CN116805172A - AIS data-driven crude oil ship destination port prediction method - Google Patents
AIS data-driven crude oil ship destination port prediction method Download PDFInfo
- Publication number
- CN116805172A CN116805172A CN202310505090.9A CN202310505090A CN116805172A CN 116805172 A CN116805172 A CN 116805172A CN 202310505090 A CN202310505090 A CN 202310505090A CN 116805172 A CN116805172 A CN 116805172A
- Authority
- CN
- China
- Prior art keywords
- ship
- lat
- lon
- port
- information extraction
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Granted
Links
- 238000000034 method Methods 0.000 title claims abstract description 33
- 239000010779 crude oil Substances 0.000 title claims abstract description 22
- 238000000605 extraction Methods 0.000 claims abstract description 67
- 230000004927 fusion Effects 0.000 claims abstract description 23
- 238000010276 construction Methods 0.000 claims abstract description 18
- 230000006399 behavior Effects 0.000 claims abstract description 16
- 239000013598 vector Substances 0.000 claims description 33
- 239000011159 matrix material Substances 0.000 claims description 17
- 229940050561 matrix product Drugs 0.000 claims description 10
- 101100182247 Caenorhabditis elegans lat-1 gene Proteins 0.000 claims description 6
- 101100182248 Caenorhabditis elegans lat-2 gene Proteins 0.000 claims description 6
- 101100021570 Caenorhabditis elegans lon-3 gene Proteins 0.000 claims description 6
- 101150061388 LON1 gene Proteins 0.000 claims description 6
- 238000013528 artificial neural network Methods 0.000 claims description 6
- 238000004364 calculation method Methods 0.000 claims description 6
- 101150004293 lon2 gene Proteins 0.000 claims description 6
- 238000007499 fusion processing Methods 0.000 claims description 3
- 238000013507 mapping Methods 0.000 claims description 3
- 238000010606 normalization Methods 0.000 claims description 3
- -1 utc2 Proteins 0.000 claims description 2
- 230000001717 pathogenic effect Effects 0.000 claims 1
- 230000000694 effects Effects 0.000 description 4
- 238000010586 diagram Methods 0.000 description 3
- 230000009286 beneficial effect Effects 0.000 description 2
- 239000003921 oil Substances 0.000 description 2
- 230000006978 adaptation Effects 0.000 description 1
- 238000004422 calculation algorithm Methods 0.000 description 1
- 238000006243 chemical reaction Methods 0.000 description 1
- 238000004590 computer program Methods 0.000 description 1
- 125000004122 cyclic group Chemical group 0.000 description 1
- 238000006073 displacement reaction Methods 0.000 description 1
- 238000011156 evaluation Methods 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 238000007781 pre-processing Methods 0.000 description 1
- 238000007637 random forest analysis Methods 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q10/00—Administration; Management
- G06Q10/04—Forecasting or optimisation specially adapted for administrative or management purposes, e.g. linear programming or "cutting stock problem"
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02A—TECHNOLOGIES FOR ADAPTATION TO CLIMATE CHANGE
- Y02A90/00—Technologies having an indirect contribution to adaptation to climate change
- Y02A90/10—Information and communication technologies [ICT] supporting adaptation to climate change, e.g. for weather forecasting or climate simulation
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Business, Economics & Management (AREA)
- Economics (AREA)
- General Physics & Mathematics (AREA)
- Human Resources & Organizations (AREA)
- Strategic Management (AREA)
- Entrepreneurship & Innovation (AREA)
- Biomedical Technology (AREA)
- Quality & Reliability (AREA)
- Tourism & Hospitality (AREA)
- Marketing (AREA)
- General Business, Economics & Management (AREA)
- Game Theory and Decision Science (AREA)
- Development Economics (AREA)
- Health & Medical Sciences (AREA)
- Life Sciences & Earth Sciences (AREA)
- Artificial Intelligence (AREA)
- Operations Research (AREA)
- Biophysics (AREA)
- Computational Linguistics (AREA)
- Data Mining & Analysis (AREA)
- Evolutionary Computation (AREA)
- General Health & Medical Sciences (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- General Engineering & Computer Science (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The invention discloses an AIS data-driven crude oil ship destination port prediction method, which comprises the following steps of S1, determining a ship navigation track based on preprocessed AIS historical data, and constructing a ship space information extraction network according to the ship navigation track; the ship navigation track comprises ship berthing in and out behaviors and ship berthing attribution; s2, based on the preprocessed AIS historical data, fusing the ship type, the ship length, the ship load and the ship construction year, and constructing a ship preference information extraction network; and S3, fusing the ship space information extraction network, the ship preference information extraction network and the ship dynamic information extraction network, constructing an information fusion network, and predicting the destination port based on the information fusion network. The advantages are that: the problem of the low accuracy of the prediction result caused by the lack of spatial information and ship transportation preference information in the crude oil ship transportation destination port prediction process is solved.
Description
Technical Field
The invention relates to the technical field of destination port prediction, in particular to an AIS data-driven crude oil ship destination port prediction method.
