Ship track prediction method based on marine environment feature fusion

Abstract

The invention discloses a ship track prediction method based on marine environment feature fusion, and the method comprises the steps: obtaining ship dynamic sequence data, ship position sequence data and marine environment information at a current moment, and inputting the data into a ship track prediction model; carrying out One-hot coding on ship dynamic and position sequence data, and carrying out space-time grid position correlation decoding on marine environment information; encoder coding is carried out on the obtained One-hot coded data of the ship dynamic sequence and the ship position sequence and the marine environment information space-time grid decoded data, and ship time sequence characteristics and marine environment time sequence characteristics are obtained; and after cross-modal alignment fusion is carried out on the two time sequence characteristics, Decoder decoding is carried out to obtain ship dynamic information and ship position information at the next moment, and the ship dynamic information and the ship position information are output after One-hot decoding. According to the method, marine environment information is utilized to supplement ship trajectory modeling, and the accuracy of a ship trajectory prediction model is improved.

Description

Technical Field

[0001] The present invention relates to the technical field of ship track prediction, and more particularly to a ship track prediction method based on the fusion of ocean environment features. Background Art

[0002] With the development of deep learning and big data processing technologies, ship trajectory prediction research has evolved from traditional statistical analysis to prediction based on deep learning models. Ship trajectory prediction is crucial for maritime route planning and scheduling, maritime traffic safety, marine environmental regulation and protection, and maritime military operations. The Automatic Identification System (AIS) provides detailed time-series data on a ship's trajectory and motion status, including its heading, speed, heading, location, cargo type, and length and width. This massive amount of accumulated ship trajectory and marine environmental information can be leveraged for deep learning modeling and analysis.

[0003] When a ship sails at sea, its motion and trajectory are profoundly influenced by the ocean environment, including meteorological and hydrological factors. Advances in ocean monitoring technology have accumulated a vast amount of ocean environmental data that can be utilized by deep learning models for modeling and analysis. Existing deep learning methods consider the ship's position and dynamics, resulting in limited accuracy in ship trajectory prediction. Comprehensively considering the ocean environment in which a ship navigates, the integration of cross-modal ocean environmental information with ship trajectory information and improving the accuracy of the ship trajectory model presents a technical challenge.

[0004] The shortcomings of existing ship trajectory prediction methods mainly include:

[0005] (1) Traditional statistical learning methods based on AIS data and single-modal deep learning methods only consider single-modal AIS data in the problem of ship trajectory prediction. They do not fully utilize the ocean environment information in which the ship is sailing, including marine meteorological and hydrological information. The modeling accuracy is limited, and the predicted ship trajectory has a large deviation.

[0006] (2) The Transformer method based on a single encoder-decoder architecture belongs to the seq2seq method, which only supports data input and output of the same modality, does not support the input of cross-modal data, and does not support the fusion of cross-modal information of different dimensions.

[0007] (3) Existing deep learning models use the mean square error and its variant loss function based on numerical regression modeling to model and predict continuous value outputs. The predicted outputs are significantly offset from the true values.

[0008] Therefore, how to use marine meteorological and hydrological information to supplement ship trajectory modeling and improve the accuracy of ship trajectory prediction models is an urgent problem that technical personnel in this field need to solve. Summary of the Invention

[0009] In view of the above problems, the present invention provides a ship track prediction method based on the fusion of ocean environment characteristics to at least solve some of the technical problems mentioned in the above background technology.

[0010] In order to achieve the above object, the present invention adopts the following technical solutions:

[0011] The present invention provides a ship track prediction method based on the fusion of ocean environment characteristics, comprising:

[0012] Obtain the current ship dynamic sequence data and ship position sequence data;

[0013] Acquiring ocean environment information based on the ship position sequence data;

[0014] The acquired ship dynamic sequence data, ship position sequence data and ocean environment information are input into a ship trajectory prediction model based on the fusion of ocean environment features;

[0015] In the ship trajectory prediction model based on the fusion of ocean environment features:

[0016] Through the ship sequence information one-hot encoding module, the ship dynamic sequence data and the ship position sequence data are one-hot encoded respectively to obtain corresponding ship dynamic sequence one-hot encoded data and ship position sequence one-hot encoded data;

[0017] The marine environment information spatiotemporal grid position association decoding module decodes the marine environment information based on the ship position sequence data to obtain marine environment information spatiotemporal grid decoding data;

