A regional ship traffic flow prediction method and system

Abstract

The application provides a regional ship traffic flow prediction method and system, which comprises the following steps: obtaining AIS data of different types of ships and preprocessing and drawing a density heat map; multiple target areas are framed in the density heat map; the current latitude and longitude coordinates of the ships in the preprocessed AIS data and the coordinates of the vertices of each target area are used to automatically determine the number of ships in each target area by using the ray method every certain period of time; the ship coefficient conversion of the number of ships of different types of ships in each target area is carried out based on the AIS data, so as to obtain the ship flow data of each type of ship in each target area; finally, a ship flow prediction model is established according to the ship flow data based on Chebyshev network and long short-term memory network, the ship flow of each type of ship in each area in a future period of time is predicted according to the ship flow prediction model, and the accuracy of the regional ship traffic flow prediction is effectively improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of ship traffic flow prediction, in particular to a regional ship traffic flow method and system. BACKGROUND

[0002] Ship traffic flow prediction is an important content of ship traffic flow research. Mining water traffic patterns and making accurate ship traffic flow prediction can provide certain reference basis for port and channel capacity planning. Existing research on ship traffic flow prediction is mostly based on fixed cross-section, and the flow through the fixed cross-section is counted and predicted. There are few researches on counting the ship traffic in a specified area within a certain period of time. When the ship density in a region is large, the ships influence each other, and the speed and heading will change with the flow.

[0003] Through summarizing the existing ship traffic flow prediction model, it is found that the existing traffic prediction model does not consider predicting the ship traffic in a specified area within a certain period of time, and does not consider the mutual influence between ships when the ship density in a region is large, resulting in low prediction accuracy of the existing traffic prediction model. SUMMARY

[0004] To solve the problem of low traffic flow prediction accuracy caused by less consideration of ship navigation behavior in complex traffic environment in the current ship traffic flow prediction process, the present application provides a regional ship traffic flow prediction method, which draws a density heat map based on ship AIS data, selects multiple target areas, uses the ray method and ship coefficient conversion method to obtain ship traffic data, and establishes a ship traffic flow prediction model based on Chebyshev network and long short-term memory network to predict the flow of different types of ships in the target water area, effectively improving the accuracy of regional ship traffic flow prediction. The present application also relates to a regional ship traffic flow prediction system.

[0005] The technical scheme of the present application is as follows:

[0006] A regional ship traffic flow prediction method, characterized in that it comprises the following steps:

[0007] Data acquisition and preprocessing step: acquiring AIS data of different types of ships, and preprocessing the AIS data;

[0008] Heat map drawing step: drawing a density heat map according to the preprocessed AIS data, and framing multiple target areas in the density heat map;

[0009] The ship quantity judgment step: according to the pre-processed AIS data of the current latitude and longitude position coordinates of the ship and the vertex coordinates of each target region, the number of ships in each target region is automatically judged by using the ray method every certain period of time;

[0010] The coefficient conversion step: based on the AIS data, the ship quantity of different types of ships in each target region is converted by data mining technology, and the ship flow data of each type of ship in each target region is obtained according to the converted ship coefficient;

[0011] The flow prediction step: according to the ship flow data, a ship flow prediction model is established based on Chebyshev network and long short-term memory network, and the ship flow of each type of ship in each region in a future period of time is predicted according to the ship flow prediction model.

[0012] Preferably, in the flow prediction step, the Chebyshev network uses Chebyshev polynomials as the convolution kernel of spectral graph convolution to convert global convolution into local convolution to reduce the calculation time of the model.

[0013] Preferably, in the data acquisition and preprocessing step, the pre-processing of the AIS data includes deleting the AIS data of the ship with a speed less than a preset speed threshold in the navigation.

[0014] Preferably, in the flow prediction step, an evaluation index is also established according to the predicted ship flow and the actual ship flow, the ship flow prediction model is evaluated according to the evaluation index and by using a comparative model, and the ship flow prediction model is verified as the optimal model;

[0015] The evaluation index includes root mean square error, mean absolute percentage error, mean absolute error, symmetric mean absolute percentage error, and mean square error, and the comparative model includes a long short-term memory network model, a graph convolution network model, and a graph attention network model.

