CN113240198A - Port ship track prediction method based on TCN model - Google Patents
Port ship track prediction method based on TCN model Download PDFInfo
- Publication number
- CN113240198A CN113240198A CN202110630771.9A CN202110630771A CN113240198A CN 113240198 A CN113240198 A CN 113240198A CN 202110630771 A CN202110630771 A CN 202110630771A CN 113240198 A CN113240198 A CN 113240198A
- Authority
- CN
- China
- Prior art keywords
- tcn
- model
- prediction
- ship
- track
- 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
Links
Images
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
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/20—Information retrieval; Database structures therefor; File system structures therefor of structured data, e.g. relational data
- G06F16/29—Geographical information databases
-
- 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
- G06N3/045—Combinations of networks
-
- 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
- G06Q10/06—Resources, workflows, human or project management; Enterprise or organisation planning; Enterprise or organisation modelling
- G06Q10/067—Enterprise or organisation modelling
Abstract
The invention relates to a TCN model-based port ship track prediction method, which comprises the following steps: s1, acquiring port AIS data, preprocessing the port AIS data to obtain a track prediction data set; s2, constructing a time sequence convolution block based on expansion causal convolution and residual connection, constructing a TCN model through time sequence convolution block stacking and one-dimensional convolution layers, using single-step output predicted by the TCN model as input of the next moment, continuously iterating to realize multi-step track prediction, and constructing a port ship track prediction model based on TCN; and S3, evaluating the prediction effect of the TCN model through three indexes of average absolute error, mean square error and goodness of fit to obtain a prediction result. The ship track prediction model based on the TCN is higher in prediction precision, and provides more accurate technical support for port safe production and efficient management in practical application.
Description
Technical Field
The invention relates to the technical field of intelligent prediction of ship tracks, in particular to a TCN (thyristor controlled network) model-based port ship track prediction method.
Background
With the rapid development of economy, the shipping industry has met with tremendous changes and the number of ships has been growing, thereby creating many areas of intense shipping. Although the proliferation of the number of ships brings prosperity of maritime trade, the problems of water traffic safety are easily caused: the air route is overloaded, the navigation channel is more crowded, and the self problems of the ship and accidents caused by human factors happen occasionally, so that the life and property safety of the crew and passengers are greatly threatened. Therefore, the ship must be effectively monitored, abnormal behaviors of the ship can be found timely, risks of traffic accidents on water can be reduced, and the monitoring method becomes a key problem for research of related personnel.
And establishing a ship navigation dynamic prediction model by using the three-layer BP neural network, and performing on-line dynamic longitude and latitude prediction for one minute in the future on a common sea area, wherein although the prediction precision meets the requirement, the dynamic characteristics such as the navigation speed, the course and the like are not predicted. The BP neural network is adopted to predict the future one minute including the navigational speed and the course, but the prediction time is too short, and the BP neural network is difficult to learn aiming at the problem of complex track prediction. With the continuous improvement of the heat of the LSTM on a time series prediction task, various improved LSTM prediction models appear in the aspect of ship track prediction, the bidirectional LSTM ship track prediction model proposed in 2018 by Miao Gao and the VLSTM model proposed in 2020 by Mengzhen Ding are based on a variational self-encoder, the long-time track iterative prediction including the position, the navigational speed and the heading is realized, and the longitude and latitude prediction accuracy is improved compared with that of the LSTM. However, the LSTM and the improved model thereof can only realize one-step prediction, and for realizing long-time multi-step multivariable track prediction, multiple times of iterative prediction are needed, and prediction errors are accumulated in the iterative process. The existing navigation track prediction research aims at the common sea area, and the track prediction research of the port area is very little. The port area differs from the ordinary sea area in that: in a relatively wide common sea area, the ship navigation freedom degree is high, the change of the speed and the course is smooth, the motion mode is relatively single, in a port area, the ship comes and goes frequently, the navigation is influenced by the port traffic state and the ship scheduling command, the motion mode is complex and changeable, the change of the speed and the course is frequent, the difficulty is increased compared with the common sea area in the prediction of the ship track in the port area, and the research method needs to be further researched.