[0018] Performing encoder encoding processing on the one-hot encoded data of the ship dynamic sequence and the one-hot encoded data of the ship position sequence through N encoder encoding modules to obtain ship time series features;

[0019] Performing encoder encoding processing on the ocean environment information spatiotemporal grid decoded data through the N encoder encoding modules to obtain ocean environment time series characteristics;

[0020] The ship time series features and the ocean environment time series features are cross-modally aligned and fused by a cross-modal alignment and fusion module to obtain a ship time series feature fused based on the ocean environment features;

[0021] Through N decoder decoding modules, combined with the ship information output sequence, the ship time series features based on the fusion of ocean environment features are decoded to obtain the predicted ship dynamic information and ship position information at the next moment;

[0022] Through the ship sequence information one-hot decoding module, the predicted ship dynamic information and ship position information at the next moment are one-hot decoded and output.

[0023] Furthermore, the obtaining of the ship dynamic sequence data and the ship position sequence data at the current moment specifically includes:

[0024] Obtaining the AIS ship track data field at the current moment from the ship automatic identification system; the AIS ship track data field includes an AIS ship dynamic data field and an AIS ship position data field;

[0025] Preprocessing the AIS ship trajectory data field;

[0026] The pre-processed AIS ship track data fields are sorted by time according to the MMSI identifier, and all data with the same MMSI identifier are filtered out, that is, all the navigation sequence information of the same ship;

[0027] The navigation sequence information with a time interval of more than half an hour for the same MMSI is segmented, and each segment of data is the navigation track sequence information of the same ship;

[0028] The segmented data are sampled at preset time intervals, and the pointless data are supplemented using an interpolation algorithm to finally obtain the ship trajectory sequence data; the ship trajectory sequence data includes ship dynamic sequence data and ship position sequence data.

[0029] Furthermore, the pre-processing of the AIS ship trajectory data specifically includes:

[0030] Cleaning the missing track data in the AIS ship track data field;

[0031] Eliminate erroneous trajectory data in the AIS ship trajectory data field;

[0032] The repeated track data in the AIS ship track data field is removed.

[0033] Furthermore, the acquiring of ocean environment information based on the ship position sequence data specifically includes:

[0034] In the target sea surface area, obtaining ocean absolute geostrophic current data and sea surface wind field data associated with the ship position sequence data;

[0035] According to preset longitude and latitude intervals, the ocean absolute geostrophic current data and the sea surface wind field data are respectively subjected to spatiotemporal gridding processing to obtain corresponding absolute geostrophic current grid data and sea surface wind field grid data;

[0036] The absolute geostrophic flow grid data and the sea surface wind field grid data are normalized, and the normalized absolute geostrophic flow grid data and the sea surface wind field grid data are used as ocean environment information for ship trajectory prediction.

[0037] Furthermore, the absolute geostrophic flow grid data geo_flow is selected by an absolute geostrophic flow selection function; it is expressed as:

[0038] geo_flow u,v =g(lon,lat,time,depth)

[0039] Where g(·) represents the absolute geostrophic flow selection function; lon represents the longitude of the target sea surface area; lat represents the latitude of the target sea surface area; time represents the current time; depth represents the depth of the ocean absolute geostrophic flow data; geo_flow u,v It represents the velocity of the ocean's absolute geostrophic current in the east-west direction u and the north-south direction v.

[0040] Furthermore, the sea surface wind field grid data wind is selected by an ocean wind field selection function; it is expressed as:

[0041] wind u,v =w(lon,lat,time,height)

[0042] Wherein, w(·) represents the ocean wind field selection function; lon represents the longitude of the target sea surface area; lat represents the latitude of the target sea surface area; time represents the current time; height represents the height of the sea surface wind field data; wind u,v It represents the wind speed in the east-west direction u and the north-south direction v of the sea surface wind field.

[0043] Furthermore, the cross-modal alignment and fusion module includes: multiple feedforward neural network modules based on residual connections, 1×1 convolution modules and fusion modules; wherein:

[0044] Each feedforward neural network module based on residual connection is composed of Linear and RELU, which is used to further perform deep learning on the ocean environment time series characteristics to obtain deep time series characteristics of the ocean environment;

[0045] Performing an alignment operation on the ocean environment depth time series features and the ship time series features through the 1×1 convolution module;

[0046] Through the fusion module, the aligned ocean environment depth time series features and the ship time series features are added and fused to obtain the ship time series features based on the fusion of ocean environment features.