[0016] Preferably, in the data acquisition and preprocessing step, the AIS data includes static data and dynamic data, the static data includes ship mobile service identification code, ship length, ship width, and ship name, and the dynamic data includes ship latitude and longitude position information, navigation state, time stamp, ground speed, draft, departure port, destination port, and estimated arrival time.

[0017] A regional ship traffic flow prediction system, characterized by comprising a data acquisition and preprocessing module, a heat map drawing module, a ship quantity judgment module, a coefficient conversion module and a flow prediction module connected in sequence,

[0018] The data acquisition and preprocessing module acquires AIS data of different types of ships and pre-processes the AIS data;

[0019] The heat map drawing module draws a density heat map according to the preprocessed AIS data, and frames a plurality of target regions in the density heat map;

[0020] The ship quantity judgment module automatically judges the number of ships in each target region by using a ray method every certain time according to the current longitude and latitude position coordinates of the ships in the preprocessed AIS data and the coordinates of each vertex constituting each target region.

[0021] The coefficient conversion module converts the number of ships of different types in each target region by using a data mining technology based on the AIS data, and obtains ship flow data of each type of ship in each target region according to the converted ship coefficients.

[0022] The flow prediction module establishes a ship flow prediction model based on Chebyshev networks and long short-term memory networks according to the ship flow data, and predicts the ship flow of each type of ship in each region in a future time period according to the ship flow prediction model.

[0023] Preferably, in the flow prediction module, the Chebyshev network uses Chebyshev polynomials as a convolution kernel for spectral graph convolution, converts global convolution into local convolution to reduce the calculation time of the model.

[0024] Preferably, in the data acquisition and preprocessing module, the preprocessing of the AIS data includes deleting AIS data of ships with a speed less than a preset speed threshold in navigation.

[0025] Preferably, in the flow prediction module, an evaluation index is further established according to the predicted ship flow and the actual ship flow, the ship flow prediction model is evaluated according to the evaluation index and by using a comparative model, and the ship flow prediction model is verified as an optimal model.

[0026] The evaluation index includes root mean square error, mean absolute percentage error, mean absolute error, symmetric mean absolute percentage error, and mean square error, and the comparative model includes a long short-term memory network model, a graph convolution network model, and a graph attention network model.

[0027] Preferably, the AIS data includes static data and dynamic data, the static data includes a ship mobile service identification code, a ship length, a ship width, and a ship name, and the dynamic data includes ship longitude and latitude position information, a navigation state, a time stamp, an over-the-ground speed, a draft, a departure port, a destination port, and a predicted arrival time.

[0028] The present application has the following advantages:

[0029] The application provides a regional ship traffic flow prediction method, which comprises the following steps in sequence: data acquisition and preprocessing, heat map drawing, ship quantity judgment, coefficient conversion and flow prediction.

[0030] The application also relates to a regional ship traffic flow prediction system, which corresponds to the above-mentioned regional ship traffic flow prediction method and can be understood as a system for realizing the above-mentioned regional ship traffic flow prediction method, and comprises a data acquisition and preprocessing module, a heat map drawing module, a ship quantity judgment module, a coefficient conversion module and a flow prediction module which are connected in sequence, and the modules work cooperatively, the density heat map is drawn based on the ship AIS data and a plurality of target regions are selected, the ship flow data is obtained by using the ray method and the ship coefficient conversion method, and the ship flow prediction model is established based on the Chebyshev network and the long short-term memory network to predict the ship flow in the target water area, which can effectively increase the comparability and accuracy of the maritime traffic research, make the ship quantity statistical result more realistic and persuasive, greatly reduce the calculation time of the model, and effectively improve the prediction speed and accuracy of the regional ship traffic flow. BRIEF DESCRIPTION OF DRAWINGS

[0031] Figure 1 It is a flow chart of the regional ship traffic flow prediction method of the application.

[0032] Figure 2 is a schematic diagram of the ray method of the present application.

[0033] Figures 3-5 is an effect diagram of the ship traffic flow prediction model of the present application and other comparative models predicting the ship traffic flow of each area. DETAILED DESCRIPTION

[0034] The present application will be described below in conjunction with the accompanying drawings.