Disclosure of Invention
The method utilizes the convolution neural network-time sequence convolution network with a specific structure to extract the characteristics of the track sequence from a space angle, constructs a port ship track prediction model, and realizes accurate multi-step prediction of the ship navigation track.
In order to achieve the purpose, the invention provides the following scheme:
a TCN model-based port ship track prediction method comprises the following steps:
s1, acquiring port AIS data, and preprocessing the acquired port AIS data to acquire a track prediction data set;
s2, constructing a time sequence convolution block based on expansion causal convolution and residual connection, constructing a TCN model through time sequence convolution block stacking and one-dimensional convolution layers, using single-step output predicted by the TCN model as input of the next moment, continuously iterating to realize multi-step track prediction, and constructing a port ship track prediction model based on TCN;
and S3, evaluating the prediction effect of the TCN model through three indexes of average absolute error, mean square error and goodness of fit to obtain a prediction result.
Preferably, in S2, the specific steps of constructing the TCN port ship trajectory multi-step prediction model include:
s1.1, defining characteristics of a ship navigation track at time intervals, wherein input track characteristics X (t) at t moment are as follows:
X(t)＝{ΔTimet,LATt,LONt,SOGt,COGt,VesselTypet}，
the predicted output trajectory feature y (t) at time t +1 is represented as:
Y(t+1)＝{ΔTimet+1,LATt+1,LONt+1,SOGt+1,COGt+1}；
s1.2, taking ship track characteristics X (t-L +1), …, X (t-1) and X (t) at continuous L moments as the input of the TCN model, processing input data into the input form of the TCN model, and predicting and outputting track characteristics Y (t +1) at the moment of t + 1;
s1.3, splicing and updating the track characteristic Y (t +1) at the t +1 moment and the ship type characteristic, constructing an input sequence predicted at the next moment together with the X (t-L +2), the …, the X (t-1) and the X (t-1), predicting the track at the t +2 moment, and sequentially iterating in the way to construct a TCN-based port ship track prediction model.
Preferably, in S2, the time-series convolution block includes two layers of dilation-causal convolution, two layers of non-linear mapping, and one residual concatenation.
Preferably, the nonlinear mapping adopts a Leaky Relu activation function, a WeightNorm is added after the dilation causal convolution in each time sequence convolution layer to carry out normalization processing on the weight, and a Dropout layer is also added for regularizing the network and reducing the overfitting phenomenon.
Preferably, the TCN model includes three time-series convolutional blocks and one-dimensional convolutional layer, and is used to construct a long-term memory, so that the TCN model generates an output sequence with the same length as an input sequence.
Preferably, the input features of the TCN model further include: time interval, latitude and longitude, speed to ground, heading to ground, ship type.
Preferably, an optimizer in the TCN port ship track prediction model adopts an Adam function.
Preferably, the TCN port ship track prediction model sets a gradient threshold value and limits a gradient upper limit through a gradient cutting strategy, so as to avoid a gradient explosion phenomenon.
Preferably, the weight initialization in the TCN port ship trajectory prediction model adopts He initialization instead of the conventional initialization complying with normal distribution, so as to ensure that the input-output variance of each layer is unchanged.
The invention has the beneficial effects that:
compared with a multi-step iterative port track prediction model, the ship track prediction model based on the TCN has higher prediction precision, proves that the convolutional neural network TCN in a specific form is also suitable for time sequence prediction tasks such as ship track prediction, provides more model structure innovation ideas for port ship track prediction, and provides more accurate technical support for port safe production and high-efficiency management in practical application.
Drawings
In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings needed to be used in the embodiments will be briefly described below, and it is obvious that the drawings in the following description are only some embodiments of the present invention, and it is obvious for those skilled in the art to obtain other drawings without inventive exercise.
