.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN114548411B_ABST
    Figure CN114548411B_ABST
Patent Text Reader

Abstract

The present invention discloses a method and device for dynamic reasoning of ship behavior based on semantics, the method comprising: collating ship navigation data and extracting semantic elements; establishing a ship behavior semantic network using abstractly extracted semantic elements, the ship behavior semantic network refers to establishing semantic relations between various semantic elements and instantiating them to realize structured expression of ship behavior semantics; converting the ship behavior semantic network into a static Bayesian network; converting the static Bayesian network into a dynamic Bayesian network by assigning time series information; and outputting the hidden ship behavior inferred by the dynamic Bayesian network. The present invention can infer potential ship behavior from the perspective of human cognition and effectively provide decision support for water traffic management.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention belongs to the technical field of ship behavior recognition, and in particular relates to a semantic-based ship behavior dynamic reasoning method and device. Background Art

[0002] The rapid development of the Global Positioning System and the Automatic Identification System (AIS) has led to an explosive growth in water traffic data, which has the characteristics of large data volume, fast real-time update, multiple data types and high degree of fragmentation. This makes it very difficult to identify and infer uncertain ship behavior from raw data. In addition, due to the human brain's lack of ability to collect, process and analyze information, it is difficult to have a deeper understanding of ships from a large amount of data. Semantics is a new knowledge discovery model based on people's expression habits and knowledge understanding habits, moving from the pure data level to the knowledge level. Therefore, identifying, analyzing and expressing ship behavior from a semantic perspective can better fit the human brain's way of thinking and improve the efficiency of water traffic management.

[0003] At present, the methods for identifying and reasoning ship behaviors mainly include data-driven methods and knowledge-based modeling methods. However, the above two methods have the following shortcomings: First, the data-driven method can only realize the identification of ship behaviors from the data level, but the behavior of ships is constrained by factors such as the navigation environment. It is difficult to reflect the interference of the ship by the surrounding environment only from the data level, and it is impossible to accurately express the behavior implied in the ship's trajectory; second, the knowledge-based modeling method can identify and express ship behaviors from a semantic perspective, but it does not take into account the uncertainty of ship behaviors, so it is not accurate enough in terms of ship identification and reasoning. In summary, how to identify and reason about uncertain ship behaviors from the semantic level from massive ship navigation data is a problem faced by current ship behavior research. Summary of the invention

[0004] In view of the problems existing in the prior art, the present invention provides a semantic-based ship behavior dynamic reasoning method and device to solve the problem of identifying and reasoning the uncertain behavior of ships.

[0005] To achieve the above object, the present invention provides a ship behavior dynamic reasoning method, comprising:

[0006] Ship navigation data collation and element extraction;

[0007] The abstracted and extracted semantic elements are used to establish a ship behavior semantic network; the ship behavior semantic network refers to establishing semantic relations between semantic elements and instantiating them to achieve a structured expression of the ship behavior semantics;

[0008] Convert the established ship behavior semantic network into a static Bayesian network;

[0009] The obtained static Bayesian network is given time series information to obtain a dynamic Bayesian network;

[0010] Output of the dynamic Bayesian network inference of ship hidden behaviors.

[0011] Furthermore, navigation data refers to the data recorded by the ship's automatic identification system, which mainly includes static data and dynamic data. Dynamic data mainly includes the geographical longitude, geographical latitude, ground heading, ground speed and ship heading corresponding to a certain timestamp of the ship; static data mainly includes the ship's unique identifier (MMSI) and ship type.

[0012] Furthermore, feature extraction refers to extracting semantic information from data based on ship navigation data.

[0013] The geographical latitude and longitude data of the ship are extracted as the ship location semantics.

[0014] According to the different hidden behaviors of ships when sailing in different functional waters, the location semantics are specifically divided into general waters, dock waters, anchorage waters and channel waters.

[0015] The timestamp data corresponding to a certain moment of a ship is extracted as time semantics. Time semantics can be further divided into time period and time point semantics. Time point semantics can be further divided into start time semantics and end time semantics.

[0016] All dynamic data corresponding to a ship at a certain moment are extracted as the ship's navigation state semantics, which is formally expressed as S = {S 1 ,S 2 ,…,S m The same navigation state semantics of ships has different semantic behaviors hidden in different location semantics. According to the different location semantics, it is divided into the state semantics of general waters, the state semantics of dock waters, the state semantics of anchorage waters, and the state semantics of waterway waters.