[0035] The present application relates to a regional ship traffic flow prediction method, and a flowchart of the method is shown in Figure 1 and sequentially comprises the following steps:

[0036] The data acquisition and preprocessing step is used for data analysis on the original AIS data in the water area, selection of a study area, and obtaining screened AIS data according to the selected study area, including acquisition of AIS data of different types of ships and preprocessing of the AIS data.

[0037] Specifically, first, the AIS data of different types of ships are acquired, and the AIS data in the original water area are processed and analyzed, the ships in navigation are taken as the statistical objects for preprocessing of the AIS data, and the AIS data of the ships with a speed less than a preset speed threshold in navigation are deleted, i.e., the AIS data of the ships with a speed < 1 knot are deleted.

[0038] The heat map drawing step is to draw a density heat map according to the preprocessed AIS data, and to frame a plurality of target areas in the density heat map.

[0039] Specifically, the density heat map is drawn according to the preprocessed AIS data, and a plurality of target areas with a relatively large data quantity density are framed on the density heat map, for example, three representative target areas are selected, and the selected three target areas are respectively corresponded to the shipnet map, respectively corresponding to the Yongjiang River estuary water area, the Shenjiamen port water area and the Xiazhimen water area. The three target areas not only have a relatively large data quantity density, but also are relatively representative. Among them, the Yongjiang River estuary water area has complex lines and many turns, and has certain requirements for the flexibility of the ships and the skills of the crew; the Xiazhimen water area is an important and busy waterway in Zhoushan water area, and is also an important channel for large ships to enter and exit the Zhoushan water area; and the Shenjiamen port water area is not only an important fishing port in China, but also a transit waterway, and there are many ships entering and exiting the port.

[0040] Ship quantity judgment step: according to the pre-processed AIS data of the current latitude and longitude position coordinates of the ship and the latitude and longitude position coordinates of each vertex constituting each target region, the number of ships in each target region is automatically judged by using the ray method every certain period of time. The AIS data of each target region is screened by using the ray method, that is, a ray is drawn from the current latitude and longitude position coordinate point of the ship to be judged in a certain direction, the number of intersection points with the target region is calculated, if the number is even or 0, the current coordinate point of the ship is outside the target region, if it is odd, the current coordinate point of the ship is inside the target region.

[0041] Specifically, as shown in Figure 2 , define an arbitrary point P(x0, y0), judge whether it is inside or outside the quadrilateral region (i.e. target region), the essence is to judge the number of intersection points of the ray from point P with the quadrilateral region, which can be regarded as the number of intersection points with four line segments. Define the total number of intersection points as s num , and provide that a ray is drawn horizontally to the right from an arbitrary point, which is divided into the following four cases:

[0042] 1) judge whether point P coincides with the vertex or not, which is expressed as follows:

[0043] x0=x i (i=A, B, C, D) and y0=y i (1)

[0044] 2) judge whether point P is on the edge other than the vertex or not, such as ray v2, take line segment AD as an example, the expression of AD is as follows:

[0045] (y-y D ) / (y A -y D )=(x-x D ) / (x A -x D ) (y D <y<y A ) (2)

[0046] Substitute x0 and y0, if (y0-y D ) / (y A -y D )-(x0-x D ) / (x A -x D )=0, then point P is on the edge AD, and the same for other edges.

[0047] 3) judge whether the ray from point P passes through the vertex of the quadrilateral region or not, provide that the two end points of the line segment, relative to the other end point above the vertex is called the upper end point, and the lower one is the lower end point. If the ray passes through the upper end point, s numAdd 1 if the lower endpoint is passed, s num No need to add 1. As the ray v1, pass through two upper endpoints, s num For 2 is even, the point P is outside the quadrilateral region.

[0048] 4) When the above three cases are excluded, the remaining is the general case, still taking the AD side as an example, first judge y0 whether in y A Between y D , if y D < y0 < y A , judge s num Whether the expression of adding 1 is as follows:

[0049]

[0050] Finally, 24 hours in a day is divided into 288 time intervals with 5 minutes interval, and the number of ships in each 5 minutes interval in the three regions is counted.

