CN114154619A - Ship track prediction method based on CNN and BILSTM - Google Patents

Ship track prediction method based on CNN and BILSTM Download PDF

Info

Publication number
CN114154619A
CN114154619A CN202111349534.1A CN202111349534A CN114154619A CN 114154619 A CN114154619 A CN 114154619A CN 202111349534 A CN202111349534 A CN 202111349534A CN 114154619 A CN114154619 A CN 114154619A
Authority
CN
China
Prior art keywords
data
ship
bilstm
representing
cnn
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.)
Pending
Application number
CN202111349534.1A
Other languages
Chinese (zh)
Inventor
李芃
陈赛
孙宏放
张兰勇
李奕霏
刘洪丹
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Heu Qingdao Ship Science And Technology Co ltd
Harbin Engineering University
Original Assignee
Heu Qingdao Ship Science And Technology Co ltd
Harbin Engineering University
Priority date (The priority date 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 date listed.)
Filing date
Publication date
Application filed by Heu Qingdao Ship Science And Technology Co ltd, Harbin Engineering University filed Critical Heu Qingdao Ship Science And Technology Co ltd
Priority to CN202111349534.1A priority Critical patent/CN114154619A/en
Publication of CN114154619A publication Critical patent/CN114154619A/en
Pending legal-status Critical Current

Links

Images

Classifications

    • 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
    • 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
    • G06N3/044—Recurrent networks, e.g. Hopfield networks
    • 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
    • G06N3/049—Temporal neural networks, e.g. delay elements, oscillating neurons or pulsed inputs
    • 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
    • G06N3/08—Learning methods
    • G06N3/084—Backpropagation, e.g. using gradient descent

Abstract

The invention discloses a ship track prediction method based on CNN and BILSTM. Step 1: preprocessing the information data; the information data are collected through the ship AIS and comprise longitude, latitude, course and navigational speed; step 2: dividing the information data preprocessed in the step 1 into a training set and a test set; and step 3: importing the training set in the step 2 into a Convolutional Neural Network (CNN) for feature extraction; and 4, step 4: combining the features extracted in the step 3 and the data of the training set to form input data of trajectory prediction; and 5: importing the input data in the step 4 into a BILSTM neural network model for learning to obtain a hidden ship motion rule model in the track data; step 6: and (5) predicting the track of the ship by using the model in the step 5. The method is used for solving the problem of low accuracy of ship track prediction in a complex water traffic environment.