FIG. 1 is a flow chart of the method of the present invention;
FIG. 2 is a schematic diagram of a sequential rolling block configuration according to the present invention;
FIG. 3 is a schematic diagram of the LSTM model training loss curve in the embodiment of the present invention;
FIG. 4 is a schematic diagram of the LSTM model predicting absolute errors at different time steps in the embodiment of the present invention;
FIG. 5 is a TCN model training loss curve of the present invention;
FIG. 6 is a graph of the TCN and LSTM iterative prediction error of the present invention.
Detailed Description
The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the 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.
In order to make the aforementioned objects, features and advantages of the present invention comprehensible, embodiments accompanied with figures are described in further detail below.
The invention provides a TCN model-based port ship track prediction method (as shown in figure 1), which comprises the following steps:
s1, acquiring port AIS data, and preprocessing the acquired port AIS data to acquire a track prediction available data set;
s2, constructing a time sequence convolution block based on expansion causal convolution and residual connection, constructing a TCN model through time sequence convolution block stacking and one-dimensional convolution layers, using single-step output predicted by the TCN model as input of the next moment, continuously iterating to realize multi-step track prediction, and constructing a port ship track prediction model based on TCN;
and S3, evaluating the prediction effect of the TCN model through three indexes of average absolute error, mean square error and goodness of fit to obtain a prediction result.
Further optimizing the scheme, the specific steps of constructing the TCN port ship track prediction model comprise:
s1.1, taking the time interval as the characteristic of the ship navigation track, wherein the input track characteristic X (t) at the time t is as follows:
X(t)＝{ΔTimet,LATt,LONt,SOGt,COGt,VesselTypet}，
the predicted output trajectory feature y (t) at time t +1 is represented as:
Y(t+1)＝{ΔTimet+1,LATt+1,LONt+1,SOGt+1,COGt+1}；
s1.2, taking ship track characteristics X (t-L +1), …, X (t-1) and X (t) at continuous L moments as the input of the TCN model, processing input data into the input form of the TCN model, and predicting and outputting track characteristics Y (t +1) at the moment of t + 1;
s1.3, splicing and updating the track characteristic Y (t +1) at the t +1 moment and the ship type characteristic, constructing an input sequence predicted at the next moment together with the X (t-L +2), the …, the X (t-1) and the X (t-1), predicting the track at the t +2 moment, and sequentially iterating in the way to construct a TCN-based port ship track prediction model.
In a further optimization, the sequential convolution block comprises two layers of dilated causal convolution, two layers of non-linear mapping and one residual concatenation, as shown in fig. 2. By stacking a plurality of time sequence convolution layers, the network depth can be deepened, the prediction precision is improved to a certain extent, and the parallel operation of the time sequence convolution layers enables the model calculation speed to be higher.
In FIG. 2, k denotes the kth time series convolutional layer,
represents the output of the time-series convolution layer. A time sequence convolution block comprises two layers of extended causal convolution and nonlinear mapping and one-time residual connection, wherein the nonlinear mapping adopts a Leaky Relu activation function, Weightnorm is added after the extended causal convolution in each layer of time sequence convolution layer to normalize the weight, model training acceleration is carried out through the weight of a rewriting depth network, and Dropout is added to regularize the network to reduce overfitting.
In a further optimization scheme, the TCN model includes three time-series convolutional layers and a one-dimensional convolutional layer, and is used to construct a long-term memory, so that the TCN model generates an output sequence with the same length as an input sequence.
Through stacking of 3 time sequence convolution blocks, the conventional full-connection layer is replaced by the one-dimensional convolution layer, the information of the whole input sequence is sensed to help to construct long-term memory, the output sequence with the length equal to that of the input sequence can be generated by the TCN, and single-step track characteristic prediction output is realized through the model structure. The model input features are time interval, longitude and latitude, speed to ground, heading to ground, ship type, but the input data needs to be processed into an input form suitable for the TCN model: (Sample, Feature, Timestep), the output of the model is the trajectory Feature at the next moment: time interval, longitude, latitude, speed to ground, heading to ground.