[0051] It should be noted that, considering the short transmission time interval of the ship AIS data, to avoid repeated statistics, taking the ship MMSI code as the index, the same ship is only counted once in the same region and in the same 5 minutes interval, and the preliminary ship flow result is obtained.

[0052] Coefficient conversion step: used for data mining on the screened AIS data to obtain ship flow data, including ship coefficient conversion of different types of ships in each target region based on AIS data through data mining technology, and obtaining ship flow data of each type of ship in each target region according to the converted ship coefficient.

[0053] Specifically, in water traffic, due to the different characteristic parameters of each type of ship, i.e. the different ship length, ship width and ship speed, etc., corresponding ship coefficient conversion should be carried out to increase the comparability and accuracy of the marine traffic volume research, so that the ship quantity statistical result is more realistic and persuasive. Among them, through data mining technology, the length of the ship in the used AIS data is counted, and the statistical result shows that the number of ships with length between 26-50 meters is the most, accounting for more than 50%, therefore the conversion coefficient of the ship with length between 26-50 meters is 1.00, and the conversion coefficient of other ship length is shown in Table 1.

[0054] Table 1

[0055]

[0056] After ship coefficient conversion, the ship flow data of different types of ships in each target region can be obtained according to the converted ship coefficient.

[0057] The traffic prediction step is used to capture the spatial features of the ship traffic data through the Chebnet network, and then transmit the spatial features to the LSTM network to learn the time features and make the final prediction, including establishing a ship traffic prediction model based on the Chebnet and LSTM networks according to the ship traffic data, and predicting the ship traffic of each type of ship in each region in a future period of time according to the ship traffic prediction model.

[0058] Specifically, first, the ship traffic data is input into the Chebnet to capture the spatial features of the data. The Chebnet is a neural network for extracting non-Euclidean space features, like the GCN graph convolutional neural network, mainly composed of convolutional layers, pooling layers and fully connected layers, using Relu as the activation function. The Chebnet network uses Chebyshev polynomials as the convolution kernel of spectral graph convolution, converts global convolution into local convolution, and the expression is as follows:

[0059]

[0060] Where a is the graph signal; * represents the graph convolution operation, g θ is a trainable graph filter, K is the top order of the Chebyshev polynomial, θ k is the Chebyshev coefficient vector, is the K-order Chebyshev polynomial, and u is the input feature.

[0061] The Chebyshev polynomial is a set of orthogonal polynomials defined in the interval [-1, 1], and its recursive formula is as follows:

[0062]

[0063] Where I is the identity matrix, represents the scaled and normalized Laplacian matrix, and the expression is as follows:

[0064]

[0065] Where λ max is the largest eigenvalue of the Laplacian matrix L. After the Chebnet network uses Chebyshev polynomials instead of the convolution kernel in the spectral domain, it no longer needs to perform eigenvalue decomposition on the Laplacian matrix, omitting this relatively time-consuming step and effectively reducing the calculation time of the model.

[0066] Then, the captured spatial features are fed into a Long Short-Term Memory (LSTM) network to learn temporal correlations and build a ship traffic prediction model (i.e., the Chebnet-LSTM model). Based on the Chebnet-LSTM model, the ship traffic flow of various types of ships in different regions within a future time period is predicted. The LSTM network is a variant of the RNN, effectively solving the gradient vanishing and gradient exploding problems that exist during RNN training, and has advantages in handling long-term dependencies. The LSTM network consists of an input gate, a forget gate, and an output gate, and introduces non-linear elements through an activation function. The core of LSTM lies in the introduction of a gate mechanism to control the flow and loss of features.