[0017] Furthermore, the state semantics of general waters include the combined state semantics of two adjacent trajectory points inside and outside a certain area, the heading change state semantics, and the speed change semantics. The state semantics of dock waters include the state of being inside the dock, the state of speed being 0, and the state of the ship type matching the dock type. The state semantics of anchorage waters include the state of being inside the anchorage, the state of speed being less than 1 (i.e., the speed being less than the preset speed threshold), the state of right-angle crossing, and the state of entering at a small angle. The state semantics of fairway waters include the state of deviating from the fairway, the state of sailing along the general flow direction of the ship, and the state of keeping away from the dividing line / belt.

[0018] The unique identifier data of the vessel is extracted as vessel semantics.

[0019] The ship type data is extracted as ship type semantics.

[0020] Extract all the ship dynamic and static data recorded at each timestamp in the automatic identification system as trajectory point semantics; extract the ship navigation trajectory point data within a continuous period of time as ship trajectory segment semantics; extract the trajectory segment data as the hidden behavior semantics of the ship, formally expressed as B = {B 1 ,B 2 ,…,B n}.

[0021] According to the different semantics of the ship's navigation location, the ship's hidden behavior mainly includes the hidden behavior in general waters, the hidden behavior in dock waters, the hidden behavior in anchorage waters, and the hidden behavior in waterway waters.

[0022] Furthermore, the hidden behaviors in general waters mainly include entering / leaving, turning, and changing speed. The hidden behaviors in dock waters mainly include berthing / unberthing, and mooring. The hidden behaviors in anchorage waters mainly include anchoring, approaching the channel, merging into the channel, and crossing the channel. The hidden behaviors in channel waters mainly include deviating from the channel, safe navigation, and turning.

[0023] Furthermore, the semantic relations between the semantic elements include: the relation between ship semantics and other semantics, the relation between trajectory segment semantics and trajectory point semantics, the relation between trajectory segment semantics, the relation between trajectory semantic segments and time semantics, the relation between trajectory point semantics and time semantics, the relation between behavior semantics and state semantics, the relation between behavior semantics within a single time slice, the relation between behavior semantics between adjacent time slices, and spatial topological relation.

[0024] Furthermore, the relationship between ship semantics and other semantics is the attribute relationship has State, has Behavior, in Place and is Type. The relationship between trajectory segment semantics and trajectory point semantics is the subordinate relationship has TrajectoryPoint. The relationship between trajectory segment semantics is the precedence relationship has Filiation. The relationship between trajectory segment semantics and time semantics is an attribute relationship, including the attribute relationship between trajectory segment semantics and the sub-semantic start time semantics of time semantics atBegin Time and the attribute relationship between trajectory segment semantics and the sub-semantic end time semantics of time semantics at EndTime. The relationship between trajectory point semantics and time semantics is the attribute relationship at Time. The relationship between behavior semantics and state semantics is an attribute relationship, which mainly refers to the attribute relationship has Characteristics within a certain time slice, abbreviated as hasC. The relationship between behavior semantics within a single time slice is an attribute relationship, which mainly refers to the attribute relationship has Characteristics within a certain time slice, abbreviated as hasC. The relationship between behavior semantics between adjacent time slices is the attribute relationship has Inter-slice Influence, abbreviated as hasIS. There are eight types of spatial topological relationships: separation, overlap, tangency, covering, being covered, containing, being contained, and coincidence.

[0025] The instantiation of the class includes: instantiating the ship semantics, time semantics, position semantics, type semantics, state semantics, behavior semantics, track point semantics and track segment semantics to obtain relevant instance information.

[0026] Furthermore, the transformation of the ship behavior semantic network into a static Bayesian network mainly includes: generating nodes, generating edges, and assigning corresponding probability information to nodes and edges.

[0027] Generating nodes refers to the nodes in the static Bayesian network that correspond to the semantics of states and behaviors in the ship behavior semantic network.

[0028] Generating edges refers to mapping the semantic relationships between semantics in the ship behavior semantic network to directed arcs between nodes in the static Bayesian network.

[0029] Assigning corresponding probability information to nodes and edges means that the nodes correspond to discrete random variables, assign prior probability distribution, and correspond the probability information of the edges to the conditional probability distribution between nodes.

[0030] Furthermore, assigning time series information to the static Bayesian network to obtain a dynamic Bayesian network mainly includes: generating an initial network, generating a transfer network, and assigning corresponding probability information.

[0031] The initial network refers to the static Bayesian network composed of nodes in the time slice at the initial moment.

[0032] The transition network is a Bayesian network containing two adjacent time slices under the assumption that the dynamic Bayesian network conforms to the first-order Markov process.