In the model, the optimizer adopts an Adam function, and has the advantages of high convergence rate and easiness in parameter adjustment; in addition, in order to solve the problem of Gradient explosion or Gradient dispersion which may occur in the model, firstly, a Gradient Clipping (Gradient Clipping) strategy is adopted, namely a Gradient threshold value is set, the upper limit of the Gradient is limited, and the Gradient explosion is effectively avoided; and secondly, replacing the conventional initialization which is subject to normal distribution by using He initialization in a weight initialization mode, wherein the significance of the He initialization method is that when the model uses a Leaky ReLU activation function, the input-output variance of each layer is ensured to be unchanged, the output is prevented from tending to 0, and thus the gradient dispersion condition is avoided.
In order to verify the technical effect of the invention, in the embodiment, the LSTM-based port ship track prediction model and the TCN-based port ship track prediction model are compared and analyzed.
Like the TCN port ship track prediction model, the LSTM port ship track prediction model also takes the single-step output predicted by the LSTM model as the input of the next moment, and then continuously iterates to perform multi-step track prediction, and finally constructs the LSTM port ship track-based prediction model.
Firstly, configuring an experimental environment and model evaluation indexes:
1. experimental Environment
The experimental environment comprises two parts of software configuration and hardware configuration, and configuration information is shown in table 1:
TABLE 1
2. Model evaluation index
The LSTM and TCN trajectory prediction models employ Mean Absolute Error (MAE), Mean Square Error (MSE), and goodness of fit (R)2) Three indexes are used for evaluating the model effect.The smaller the values of MAE and MSE indicate the higher the model prediction accuracy. R2Also known as determination coefficients, whose values are typically [0, 1 ]]The closer to 1, the better the model fit, and when R is2When the value is less than 0, the learned model is not as good as the prediction effect of the average value of the true values. The three evaluation indices are defined as follows:
wherein y isiIn order to be the true value of the value,
in order to predict the value of the target,
and the average value of the real values of the samples in the test set is shown, and n is the number of the samples.
Performing a single-step track prediction experiment based on the constructed LSTM network model, wherein a training set, a verification set and a test set are divided according to the following steps of: 1: 1, the optimizer adopts Adam, the loss function adopts MSE, and the evaluation criteria are MAE, MSE and R2. Other superparameters were determined by multiple experiments as: the training loss curve for the model is shown in FIG. 3, with time step set to 6, Epoch set to 20, Batch size set to 64, and learning rate set to 5 e-5.
As shown in fig. 3, loss is a loss function MSE, epochs are training rounds, and it can be seen that the loss of the training set and the verification set of the LSTM model is continuously reduced in the first three epochs, which indicates that the model is in the learning stage at this stage, and then the loss curve tends to be stable and does not decrease, which indicates that the model has converged, the model training is completed, and the trained model is used to perform single-step trajectory prediction on the test set, so as to obtain a predicted absolute error MAE of 0.031234, where the single-step performance of the prediction of a single feature is measured by three evaluation indexes, and the results are shown in table 2:
TABLE 2
As can be seen from Table 2, the LSTM-based port ship track prediction model has good effect in single-step prediction of four characteristics of latitude, longitude, speed to ground and heading to ground, the absolute errors of prediction MAE are 0.0075889, 0.0088575, 0.0170881 and 0.0213580 respectively, and the latitude and longitude prediction determination coefficient R2All reach over 0.99, and the determining coefficient R of the course of the ground2Above 0.95, the coefficient for determining the prediction of the speed of the ground is 0.8968590, and still needs to be improved. For the prediction of the time interval, the absolute error MAE reaches 0.1057801, and the coefficient R is determined20.0404516, the prediction effect on the time interval is poor, because the original data set itself has the problem of uneven distribution of the acquisition time interval.