[0067] Forgotten Gate f t This controls the memory units, specifically the degree to which information from the previous time step is forgotten. To learn longer time-series information, the memory units selectively forget previous information, thus avoiding gradient explosion and gradient vanishing problems. Input gate i t Based on the output h from the previous moment t-1 and the current input u t This controls the degree of cell state updates, thereby controlling the current candidate state information. The degree to which it is preserved. Output gate o t This determines the information that needs to be output in the current state, that is, the control information C for the current state. t The degree to which it is output. The calculation process is as follows:

[0068] f t =σ(W f ·[h t-1 ,u t ]+b f (7)

[0069] i t =σ(W i ·[h t-1 ,u t ]+b i (8)

[0070]

[0071]

[0072] o t =σ(W o ·[h t-1 ,u t ]+b o (11)

[0073] h t =o t ⊙tanh(Ct ) (12)

[0074] wherein f t is the decay coefficient, which is also the output of the forget gate, W f , W i , W c and W o represent the weight matrix of the forget gate, the input gate, the memory cell and the output gate respectively, b f , b i , b C and b o represent the bias conditions of the forget gate, the input gate, the memory cell and the output gate respectively, σ is a Sigmoid activation function, tanh is an activation function, and is a dot product operation.

[0075] In addition, in order to evaluate the prediction performance of the Chebnet-LSTM model, evaluation indexes are established according to the predicted ship flow and the actual ship flow, and the ship flow prediction model is evaluated according to the evaluation indexes and by using comparative models, wherein experiments are carried out in three different regions, and the prediction results of the Chebnet-LSTM model are compared with the prediction results of the Chebnet, LSTM, GCN and GAT models respectively, and five evaluation indexes are selected to intuitively and comprehensively evaluate the prediction accuracy and accuracy of each model, and the root mean square error (e RMSE ), the mean absolute percentage error (e MAPE ), the mean absolute error (e MAE ), the symmetric mean absolute percentage error (e SMAPE ) and the mean square error (e MSE ) and the like are selected. The calculation formulas are as follows:

[0076]

[0077]

[0078]

[0079]

[0080]

[0081] In the above formula, is the predicted ship flow value, y i is the true value of the ship flow, and n is the total number of predicted ship traffic flow data points.

[0082] For example, as Figures 3-5As shown, by predicting the ship traffic flow in three different areas of Zhoushan waters, and comparing the evaluation index values of the performance of the ship traffic flow prediction in different areas by different algorithms in Table 2, it can be concluded that the prediction effect of the GAT model in Area One and Area Three is not ideal. In terms of the comparison of the five evaluation indexes, the prediction effects of the Chebnet, LSTM and GCN models are relatively close. However, the Chebnet-LSTM combined model of the present application has obvious improvement in prediction accuracy under the joint action of the spatial feature extraction of the Chebnet model and the time feature extraction of the LSTM model, and the results of the five evaluation indexes are the best. In Area One, compared with the Chebnet model, e RMSE and e MAE increased by 5.74%, e MAPE increased by 6.78%, e SMAPE increased by 5.68%, e MSE increased by 11.16%; in Area Two, compared with other models, e RMSE increased by at least 5.24%, e MAPE increased by at least 12.79%, e MAE increased by at least 10.63%, e SMAPE increased by at least 11.24%, e MSE increased by at least 9.51% and up to 18.97%; in Area Three, compared with the Chebnet, LSTM and GCN three models, e RMSE increased by 9.24% to 12.71%, e MAPE increased by 16.17% to 25.11%, e MAE increased by 14.61% to 20.51%, e SMAPE increased by 15.8% to 21.16%, e MSE increased by 18.67% to 27.33%. This shows that a single neural network model is difficult to well consider the dual characteristics of time and space, and in comparison, the combined model can better capture the spatio-temporal characteristics. Therefore, it is reasonably believed that the ship flow prediction model (Chebnet-LSTM model) of the present application has the prediction value closest to the true value, and has high precision and accuracy in different scenarios, and has good prediction performance, which is better than other comparative models.

[0083] Table 2

[0084]

[0085] It is worth noting that the indicators of region two are higher than those of region one and region three. Through checking the data, it is found that the data fluctuation of region two is larger, and the maximum and minimum values of the flow are about 130 ships. The data fluctuation of region one and region three is smaller, and the maximum and minimum values of the flow of region one are about 60 ships, and the maximum and minimum values of the flow of region three are about 40 ships.