Description

Ship track prediction method based on CNN and BILSTM
Technical Field
The invention belongs to the technical field of time series prediction; in particular to a ship track prediction method based on CNN and BILSTM.
Background
With the continuous development of marine transportation, strengthening the safety management of ships and monitoring key targets is more urgent. In many coastal and port water areas with dense traffic and complex conditions, the accuracy and effectiveness of marine traffic accident early warning are also particularly important. The trend information of the ship can be mastered in advance, so that the occurrence of marine traffic accidents such as ship reef touch, collision and the like can be effectively reduced. The analysis of the ship sailing track can obtain useful information of the sea channel and the behavior pattern of the ship. Due to the fact that ship navigation and vehicle driving characteristics are different, obvious road network constraint is avoided, the random degree of a flight path is large, and prediction difficulty is large. A traditional ship track prediction method adopts a method for constructing a kinetic equation, the method needs professional knowledge support, and targeted modification is needed according to different ships and scenes, so that the method is poor in adaptability. At present, machine learning is adopted as a mainstream method, and parameter learning can be performed according to a historical track and a current driving track, so that a prediction model has good adaptability.
At present, as the water traffic environment is increasingly complex, a single model is difficult to meet the use requirement in the aspect of accuracy, and a new challenge is provided for solving the real-time property and accuracy of ship track prediction.
Disclosure of Invention
The invention provides a ship track prediction method based on CNN and BILSTM, which is used for solving the problem of low accuracy of ship track prediction in a complex water traffic environment.
The invention is realized by the following technical scheme:
a ship track prediction method based on CNN and BILSTM comprises the following steps:
step 1: preprocessing the information data; the information data are collected by an automatic identification system AIS of the ship and comprise longitude, latitude, course and speed;
step 2: dividing the information data preprocessed in the step 1 into a training set and a test set;
and step 3: importing the training set in the step 2 into a Convolutional Neural Network (CNN) for feature extraction;
and 4, step 4: combining the features extracted in the step 3 and the data of the training set to form input data of trajectory prediction;
and 5: importing the input data in the step 4 into a BILSTM neural network model for learning to obtain a hidden ship motion rule model in the track data;
step 6: and (5) predicting the track of the ship by using the model in the step 5.
Further, the preprocessing of step 1 is specifically to screen out information in the automatic identification system AIS of the ship and having a large influence degree on the prediction of the ship track, delete obviously wrong data in the AIS, delete all data with very sparse ship track data, and correct data with less missing of the ship track data by using a linear interpolation method.
Further, the linear interpolation method is specifically set as (t)j,pj) For missing data, the two data closest to the missing data are (t)i,pi) And (t)k,pk) Then the complemented data is:
Figure BDA0003355263220000021
further, the preprocessed information data are normalized in the step 2, and then the data are divided into a training set and a testing set according to the ratio of 8: 2.
Further, the training set in the step 3 is input into a one-dimensional convolutional neural network as input data, and data features are obtained through convolution calculation of convolution kernels in the convolutional layer.
Further, the step 4 is specifically to input the feature data output by the convolutional layer into the pooling layer, perform average pooling, reduce the size of the data matrix, and finally input the output result of the convolutional neural network into the BILSTM for further extraction.
Further, the step 5 is specifically that an input gate, an output gate and a forgetting gate of the bilst network correspond to the writing and reading of the ship track characteristic sequence and the resetting operation of the previous state respectively, Forward calculation is performed once from the time 1 to the time t on the Forward layer, the output of the Forward hidden layer at each time is obtained and stored, reverse calculation is performed once from the time t to the time 1 on the Backward layer, the output of the Backward hidden layer at each time is obtained and stored, and finally the final output is obtained by combining the output results at each time of the Forward layer and the Backward layer.
Further, the expression at time t is shown as:
ft＝σ(Wf*[ht-1,xt]+bf)
it＝σ(Wi*[ht-1,xt]+bi)
Figure BDA0003355263220000022
Figure BDA0003355263220000024
ot＝σ(Wo*[ht-1,xt]+bo)
ht＝ot*tanh(ct)
wherein, WfWeight matrix, W, representing the correspondence of forgetting gateiRepresenting the weight matrix, W, corresponding to the input gatecWeight matrix corresponding to the current input cell state, representing WoOutputting a weight matrix corresponding to the gate; bfBias term representing forgetting gate, biRepresenting the offset term of the input gate, bcBias term representing the state of the currently input cell, boA bias term representing an output gate; σ denotes the activation function, tanh is the hyperbolic tangent functionCounting; x is the number oftAn input representing time t;
Figure BDA0003355263220000023
indicating the currently input unit state; c. CtRepresenting the state of the cell at the current time; f. oftRepresenting a forgetting gate; otRepresenting the output gate.
Further, step 6 specifically includes comparing the tested prediction result with the true value, and evaluating the accuracy of the prediction by using a mean square error, where the mean square error is an average of a sum of squares of differences between the predicted value and the true value, and the expression of MSE is:
Figure BDA0003355263220000031