In a time series prediction task, the LSTM prediction effect depends on the size of a time step to a great extent, so that the popular understanding is that the value to be predicted is related to information of previous time points, and the prediction precision is influenced by too little information provided due to too small time step. Meanwhile, the larger the time step is, the better the prediction effect is, although the larger the time step is, the larger the information quantity provided by the time step is, the smaller the correlation between the information of the forward time point and the information of the prediction time point is, the historical time information which exceeds a certain time step and is forward is useless redundant information, which is not helpful for improving the prediction precision, and for the LSTM model, the longer the time step is, the higher the risk of gradient disappearance or explosion is.
The setting of the time step parameters of the LSTM model is selected according to the experimental results under different time steps, integers are taken between [3, 9] for the time steps, each time step is subjected to multiple experiments, the experimental results are evaluated by using MAE, the average absolute prediction errors under different time steps are obtained, and the average absolute prediction errors are shown in the attached figure 4, and it can be known that the prediction errors are the minimum when the time step is 6.
TCN-based port ship track prediction experiment result analysis:
TCN Single step predictive analysis
The single-step ship navigation track prediction experiment is carried out based on the built TCN model, and like the LSTM single-step prediction model, the method still comprises the following steps of: 1: 1, dividing a data set into a training set, a verification set and a test set, wherein MSE is used as a loss function, and evaluation criteria are MAE, MSE and R2. Obtaining the optimal hyper-parameter setting of the model through multiple experiments: the time step is 8, the Batch size is 64, the dropout is 0.2, the gradient clipping threshold is 0.2, the convolution kernel size is 3, the learning rate is 1e-4, the number of nodes of the hidden layer is 32, and the loss curve of the model training is shown in fig. 5:
as shown in FIG. 5, the loss of the training set and the verification set of the TCN model is sharply reduced in the first Epoch, and the loss curve tends to converge smoothly in the two epochs, so that the convergence rate is faster than that of the LSTM model. The trained model is used for single-step track prediction on a test set to obtain a prediction absolute error MAE of 0.0285365, which is lower than that of an LSTM model, wherein the prediction performance of a single feature is implemented by MSE, MAE and R2Three evaluation indexes were evaluated, and the results are shown in table 3:
TABLE 3
As can be seen from the comparison of the TCN model single-step prediction results obtained in the table 3 and the LSTM model single-step prediction results in the tables 3-3, the prediction errors of the TCN model to time, latitude, longitude, navigational speed and course are all lower than the prediction errors of the LSTM model to the five characteristics, and the determination coefficients are all improved, wherein the determination coefficients for the real-time latitude and longitude prediction of the ship even reach more than 0.999, which also shows that the TCN model has better performance in the aspect of ship single-step trajectory prediction.
Comparing multi-step track prediction errors based on LSTM and TCN models:
based on the trained LSTM single-step ship track prediction model and the trained TCN single-step ship track prediction model, track prediction experiments at the next 8 continuous moments are carried out in a recursive iteration mode, so that the influence of iteration times, namely prediction step length N, on model track prediction errors is discussed. The absolute error MAE of the two model predictions at different iterations is shown in fig. 6, and as can be seen from fig. 6, the absolute error of both the TCN and LSTM model predictions increases with the number of iterations, which is an error accumulation process.
For the prediction of the ship track of the port, the port range is small, so the requirement on the prediction step length N is not too long, the track of 5 points in the future which is close to 6 minutes is predicted to basically meet the requirement of the port, and the ship sailing track prediction performance with the prediction step length N of 5 is concerned. Respectively carrying out iterative prediction by using the trained LSTM and TCN single-step track prediction models to realize track point prediction of the port ship at 5 continuous moments in the future, wherein the prediction evaluation results of each characteristic of five track points are shown in Table 4:
TABLE 4
As can be seen from the table, the two models achieve high prediction accuracy on the position, latitude and longitude of the ship, and in general, track point characteristic prediction errors at five future moments based on the TCN model are lower than those predicted based on the LSTM model, the fitting coefficient is higher, and the track prediction performance is better. Compared with tables 2 and 3, the prediction errors of the two models are increased and the fitting coefficient is reduced after multi-step iteration, which shows that the prediction errors are accumulated and increased along with the increase of the iteration times by realizing multi-step prediction in an iterative prediction mode, the longer the prediction time is, the larger the errors are, and thus the port ship track prediction model based on the LSTM and the TCN is only suitable for track prediction in a shorter time in the future.