[0033] The corresponding probability information is mainly assigned to the probability distribution information of the nodes in the initial network and the transition probability information of the transition network.

[0034] Furthermore, reasoning refers to posterior probability reasoning, which is based on the instance information of observable state semantics and uses the state semantic information under the time series to calculate the probability distribution of each hidden behavior in the current time slice, and then infers the hidden behavior with the largest probability value as the hidden behavior in the current time slice.

[0035] For a time series with T time slices, the posterior probability inference can be expressed as:

[0036]

[0037] According to the Bayesian formula:

[0038]

[0039] And the conditional independence assumption of the Bayesian network can be obtained:

[0040]

[0041] Where B represents the hidden behavior of the ship, S represents the navigation state of the ship, m is the number of state elements in the navigation state set, n is the number of behavior elements in the hidden behavior set, u = 1, 2, ..., T, v = 1, 2, ..., m, p = 1, 2, ..., T, q = 1, 2, ..., n, for A value state of is the evidence variable The parent node set of , transforms the reasoning of hidden behavior into the calculation of probability.

[0042] The present invention also provides a semantic-based ship behavior dynamic reasoning device for implementing the above method, comprising a data interface, a power supply, a circuit board, a data processing module, a semantic network generation module, a ship hidden behavior reasoning module and a user interface;

[0043] The data processing module pre-processes the input data of the ship automatic identification system and extracts semantic elements. The semantic network generation module transforms the semantic elements into a semantic network. The ship semantic behavior dynamic reasoning module transforms the semantic network into a dynamic Bayesian network and infers the hidden behavior of the ship.

[0044] The power supply is used to supply power to the device, the circuit board is used to install the data processing module, the semantic network generation module and the ship hidden behavior reasoning module, and the user interface is used to output the hidden behavior of the ship.

[0045] Compared with the prior art, the present invention has the following advantages and beneficial effects:

[0046] The present invention can infer potential ship behaviors from the perspective of human cognition and effectively provide decision support for water traffic management. BRIEF DESCRIPTION OF THE DRAWINGS

[0047] Figure 1 A schematic flow chart of the semantic-based ship behavior dynamic reasoning method provided by the present invention.

[0048] Figure 2 This is a schematic diagram of the ship behavior semantic network provided by the present invention.

[0049] Figure 3 Schematic diagram of ship status in different waters and examples provided by the present invention.

[0050] Figure 4 Schematic diagram of ship behaviors in different waters and examples provided by the present invention.

[0051] Figure 5 A schematic diagram of a process for converting a ship behavior semantic network into a static Bayesian network provided by the present invention.

[0052] Figure 6 A schematic diagram of a process of obtaining a dynamic Bayesian network by assigning time series information to a static Bayesian network provided by the present invention.

[0053] Figure 7 A schematic diagram of the process of reasoning about hidden ship behavior provided by the present invention.

[0054] Figure 8 A schematic diagram of the semantics-based ship behavior dynamic reasoning device provided by the present invention. DETAILED DESCRIPTION

[0055] In order to make the purpose, technical solutions and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention. In addition, the technical features involved in the various embodiments of the present invention described below can be combined with each other as long as they do not conflict with each other.

[0056] The semantic-based dynamic reasoning method of ship behavior in the embodiment of the present invention is as follows: Figure 1 As shown, the following steps are included:

[0057] Step S1, collating ship navigation data and extracting elements;

[0058] Step S2, abstracting the extracted semantic elements to establish a ship behavior semantic network;

[0059] Step S3, converting the established ship behavior semantic network into a static Bayesian network;

[0060] Step S4, assigning time series information to the established static Bayesian network to obtain a dynamic Bayesian network;

[0061] Step S5, outputting the hidden behavior of the ship inferred by the dynamic Bayesian network.

[0062] Among them, the navigation data in step S1 refers to the data recorded by the ship automatic identification system, which mainly includes static data and dynamic data. The dynamic data mainly includes the geographical longitude, geographical latitude, ground heading, ground speed and ship heading corresponding to a certain timestamp of the ship. The static data mainly includes the ship's unique identifier MMSI and the ship type.

[0063] Feature extraction refers to extracting semantic information from data based on ship navigation data:

[0064] The geographical latitude and longitude data of the ship are extracted as the ship location semantics.

[0065] According to the different hidden behaviors of ships when sailing in different functional waters, the location semantics are specifically divided into general waters, dock waters, anchorage waters and channel waters.