[0047] Furthermore, a cross-attention mechanism is introduced into the Decoder decoding module.

[0048] Furthermore, the method further includes: calculating the loss of the ship trajectory prediction model based on the fusion of ocean environment features by using a multi-granularity fusion loss function based on one-hot encoding;

[0049] The multi-granularity fusion loss function L based on One-hot encoding multi_grain Expressed as:

[0050]

[0051] Among them, L CE represents the cross entropy loss function; y fine Represents fine-grained one-hot representation of the true value; Represents the fine-grained prediction output One-hot representation; y med Represents the one-hot representation of the true value at medium granularity; Represents the medium-granularity prediction output One-hot representation; y coarse Represents the coarse-grained one-hot representation of the true value; It represents the one-hot representation of the coarse-grained prediction output; α represents the weight coefficient of the medium-grained prediction output; β represents the weight coefficient of the coarse-grained prediction output.

[0052] It can be seen from the above technical solutions that, compared with the prior art, the present invention discloses a ship track prediction method based on the fusion of marine environment features, which has the following beneficial effects:

[0053] The present invention utilizes a multi-encoding-single decoding module architecture to effectively process the meteorological and hydrological information characteristics of the marine environment, and uses marine environmental information to supplement ship trajectory modeling, thereby improving the accuracy of trajectory prediction.

[0054] The present invention utilizes a cross-modal alignment and fusion module to integrate and process marine environmental meteorological and hydrological information, including absolute geostrophic current, sea surface wind field and other information, and improves the accuracy of modeling through spatiotemporal gridding processing and modal alignment and fusion.

[0055] The present invention designs a multi-granularity fusion loss function based on One-hot encoding, constructs a trajectory prediction model as a data quantization classification problem, effectively limits the output value range of the ship trajectory model, and limits the trajectory prediction area and state value range, thereby improving the accuracy and effectiveness of the model prediction.

[0056] The technical solution of the present invention is further described in detail below through the accompanying drawings and embodiments. BRIEF DESCRIPTION OF THE DRAWINGS

[0057] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are merely embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on the provided drawings without paying any creative work.

[0058] Figure 1 Schematic diagram of a ship track prediction method based on ocean environment feature fusion provided by an embodiment of the present invention.

[0059] Figure 2 A schematic diagram of the process of obtaining ship trajectory sequence data provided by an embodiment of the present invention.

[0060] Figure 3 A schematic diagram of the process of obtaining marine environmental information provided by an embodiment of the present invention.

[0061] Figure 4 A schematic diagram of the multi-granularity one-hot encoding representation provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0062] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0063] The embodiment of the present invention discloses a ship track prediction method based on the fusion of ocean environment characteristics, see Figure 1 As shown, the specific steps include:

[0064] Obtain the current ship dynamic sequence data and ship position sequence data;

[0065] Obtain ocean environment information based on ship position sequence data;

[0066] The acquired ship dynamic sequence data, ship position sequence data and ocean environment information are input into a ship trajectory prediction model based on the fusion of ocean environment features;

[0067] The following operations are performed in the ship trajectory prediction model based on the fusion of ocean environment features:

[0068] Through the ship sequence information one-hot encoding module, the ship dynamic sequence data and the ship position sequence data are one-hot encoded respectively to obtain the corresponding ship dynamic sequence one-hot encoded data and ship position sequence one-hot encoded data;

[0069] Through the ocean environment information spatiotemporal grid position correlation decoding module, the ocean environment information is decoded based on the ship position sequence data to obtain the ocean environment information spatiotemporal grid decoding data;

[0070] Through N encoder modules, the one-hot encoded data of the ship dynamic sequence and the one-hot encoded data of the ship position sequence are processed by encoder to obtain the ship time series features;

[0071] Through N encoder modules, the ocean environment information spatiotemporal grid decoded data is processed by encoder to obtain the ocean environment time series characteristics;

[0072] Through the cross-modal alignment and fusion module, the ship time series characteristics and the ocean environment time series characteristics are cross-modally aligned and fused to obtain the ship time series characteristics based on the fusion of ocean environment characteristics;

[0073] Through N decoder modules, combined with the ship information output sequence, the ship time series features based on the fusion of ocean environment features are decoded and processed to obtain the predicted ship dynamic information and ship position information at the next moment;

[0074] Through the ship sequence information one-hot decoding module, the predicted ship dynamic information and ship position information at the next moment are one-hot decoded and output.