The invention builds the method of LSTM and TCN model through iterative predictionThe formula realizes the ship track prediction at five future moments, and uses mean square error MSE, absolute error MAE and determination coefficient R2The evaluation compares the experimental results of the two models, and the experiment shows that the prediction accuracy of the ship track prediction model based on the TCN is higher than that of the ship track prediction model based on the LSTM, so that the convolutional neural network TCN in a specific form is proved to be also suitable for time sequence prediction tasks such as ship track prediction.
Compared with a multi-step iterative port track prediction model, the ship track prediction model based on the TCN has higher prediction precision, proves that the convolutional neural network TCN in a specific form is also suitable for time sequence prediction tasks such as ship track prediction, provides more model structure innovation ideas for port ship track prediction, and provides more accurate technical support for port safe production and high-efficiency management in practical application.
The above-described embodiments are merely illustrative of the preferred embodiments of the present invention, and do not limit the scope of the present invention, and various modifications and improvements of the technical solutions of the present invention can be made by those skilled in the art without departing from the spirit of the present invention, and the technical solutions of the present invention are within the scope of the present invention defined by the claims.
Claims (9)
1. A TCN model-based port ship track prediction method is characterized by comprising the following steps:
s1, acquiring port AIS data, and preprocessing the acquired port AIS data to acquire a track prediction data set;
s2, constructing a time sequence convolution block based on expansion causal convolution and residual connection, constructing a TCN model through time sequence convolution block stacking and one-dimensional convolution layers, using single-step output predicted by the TCN model as input of the next moment, continuously iterating to realize multi-step track prediction, and constructing a port ship track prediction model based on TCN;
and S3, evaluating the prediction effect of the TCN model through three indexes of average absolute error, mean square error and goodness of fit to obtain a prediction result.
2. The TCN model-based harbor ship track prediction method as claimed in claim 1, wherein in S2, the specific steps of constructing TCN harbor ship track multi-step prediction model comprise:
s1.1, defining characteristics of a ship navigation track at time intervals, wherein input track characteristics X (t) at t moment are as follows:
X(t)＝{ΔTimet,LATt,LONt,SOGt,COGt,VesselTypet}，
the predicted output trajectory feature y (t) at time t +1 is represented as:
Y(t+1)＝{ΔTimet+1,LATt+1,LONt+1,SOGt+1,COGt+1}；
s1.2, taking ship track characteristics X (t-L +1), …, X (t-1) and X (t) at continuous L moments as the input of the TCN model, processing input data into the input form of the TCN model, and predicting and outputting track characteristics Y (t +1) at the moment of t + 1;
s1.3, splicing and updating the track characteristic Y (t +1) at the t +1 moment and the ship type characteristic, constructing an input sequence predicted at the next moment together with the X (t-L +2), the …, the X (t-1) and the X (t-1), predicting the track at the t +2 moment, and sequentially iterating in the way to construct a TCN-based port ship track prediction model.
3. The TCN model-based harbor ship trajectory prediction method according to claim 2, wherein in S2, the time series convolution block comprises two layers of dilation causal convolution, two layers of non-linear mapping and one residual join.
4. The TCN model-based port ship trajectory prediction method of claim 3, wherein a Leaky Relu activation function is adopted in the nonlinear mapping, Weiightnorm is added after the dilation causal convolution in each time sequence convolution layer to normalize the weights, and a Dropout layer is added for regularization of a network to reduce an overfitting phenomenon.
5. The TCN model-based harbor vessel trajectory prediction method of claim 2, wherein the TCN model comprises three sequential convolution blocks and one-dimensional convolution layer for constructing long-term memory, so that the TCN model generates an output sequence with the same length as an input sequence.