where N represents the total duration of the prediction, t is the time sequence number, pretPredicted value, real, representing trajectorytRepresenting the true value of the trace.
The invention has the beneficial effects that:
the invention conveniently adopts the existing AIS data and the deep learning framework to construct and predict the model. According to the method, the acquired AIS data are preprocessed, the processed AIS data are used for preparation of model construction and training, and then the deep learning frame Keras is used for model construction, so that the cost is low, and the method is easy to implement.
The invention is different from the prior numerical prediction method, statistical prediction method and traditional machine learning prediction method. The invention constructs a deep learning model aiming at the ship track by utilizing the capability of deep learning to automatically extract the nonlinear relation between variables, the strong characteristic extraction capability of CNN and the strong time characteristic extraction capability of BILSTM, and realizes the deep characteristic extraction of the ship track characteristic.
The method effectively fuses two neural network algorithms, has stronger generalization capability and higher prediction precision, and can keep excellent prediction quality and stability on different AIS data sets.
The built model is trained by using tens of thousands of ship track data samples, and the trained model can be used for effectively improving the accuracy of ship track prediction.
The invention can set different prediction targets according to different requirements, and has strong expansibility.
Drawings
FIG. 1 is a flowchart of a ship trajectory prediction method according to the present invention.
FIG. 2 is a sparse graph of ship trajectories in the present invention.
FIG. 3 is a linear fitting graph of the ship track points in the invention.
FIG. 4 is a diagram of the convolution process in the convolutional neural network of the present invention.
FIG. 5 is a diagram of pooling in a convolutional neural network of the present invention.
Fig. 6 is a schematic diagram of a bidirectional long-term and short-term neural network model in the present invention.
Detailed Description
The technical solutions in the embodiments of the present invention will be described clearly and completely with reference to the accompanying drawings in the embodiments of the present invention, and it is obvious that the described embodiments are only a part of the embodiments of the present invention, and not all of the embodiments. All other embodiments, which can be derived by a person skilled in the art from the embodiments given herein without making any creative effort, shall fall within the protection scope of the present invention.
A ship track prediction method based on CNN and BILSTM comprises the following steps:
step 1: preprocessing the information data; the information data are collected by an automatic identification system AIS of the ship and comprise longitude, latitude, course and speed;
the preprocessing of the step 1 is to screen out information which is in the automatic identification system AIS of the ship and has a large influence degree on the prediction of the ship track, delete obviously wrong data in the AIS, delete all data with sparse ship track data, and correct data with less loss of the ship track data by adopting a linear interpolation method.
Because the data information recorded by the ship AIS data is more, in order to avoid the interference of other useless information, the information with larger influence degree with the ship track prediction in the AIS, such as longitude, latitude, course, navigational speed, time, unique identification of the ship and the like, is firstly screened out, besides, due to external factors such as human factors or weather, some error data may exist in the collected AIS data, the data can cause the accuracy of subsequent prediction, the original data needs to be processed, and the error data is deleted. The obvious errors in the AIS data include the following categories:
(1) the longitude of the ship is more than 180 degrees or is a negative value, and the latitude of the ship is more than 90 degrees or is a negative value;
(2) the ship speed is a negative value, and the ship course is more than 360 degrees or is a negative value;
(3) the acquisition time of the ship is not within a reasonable range (such as 1 month and 1 day 2022), and the marine Mobile Service identification code (MMSI) of the ship is not 9 digits.
The data after being removed can be observed, the AIS data has the condition of data missing, and the data with more data missing is completely deleted as shown in FIG. 2; the data with less data loss is corrected by linear interpolation as shown in fig. 3.
The linear interpolation method is specifically to set (t)j,pj) For missing data, the two data closest to the missing data are (t)i,pi) And (t)k,pk) Then the complemented data is:
Figure BDA0003355263220000041
step 2: dividing the information data preprocessed in the step 1 into a training set and a test set;
original data are converted into relative values in a specific range, the calculation complexity is reduced in the calculation process, the speed of seeking an optimal solution is improved, and the program operation time is reduced; on the other hand, the values of the attributes of the data are limited to the same specific range, so that the size difference of the data and the deviation generated in the calculation process are reduced, and the model prediction can be improvedAnd (6) measuring the precision. The normalized data is then divided into training and test sets in 8: 2. The invention adopts a normalization formula:
Figure BDA0003355263220000051
in the formula x*Representing the normalized value, x representing the data to be normalized, xminDenotes the minimum value, x, in the datamaxRepresenting the maximum value in the data.
And step 3: importing the training set in the step 2 into a Convolutional Neural Network (CNN) for feature extraction;
and (3) inputting the training set in the step (3) serving as input data into a one-dimensional convolution neural network, and obtaining data characteristics through convolution calculation of convolution kernels in the convolution layer.