[0066] The trajectory segment data is extracted as the hidden behavior semantics of the ship, which is formally expressed as B = {B 1 ,B 2 ,…,B n According to the different semantics of the ship's location, the hiding behavior can be divided into the hiding behavior in general waters, the hiding behavior in dock waters, the hiding behavior in anchorage waters, and the hiding behavior in fairway waters.

[0067] The hidden behavior of the ship is extracted by combining the kinematic characteristics of the ship and the provisions on ship behavior in the International Regulations for Preventing Collisions at Sea, 1972. Figure 4 Come and see, Figure 4 Schematic diagram of ship behaviors in different waters and examples provided by the present invention.

[0068] Specifically, the hidden behaviors of general waters include entry / exit (E / L), turn (Tu), and speed change (SC). More specifically, entry / exit includes entry (E), exit (L), non-entry and exit Turning includes turning left (TS), turning right (TP), and going straight / stopping (GS / S). Speed ​​changing includes accelerating (SU), decelerating (SD), and sailing at a constant speed / stopping (R / S).

[0069] Hidden behaviors in the dock waters include berthing / unberthing and mooring. More specifically, berthing / unberthing includes berthing (Ar), unberthing (De), and non-berthing / unberthing. The “-” before a letter means “negative”.

[0070] Hidden behaviors in anchorage waters include anchoring (An), approaching the channel (Ap), merging into the channel (J), and crossing the channel (Cr).

[0071] Hidden behaviors in the waterway include deviation from the channel (D), safe navigation (isS), and turn to (STto). More specifically, deviation from the channel includes deviation to port (DtoP), deviation to starboard (DtoS), and no deviation. Safe navigation includes safe navigation behavior (isSafe) and unsafe navigation behavior (isUnSafe). The required turns include turn to port (STP), turn to starboard (STS), and keep straight (SGS).

[0072] The timestamp data corresponding to a certain moment of a ship is extracted as time semantics. Time semantics can be further divided into time period and time point semantics. Time point semantics can be further divided into start time semantics and end time semantics.

[0073] All dynamic data corresponding to a ship at a certain moment are extracted as the ship's navigation state semantics, which is formally expressed as S = {S 1 ,S 2 ,…,S m}. Combine Figure 3 Come and see, Figure 3 The ship status semantics and example diagrams of different waters provided by the present invention. The status representation refers to the set of ship status semantics at a specific moment. According to the location of the ship, the status can be divided into the status of general waters, the status of dock waters, the status of anchorage waters, and the status of waterway waters.

[0074] Specifically, the general water state includes the combined state semantics of two adjacent trajectory points inside and outside a certain area (i&o), the heading change state semantics (dc), and the speed change state semantics (sv). More specifically, the combined state of two adjacent trajectory points in the general water area inside and outside a certain area includes the combined state of inside and inside (i^i), the combined state of inside and outside (i^o), the combined state of outside and inside (o^i), and the combined state of outside and outside (o^o). The general water heading change state semantics include turning left (tp), turning right (ts), and going straight / stopping (gs / s). The general water speed change state semantics include speed increase (su), speed decrease (sd), and constant speed / stopping (r / s).

[0075] The semantics of the dock water area state include the state within the dock (Qu), the state of zero speed (sp=0), and the state that the ship type matches the dock type (Ty).

[0076] The semantics of anchorage water state include the state in anchorage (Anc), the state of speed less than 1 (sp<1), the state of right-angle crossing (RA), and the state of small-angle approach (SA).

[0077] The semantics of the waterway state include the state of deviating from the waterway (dev), the state of sailing along the general flow direction of the ship (inGD), and the state of keeping away from the separation line / belt (KC).

[0078] More specifically, according to Article 10.2 of the International Regulations for Preventing Collisions at Sea, 1972, which stipulates that "the ship shall sail in the corresponding traffic lane along the general flow of ships in that lane", this state refers to the ship sailing along the general flow of ships in the lane. The state of sailing along the general flow of ships is semantically divided into sailing along the general flow of ships I (inGDI), sailing along the general flow of ships II (inGDII), sailing along the general flow of ships III (inGDIII) and sailing along the general flow of ships IV (inGDIV).

[0079] According to Article 9.1 of the 1972 International Regulations for Preventing Collisions at Sea, "a vessel proceeding in a narrow channel or fairway shall, whenever safe and practicable, keep as close as possible to the outer edge of that channel or fairway on her starboard side" and Article 10.2(2) of the 1972 International Regulations for Preventing Collisions at Sea, "keep as far as practicable from separating lines or strips". The Keep Away Separation Line / Strip state semantics indicates the degree to which a vessel keeps away from separating lines or strips. The Keep Away Separation Line / Strip state is divided into Keep Away Separation Line / Strip I (KCI), Keep Away Separation Line / Strip II (KCII), and Keep Away Separation Line / Strip III (KCIII).