[0086] The application also relates to a regional ship traffic flow prediction system corresponding to the above-mentioned regional ship traffic flow prediction method, which can be understood as a system for realizing the above-mentioned method. The system comprises a data acquisition and preprocessing module, a heat map drawing module, a ship quantity judgment module, a coefficient conversion module and a flow prediction module connected in sequence. Specifically,

[0087] The data acquisition and preprocessing module acquires AIS data of different types of ships and pre-processes the AIS data.

[0088] The heat map drawing module draws a density heat map according to the pre-processed AIS data, and frames a plurality of target regions in the density heat map.

[0089] The ship quantity judgment module automatically judges the number of ships in each target region by using the ray method every certain period of time according to the current latitude and longitude position coordinates of the ships in the pre-processed AIS data and the coordinates of each vertex constituting each target region.

[0090] The coefficient conversion module converts the number of ships of different types of ships in each target region through data mining technology based on the AIS data, and obtains the ship flow data of each type of ship in each target region according to the converted ship coefficient.

[0091] The flow prediction module establishes a ship flow prediction model based on Chebyshev network and long short-term memory network according to the ship flow data, and predicts the ship flow of each type of ship in each region in a future period of time according to the ship flow prediction model.

[0092] Preferably, in the flow prediction module, the Chebyshev network uses Chebyshev polynomials as the convolution kernel of spectral convolution to convert global convolution into local convolution to reduce the calculation time of the model.

[0093] Preferably, in the data acquisition and preprocessing module, the pre-processing of the AIS data includes deleting the AIS data of ships with a speed less than a preset speed threshold in navigation.

[0094] Preferably, in the flow prediction module, an evaluation index is further established according to the predicted ship flow and the actual ship flow, the ship flow prediction model is evaluated according to the evaluation index and by using a comparison model, and it is verified that the ship flow prediction model is an optimal model.

[0095] The evaluation index includes a root mean square error, a mean absolute percentage error, a mean absolute error, a symmetric mean absolute percentage error and a mean square error, and the comparison model includes a long short-term memory network model, a graph convolution network model and a graph attention network model.

[0096] Preferably, the AIS data includes static data and dynamic data, the static data includes a ship movement service identification code, a ship length, a ship width and a ship name, and the dynamic data includes ship latitude and longitude position information, a navigation state, a time stamp, a ground speed, a draft, a departure port, a destination port and a predicted arrival time.

[0097] The application provides an objective and scientific regional ship traffic flow prediction method and system, draws a density heat map based on ship AIS data, selects a plurality of target regions, obtains ship flow data by using a ray method and a ship coefficient conversion method, captures spatial features of data based on a Chebyshev network, transmits the features to a long short-term memory network to learn time correlation, and finally predicts the flow of different types of ships in the target water area by establishing a ship flow prediction model, and a plurality of evaluation indexes can be further selected to intuitively and comprehensively evaluate the prediction accuracy and accuracy of each model, effectively improving the accuracy of regional ship traffic flow prediction, effectively extracting the space-time features of ship flow data, and showing high precision and accuracy in different scenarios, greatly improving the prediction accuracy, and providing data support for intelligent navigation of water traffic.

[0098] It should be noted that the above specific embodiments can enable those skilled in the art to more fully understand the present application, but do not limit the present application in any way. Therefore, although the present application has been described in detail with reference to the drawings and examples, those skilled in the art should understand that the present application can still be modified or replaced by equivalents, in short, all technical solutions and improvements that do not deviate from the spirit and scope of the present application should be covered in the protection scope of the present application.

Claims

1. A regional ship traffic flow prediction method, characterized by, The method comprises the following steps: a data acquisition and preprocessing step of acquiring AIS data of different types of ships and preprocessing the AIS data; a heat map drawing step of drawing a density heat map according to the preprocessed AIS data and framing a plurality of target regions in the density heat map; a ship quantity judgment step of automatically judging the number of ships in each target region by using a ray method every certain period of time according to the current latitude and longitude coordinates of the ships in the preprocessed AIS data and the coordinates of each vertex constituting each target region; a coefficient conversion step of converting the number of ships of different types of ships in each target region into a ship coefficient based on the AIS data by using a data mining technology, and obtaining ship flow data of each type of ship in each target region according to the converted ship coefficient; a flow prediction step of establishing a ship flow prediction model based on a Chebyshev network and a long short-term memory network according to the ship flow data, and predicting the ship flow of each type of ship in each region in a future period of time according to the ship flow prediction model.