Background
The existing destination port prediction method mainly takes the original destination port prediction in a local area as a main part, and uses random forests, naive Bayes and a cyclic neural network to realize the prediction, so that the crude oil transportation ship destination port prediction cannot be effectively carried out in a global scope; the identification of space information and ship transportation preference is lacked, and the prediction accuracy is low.
Disclosure of Invention
The invention aims to provide an AIS data-driven crude oil ship destination port prediction method, so as to solve the problems in the prior art.
In order to achieve the above purpose, the technical scheme adopted by the invention is as follows:
an AIS data-driven crude oil ship destination port prediction method comprises the following steps,
s1, determining a ship navigation track based on preprocessed AIS historical data, and constructing a ship space information extraction network according to the ship navigation track; the ship navigation track comprises ship berthing in and out behaviors and ship berthing attribution;
s2, based on the preprocessed AIS historical data, fusing the ship type, the ship length, the ship load and the ship construction year, and constructing a ship preference information extraction network;
and S3, fusing the ship space information extraction network, the ship preference information extraction network and the ship dynamic information extraction network, constructing an information fusion network, and predicting the destination port based on the information fusion network.
Preferably, the vessel berthing and out behavior is determined by the following steps,
a11 AIS track set T with length N from given single ship according to speed i，N Track points with the ship speed smaller than the preset section are screened out to form a track set T with the length of m i，m ；
Wherein T is i，N ＝{t 1 (u，s，c，lon，lat)，t 2 (u，s，c，lon，lat)，...，t N (u, s, c, lon, lat) }; u is world universal time, utc; s is the speed, lon is the longitude, lat is the latitude, and i is the independent unique id named for the ship; t (T) i，m ＝{t 1 (u，s，c，lon，lat)，t 2 (u，s，c，lon，lat)，...，t m (u，s，c，lon，lat)}；
A12, calculating the set T in turn i，m The time difference and the distance between the two track points; when the time difference between the two points is smaller than the preset time threshold value and the distance between the two points is smaller than the preset distance threshold value, the track point t 1 And locus point t 2 May be clustered into one cluster;
the time difference between two trace points is calculated as,
utcDifference＝u 2 -u 1
wherein utcDifference is utc time difference; u (u) 1 Is the 1 st track point t 1 Corresponding time; u (u) 2 Is the 2 nd track point t 2 Corresponding time;
the distance between two track points is calculated by the formula,
distance(lon 2 ，lat 2 ，lon 1 ，lat 1 )
＝R 0 *arccos(cos(lat 1 )
*coS(lat 2 )*cos(lon 2 -lon 1 )+sin(lat 2 )*sin(lat 1 ))
wherein lon 1 ，lat 1 Is the 1 st track point t 1 Corresponding longitude and latitude; lon (lon) 2 ，lat 2 Respectively the 2 nd track point t 2 Corresponding longitude and latitude; r is R 0 Is the earth radius;
a13, based on the judgment rule in A12, collecting T i，m Dividing into sets of subsets
T i，m，sub ＝{sub 1 (t 1 (u，s，c，lon，lat)，t 2 (u，s，c，lon，lat))，...，sub K (t K (u，s，c，lon，lat)，t K+1 (u，s，c，lon，lat))，
Wherein sub is a subset; k is the total number of subsets, each subset representing a stop state of the vessel.
Preferably, the determination of the berthing attribution of the ship is that,
a21, assuming that the stopped states of the two ships are S respectively a = (u 1, lon1, lat1, utc2, lon2, lat 2) and S b = (u 3, lon3, lat3, utc4, lon4, lat 4) and all at the first berth B 1 = (blon 1, blat 1) and second berth B 2 Vicinity of = (blon 2, blat 2);
wherein u1 and u3 are the starting moments of the two ship stopping states respectively; u2 and u4 are the end times of the two ship stop states respectively; lon1, lat1, lon3 and lat3 are longitude and latitude corresponding to u1 and u3 respectively; lon2, lat2, lon4 and lat4 are longitude and latitude corresponding to u2 and u4 respectively; blon1 and blat1 are the longitude and latitude of the first berth respectively; blon2 and blat2 are the longitude and latitude of the second berth respectively;
a22, respectively calculate S a 、S b And B is connected with 1 、B 2 If u1 is smaller than u3, and distance (S a ，B 1 )<distance(S b ，B 1 ) Between u1 and u2, the first berth B 1 Quilt S a Occupied S b Should rest at the second berth B 2 。
Preferably, the fusion process in step S2 is specifically,
according to IMO matching ship type, ship length, ship load and ship construction year data, respectively classifying the ship type and the ship length and identifying the ship type and the ship length by binary codes; mapping between 0 and 1 is performed by maximum and minimum normalization on ship load and ship construction year.
Preferably, in step S3, the specific procedure of destination port prediction based on the information fusion network is that,
s31, extracting a vector containing 504 elements according to the last berthing record of the ship based on the ship space information extraction network;
s32, extracting each attribute of the ship as a 256-dimensional vector based on the ship preference information extraction network, and fusing different attributes through the linear neural network layer to form a 504-dimensional vector;
s33, extracting real-time track data of the ship based on the ship dynamic information extraction network;
s34, fusing the three extracted data to obtain a vector with 504 dimensions;
s35, calculating the cosine distance between the vector obtained in S34 and the 504-dimensional one-hot encoding vector of the port, wherein the port corresponding to the minimum cosine distance is the destination port.