[0080] Among them, step S2 of the ship behavior semantic network refers to establishing the semantic relationship between the semantic elements and instantiating them to achieve a structured expression of the ship behavior semantics. Figure 2 Come and see, Figure 2This is a schematic diagram of the ship behavior semantic network provided by the present invention.

[0081] The semantic relations mainly include: the relation between ship semantics and other semantics, the relation between trajectory segment semantics and trajectory point semantics, the relation between trajectory segment semantics, the relation between trajectory segment semantics and time semantics, the relation between trajectory point semantics and time semantics, the relation between behavior semantics and state semantics, the attribute relation between behavior semantics within a single time slice, the relation between behavior semantics between adjacent time slices, and spatial topological relation.

[0082] Specifically, the relationship between ship semantics and other semantics is an attribute relationship, including has State, has Behavior, in Place, and is Type.

[0083] The relationship between trajectory segment semantics and trajectory point semantics is the subordinate relationship has Trajectory Point.

[0084] The semantic relationship between trajectory segments is the has Filiation relationship.

[0085] The relationship between the track segment semantics and the time semantics is an attribute relationship, including the attribute relationship between the track segment semantics and the sub-semantics of the time semantics, the start time semantics at Begin Time, and the attribute relationship between the track segment semantics and the sub-semantics of the time semantics, the end time semantics at End Time.

[0086] The relationship between trajectory point semantics and time semantics is the attribute relationship at Time.

[0087] The relationship between behavioral semantics and state semantics is the attribute relationship has Characteristics, abbreviated as hasC.

[0088] The relationship between behavioral semantics within a single time slice is the attribute relationship has Characteristics, abbreviated as hasC.

[0089] The relationship between the behavior semantics between adjacent time slices is the attribute relationship has Inter-slice Influence, abbreviated as hasIS.

[0090] The spatial topological relationship consists of RCC8, which expresses 8 spatial topological relationships in the regional connectivity calculus theory, including separation (DC), overlap (PO), tangency (EC), covering (TPP-), covered (TPP), containing (NTPP-), contained (NTPP) and coincidence (EQ).

[0091] Among them, step S2 "instantiation of semantics" refers to the instantiation of ship semantics, time semantics, position semantics, type semantics, state semantics, behavior semantics, trajectory point semantics and trajectory segment semantics in the ship behavior semantics to obtain relevant instance information.

[0092] Specifically, the instantiation of ship semantics refers to mapping a ship to a specific ship unique identifier MMSI.

[0093] The instantiation of time semantics refers to assigning time points and time periods to accurate real-world time, including the start and end times and a specific time, such as 2017-06-15T23:13:30+08:00.

[0094] The instantiation of location refers to the correspondence of location semantics to specific general waters, anchorages, docks and waterway areas. Such specific locations have their own names and longitude and latitude ranges, such as Anchorage No. 1.

[0095] The instantiation of ship state semantics refers to identifying the corresponding ship state from trajectory data, which is divided into the following parts:

[0096] The longitude and latitude coordinates of the WGS-84 coordinate system in the automatic identification system are converted into plane coordinates using the Mercator projection method.

[0097] Status identification of docks, anchorages and general waters. Specifically, the position of the ship is regarded as a point, and the ray method is used to identify the topological relationship between the ship and the dock and anchorage waters, and then the topological state semantics of the ship and the navigation waters are obtained. The main principle of the ray method is to draw a ray from the trajectory point to the left and calculate the number of edges crossed by the ray. If the number of edges crossed is an odd number, the trajectory point is within the area, and if the number of edges crossed is an even number, the trajectory point is outside the area.

[0098] The semantic recognition of heading change state adopts the cross product formula based on two-dimensional vector: is the front line vector of the adjacent trajectory segment, is the back segment vector. If z is a positive number, it means arrive If it is counterclockwise, it means the heading state semantics is left turn. If it is a negative number, it means the heading state semantics is right turn. If the result is 0, it means the heading state is straight ahead.

[0099] The speed change state semantics recognition uses acceleration judgment. If the acceleration is positive, it is the speed increase state semantics, if it is negative, it is the speed decrease state semantics, and if it is 0, it is the constant speed navigation state semantics.

[0100] The semantics of the "right-angle crossing" and "small-angle entry" states are determined by the angle between the ship's heading and the channel boundary. The "right-angle crossing" state means that the acute angle between the ship's heading and the channel boundary is between 60° and 90°, and the "small-angle entry" means that the acute angle between the ship's heading and the channel boundary is between 0° and 60°.