[0075] Next, each of the above steps will be described separately.

[0076] Step 1: Get the current ship dynamic sequence data and ship position sequence data, such as Figure 2 shown.

[0077] (1) Obtaining the AIS ship track data field at the current moment from the Automatic Identification System (AIS); the AIS ship track data field includes the AIS ship dynamic data field and the AIS ship position data field; specifically:

[0078] AIS ship dynamic data fields include: Maritime Mobile Service Identity (MMSI) field, speed over ground (SOG), course angle (COG), heading angle (Heading), and accumulated sailing distance (acc_dist);

[0079] The AIS ship position data fields include: message timestamp (BaseDateTime), latitude coordinate (LAT) and longitude coordinate (LON) fields;

[0080] (2) Preprocessing the AIS ship trajectory data field; specifically including:

[0081] The missing track data in the AIS ship track data field is cleaned; namely: ① the missing unique identifier data is removed, including the Maritime Mobile Service Identifier (MMSI) field being empty; ② the missing dynamic information data is removed, including the message timestamp (BaseDateTime), latitude coordinates (LAT) and longitude coordinates (LON), speed over ground (SOG), heading angle (COG), and heading angle (Heading) fields being empty;

[0082] Remove erroneous track data from the AIS ship track data field. Common erroneous track data includes the following: ① The MMSI field is 0; ② The speed over ground (SOG) is greater than the reasonable value. Aircraft carriers and missile destroyers generally have a speed of about 30 knots, small ships such as missile boats and torpedo boats have a speed of 40-60 knots, ordinary ships generally have a speed of less than 15 knots, and a few large cargo ships have a speed close to 20 knots. The set sailing speed is greater than 70 knots.

[0083] Remove duplicate track data in the AIS ship track data field; that is, when the timestamp information and longitude and latitude information of multiple AIS data are the same, delete the duplicate data and retain one;

[0084] (3) The pre-processed AIS ship trajectory data fields are sorted by time according to the MMSI identifier, and all data with the same MMSI identifier are filtered out, that is, all navigation sequence information of the same ship;

[0085] Since the ship trajectory data reported by the AIS system is mixed with the position and status information of all ships in the area at the current time point, the data of the same ship can be selected through the MMSI identifier to facilitate subsequent processing;

[0086] (4) The navigation sequence information with a time interval of more than half an hour for the same MMSI is segmented, and each segment of data is the navigation track sequence information of the same ship;

[0087] (5) The segmented data are sampled according to the preset time interval, and the pointless data are supplemented by the interpolation algorithm, and finally the ship trajectory sequence data is obtained; the ship trajectory sequence data includes the ship dynamic sequence data and the ship position sequence data.

[0088] Step 2: Based on the ship position sequence data, obtain the ocean environment information; Figure 3 shown.

[0089] The current ship trajectory prediction model only considers the ship's position information for trajectory prediction, without considering the impact of the ship and the surrounding ocean environment on the ship's trajectory and attitude. The prediction accuracy is limited. The present invention uses ocean environment data, including ocean absolute geostrophic current data and sea surface wind field data, to improve the accuracy of ship trajectory and attitude prediction. Specifically:

[0090] (1) Obtain ocean absolute geostrophic current data and sea surface wind field data associated with the ship position sequence data within the target sea surface area;

[0091] Since ships are affected by geostrophic currents when sailing on the sea, they need to change their heading and speed to adapt to the currents. In addition, the maximum draft of modern ships is about 10 meters. Modeling ship trajectories based on ship position and geostrophic current data can help improve the prediction accuracy of ship speed and heading.

[0092] Since sea surface wind speed affects the ship's navigation attitude and speed, correlating sea surface wind field data with the ship's position can improve the ship's speed and heading prediction capabilities;

[0093] (2) performing spatiotemporal gridding processing on the ocean absolute geostrophic current data and the sea surface wind field data according to preset latitude and longitude intervals, respectively, to obtain corresponding absolute geostrophic current grid data and sea surface wind field grid data; specifically:

[0094] Select the target sea surface area to analyze the relevant marine environment information. The longitude of the target sea surface area is lon∈[lon min ,lon max ], latitude is lat∈[lat min ,lat max ].