Designing a CNN-BILSTM network, wherein the CNN is used for receiving input data, compressing and extracting important features of the data, the BILSTM is used for receiving the output of the CNN, capturing long-time dependency relationship, extracting time features and generating a prediction result;
the convolution layer in the convolutional neural network processes the data input in the step S2 through convolution kernel convolution, and outputs the data characteristics after convolution, and the formula of the convolution operation is:
Figure BDA0003355263220000052
wherein i represents the ith convolution kernel; g (i) denotes the ith feature map, a denotes input data, x, y, z denotes dimensions of the input data,
Figure BDA0003355263220000053
weight vector representing i-th layer convolution kernel, biIndicating the offset of the i-th layer convolution kernel.
The present invention employs a modified linear unit (Relu) as the activation function of neurons. FIG. 4 shows a one-dimensional convolution process, where x1～x6Is an input, c1～c4Is a feature vector. c. C1～c4Is obtained by a convolution operation, c1By x1～x3Calculated by convolution, and c1And x1～x3Have different weights between them, in calculating c2When these 3 weights correspond to x2～x4. The single hidden unit of one-dimensional convolution connects 3 to three inputs of the input layer, and this connection greatly reduces the number of parameters and also speeds up the training process of the neural network.
And 4, step 4: combining the features extracted in the step 3 and the data of the training set to form input data of trajectory prediction;
and step 4, specifically, inputting the characteristic data output by the convolutional layer into the pooling layer, performing average pooling, reducing the size of the data matrix, and finally inputting the output result of the convolutional neural network into the BILSTM for further extraction.
FIG. 5 shows a one-dimensional pooling process, c1～c4Is a feature vector, p, of the convolution layer calculationdIs a filler element, p1～p3Is the output of the pooling layer with a stride of 2. As shown in FIG. 5, p1By pooling pdAnd c1And (6) calculating. The invention adopts an average pooling method, and the formula is as follows:
Figure BDA0003355263220000054
wherein a isl(i, t) denotes the t neuron of the i characteristic map in the l layer, w denotes the width of the convolution kernel; j denotes the jth pooling core, and avg denotes the averaging.
P is pooled according to the average methoddAnd c1Average value of (1) is defined as p1Value of (a), p2And p3The value calculation method is also the same. The pooling layer will get an output p1～p3After passing through the convolutional layer and the pooling layer, the output result is input into BILSTM for further extraction.
And 5: importing the input data in the step 4 into a BILSTM neural network model for learning to obtain a hidden ship motion rule model in the track data;
the step 5 is specifically that an input gate, an output gate and a forgetting gate of the BILSTM network respectively correspond to the writing and reading of the ship track characteristic sequence and the resetting operation of the previous state in the invention; the input gate decides which information is important for memorizing; the output gate determines the information to be transferred; the forgetting gate decides which information should be remembered or forgotten; and performing Forward calculation once from the time 1 to the time t on the Forward layer to obtain and store the output of the Forward hidden layer at each time, performing Backward calculation once from the time t to the time 1 on the Backward layer to obtain and store the output of the Backward hidden layer at each time, and finally obtaining final output by combining the output results at each time of the Forward layer and the Backward layer.
The values of the 3 gates are set between 0 and 1 and the BILSTM network also uses the modified linear element Relu as an activation function for the neuron.
The expression at time t is shown as:
ft＝σ(Wf*[ht-1,xt]+bf)
it＝σ(Wi*[ht-1,xt]+bi)
Figure BDA0003355263220000061
Figure BDA0003355263220000062
ot＝σ(Wo*[ht-1,xt]+bo)
ht＝ot*tanh(ct)
wherein, WfWeight matrix, W, representing the correspondence of forgetting gateiRepresenting the weight matrix, W, corresponding to the input gatecWeight matrix corresponding to the current input cell state, representing WoOutputting a weight matrix corresponding to the gate; bfBias term representing forgetting gate, biRepresenting the offset term of the input gate, bcBias term representing the state of the currently input cell, boTo representAn offset term of the output gate; σ represents an activation function, and tanh is a hyperbolic tangent function; x is the number oftAn input representing time t;
Figure BDA0003355263220000063
indicating the currently input unit state; c. CtRepresenting the state of the cell at the current time; f. oftRepresenting a forgetting gate which determines the state c of the cell at the previous momentt-1How much to keep current time ct；itRepresenting an input gate which determines the input x of the network at the current momenttHow many cells to save to cell state ct；otRepresenting output gates which control the cell state ctHow many outputs to the current output value ht；htThe final output is determined by the output gate and the cell state.
Step 6: and (5) predicting the track of the ship by using the model in the step 5.
The step 6 is specifically to compare the tested prediction result with the actual value, and evaluate the accuracy of the prediction by adopting a mean square error, wherein the mean square error is an average value of the sum of squares of the difference between the predicted value and the actual value, and the expression of the MSE is as follows:
Figure BDA0003355263220000071
where N represents the total duration of the prediction, t is the time sequence number, pretPredicted value, real, representing trajectorytRepresenting the true value of the trace.
Aiming at the defects of the traditional ship track prediction method, the invention fully utilizes the existing research results and provides a prediction model based on the fusion of two deep neural networks. The model takes CNN as a bottom layer, extracts important features of input data, takes an output result as the input of a high-layer BILSTM, extracts time series features of a ship track, can fully consider the time relevance of the ship track, and obtains a more accurate prediction result.