[0101] The state semantics of "speed is less than 1" and "speed is 0" can be determined by the ship speed being less than 1 knot and the speed being less than 0.

[0102] The state semantics of "sailing along the general flow direction of the ship" is identified by calculating the angle between the ship's heading to the ground and the general flow direction of the ship. There are four types in total. The state semantics of sailing along the general flow direction of the ship are divided into sailing along the general flow direction of the ship I (abbreviated as inGDI), sailing along the general flow direction of the ship II (abbreviated as inGDII), sailing along the general flow direction of the ship III (abbreviated as inGDIII), and sailing along the general flow direction of the ship IV (abbreviated as inGDIV). inGDⅠ means that the angle between the ship's heading and the general flow direction of the ship is 30° to 90° to the left. inGDⅡ means 0 to 30° to the left. inGDⅢ means 0 to 30° to the right, and inGDⅣ means 30° to 90° to the right.

[0103] The "keep away from the dividing line / strip" state is identified by calculating the distance between the ship's position and the channel dividing strip / line. There are three types of "keep away from the dividing line / strip" state semantics: KCI, KCII, and KCIII. KCIⅠ refers to the navigation range from the left boundary to the right boundary of the channel with the ship's navigation direction as the front, and from 3 / 4 of the channel width to the right boundary as the positive direction. KCⅡ refers to the navigation range from 1 / 2 to 3 / 4 of the channel width. KCⅢ refers to the navigation range from the left boundary to 1 / 2 of the channel width. According to the International Regulations for Preventing Collisions at Sea, when a ship is navigating in a channel, it should try to sail on the right side of the channel and keep away from the channel dividing line / strip. Therefore, when a ship is sailing within the range of KCⅡ and KCⅢ, it needs to try to sail within the range of KCⅠ.

[0104] The deviation length (DL) is the track length between the current position of the ship and the position when it crosses the channel boundary according to the current trajectory. When the deviation length exceeds the threshold, the track point is considered to have the "deviation from channel" state semantics.

[0105] Among them, step S3 "converting the ship behavior semantic network into a static Bayesian network" mainly includes step S31, step S32, and step S33. Figure 5 As shown, Figure 5 A schematic diagram of a process for converting a ship behavior semantic network into a static Bayesian network provided by the present invention.

[0106] Step S31 is mainly to generate nodes, that is, to correspond the semantics of states and behaviors in the ship behavior semantic network to nodes in the Bayesian network;

[0107] Step S32 is mainly to generate edges, that is, to correspond the semantic relationships between semantics in the ship behavior semantic network to directed arcs between nodes in the Bayesian network;

[0108] Step S33 is mainly to assign corresponding probability information to nodes and edges, that is, to correspond nodes to discrete random variables, assign prior probability distribution, and correspond the probability information of edges to conditional probability distribution between nodes;

[0109] Specifically, both the prior probability distribution and the conditional probability distribution are obtained through the maximum likelihood estimation method.

[0110] Among them, the step S4 of "adding time series information to the static Bayesian network to obtain a dynamic Bayesian network" refers to establishing a Bayesian network under the time slice sequence, which mainly includes steps S41, S42, and S43. Figure 6 As shown, Figure 6 A schematic diagram of a process of obtaining a dynamic Bayesian network by assigning time series information to a static Bayesian network provided by the present invention.

[0111] Step S41 is mainly to generate an initial network, which refers to the distribution of nodes at the initial moment, and the distribution refers to the probability of each node occurring at the initial moment;

[0112] Step S42 is mainly to generate a transfer network, that is, a Bayesian network containing two adjacent time slices under the assumption that the dynamic Bayesian network conforms to the first-order Markov process, that is, the probability of the hidden behavior in one time slice being transferred to the corresponding hidden behavior in another time slice;

[0113] Step S43 is mainly to assign corresponding probability information, mainly the probability information of node distribution in the initial network and the probability information of the transfer network.

[0114] More specifically, the probability information is obtained through the maximum likelihood estimation method.

[0115] The "hidden behavior" in step S5 refers to the ship behavior that is difficult to obtain directly through observation data. It is the potential behavior of the ship hidden under the observation data. It is necessary to infer the hidden behavior of the ship through the state semantic information of the ship.

[0116] Specifically, “inference of ship hidden behavior” is mainly divided into steps S51 and S52. Figure 7 As shown, Figure 7 A schematic diagram of the process of reasoning about hidden ship behavior provided by the present invention.