Preferably, the extracting process in step S31 specifically includes constructing a set S of ship berthing time and place based on the behavior of the ship going in and out of berthing and the attribution of the berthing, and extracting information from the set S:
for s= { S 1 ((u a ，p m )，...，(u b ,p k ))，…，s n ((u c ,p o )，...，(u d ,p q ) Each element s) in } t ，And->Forming an OD pair, wherein O is used as a starting point, and D is used as a destination; the set S is de-duplicated to form a node set V= { V 1 ，v 2 ，…，v n All OD pairs in set S get edge set +.> The number of vessels sailing from O to D constitutes the weight of a certain node +.> The matrix product of V and W forms a space information matrix to realize the extraction of ship space information；
Wherein u is a And u b Respectively, vessels s 1 P m And p k Is a ship s 1 Different harbors of the ship; u (u) c And u d Respectively, vessels s n Is provided for the different parking times; p is p o And p q Is a ship s n Different harbors of the ship; v n Is the nth node, namely the nth port;to connect the nth node and the edge of the n-1 th node +.>Weights for the edge;
preferably, the extraction process of step S32 is specifically,
s321, for a given crude oil tanker set z= { Z 1 (shipType，weight,year，length)，...，z m (ship, weight, year, length) }, dividing the set Z into three parts, respectively a ship embedded matrixShip characteristic embedding matrix-> Harbor embedded matrix>
Wherein, shipType, weight, year, length are respectively the ship type, ship load, ship construction year and ship length; d is the potential embedding dimension; m is m ship ，l feature And n port The number of vessel, port and vessel properties, respectively; u (u) j ，p j And f j The j-th row in the matrixes U, P and F respectively; r is the real space;
s322, calculate p j And f j Matrix product between each feature of (a), i.e. for a given p j ＝{a 1 ，a 2 ，…，a d Sum ofThe matrix product of the two is +.> Calculation of p j And f j All matrix products between>
S323, p j And f j The matrix product between is added to u j In (u) j And p j *f j Forming a tanker preference vector x j ，x j ＝u j +p j *f j ；
S324、x j And p j The matrix inner product between them is taken as the final output of the ship preference information extraction network, output=p j *x j 。
Preferably, the loss function of the information fusion network is a cross entropy loss function,
wherein k is a class; r is the total number of categories, and is the independent thermal coding of the port in the ship destination port prediction; h is a k A true tag for the kth category; g k Is the Softmax probability corresponding to the kth category.
The beneficial effects of the invention are as follows: 1. through AIS data, the ship is subjected to position tracking after the ship starts from the oil loading port, and the destination port is predicted, so that the problem of low accuracy of a prediction result caused by the lack of spatial information and ship transportation preference information in the crude oil ship transportation destination port prediction process is solved. 2. The ship space information extraction network and the ship preference information extraction network are constructed, and the ship space information extraction network and the ship dynamic information extraction network are fused to construct an information fusion network to predict the destination port, so that the prediction accuracy is improved.
Drawings
FIG. 1 is a flow chart of a prediction method in an embodiment of the invention;
FIG. 2 is a schematic diagram of a marine spatial information extraction network according to an embodiment of the present invention;
FIG. 3 is a schematic diagram of extracting a network for extracting preference information of a ship according to an embodiment of the present invention;
fig. 4 is a schematic diagram of information fusion network extraction in an embodiment of the present invention.
Fig. 5 is a graph comparing the prediction effect of the information fusion network with that of other network models according to the embodiment of the present invention.
Detailed Description
The present invention will be described in further detail with reference to the accompanying drawings, in order to make the objects, technical solutions and advantages of the present invention more apparent. It should be understood that the detailed description is presented by way of example only and is not intended to limit the invention.
As shown in fig. 1, in this embodiment, there is provided an AIS data driven crude oil ship destination port prediction method, comprising the steps of,
s1, determining a ship navigation track based on preprocessed AIS historical data, and constructing a ship space information extraction network according to the ship navigation track;
s2, based on the preprocessed AIS historical data, fusing the ship type, the ship length, the ship load and the ship construction year, and constructing a ship preference information extraction network;
and S3, fusing the ship space information extraction network, the ship preference information extraction network and the ship dynamic information extraction network, constructing an information fusion network, and predicting the destination port based on the information fusion network.
The method mainly comprises three parts of contents, namely, construction of a ship space information extraction network, construction of a ship preference information extraction network and prediction of a destination port by using an information fusion network.
Before the three parts of contents are carried out, the AIS historical data are required to be preprocessed, and the specific content of the preprocessing is that the AIS historical data with different time intervals are converted into AIS data with preset duration intervals through linear interpolation. The preset time length can be determined according to the actual situation so as to better meet the actual requirement, and in the embodiment, the preset time length is set to be 10 minutes.