[0095] According to the preset longitude and latitude intervals (for example, 0.5 degree longitude and latitude intervals), the ocean environment information is gridded to obtain the ocean environment grid data Sea_grid lat，lon , expressed as:

[0096] Sea_grid lat，lon ={wind,geo_flow}

[0097] In this formula:

[0098] ① The absolute geostrophic flow grid data geo_flow is selected by the absolute geostrophic flow selection function; it is expressed as:

[0099] geo_flow u,v =g(lon,lat,time,depth)

[0100] Where g(·) represents the absolute geostrophic flow selection function; lon represents the longitude of the target sea surface area; lat represents the latitude of the target sea surface area; time represents the current time; depth represents the depth of the ocean absolute geostrophic flow data (which can be 1m, 5m, 10m, and 20m); geo_flow u,v It represents the velocity of the ocean's absolute geostrophic current in the east-west direction u and the north-south direction v.

[0101] ② The sea surface wind field grid data wind is selected through the ocean wind field selection function; it is expressed as:

[0102] wind u,v =w(lon,lat,time,height)

[0103] Where w(·) represents the ocean wind field selection function; lon represents the longitude of the target sea surface area; lat represents the latitude of the target sea surface area; time represents the current time; height represents the height of the sea surface wind field data (which can be 10m); wind u,v It represents the wind speed in the east-west direction u and the north-south direction v of the sea surface wind field.

[0104] (3) The absolute geostrophic flow grid data and the sea surface wind field grid data are normalized, and the normalized value range is [0,1]; it is expressed as:

[0105]

[0106] Where, e represents geostrophic flow grid data geo_flow or ocean wind field grid data wind; i represents east-west direction u or north-south direction v; e min Indicates the minimum value of geostrophic flow grid data or ocean wind field grid data; e max Indicates the maximum value of geostrophic flow grid data or ocean wind field grid data.

[0107] The normalized absolute geostrophic flow grid data and sea surface wind field grid data are used as the ocean environment information for ship trajectory prediction.

[0108] Step 3: Input the above-obtained ship dynamic sequence data, ship position sequence data and ocean environment information into the ship trajectory prediction model based on the fusion of ocean environment features;

[0109] This ship trajectory prediction model based on ocean environment feature fusion is designed based on a typical Transformer architecture for cross-modal fusion. It specifically includes a one-hot encoding module for ship sequence information, a spatiotemporal grid decoding module for ocean environment information, N encoder encoding modules, a cross-modal alignment and fusion module, N decoder decoding modules, and a one-hot decoding module for ship sequence information.

[0110] Step 4: Use the ship sequence information one-hot encoding module to perform one-hot encoding on the ship dynamic sequence data and the ship position sequence data to obtain the corresponding ship dynamic sequence one-hot encoded data and ship position sequence one-hot encoded data; specifically:

[0111] (1) Constructing a one-hot encoding table for ship sequence information

[0112] First, the maximum and minimum values ​​of ship dynamic sequence data and ship position sequence data are counted respectively; the value range of each type of data [x min ,x max ], x is a specific field in the ship dynamic sequence data or ship position sequence data, that is, x∈{lon,lat,SOG,COG,Heading,acc_dist};

[0113] Second, set the dimension D of the One-hot encoding one-hot Construct a coding table; the coding table contains id, x one-hot Two fields; id is the serial number of the encoding table, and id∈[0,D one-hot -1];x one-hot is the encoding value, the id and x in the encoding table one -hot It is a one-to-one mapping relationship and supports reverse lookup.

[0114] (2) One-hot encoding of ship sequence information

[0115] The corresponding x real value is quantized and mapped to D by the following formula one-hot In the One-hot encoding, we get x one-hot Encoded representation:

[0116]

[0117] Step 5: Decode the ocean environment information based on the ship position sequence data through the ocean environment information spatiotemporal grid position association decoding module to obtain ocean environment information spatiotemporal grid decoding data;

[0118] Specifically, based on the ship position sequence data, the ocean environment grid point data (including absolute geostrophic flow grid point data and sea surface wind field grid point data) corresponding to the time and space coordinates are associated to achieve decoding of the ocean environment information.