[0117] Step S51 is mainly for parameter learning. The parameters are the conditional probability distribution parameters between hidden behaviors and state semantics in the Bayesian network within a single time slice, the prior probability distribution parameters of each node in the initial network, and the parameters in the transition probability distribution between dynamic Bayesian networks under time series. The maximum likelihood estimation method is used for parameter learning to obtain the probability distribution under the corresponding parameters.

[0118] Step S52 is mainly the reasoning of the posterior probability, which is mainly based on the instance information of the observable state semantics. The Bayesian formula is used to calculate the probability distribution of each hidden behavior in the current time slice, and then the hidden behavior with the largest probability value is taken as the hidden behavior in the current time slice. The dynamic Bayesian network under the time series is used to infer the hidden behavior of the ship pressed in continuous time.

[0119] Specifically, for a time series with T time slices, the posterior probability inference can be expressed as:

[0120]

[0121] According to the Bayesian formula:

[0122]

[0123] And the conditional independence assumption of the Bayesian network can be obtained:

[0124]

[0125] Where B represents the hidden behavior of the ship, S represents the navigation state of the ship, m is the number of state elements in the state set, n is the number of behavior elements in the hidden behavior set, u = 1, 2, ..., T, v = 1, 2, ..., m, p = 1, 2, ..., T, q = 1, 2, ..., n, for A value state of is the evidence variable The parent node set of , finally converts the behavior reasoning into probability calculation to obtain the ship hidden behavior.

[0126] The present invention also discloses a semantic-based ship behavior dynamic reasoning device, such as Figure 8As shown, it is composed of a power supply, a circuit board, a data interface, a user interface, a data processing module, a semantic network generation module, and a behavior reasoning module. First, the data processing module performs data sorting and semantic element extraction on the input automatic identification system data. Then, the obtained semantic elements are input into the semantic network generation module to obtain the ship behavior semantic network. Furthermore, the obtained ship behavior semantic network is input into the behavior reasoning module for semantic network transformation and dynamic Bayesian network generation. Finally, the hidden behavior of the ship is inferred, and the hidden behavior of the ship is output by the user interface. The power supply supplies power to the device. The circuit board integrates the data processing module, the semantic network generation module and the behavior reasoning module. The data interface is used to introduce exogenous AIS data.

[0127] It will be easily understood by those skilled in the art that the above are only preferred embodiments of the present invention and are not intended to limit the present invention. Any modifications, equivalent substitutions and improvements made within the spirit and principles of the present invention should be included in the protection scope of the present invention.

Claims

1. A semantics-based ship behavior dynamic reasoning method, characterized in that: The following steps are involved: Ship navigation data collation and semantic element extraction; The abstracted and extracted semantic elements are used to establish a ship behavior semantic network; the ship behavior semantic network refers to establishing semantic relations between semantic elements and instantiating them to achieve a structured expression of the ship behavior semantics; Convert the established ship behavior semantic network into a static Bayesian network; The obtained static Bayesian network is given time series information to obtain a dynamic Bayesian network; Inferring hidden ship behaviors using dynamic Bayesian networks; Among them, ship navigation data refers to the data recorded by the ship's automatic identification system, including static data and dynamic data; dynamic data includes the geographical longitude, geographical latitude, ground heading, ground speed and ship heading corresponding to a certain timestamp of the ship; static data includes the ship's unique identifier and ship type; Feature extraction refers to extracting the respective semantic information in the data based on the ship navigation data; the geographical longitude and latitude data of the ship are extracted as the ship location semantics; according to the different hidden behaviors of ships when sailing in different functional waters, the location semantics are divided into general waters, dock waters, anchorage waters and channel waters; the timestamp data corresponding to a ship at a certain moment is extracted as time semantics; time semantics are divided into time period and time point semantics; time point semantics are divided into start time semantics and end time semantics; all dynamic data corresponding to a ship at a certain moment are extracted as ship navigation status semantics; the unique identifier data of the ship is extracted as ship semantics; the type data of the ship is extracted as ship type semantics; all dynamic and static data of the ship recorded at each timestamp in the automatic identification system are extracted as track point semantics; the ship navigation track point data within a continuous period of time is extracted as ship track segment semantics; the track segment data is extracted as the hidden behavior semantics of the ship; The semantic behaviors hidden in different location semantics of the same navigation state semantics of ships are obviously different. According to the different location semantics, they are divided into the state semantics of general waters, the state semantics of dock waters, the state semantics of anchorage waters, and the state semantics of waterway waters; the state semantics of general waters include the combined state semantics of two adjacent trajectory points inside and outside a certain area, the state semantics of heading change, and the state semantics of speed change; the state semantics of dock waters include the state of being in the dock, the state of speed being 0, and the state of the ship type being consistent with the dock type; the state semantics of anchorage waters include the state of being in the anchorage, the state of speed being less than the preset threshold, the state of right-angle crossing, and the state of entering at a small angle; the state semantics of waterway waters include the state of deviating from the waterway, the state of sailing along the general flow direction of the ship, and the state of leaving the dividing line / belt; According to the different semantics of the ship's navigation location, the ship's hidden behavior includes hidden behavior in general waters, hidden behavior in dock waters, hidden behavior in anchorage waters, and hidden behavior in fairway waters; Hidden behaviors in general waters include entering / leaving, turning, and changing speed; hidden behaviors in dock waters include arriving / leaving, and mooring; hidden behaviors in anchorage waters include anchoring, approaching the waterway, merging into the waterway, and crossing the waterway; hidden behaviors in waterway waters include deviating from the waterway, safe navigation, and turning; Reasoning refers to posterior probability reasoning, which is based on the instance information of observable state semantics, and uses the state semantic information under the time series to calculate the probability distribution of each hidden behavior in the current time slice, and then infers the hidden behavior with the largest probability value as the hidden behavior in the current time slice: For a time series with T time slices, the posterior probability inference can be expressed as: According to the Bayesian formula: And the conditional independence assumption of the Bayesian network: Where B represents the hidden behavior of the ship, S represents the navigation state of the ship, m is the number of navigation state elements in the navigation state set, n is the number of behavior elements in the hidden behavior set, u = 1, 2, ..., T, v = 1, 2, ..., m, p = 1, 2, ..., T, q = 1, 2, ..., n, for A value state of is the evidence variable The parent node set of , transforms the reasoning of hidden behavior into the calculation of probability.