The following description is made for the three parts of the above description:
1. construction of ship space information extraction network
The part corresponds to step S1, and a ship passing record (ship sailing track) between different ports is extracted from the preprocessed AIS history data, so as to establish a ship space information extraction network, specifically, the ports are directly connected and the connection degree between the ports, namely, the number of ships passing between the two ports is extracted from the AIS history data in the past preset years. The trade connection between ports can be known through the extracted information, so that the prediction range of the destination port is reduced, and the prediction accuracy is improved. The preset years can be selected according to actual conditions so as to better meet actual demands, and the preset years are set to be 3 years in the embodiment.
Because the ship navigation track comprises the behaviors of the ship, the berthing entering and exiting behaviors and the time of the ship entering and exiting, the determination of the behaviors of the ship entering and exiting berthing and the determination of the attribution of the berthing of the ship are needed to be carried out, and the following steps are respectively carried out:
1. the ship berthing and out behavior is determined by,
(1) AIS track set T of length N from a given single vessel according to speed i，N The track points with the ship speed smaller than the preset section are screened out to form the track points with the length of mTrack set T i，m The method comprises the steps of carrying out a first treatment on the surface of the The preset section can be selected according to actual conditions so as to better meet actual requirements, and in the embodiment, the preset section is set to be 0.5 section.
Wherein T is i，N ＝{t 1 (u，s，c，lon，lat)，t 2 (u，s，c，lon，lat)，...，t N (u, s, c, lon, lat) }; u is world universal time, utc; s is the speed, lon is the longitude, lat is the latitude, and i is the independent unique id named for the ship; t (T) i，m ＝{t 1 (u，s，c，lon，lat)，t 2 (u，s，c，lon，lat)，...，t m (u，s，c，lon，lat)}；
(2) Sequentially calculating a set T i，m The time difference and the distance between the two track points; when the time difference between the two points is smaller than the preset time threshold value and the distance between the two points is smaller than the preset distance threshold value, the track point t 1 And locus point t 2 May be clustered into one cluster; v (V) 1 And V 2 The value of (2) can be selected according to the actual situation so as to better meet the actual demands. In this embodiment, verified, V 1 = 864000 seconds, V 2 =200 meters.
The time difference between two trace points is calculated as,
utcDifference＝u 2 -u 1
wherein utcDifference is utc time difference; u (u) 1 Is the 1 st track point t 1 Corresponding time; u (u) 2 Is the 2 nd track point t 2 Corresponding time;
the distance between two track points is calculated by the formula,
distance(lon 2 ，lat 2 ，lon 1 ，lat 1 )
＝R 0 *arccos(cos(lat 1 )
*cos(lat 2 )*cos(lon 2 -lon 1 )+sin(lat 2 )*sin(lat 1 ))
wherein lon 1 ，lat 1 Is the 1 st track point t 1 Corresponding longitude and latitude; lon (lon) 2 ，lat 2 Respectively the 2 nd track point t 2 Corresponding longitude and latitude; r is R 0 Is the earth radius;
(3) Based on the judgment rule in A12, the set T is assembled i，m Dividing into sets of subsets
T i，m,sub ＝{sub 1 (t 1 (u，s，c，lon，lat)，t 2 (u，s，c，lon，lat))，...，sub K (t K (u，s，c，lon，lat)，t K+1 (u，s，c，lon，lat))，
Wherein sub is a subset; k is the total number of subsets, each subset representing a stop state of the vessel.
The navigation track of the ship between the countries is the best basis for establishing the ship space information extraction network, so that calculating when the ship is berthed in which country has great significance for constructing the ship space information extraction network.
In this embodiment, the computer program code for determining the berthing behavior of a ship is shown in table 1.
Table 1 Ship berthing and berthing behavior determination algorithm
After the ship berthing and moving in/out behavior is determined, whether the ship is berthed or not needs to be further determined according to an application formula of the distance between the ship stopping state (berthing and moving in/out behavior) and the berthing. However, due to the small space between berths when the ship is berthed (particularly a container ship), the final unique berthing attribution, i.e. the ship berthing attribution determination, needs to be determined for one stopped state.
2. The determination process of the ship berthing attribution comprises the following steps of,
(1) Assume that the stopped states of the two vessels are S a = (u 1, lon1, lat1, u2, lon2, lat 2) and S b = (u 3, lon3, lat3, u4, lon4, lat 4) and all at the first berth B 1 = (blon 1, blat 1) and second berth B 2 Vicinity of = (blon 2, blat 2);
wherein u1 and u3 are the starting moments of the two ship stopping states respectively; u2 and u4 are the end times of the two ship stop states respectively; lon1, lat1, lon3 and lat3 are longitude and latitude corresponding to u1 and u3 respectively; lon2, lat2, lon4 and lat4 are longitude and latitude corresponding to u2 and u4 respectively; blon1 and blat1 are the longitude and latitude of the first berth respectively; and blon2 and blat2 are the longitude and latitude of the second berth respectively.
(2) Respectively calculate S a 、S b And B is connected with 1 、B 2 If u1 is smaller than u3, and distance (S a ，B 1 )<distance(S b ，B 1 ) Between u1 and u2, the first berth B 1 Quilt S a Occupied S b Should rest at the second berth B 2 。
2. Construction of ship preference information extraction network
In the ship space information extraction network, there is a certain preference for the destination ports of different ships. The ship preference is classified into a ship attribute preference, which is extracted by attributes such as a length, a width, and a draft of the ship, and a general preference, which is based on embedding of the identity of the displacement ship. Based on the ship attribute preference, general preference and port item attribute, a user attribute preference information extraction network is constructed to rapidly extract the ship preference information.