[0119] Step 6: Encode the one-hot encoded data of the ship dynamic sequence and the one-hot encoded data of the ship position sequence using N encoder encoding modules to obtain ship time series features; and also use the N encoder encoding modules to encode the spatiotemporal grid decoded data of the ocean environment information to obtain ocean environment time series features.

[0120] N can be set to a value between 3 and 12, and the computational complexity, computational time, and computational accuracy are comprehensively optimized in combination with the training process. It is generally set to 6.

[0121] Step 7: Through the cross-modal alignment and fusion module, the ship time series features and the ocean environment time series features are cross-modally aligned and fused to obtain the ship time series features based on the fusion of ocean environment features;

[0122] The cross-modal alignment and fusion module includes multiple (for example, M ff A feedforward neural network module based on residual connection, a 1×1 convolution module and a fusion module; wherein:

[0123] Each feedforward neural network module based on residual connection is composed of Linear and RELU, which is used to further deeply learn the time series characteristics of the ocean environment and obtain the deep time series characteristics of the ocean environment; the number of feedforward neural network modules M ff It can be 1-5 and can be adjusted according to the training effect.

[0124] Through the 1×1 convolution module, the ocean environment depth time series features and the ship time series features are aligned to make the dimensions of the two consistent;

[0125] Through the fusion module, the aligned ocean environment depth time series features and the ship time series features are added and fused to obtain the ship time series features based on the fusion of ocean environment features;

[0126] In summary, in this cross-modal alignment and fusion module, the temporal characteristics of the ocean environment are used to perturb the ship's temporal feature encoding data, making the trajectory prediction model more comprehensive and accurate.

[0127] Step 8: Through N decoder decoding modules, combined with the ship information output sequence, the ship time series features based on the fusion of ocean environment features are decoded to obtain the predicted ship dynamic information and ship position information at the next moment. At this time, the predicted ship dynamic information and ship position information are in one-hot encoding mode;

[0128] The introduction of the cross-attention mechanism in the Decoder decoding module can effectively enhance the feature expression capability; in addition, the number of the Decoder decoding modules should be consistent with the number of the Encoder encoding modules mentioned above.

[0129] Step 9: Through the ship sequence information one-hot decoding module, the predicted ship dynamic information and ship position information at the next moment are one-hot decoded and output;

[0130] Specifically, the model output y one-hot The value is reverse mapped to the one-hot encoded sequence number y by looking up the encoding table id The final ship trajectory information prediction value y^ is calculated by the following formula.

[0131]

[0132] Among them, y max 、y min are the maximum and minimum values ​​of the field data defined in the construction code table, D one-hot Indicates the code table dimension of this type of data.

[0133] In an embodiment of the present invention, one-hot encoding is performed on the input ship dynamic sequence data and ship position sequence data, and one-hot decoding is performed on the output to obtain the predicted ship dynamic information and ship position information; this can limit the input range of the model, and then train the model output within the limited range, thereby improving the efficiency and accuracy of training.

[0134] Step 10: Multi-granularity fusion loss function based on one-hot encoding

[0135] In the prior art, the loss of output calculation is calculated by using a single fine-grained One-hot encoding, which easily leads to errors in predicting similar data, and makes the model easily oscillate and fail to converge during training. To address this problem, the embodiment of the present invention designs a multi-granularity fusion loss function to reduce the problem of difficult training convergence caused by single fine-grained One-hot encoding. Multi-granularity One-hot encoding is represented as follows: Figure 4 As shown in the figure, three representations are listed: coarse-grain, medium-grain, and fine-grain. They can be expanded to more granularity representations according to training needs. The particles of the coarse-grain representation are the particles of the fine-grained M c times, the medium-sized particles are the fine-sized particles M m times, M c 、M m Different values ​​can be set according to training, generally set to 5 or 3.

[0136] Based on three types of encoding representation, multi-granularity fusion loss function L based on One-hot encoding multi_grain Expressed as:

[0137]

[0138] Among them, L CE represents the cross entropy loss function; y fine Represents fine-grained one-hot representation of the true value; Represents the fine-grained prediction output One-hot representation; y med Represents the one-hot representation of the true value at medium granularity; Represents the medium-granularity prediction output One-hot representation; y coarse Represents the coarse-grained one-hot representation of the true value; The coarse-grained prediction output is represented by a one-hot representation; α represents the weight coefficient for medium-grained predictions; and β represents the weight coefficient for coarse-grained predictions. α and β are generally set to 0.3. You can set different hyperparameters to train multiple models for optimal results.