Claims (9)

1. A ship track prediction method based on CNN and BILSTM is characterized by comprising the following steps:
step 1: preprocessing the information data; the information data are collected by an automatic identification system AIS of the ship and comprise longitude, latitude, course and speed;
step 2: dividing the information data preprocessed in the step 1 into a training set and a test set;
and step 3: importing the training set in the step 2 into a Convolutional Neural Network (CNN) for feature extraction;
and 4, step 4: combining the features extracted in the step 3 and the data of the training set to form input data of trajectory prediction;
and 5: importing the input data in the step 4 into a BILSTM neural network model for learning to obtain a hidden ship motion rule model in the track data;
step 6: and (5) predicting the track of the ship by using the model in the step 5.
2. The CNN and BILSTM based ship trajectory prediction method of claim 1, wherein: the preprocessing of the step 1 is to screen out information which is in the automatic ship identification system AIS and has a large influence degree on the prediction of the ship track, delete obviously wrong data in the automatic ship identification system AIS, delete all data with very sparse ship track data, and correct data with less ship track data loss by adopting a linear interpolation method.
3. The CNN and BILSTM based ship trajectory prediction method of claim 2, wherein: the linear interpolation method is specifically to set (t)j,pj) For missing data, the two data closest to the missing data are (t)i,pi) And (t)k,pk) Then the complemented data is:
Figure FDA0003355263210000011
4. the CNN and BILSTM based ship trajectory prediction method of claim 1, wherein: and 2, carrying out normalization processing on the preprocessed information data, and dividing the data into a training set and a test set according to a ratio of 8: 2.
5. The CNN and BILSTM based ship trajectory prediction method of claim 1, wherein: and (3) inputting the training set in the step (3) serving as input data into a one-dimensional convolution neural network, and obtaining data characteristics through convolution calculation of convolution kernels in the convolution layer.
6. The CNN and BILSTM based ship trajectory prediction method of claim 1, wherein: and step 4, specifically, inputting the characteristic data output by the convolutional layer into the pooling layer, performing average pooling, reducing the size of the data matrix, and finally inputting the output result of the convolutional neural network into the BILSTM for further extraction.
7. The CNN and BILSTM based ship trajectory prediction method of claim 1, wherein: the step 5 is specifically that an input gate, an output gate and a forgetting gate of the BILSTM network respectively correspond to the writing and reading of the ship track characteristic sequence and the resetting operation of the previous state, Forward calculation is performed once from the moment 1 to the moment t on the Forward layer, the output of the Forward hidden layer at each moment is obtained and stored, reverse calculation is performed once from the moment t to the moment 1 on the Backward layer, the output of the Backward hidden layer at each moment is obtained and stored, and finally the final output is obtained by combining the output results at each moment of the Forward layer and the corresponding moment of the Backward layer.
8. The CNN and BILSTM based ship trajectory prediction method of claim 7, wherein: the expression at time t is shown as:
ft＝σ(Wf*[ht-1,xt]+bf)
it＝σ(Wi*[ht-1,xt]+bi)
Figure FDA0003355263210000021
Figure FDA0003355263210000022
ot＝σ(Wo*[ht-1,xt]+bo)
ht＝ot*tanh(ct)
wherein, WfWeight matrix, W, representing the correspondence of forgetting gateiRepresenting the weight matrix, W, corresponding to the input gatecWeight matrix corresponding to the current input cell state, representing WoOutputting a weight matrix corresponding to the gate; bfBias term representing forgetting gate, biRepresenting the offset term of the input gate, bcBias term representing the state of the currently input cell, boA bias term representing an output gate; σ represents an activation function, and tanh is a hyperbolic tangent function; x is the number oftAn input representing time t;
Figure FDA0003355263210000023
indicating the currently input unit state; c. CtRepresenting the state of the cell at the current time; f. oftRepresenting a forgetting gate; i.e. itRepresenting an input gate; otRepresenting the output gate.
9. The CNN and BILSTM based ship trajectory prediction method of claim 7, wherein: the step 6 is specifically to compare the tested prediction result with the actual value, and evaluate the accuracy of the prediction by adopting a mean square error, wherein the mean square error is an average value of the sum of squares of the difference between the predicted value and the actual value, and the expression of the MSE is as follows:
Figure FDA0003355263210000024
where N represents the total duration of the prediction, t is the time sequence number, pretPredicted value, real, representing trajectorytRepresenting the true value of the trace.
CN202111349534.1A 2021-11-15 2021-11-15 Ship track prediction method based on CNN and BILSTM Pending CN114154619A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202111349534.1A CN114154619A (en) 2021-11-15 2021-11-15 Ship track prediction method based on CNN and BILSTM

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202111349534.1A CN114154619A (en) 2021-11-15 2021-11-15 Ship track prediction method based on CNN and BILSTM

Publications (1)