The part corresponds to the step S2, and a ship preference information extraction network is established by extracting the past arrival records of the ship from AI S historical data of preset years, and integrating the information of the ship type, the ship length, the ship load, the ship construction year and the like. The ship traveling preference is crucial to the prediction of the destination port, and the accuracy of the prediction is further improved by constructing a ship preference information extraction network.
In this embodiment, the fusion process is specifically,
according to the I MO matching ship type, ship length, ship load and ship construction year data, respectively classifying the ship type and the ship length and identifying the ship type and the ship length by binary codes; mapping between 0 and 1 is performed by maximum and minimum normalization on ship load and ship construction year.
The classification quantity of the ship types and the ship lengths can be selected according to actual conditions so as to better meet actual demands. In this embodiment, the types of the ships are classified into 3 types, VLCC, SUZEMAX, AFRAMAX types, and 5 types, 200 meters to 230 meters, 230 meters to 250 meters, 250 meters to 270 meters, 270 meters to 300 meters, and 300 meters or more.
3. Port prediction using information fusion network
By means of the continuous motion state of the vessel, some information can be extracted which is helpful for predicting the destination port. In the embodiment, 6 AI S data of a certain ship are continuously collected and fused with other data, so that the extraction of the ship dynamic information is realized. The total information includes three parts, respectively: (1) spatial information: extracting a vector containing 504 elements according to the last berthing record of the ship through a ship space information extraction network; (2) preference information: each attribute of the ship is extracted into a 256-dimensional vector through the ship preference information extraction network, and different attributes are fused through the linear neural network layer to form a 504-dimensional vector. The ship preference information extraction network is composed of a ship embedded layer (for encoding the vertex ship I D), a feature embedded layer, and a port embedded layer (for encoding the port I D). (3) Ship dynamic information: and acquiring the dynamic information of the ship in real time through a ship dynamic information extraction network, wherein the ship dynamic information extraction network consists of an LSTM layer and a linear layer. And fusing the three extracted information through a linear layer to obtain a vector with 504 dimensions, and then carrying out cosine distance calculation on the vector and the 504 dimension one-hot coding vector of the port, wherein the port corresponding to the nearest cosine distance is the destination port.
The cosine distance is calculated as,
wherein x and y are vectors x= (x) respectively 1 ，x 2 ，…，x p ) And y= (y) 1 ，y 2 ，…，y p ) Euclidean range of (2)A number.
This part corresponds to step S3, where, as shown in fig. 4, the specific procedure of performing destination port prediction based on the information fusion network is that,
1. extracting a vector containing 504 elements according to the last berthing record of the ship based on the ship space information extraction network; referring to fig. 2, the extraction process specifically includes constructing a set S of ship berthing time and place based on the ship berthing in and out behavior and the ship berthing attribution, and extracting information from the set S:
for s= { S 1 ((u a ，p m )，...，(u b ,p k ))，…，s n ((u c ,p o )，...，(u d ,p q ) Each element s) in } t ，And->Forming an OD pair, wherein O is used as a starting point, and D is used as a destination; the set S is de-duplicated to form a vertex set V= { V 1 ，v 2 ，…，v n All OD pairs in set S get edge set +.> The number of vessels sailing from O to D constitutes the weight of a certain node +.> The matrix products of V and W form a space information matrix to realize the extraction of ship space information;
wherein n is the ship index in the ship file, u a And u b Respectively, vessels s 1 P m And p k Is a ship s 1 Different harbors of the ship; u (u) c And u d Respectively, vessels s n Is provided for the different parking times; p is p o And p q Is a ship s n Different harbors of the ship; v n Is the nth node, namely the nth port;to connect the nth node and the edge of the n-1 th node +.>Is the weight of the edge.
2. Extracting each attribute of the ship as a 256-dimensional vector based on the ship preference information extraction network, and fusing different attributes through a linear neural network layer to form a 504-dimensional vector; referring to fig. 3, the extraction process is specifically,
(1) For a given crude tanker aggregate z= { Z 1 (shipType，weight,year，length)，...，z m (ship, weight, year, length) }, dividing the set Z into three parts, respectively a ship embedded matrixShip characteristic embedding matrix-> Harbor embedded matrix>
Wherein, shipType, weight, year, length are respectively the ship type, ship load, ship construction year and ship length; d is the potential embedding dimension; m is m ship ，l feature And n port The number of vessel, port and vessel properties, respectively; u (u) j ，p j And f j The j-th row in the matrixes U, P and F respectively; r is the real space.
In this embodiment, two ships are assumed, and the two ships are converted into vectors (0, 1) and (1, 0) through one-hot revolution, the (0, 1) can be converted into 256-dimensional vectors (0.1, 0.2, …, 0.8) through one 2×256 vector, and the process is called embedding (embedding) for performing dimension conversion on the vectors, so that the diagonal element is not 0, the method has more calculation significance, and the neural network calculation effect is improved. The other embedding matrices have the same embedding meaning, the form of which is shown in fig. 3.