6. The TCN model-based port vessel trajectory prediction method of claim 2, wherein the input features of the TCN model further comprise: time interval, latitude and longitude, speed to ground, heading to ground, ship type.
7. The TCN model-based port ship trajectory prediction method of claim 1, wherein an optimizer in the TCN port ship trajectory prediction model employs an Adam function.
8. The TCN model-based port ship trajectory prediction method of claim 1, wherein the TCN port ship trajectory prediction model sets a gradient threshold value and limits an upper gradient limit through a gradient clipping strategy to avoid a gradient explosion phenomenon.
9. The TCN model-based port ship trajectory prediction method of claim 1, wherein weight initialization in the TCN port ship trajectory prediction model adopts He initialization instead of conventional initialization complying with normal distribution, so as to ensure that the input-output variance of each layer is unchanged.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110630771.9A CN113240198A (en) | 2021-06-07 | 2021-06-07 | Port ship track prediction method based on TCN model |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110630771.9A CN113240198A (en) | 2021-06-07 | 2021-06-07 | Port ship track prediction method based on TCN model |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN113240198A true CN113240198A (en) | 2021-08-10 |
Family
ID=77136981
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202110630771.9A Pending CN113240198A (en) | 2021-06-07 | 2021-06-07 | Port ship track prediction method based on TCN model |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN113240198A (en) |
Cited By (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN115293297A (en) * | 2022-10-08 | 2022-11-04 | 武汉理工大学 | Method for predicting track of ship driven by intention |
| CN116232772A (en) * | 2023-05-08 | 2023-06-06 | 中国人民解放军国防科技大学 | Unsupervised network data intrusion detection method based on ensemble learning |
| CN116342657A (en) * | 2023-03-29 | 2023-06-27 | 西安电子科技大学 | TCN-GRU ship track prediction method, system, equipment and medium based on coding-decoding structure |
| CN116342657B (en) * | 2023-03-29 | 2024-04-26 | 西安电子科技大学 | TCN-GRU ship track prediction method, system, equipment and medium based on coding-decoding structure |
Citations (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110909926A (en) * | 2019-11-18 | 2020-03-24 | 中国计量大学 | TCN-LSTM-based solar photovoltaic power generation prediction method |
| CN111114554A (en) * | 2019-12-16 | 2020-05-08 | 苏州智加科技有限公司 | Method, device, terminal and storage medium for predicting travel track |
| CN111292360A (en) * | 2020-01-21 | 2020-06-16 | 四川省交通勘察设计研究院有限公司 | Method and system for recommending ship driving route |
| US20210001897A1 (en) * | 2019-07-03 | 2021-01-07 | Waymo Llc | Agent trajectory prediction using anchor trajectories |
| CN112561148A (en) * | 2020-12-09 | 2021-03-26 | 北京京航计算通讯研究所 | Ship track prediction method and system based on one-dimensional convolutional neural network and LSTM |
| KR102235787B1 (en) * | 2020-01-09 | 2021-04-05 | 씨드로닉스(주) | Device and method for monitoring a berthing |
| CN112906858A (en) * | 2021-01-26 | 2021-06-04 | 武汉工程大学 | Real-time prediction method for ship motion trail |
-
2021
- 2021-06-07 CN CN202110630771.9A patent/CN113240198A/en active Pending
Patent Citations (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20210001897A1 (en) * | 2019-07-03 | 2021-01-07 | Waymo Llc | Agent trajectory prediction using anchor trajectories |
| CN110909926A (en) * | 2019-11-18 | 2020-03-24 | 中国计量大学 | TCN-LSTM-based solar photovoltaic power generation prediction method |
| CN111114554A (en) * | 2019-12-16 | 2020-05-08 | 苏州智加科技有限公司 | Method, device, terminal and storage medium for predicting travel track |
| KR102235787B1 (en) * | 2020-01-09 | 2021-04-05 | 씨드로닉스(주) | Device and method for monitoring a berthing |
| CN111292360A (en) * | 2020-01-21 | 2020-06-16 | 四川省交通勘察设计研究院有限公司 | Method and system for recommending ship driving route |