Publication Number Publication Date
CN114154619A true CN114154619A (en) 2022-03-08

Family

ID=80460030

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202111349534.1A Pending CN114154619A (en) 2021-11-15 2021-11-15 Ship track prediction method based on CNN and BILSTM

Country Status (1)

Country Link
CN (1) CN114154619A (en)

Cited By (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN115905434A (en) * 2022-10-26 2023-04-04 南京航空航天大学 Road network track completion method based on learning interpolation prediction
CN116306790A (en) * 2023-01-16 2023-06-23 西安电子科技大学 Offshore ship track real-time prediction method, system, equipment and medium based on CNN-GRU and attention mechanism
CN116451177A (en) * 2023-06-15 2023-07-18 创意信息技术股份有限公司 Track association method and device
CN116738565A (en) * 2023-05-18 2023-09-12 哈尔滨工程大学 Ship single break soaking time prediction method of convolution-circulation composite neural network
CN116738324A (en) * 2023-08-11 2023-09-12 太极计算机股份有限公司 Model training method and identification method for single-towing operation behavior of fishing boat

Cited By (8)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN115905434A (en) * 2022-10-26 2023-04-04 南京航空航天大学 Road network track completion method based on learning interpolation prediction
CN115905434B (en) * 2022-10-26 2023-10-13 南京航空航天大学 Road network track completion method based on learning interpolation prediction
CN116306790A (en) * 2023-01-16 2023-06-23 西安电子科技大学 Offshore ship track real-time prediction method, system, equipment and medium based on CNN-GRU and attention mechanism
CN116738565A (en) * 2023-05-18 2023-09-12 哈尔滨工程大学 Ship single break soaking time prediction method of convolution-circulation composite neural network
CN116451177A (en) * 2023-06-15 2023-07-18 创意信息技术股份有限公司 Track association method and device
CN116451177B (en) * 2023-06-15 2023-09-12 创意信息技术股份有限公司 Track association method and device
CN116738324A (en) * 2023-08-11 2023-09-12 太极计算机股份有限公司 Model training method and identification method for single-towing operation behavior of fishing boat
CN116738324B (en) * 2023-08-11 2023-12-22 太极计算机股份有限公司 Model training method and identification method for single-towing operation behavior of fishing boat

Similar Documents

Publication Publication Date Title
CN114154619A (en) Ship track prediction method based on CNN and BILSTM
CN112561148B (en) Ship track prediction method and system based on one-dimensional convolutional neural network and LSTM
CN110648014B (en) Regional wind power prediction method and system based on space-time quantile regression
CN112101363B (en) Full convolution semantic segmentation system and method based on cavity residual error and attention mechanism
CN112085163A (en) Air quality prediction method based on attention enhancement graph convolutional neural network AGC and gated cyclic unit GRU
CN111814956B (en) Multi-task learning air quality prediction method based on multi-dimensional secondary feature extraction
CN112838946B (en) Method for constructing intelligent sensing and early warning model based on communication network faults
CN110909909A (en) Short-term traffic flow prediction method based on deep learning and multi-layer spatiotemporal feature map
CN110555551B (en) Air quality big data management method and system for smart city
CN108960421B (en) Improved online forecasting method for speed of unmanned surface vehicle based on BP neural network
CN111142538B (en) Unmanned ship operation strategy real-time decision-making method for water environment monitoring
CN112036075A (en) Abnormal data judgment method based on environmental monitoring data association relation
CN113240199B (en) Port ship track prediction method based on DILATE _ TLSTM
CN113496104A (en) Rainfall forecast correction method and system based on deep learning
CN115220133B (en) Rainfall prediction method, device and equipment for multiple meteorological elements and storage medium
CN111141879B (en) Deep learning air quality monitoring method, device and equipment
CN113449919B (en) Power consumption prediction method and system based on feature and trend perception
CN114357670A (en) Power distribution network power consumption data abnormity early warning method based on BLS and self-encoder
CN116976227B (en) Storm water increasing forecasting method and system based on LSTM machine learning
CN113128769A (en) Intelligent flight delay prediction method based on deep learning
CN108171002A (en) A kind of polypropylene melt index Forecasting Methodology based on semi-supervised mixed model
CN116611580A (en) Ocean red tide prediction method based on multi-source data and deep learning
CN114648095A (en) Air quality concentration inversion method based on deep learning
CN113723670A (en) Photovoltaic power generation power short-term prediction method with variable time window
CN116342657B (en) TCN-GRU ship track prediction method, system, equipment and medium based on coding-decoding structure

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