(2) First, p is calculated j And f j Matrix product between each feature of (a), i.e. for a given p j ＝{a 1 ，a 2 ，…，a d Sum ofThe matrix product of the two is +.> Recalculating p j And f j All matrix products between>
After embedding, each element is mapped to a higher space, p j ＝{a 1 ，a 2 ，…，a d Sum ofThen it is a ship-like embedded 256 bit vector, { a 1 ，a 2 ，…，a d }、{b 1 ，b 2 ，…，b d And the 1 st to d th components corresponding thereto, respectively.
(3) Will p j And f j The matrix product between is added to u j In (u) j And p j *f j Forming a tanker preference vector x j ，x j ＝u j +p j *f j ；
(4)、x j And p j The matrix inner product between them is taken as the final output of the ship preference information extraction network, output=p j *x j 。
In this embodiment, the loss function of the information fusion network is a cross entropy loss function,
wherein k is a class; r is the total number of categories, and is the independent thermal coding of the port in the ship destination port prediction; h is a k A true tag for the kth category; g k Is the Softmax probability corresponding to the kth category.
3. Extracting real-time track data of the ship based on the ship dynamic information extraction network;
4. fusing the three extracted data to obtain a vector of 504 dimensions;
5. and (3) calculating the cosine distance between the vector obtained in the step (4) and the 504-dimensional one-hot coding vector of the port, wherein the port corresponding to the minimum cosine distance is the destination port.
In this embodiment, the prediction accuracy rate may be used as an evaluation index, so as to determine that the information fusion network in the present invention is more superior to other destination port prediction models, and the prediction is more accurate.
Wherein Accuracy is the prediction Accuracy; n is n corrrect To predict the correct number of samples, n total Is the total number of samples.
In the embodiment, the prediction accuracy of the information fusion network in the method is 74.3% obtained through specific data calculation, and the effect is improved by at least more than 5% compared with the existing model. Referring to fig. 5, specific predictive effects of the information fusion network and several other models are shown.
By adopting the technical scheme disclosed by the invention, the following beneficial effects are obtained:
the invention provides an AIS data-driven crude oil ship destination port prediction method, which is characterized in that position tracking is carried out on a ship after the ship starts from an oil loading port through AIS data, and the destination port is predicted, so that the problem of low accuracy of a prediction result caused by the lack of spatial information and ship transportation preference information in the crude oil ship transportation destination port prediction process is solved. The ship space information extraction network and the ship preference information extraction network are constructed, and the ship space information extraction network and the ship dynamic information extraction network are fused to construct an information fusion network to predict the destination port, so that the prediction accuracy is improved.
The foregoing is merely a preferred embodiment of the present invention and it should be noted that modifications and adaptations to those skilled in the art may be made without departing from the principles of the present invention, which is also intended to be covered by the present invention.
Claims (8)
1. An AIS data-driven crude oil ship destination port prediction method is characterized by comprising the following steps of: comprises the following steps of the method,
s1, determining a ship navigation track based on preprocessed AIS historical data, and constructing a ship space information extraction network according to the ship navigation track; the ship navigation track comprises ship berthing in and out behaviors and ship berthing attribution;
s2, based on the preprocessed AIS historical data, fusing the ship type, the ship length, the ship load and the ship construction year, and constructing a ship preference information extraction network;
and S3, fusing the ship space information extraction network, the ship preference information extraction network and the ship dynamic information extraction network, constructing an information fusion network, and predicting the destination port based on the information fusion network.
2. The AIS data driven crude oil vessel port of destination prediction method of claim 1, wherein: the ship berthing and out behavior is determined by,
a11 AIS track set T with length N from given single ship according to speed i，N Track points with the ship speed smaller than the preset section are screened out to form a track set T with the length of m i，m ；
Wherein T is i，N ＝{t 1 (u，s，c，lon，lat)，t 2 (u，s，c，lon，lat)，...，t N (u, s, c, lon, lat) }; u is world universal time, utc; s is the speed, lon is the longitude, lat is the latitude, and i is the independent unique id named for the ship; t (T) i，m ＝{t 1 (u，s，c，lon，lat)，t 2 (u，s，c，lon，lat)，...，t m (u，s，c，lon，lat)}；
A12, calculating the set T in turn i，m The time difference and the distance between the two track points; when the time difference between the two points is smaller than the preset time threshold value and the distance between the two points is smaller than the preset distance threshold value, the track point t 1 And locus point t 2 May be clustered into one cluster;
the time difference between two trace points is calculated as,
utcDifference＝u 2 -u 1
wherein utcDifference is utc time difference; u (u) 1 Is the 1 st track point t 1 Corresponding time; u (u) 2 Is the 2 nd track point t 2 Corresponding time;
the distance between two track points is calculated by the formula,
distance(lon 2 ，lat 2 ，lon 1 ，lat 1 )＝R 0 *arccos(cos(lat 1 )*cos(lat 2 )*cos(lon 2 -lon 1 )+sin(lat 2 )*sin(lat 1 ))
wherein lon 1 ，lat 1 Is the 1 st track point t 1 Corresponding longitude and latitude; lon (lon) 2 ，lat 2 Respectively the 2 nd track point t 2 Corresponding longitude and latitude; r is R 0 Is the earth radius;
a13, based on the judgment rule in A12, collecting T i，m Dividing into sets of subsets
T i，m,sub ＝{sub 1 (t 1 (u，s，c，lon，lat)，t 2 (u，s，c，lon，lat))，...，sub K (t K (u，s，c，lon，lat)，t K+1 (u，s，c，lon，lat))，
Wherein sub is a subset; k is the total number of subsets, each subset representing a stop state of the vessel.