2. The regional ship traffic flow prediction method according to claim 1, characterized by, In the flow prediction step, the Chebyshev network uses a Chebyshev polynomial as a convolution kernel of spectral graph convolution to convert global convolution into local convolution to reduce the calculation time of the model.

3. The regional ship traffic flow prediction method according to claim 1, characterized by, In the data acquisition and preprocessing step, the preprocessing of the AIS data includes deleting the AIS data of ships with a speed less than a preset speed threshold in navigation.

4. The regional ship traffic flow prediction method according to one of claims 1 to 3, characterized in that, In the flow prediction step, an evaluation index is further established according to the predicted ship flow and the actual ship flow, the ship flow prediction model is evaluated according to the evaluation index and by using a comparative model, and it is verified that the ship flow prediction model is an optimal model. The evaluation index includes a root mean square error, a mean absolute percentage error, a mean absolute error, a symmetric mean absolute percentage error, and a mean square error, and the comparative model includes a long short-term memory network model, a graph convolution network model, and a graph attention network model.

5. The regional ship traffic flow prediction method according to claim 1, characterized by, In the data acquisition and preprocessing step, the AIS data includes static data and dynamic data, the static data includes a ship mobile service identity, a ship length, a ship width, and a ship name, and the dynamic data includes ship latitude and longitude position information, a navigation state, a time stamp, an overwater speed, a draft, a departure port, a destination port, and a predicted arrival time.

6. A regional ship traffic flow prediction system, characterized by, The method comprises a data acquisition and preprocessing module, a heat map drawing module, a ship quantity judgment module, a coefficient conversion module, and a flow prediction module connected in sequence, The data acquisition and preprocessing module acquires AIS data of different types of ships and preprocesses the AIS data; The heat map drawing module draws a density heat map according to the preprocessed AIS data and frames a plurality of target regions in the density heat map; The ship quantity judgment module automatically judges the number of ships in each target region by using a ray method every certain period of time according to the current latitude and longitude coordinates of the ships in the preprocessed AIS data and the coordinates of each vertex constituting each target region; The coefficient conversion module converts the number of ships of different types of ships in each target region into a ship coefficient based on the AIS data by using a data mining technology, and obtains ship flow data of each type of ship in each target region according to the converted ship coefficient. The coefficient conversion module uses data mining technology to convert the number of different types of ships in each target area based on AIS data, and obtains the ship traffic data of each type of ship in each target area based on the converted ship coefficients. The flow prediction module establishes a ship flow prediction model based on Chebyshev network and Long Short-Term Memory network according to ship flow data, and predicts the ship flow of various types of ships in various regions within a certain period of time in the future based on the ship flow prediction model.

7. The zone-oriented vessel traffic flow prediction system according to claim 6, wherein, In the traffic prediction module, the Chebyshev network uses Chebyshev polynomials as the convolution kernel for spectral graph convolution, converting global convolution into local convolution to reduce the computation time of the model.

8. The zone-oriented vessel traffic flow prediction system according to claim 6, wherein, The data acquisition and preprocessing module preprocesses AIS data by deleting AIS data of ships whose speed is less than a preset speed threshold during navigation.

9. The regional ship traffic flow prediction system according to one of claims 6 to 8, characterized in that In the flow prediction module, an evaluation index is established based on the predicted ship flow and the actual ship flow. The ship flow prediction model is evaluated based on the evaluation index and a comparison model to verify that the ship flow prediction model is the optimal model. The evaluation metrics include root mean square error, mean absolute percentage error, mean absolute error, symmetric mean absolute percentage error, and mean square error. The comparison models include long short-term memory network model, graph convolutional network model, and graph attention network model.

10. The zone-oriented vessel traffic flow prediction system according to claim 6, wherein, The AIS data includes static data and dynamic data. The static data includes the vessel's mobile service identification code, length, beam, and name. The dynamic data includes the vessel's latitude and longitude location information, navigation status, timestamp, speed over land, draft, port of departure, port of destination, and estimated arrival time.

Images