[0139] Based on this loss function, in an embodiment of the present invention, the ship trajectory prediction model based on the fusion of marine environment features adopts a general transformer architecture model back-propagation training method during the training process, designs input and output of multiple granularities according to the multi-granularity fusion strategy, and uses the corresponding loss function for back-propagation training. By setting different hyperparameters, the optimal model is selected for ship trajectory prediction.

[0140] The above steps 1 to 10 are only for the convenience of explanation and do not limit the specific implementation order of each step.

[0141] In summary, in response to the problems existing in the existing ship trajectory prediction, such as insufficient accuracy of single-modal data modeling, difficulty in cross-modal information fusion processing, and large deviation in prediction output of training methods based on numerical regression, the present invention provides a ship track prediction method based on the fusion of marine environment features, designs a multi-encoding-single decoding module architecture and a cross-modal alignment and fusion module for marine environment feature fusion, solves the problem of fusion modeling of marine environment features and trajectory features, and improves the accuracy of modeling; designs ship sequence One-hot encoding and multi-granularity fusion loss, constructs a trajectory prediction model as a data quantization classification problem, and effectively improves the effectiveness and accuracy of the model prediction output. The method provided by the present invention can be applied in the fields of maritime route planning and scheduling, maritime traffic safety, marine environmental supervision and protection, maritime military operations, etc., to improve the effectiveness and accuracy of ship trajectory prediction, and help improve the efficiency of supervision in related fields.

[0142] The various embodiments in this specification are described in a progressive manner, and each embodiment focuses on the differences from other embodiments. The same or similar parts between the various embodiments can be referenced to each other.

[0143] The above description of the disclosed embodiments is intended to enable one skilled in the art to implement or use the present invention. Various modifications to these embodiments will be readily apparent to one skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present invention. Therefore, the present invention is not limited to the embodiments shown herein but is intended to conform to the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A ship track prediction method based on the fusion of ocean environment characteristics, characterized in that: include: Obtain the current ship dynamic sequence data and ship position sequence data; Based on the ship position sequence data, obtaining marine environment information; The acquired ship dynamic sequence data, ship position sequence data and ocean environment information are input into a ship trajectory prediction model based on the fusion of ocean environment characteristics; In the ship trajectory prediction model based on the fusion of marine environment features: Through the ship sequence information One-hot encoding module, the ship dynamic sequence data and the ship position sequence data are respectively One-hot encoded to obtain corresponding ship dynamic sequence One-hot encoded data and ship position sequence One-hot encoded data; By means of a marine environment information spatiotemporal grid position associated decoding module, the marine environment information is decoded based on the ship position sequence data to obtain marine environment information spatiotemporal grid decoding data; Through N Encoder encoding modules, the One-hot encoded data of the ship dynamic sequence and the One-hot encoded data of the ship position sequence are processed by Encoder to obtain ship time series features; Through the N Encoder encoding modules, the ocean environment information spatiotemporal grid decoded data is subjected to Encoder encoding processing to obtain ocean environment time series characteristics; Through the cross-modal alignment and fusion module, the ship time series characteristics and the ocean environment time series characteristics are cross-modally aligned and fused to obtain the ship time series characteristics based on the fusion of ocean environment characteristics; Through N decoder decoding modules, combined with the ship information output sequence, the ship time series features based on the fusion of ocean environment features are decoded and processed by decoder to obtain the predicted ship dynamic information and ship position information at the next moment; Through the ship sequence information One-hot decoding module, the predicted ship dynamic information and ship position information at the next moment are processed by One-hot decoding and output.

2. The ship track prediction method based on the fusion of marine environment characteristics according to claim 1 is characterized in that: The obtaining of the ship dynamic sequence data and the ship position sequence data at the current moment specifically includes: Acquire the AIS ship track data field at the current moment from the ship automatic identification system; the AIS ship track data field includes an AIS ship dynamic data field and an AIS ship position data field; Preprocessing the AIS ship track data field; The pre-processed AIS ship track data fields are sorted by time according to the MMSI identifier, and all data with the same MMSI identifier are screened out, that is, all navigation sequence information of the same ship; The navigation sequence information with a time interval of more than half an hour for the same MMSI mark is segmented, and each segment of data is the navigation track sequence information of the same ship; The segmented data are sampled at preset time intervals, and the pointless data are supplemented by an interpolation algorithm, so as to finally obtain the ship trajectory sequence data; the ship trajectory sequence data includes ship dynamic sequence data and ship position sequence data.