3. The AIS data driven crude oil vessel port of destination prediction method of claim 1, wherein: the determination process of the ship berthing attribution comprises the following steps of,
a21, assuming that the stopped states of the two ships are S respectively a = (u 1, lon1, lat1, utc2, lon2, lat 2) and S b = (u 3, lon3, lat3, utc4, lon4, lat 4) and all at the first berth B 1 = (blon 1, blat 1) and second berth B 2 Vicinity of = (blon 2, blat 2);
wherein u1 and u3 are the starting moments of the two ship stopping states respectively; u2 and u4 are the end times of the two ship stop states respectively; lon1, lat1, lon3 and lat3 are longitude and latitude corresponding to u1 and u3 respectively; lon2, lat2, lon4 and lat4 are longitude and latitude corresponding to u2 and u4 respectively; blon1 and blat1 are the longitude and latitude of the first berth respectively; blon2 and blat2 are the longitude and latitude of the second berth respectively;
a22, respectively calculate S a 、S b And B is connected with 1 、B 2 If u1 is smaller than u3, and distance (S a ，B 1 )＜distance(S b ，B 1 ) Between u1 and u2, the first berth B 1 Quilt S a Occupied S b Should rest at the second berth B 2 。
4. The AIS data driven crude oil vessel port of destination prediction method of claim 1, wherein: the fusion process in step S2 is specifically that,
according to IMO matching ship type, ship length, ship load and ship construction year data, respectively classifying the ship type and the ship length and identifying the ship type and the ship length by binary codes; mapping between 0 and 1 is performed by maximum and minimum normalization on ship load and ship construction year.
5. The AIS data driven crude oil vessel port of destination prediction method of claim 1, wherein: in step S3, the specific process of destination port prediction based on the information fusion network is that,
s31, extracting a vector containing 504 elements according to the last berthing record of the ship based on the ship space information extraction network;
s32, extracting each attribute of the ship as a 256-dimensional vector based on the ship preference information extraction network, and fusing different attributes through the linear neural network layer to form a 504-dimensional vector;
s33, extracting real-time track data of the ship based on the ship dynamic information extraction network;
s34, fusing the three extracted data to obtain a vector with 504 dimensions;
s35, calculating the cosine distance between the vector obtained in S34 and the 504-dimensional one-hot encoding vector of the port, wherein the port corresponding to the minimum cosine distance is the destination port.
6. The AIS data driven crude oil vessel port of destination prediction method of claim 6, wherein: the extraction process of the step S31 specifically includes constructing a set S of ship berthing time and place based on the ship berthing in and out behavior and the ship berthing attribution, and extracting information from the set S:
for s= { S 1 ((u a ，p m )，...，(u b ,p k ))，...，s n ((u c ,p o )，...，(u d ,p q ) Each element st in) is not shown,and->Forming an OD pair, wherein O is used as a starting point, and D is used as a destination; the set S is de-duplicated to form a node set V= { V 1 ，v 2 ，…，v n All OD pairs in set S get edge set +.> The number of vessels sailing from O to D constitutes the weight of a certain node +.> The matrix products of V and W form a space information matrix to realize the extraction of ship space information;
wherein u is a And u b Respectively, vessels s 1 P m And p k Is a ship s 1 Different harbors of the ship; u (u) c And u d Respectively, vessels s n Is provided for the different parking times; p is p o And p q Is a ship s n Different harbors of the ship; v n Is the nth node, namely the nth port;to connect the nth node and the edge of the n-1 th node +.>Is the weight of the edge.