| CN112561148A (en) * | 2020-12-09 | 2021-03-26 | 北京京航计算通讯研究所 | Ship track prediction method and system based on one-dimensional convolutional neural network and LSTM |
| CN112906858A (en) * | 2021-01-26 | 2021-06-04 | 武汉工程大学 | Real-time prediction method for ship motion trail |
Cited By (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN115293297A (en) * | 2022-10-08 | 2022-11-04 | 武汉理工大学 | Method for predicting track of ship driven by intention |
| CN115293297B (en) * | 2022-10-08 | 2023-01-20 | 武汉理工大学 | Method for predicting track of ship driven by intention |
| CN116342657A (en) * | 2023-03-29 | 2023-06-27 | 西安电子科技大学 | TCN-GRU ship track prediction method, system, equipment and medium based on coding-decoding structure |
| CN116342657B (en) * | 2023-03-29 | 2024-04-26 | 西安电子科技大学 | TCN-GRU ship track prediction method, system, equipment and medium based on coding-decoding structure |
| CN116232772A (en) * | 2023-05-08 | 2023-06-06 | 中国人民解放军国防科技大学 | Unsupervised network data intrusion detection method based on ensemble learning |
| CN116232772B (en) * | 2023-05-08 | 2023-07-07 | 中国人民解放军国防科技大学 | Unsupervised network data intrusion detection method based on ensemble learning |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN111832814B (en) | Air pollutant concentration prediction method based on graph attention mechanism | |
| CN113240198A (en) | Port ship track prediction method based on TCN model | |
| CN112116162B (en) | Power transmission line icing thickness prediction method based on CEEMDAN-QFAOA-LSTM | |
| CN111310965A (en) | Aircraft track prediction method based on LSTM network | |
| CN113126676B (en) | Livestock and poultry house breeding environment parameter intelligent control system | |
| Afshin et al. | Long term rainfall forecasting by integrated artificial neural network-fuzzy logic-wavelet model in Karoon basin | |
| CN114912077B (en) | Sea wave forecasting method integrating random search and mixed decomposition error correction | |
| CN112257847A (en) | Method for predicting geomagnetic Kp index based on CNN and LSTM | |
| CN112364560A (en) | Intelligent prediction method for working hours of mine rock drilling equipment | |
| Sylaios et al. | A fuzzy inference system for wind-wave modeling | |
| CN115422687A (en) | Service life prediction method of rolling bearing | |
| CN114911185A (en) | Security big data Internet of things intelligent system based on cloud platform and mobile terminal App | |
| Jiang et al. | Deterministic and probabilistic multi-time-scale forecasting of wind speed based on secondary decomposition, DFIGR and a hybrid deep learning method | |
| CN113128666A (en) | Mo-S-LSTMs model-based time series multi-step prediction method | |
| CN114970745B (en) | Intelligent security and environment big data system of Internet of things | |
| Han et al. | A Hybrid BPNN-GARF-SVR PredictionModel Based on EEMD for Ship Motion. | |
| CN114358244B (en) | Big data intelligent detection system of pressure based on thing networking | |
| CN115659201A (en) | Gas concentration detection method and monitoring system for Internet of things | |
| CN113255963A (en) | Road surface use performance prediction method based on road element splitting and deep learning model LSTM | |
| Yang et al. | ELM weighted hybrid modeling and its online modification | |
| Jin | Feasibility Analysis and Countermeasures of Psychological Health Training Methods for Volleyball Players Based on Artificial Intelligence Technology | |
| CN113688774B (en) | Advanced learning-based high-rise building wind induced response prediction and training method and device | |
| CN115016275B (en) | Intelligent feeding and livestock house big data Internet of things system | |
| Zhan et al. | Long Short-term Memory modeling method with monotonicity analysis as constraints base on Spearman coefficient | |
| Wu et al. | A BP neural network model for the demand forecasting of road freight transportation system |
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 |