3. The ship track prediction method based on the fusion of marine environment characteristics according to claim 2 is characterized in that: The preprocessing of the AIS ship trajectory data specifically includes: Cleaning the missing track data in the AIS ship track data field; Eliminating erroneous track data in the AIS ship track data field; The repeated track data in the AIS ship track data field is removed.

4. The ship track prediction method based on the fusion of marine environment characteristics according to claim 1 is characterized in that: The acquiring of ocean environment information based on the ship position sequence data specifically includes: In the target sea surface area, obtaining ocean absolute geostrophic current data and sea surface wind field data associated with the ship position sequence data; According to the preset longitude and latitude intervals, the ocean absolute geostrophic current data and the sea surface wind field data are respectively subjected to spatiotemporal gridding processing to obtain corresponding absolute geostrophic current grid data and sea surface wind field grid data; The absolute geostrophic flow grid data and the sea surface wind field grid data are normalized, and the normalized absolute geostrophic flow grid data and the sea surface wind field grid data are used as the ocean environment information for ship trajectory prediction.

5. The ship track prediction method based on the fusion of marine environment characteristics according to claim 4 is characterized in that: The absolute geostrophic flow grid data geo_flow is selected by the absolute geostrophic flow selection function; it is expressed as: geo_flow u,v =g(lon,lat,time,depth) Where g(·) represents the absolute geostrophic flow selection function; lon represents the longitude of the target sea surface area; lat represents the latitude of the target sea surface area; time represents the current time; depth represents the depth of the ocean absolute geostrophic flow data; geo_flow u,v It represents the velocity of the absolute geostrophic current in the east-west direction u and the north-south direction v of the ocean.

6. The ship track prediction method based on the fusion of ocean environment characteristics according to claim 4 is characterized in that: The sea surface wind field grid data wind is selected by the ocean wind field selection function; it is expressed as: wind u,v =w(lon,lat,time,height) Wherein, w(·) represents the ocean wind field selection function; lon represents the longitude of the target sea surface area; lat represents the latitude of the target sea surface area; time represents the current time; height represents the height of the sea surface wind field data; wind u,v It represents the wind speed in the east-west direction u and the north-south direction v of the sea surface wind field.

7. The ship track prediction method based on the fusion of ocean environment characteristics according to claim 1 is characterized in that: The cross-modal alignment and fusion module includes: a plurality of feed-forward neural network modules based on residual connections, a 1×1 convolution module and a fusion module; wherein: Each residual connection-based feedforward neural network module is composed of Linear and RELU, and is used to further perform deep learning on the ocean environment time series characteristics to obtain deep time series characteristics of the ocean environment; By means of the 1×1 convolution module, an alignment operation is performed on the ocean environment depth time series feature and the ship time series feature; Through the fusion module, the aligned ocean environment depth time series features and the ship time series features are added and fused to obtain the ship time series features fused based on the ocean environment features.

8. The ship track prediction method based on the fusion of ocean environment characteristics according to claim 1 is characterized in that: A cross attention mechanism is introduced into the Decoder decoding module.

9. The ship track prediction method based on ocean environment feature fusion according to claim 1 is characterized in that: Also includes: For the ship trajectory prediction model based on the fusion of marine environment features, the loss is calculated by a multi-granularity fusion loss function based on One-hot encoding; The multi-granularity fusion loss function L based on One-hot encoding multi_grain It is expressed as: Among them, L CE represents the cross entropy loss function; y fine Represents fine-grained one-hot representation of the true value; Represents the fine-grained prediction output One-hot representation; y med One-hot representation of the true value at medium granularity; Represents the medium-granularity prediction output One-hot representation; y coarse Represents a coarse-grained one-hot representation of the true value; It represents the one-hot representation of the coarse-grained prediction output; α represents the weight coefficient of the medium-grained weight; β represents the coarse-grained weight coefficient.

Images