2. The method for dynamic reasoning of ship behavior based on semantics according to claim 1, characterized in that: The semantic relations between the semantic elements include: the relation between ship semantics and other semantics, the relation between track segment semantics and track point semantics, the relation between track segment semantics, the relation between track semantic segments and time semantics, the relation between track point semantics and time semantics, the relation between behavior semantics and state semantics, the relation between behavior semantics within a single time slice, the relation between behavior semantics between adjacent time slices, and spatial topological relations; Instantiation includes instantiating the ship semantics, time semantics, location semantics, type semantics, state semantics, behavior semantics, track point semantics and track segment semantics to obtain relevant instance information.

3. The semantic-based ship behavior dynamic reasoning method according to claim 1 is characterized in that: The conversion of ship behavior semantic network into static Bayesian network includes: generating nodes, generating edges, and assigning corresponding probability information to nodes and edges; Among them, generating nodes means corresponding the semantics of states and behaviors in the ship behavior semantic network to nodes in the static Bayesian network; generating edges means corresponding the semantic relationships between semantics in the ship behavior semantic network to directed arcs between nodes in the static Bayesian network; assigning corresponding probability information to nodes and edges means corresponding nodes to discrete random variables, assigning prior probability distributions, and corresponding the probability information of edges to conditional probability distributions between nodes.

4. The semantic-based ship behavior dynamic reasoning method according to claim 1 is characterized in that: The static Bayesian network is given time series information to obtain a dynamic Bayesian network, including: generating an initial network, generating a transfer network and giving corresponding probability information; Among them, the initial network refers to the static Bayesian network composed of nodes in the time slice at the initial moment; the transfer network is a Bayesian network containing two adjacent time slices under the assumption that the dynamic Bayesian network conforms to the first-order Markov process; the corresponding probability information is the probability distribution information of the nodes in the initial network and the transfer probability information of the transfer network.

5. A semantics-based ship behavior dynamic reasoning device for implementing the semantics-based ship behavior dynamic reasoning method according to any one of claims 1 to 4, characterized in that: include: Power supply, circuit board, data interface, user interface, data processing module, semantic network generation module, behavior reasoning module; Firstly, the data processing module sorts the input ship navigation data and extracts semantic elements. Then, the obtained semantic elements are input into the semantic network generation module to obtain the ship behavior semantic network. Then, the obtained ship behavior semantic network is input into the behavior reasoning module for semantic network transformation and dynamic Bayesian network generation. Finally, the hidden behavior of the ship is inferred and outputted from the user interface. The power supply provides power for the device, the circuit board integrates a data processing module, a semantic network generation module and a behavior reasoning module, and the data interface is used to introduce external source ship navigation data.