7. The AIS data driven crude oil vessel port of destination prediction method of claim 6, wherein: the extraction process of step S32 is specifically that,
s321, for a given crude oil tanker set z= { Z 1 (shipType，weight,year，length)，...，z m (ship, weight, year, length) }, dividing the set Z into three parts, respectively a ship embedded matrixShip characteristic embedding matrix-> Harbor embedded matrix>
Wherein, shipType, weight, year, length are respectively the ship type, ship load, ship construction year and ship length; d is the potential embedding dimension; m is m ship ，l feature And n port The number of vessel, port and vessel properties, respectively; u (u) j ，p j And f j The j-th row in the matrixes U, P and F respectively; r is the real space;
s322, calculate p j And f j Matrix product between each feature of (a), i.e. for a given p j ＝{a 1 ，a 2 ，…，a d Sum ofThe matrix product of the two is +.> Calculation of p j And f j All matrix multiplication betweenAccumulation of pathogenic qi>
S323, p j And f j The matrix product between is added to u j In (u) j And p j *f j Forming a tanker preference vector x j ，x j ＝u j +p j *f j ；
S324、x j And p j The matrix inner product between them is taken as the final output of the ship preference information extraction network, output=p j *x j 。
8. The AIS data driven crude oil vessel port of destination prediction method of claim 6, wherein: the loss function of the information fusion network is a cross entropy loss function,
wherein k is a class; r is the total number of categories, and is the independent thermal coding of the port in the ship destination port prediction; h is a k A true tag for the kth category; g k Is the Softmax probability corresponding to the kth category.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202310505090.9A CN116805172B (en) | 2023-05-06 | 2023-05-06 | AIS data-driven crude oil ship destination port prediction method |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202310505090.9A CN116805172B (en) | 2023-05-06 | 2023-05-06 | AIS data-driven crude oil ship destination port prediction method |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN116805172A true CN116805172A (en) | 2023-09-26 |
| CN116805172B CN116805172B (en) | 2023-12-12 |
Family
ID=88079134
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202310505090.9A Active CN116805172B (en) | 2023-05-06 | 2023-05-06 | AIS data-driven crude oil ship destination port prediction method |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN116805172B (en) |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110706518A (en) * | 2019-08-19 | 2020-01-17 | 浙江海洋大学 | Ship port-hanging and berthing prediction technology |
| CN111581314A (en) * | 2020-04-26 | 2020-08-25 | 亿海蓝（北京）数据技术股份公司 | Berth identification method and device based on ship track |
| CN115018368A (en) * | 2022-06-28 | 2022-09-06 | 交通运输部规划研究院 | Port berth occupancy rate calculation method based on massive AIS data |
| CN115577868A (en) * | 2022-12-12 | 2023-01-06 | 亿海蓝（北京）数据技术股份公司 | Method and device for predicting destination port of in-transit ship, readable storage medium and ship |
-
2023
- 2023-05-06 CN CN202310505090.9A patent/CN116805172B/en active Active
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110706518A (en) * | 2019-08-19 | 2020-01-17 | 浙江海洋大学 | Ship port-hanging and berthing prediction technology |
| CN111581314A (en) * | 2020-04-26 | 2020-08-25 | 亿海蓝（北京）数据技术股份公司 | Berth identification method and device based on ship track |
| CN115018368A (en) * | 2022-06-28 | 2022-09-06 | 交通运输部规划研究院 | Port berth occupancy rate calculation method based on massive AIS data |
| CN115577868A (en) * | 2022-12-12 | 2023-01-06 | 亿海蓝（北京）数据技术股份公司 | Method and device for predicting destination port of in-transit ship, readable storage medium and ship |
Also Published As
| Publication number | Publication date |
|---|---|
| CN116805172B (en) | 2023-12-12 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN113553682B (en) | Data-driven multi-level ship route network construction method | |
| Li et al. | Forecaster: A graph transformer for forecasting spatial and time-dependent data | |
| CN111597340A (en) | Text classification method and device and readable storage medium | |
| CN110532546B (en) | Automatic alarm issuing method fusing geographic position and text similarity | |
| CN114943694A (en) | Defect detection method based on confrontation generation network and attention | |
| CN110674858A (en) | Traffic jam detection method based on space-time association and big data mining | |
| Chen et al. | Exploration and mining evaluation system and price prediction of uranium resources | |
| Zhu et al. | Feature selection and deep learning for deterioration prediction of the bridges | |
| CN116608861A (en) | Ship track behavior abnormality detection method, system, device and storage medium | |
| CN113393027B (en) | Navigation mark drift intelligent prediction method based on deep learning | |
| Xu et al. | Trajectory clustering for SVR-based Time of Arrival estimation | |
| CN117390141B (en) | Agricultural socialization service quality user evaluation data analysis method | |
| Mei et al. | Port recommendation system for alternative container port destinations using a novel neural language-based algorithm | |
| CN116805172B (en) | AIS data-driven crude oil ship destination port prediction method | |
| Liao et al. | Traj2Traj: A road network constrained spatiotemporal interpolation model for traffic trajectory restoration | |
| CN116665482B (en) | Parking space recommending method and device based on intelligent parking | |
| Saremi et al. | Alternative sequence classification of neural networks for bridge deck condition rating | |
| Hashemi et al. | Gaussian-based runtime detection of out-of-distribution inputs for neural networks | |
| Xia | Navigational risk analysis based on GIS spatiotemporal trajectory mining: a case study in Nanjing Yangtze River Bridge waters | |
| Zhou et al. | Estimation of shipment size in seaborne iron ore trade | |
| CN115565376A (en) | Vehicle travel time prediction method and system fusing graph2vec and double-layer LSTM | |
| CN115346363A (en) | Driver violation prediction method based on neural network | |
| CN113505937A (en) | Multi-view encoder-based legal decision prediction system and method | |
| Zhang et al. | Deep Trajectory Similarity Model: A Fast Method for Trajectory Similarity Computation | |
| Wenzel et al. | A Neural Network Approach for ETA Prediction in Inland Waterway Transport |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |