A method for predicting port ship traffic flow considering spatial dependence

Abstract

The present invention discloses a method for predicting port ship traffic flow considering spatial dependence, belonging to the field of water transportation. The method includes the following steps: Step 1: Acquisition and preprocessing of AIS data; Step 2: Identification of ships calling at ports; Step 3: Establishment of a port network topology graph; Step 4: Construction of a port feature matrix; Step 5: Prediction of port ship traffic flow based on a graph convolutional neural network model; Step 6: Prediction error test. The present invention predicts the ship traffic flow from the port network level, can evaluate the trend of the ship traffic flow in the port as a whole. At the same time, a prediction model is constructed based on the graph convolutional neural network suitable for the port network topology structure, which can fully exploit the spatial dependence between ports, thereby improving the prediction accuracy of the port ship traffic flow. The prediction of the port ship traffic flow by the present invention can provide a reference basis for the port to manage berths in advance and plan port construction, etc.

Description

Technical Field

[0001] The present invention belongs to the field of water transportation, and particularly relates to a method for predicting port ship traffic flow considering spatial dependence. Background Art

[0002] The development of the global economy has led to a continuous increase in the demand for maritime trade. As a node of maritime trade, the importance of ports has been increasing. However, the gradual growth of maritime transport demand and the number of transport ships also pose higher requirements for port operation efficiency. Predicting the ship traffic flow of ports can not only help ports conduct traffic management in advance, but also provide important reference for the long-term construction and development of ports in the future. Therefore, how to accurately predict the ship traffic flow of ports is of great significance to ports.

[0003] Regarding the research on port ship traffic flow prediction, time series models and regression models were mainly used earlier to predict traffic flow based on historical data and other relevant influencing factors. With the rapid rise of artificial intelligence, models such as machine learning have gradually replaced traditional time series models and regression models, and artificial intelligence models such as neural networks, support vector machines, and ensemble learning have become the mainstream for predicting port ship traffic flow. However, current related research mainly focuses on improving the capabilities of prediction models to extract information from the historical data of a certain port to the greatest extent, and finally realizes the prediction of port ship traffic flow. However, the ship traffic flow of ports is not only related to the historical flow of that port, but also affected by other major ports and the global shipping network. Therefore, only considering the historical flow of a certain port cannot incorporate the changes and influences of other associated ports and the entire network into the prediction model, thereby ignoring the spatial dependence between ports.

[0004] As an artificial intelligence algorithm considering graph structure, the graph convolutional neural network can aggregate the features of adjacent nodes, thereby incorporating the spatial dependence between adjacent nodes into the prediction model. Currently, some scholars have used the graph convolutional neural network to predict the traffic flow of road traffic. First, the intersections are regarded as nodes, and the connected roads are regarded as edges to form a graph structure, and then the road traffic flow is predicted based on the graph convolutional neural network. This method solves the problem that traditional traffic flow prediction methods ignore spatial dependence. The port network composed of ports as nodes and shipping lines as edges has been proven by scholars to be a complex network, and the graph convolutional neural network can also be used to predict the port ship flow. In the field of water transportation prediction, some scholars have used the graph convolutional network, taking features such as ship speed and ship density as inputs to predict the water transportation navigation density. However, there are still disadvantages when this method is directly applied to the prediction of port ship flow. The ship flow in the port network is not only related to features such as the ship's own speed, but also closely related to factors such as the overall characteristics of the network and the position of the port in the network. Ordinary graph convolutional neural networks cannot fully mine relevant features. Therefore, an improved graph convolutional neural network that better adapts to the characteristics of the port network is needed to fully mine the spatial dependence between ports, so as to achieve more accurate prediction of port ship traffic flow.

[0005] To solve the above problems, the present invention uses AIS ship big data. After data preprocessing, it identifies the port calls and constructs a port network topology map based on the identification results. On this basis, according to the complex network theory, the present invention obtains the characteristics of the port in the complex network topology structure and the common characteristics of the port complex network, and uses these two characteristics together with the historical ship traffic flow of the port as the input features of the prediction model. Then, a port ship traffic flow prediction model based on the graph convolutional neural network is established to predict the port flow, and finally, a prediction error test is carried out. This method is based on the complex network theory. On the basis of the graph convolutional neural network prediction model, the characteristics of the port in the complex network and the overall characteristics of the port network are added, so that the model fully considers the spatial dependence between ports, avoids the limitation of only using the port's own historical data for prediction, and provides an important reference basis for the traffic management of ports. Summary of the Invention

[0006] The technical problem to be solved by the present invention is: to overcome the above-mentioned deficiencies of the prior art and propose a method for predicting port ship traffic flow.

[0007] The technical solution of the present invention is as follows:

[0008] A method for predicting port ship traffic flow considering spatial dependence, comprising the following steps:

[0009] Step 1: AIS data acquisition and preprocessing

[0010] The AIS data in this step is obtained from the actual navigation of the ship. The original AIS data is parsed through the AIS decoding algorithm, and then the required AIS data fields are extracted, including information such as the ship name, speed, ship geographical coordinates, and time. Then, the data is preprocessed, including data cleaning, data outlier removal, and data completion.

[0011] Step 2: Identification of ports of call of ships

[0012] In this step, the identification of ports of call of ships is carried out based on information such as the ship's geographical coordinates, time, and speed from AIS. First, the points of the ship's geographical coordinate information are spatially connected to the surfaces of the docking ranges of all ports, and the AIS data with geographical locations within the port docking spatial range is extracted. Second, calculate the continuous stay time t of the ship in the port, and judge whether the continuous stay time t is greater than the stay time threshold t α . Third, if the stay time is greater than the time threshold, calculate the average speed of the ship during the stay time and the maximum speed v max , if the average speed is less than the average speed threshold v β , and the maximum speed v max is less than the maximum speed threshold v γ , then it is judged that the ship has successfully called at this port, and the identification of the port of call of the ship is completed.

[0013] Step 3: Establishment of the port network topology graph

[0014] Based on the port of call information identified in Step 2, a port network topology graph is established. Taking all the ports called at by ships as nodes and the ship routes between two ports as edges, the port network topology graph G is established. G=(V, E) is a directed graph. When there is a ship traveling from port A to port B, there is a route connection from port A to port B, but there is no route connection from port B to port A. The graph G contains a set of points V and a set of edges E, where:

[0015]

[0016] In the formula, N is the number of all ports in the port network, and vi is the i-th port;

[0017]

[0018] In the formula, e ij is the edge from the i-th port to the j-th port.

[0019] Step 4: Construction of the port feature matrix

[0020] In this step, based on the port network topology map established in Step 3, the feature matrix X of the port is generated. The feature matrix X of the port is an N*Q matrix, where Q represents the feature dimension of the matrix. The features of the port are divided into three parts, including the historical ship traffic flow of the port, the features of the port in the network topology structure, and the common features of the port network.

[0021] The historical ship traffic flow of the port consists of the ship traffic flows in m historical port time periods before the corresponding prediction time period.

[0022] The features of the port in the network topology are calculated based on the indicators of complex network theory. Existing research has proven that the shipping network constructed from AIS data is a complex network. Therefore, the indicators in complex networks can be used to reflect the features of the port in the network topology and quantify the degree of association between the port and other ports in the network. The selected indicators include out-degree, in-degree, weighted in-degree, weighted out-degree, betweenness centrality, and port clustering coefficient. The indicator formulas are as follows:

[0023]

[0024] In the formula, U i is the in-degree of the i-th port, and x ij indicates whether there is a connection from the i-th port to the j-th port. If connected, it is 1; otherwise, it is 0.

[0025]

[0026] In the formula, S i is the out-degree of the i-th port.

[0027]

[0028] In the formula, UW i is the weighted in-degree of the i-th port, and w ij represents the number of ships from the i-th port to the j-th port, that is, the weight of the edge.

[0029]

[0030] In the formula, SW i is the weighted out-degree of the i-th port.

[0031]

[0032] In the formula, B i is the betweenness centrality of the i-th port, and g jk represents the number of the shortest paths from the j-th port to the k-th port, and g ji(i) represents the number of shortest paths from the j-th port to the k-th port passing through the i-th port.

[0033] The clustering coefficient of a port reflects the aggregation among ports. In a directed graph, assuming that port i is connected to k i ports either unidirectionally or bidirectionally, then among these k i ports, there are at most k i (k i -1) edges. Thus, the formula for the clustering coefficient of a port is as follows:

[0034]

[0035] In the formula, C i is the clustering coefficient of the i-th port, and r i is the actual number of edges among k i ports.

[0036] The common characteristics of the port network can also be calculated using complex network metrics. Incorporating the common characteristics makes the feature matrix contain information about the overall port network. The selected metrics include the average shortest path, network efficiency, and network clustering coefficient. The metric formulas are as follows:

[0037]

[0038] In the formula, D is the average shortest path of the port network, and d ij represents the shortest path from the i-th port to the j-th port.

[0039]

[0040] In the formula, E is the network efficiency of the port network.

[0041]

[0042] In the formula, C is the network clustering coefficient of the port network.

[0043] Thus, the historical ship traffic flow of ports in m dimensions, the characteristics of ports in the network topology in 6 dimensions, and the common characteristics of the port network in 3 dimensions are obtained. Combining these characteristics yields the N*Q-dimensional feature matrix X, where Q = m + 9.

[0044] Step Five: Predict the port ship traffic flow based on the graph convolutional neural network model

[0045] In this step, the established prediction model is a two-layer graph convolutional neural network. The port feature matrix obtained in Step Four is used as the input of the model for prediction. The propagation rule of each layer of the model is as follows:

[0046]

[0047] In the formula, H (1) is the port feature of the first layer, σ is the activation function, is the concatenated adjacency matrix, represents the degree matrix of, W (l) is the parameter matrix of the first layer.

[0048]

[0049] In the formula, A is the adjacency matrix of graph G, and I is the N*N dimensional identity matrix.

[0050] Through the propagation rule, the graph convolutional neural network can aggregate the features of adjacent ports together, thereby incorporating the spatial dependence between ports into the consideration of ship traffic flow prediction. In the first layer of the model, the input matrix is the port feature matrix obtained in step three, that is, H (0) = X. When the calculation of the first layer of the graph convolutional neural network is completed, the information of the port has been updated by the information of the first-order adjacent ports and is input into the second layer of the graph convolutional neural network. The second layer completes the same calculation. At this time, the port has obtained the information of the second-order adjacent ports, and finally the softmax function is used to output the prediction result. Based on this, the output result of the model is as follows:

[0051]

[0052] In the formula, Z is the N*1 feature vector output by the model, and both softmax and ReLU are activation functions.

[0053] Step six: Prediction error test

[0054] After obtaining the predicted port ship traffic flow of the model in step five, the prediction result is tested for error to verify the prediction effect. The metrics used are the mean absolute error MAE and the mean absolute percentage error MAPE. The metric formulas are as follows:

[0055]

[0056]

[0057] In the formula, is the predicted value of the ship traffic flow of the i-th port, and y i is the actual value of the ship traffic flow of the i-th port.

[0058] The beneficial effects of the present invention are as follows: The present invention predicts the ship traffic flow from the port network level, which can evaluate the trend of the ship traffic flow in the port as a whole. At the same time, a prediction model is constructed based on the graph convolutional neural network suitable for the port network topology structure, which can fully exploit the spatial dependence between ports, thereby improving the prediction accuracy of the port ship traffic flow. In addition, according to the complex network theory, the topological features of the port in the network and the common features of the port network are jointly input into the model, so that the status of the port in the network can be further reflected in the prediction model. The prediction of the port ship traffic flow by the present invention can provide a reference basis for the port to manage berths in advance and plan port construction, etc. BRIEF DESCRIPTION OF THE DRAWINGS

[0059] Figure 1 is the overall flowchart of the prediction method provided by the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0060] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments.

[0061] Embodiment 1.

[0062] As Figure 1 shown, a method for predicting port ship traffic flow considering spatial dependence includes the following steps:

[0063] Step 1: Acquisition and preprocessing of AIS data

[0064] The AIS data in this step is obtained from the actual navigation of ships. The original AIS data is parsed through the AIS decoding algorithm, and then the required AIS data fields are extracted, including information such as ship name, speed, ship geographical coordinates, and time. Then, the data is preprocessed, including data cleaning, data outlier removal, and data completion.

[0065] First, the AIS data is cleaned, including unifying the data format and filling in the missing data in the ship name.

[0066] Then, outliers in the AIS data are removed, and the data with abnormal geographical coordinates, abnormal time, abnormal route speed, and duplicate information in the data are deleted.

[0067] Finally, the Kalman filter is used to complete the missing data such as ship speed and spatial position.

[0068] Step 2: Identification of ship calls at ports

[0069] In this step, ship port calls are identified based on information such as the ship's geographical coordinates, time, and speed from AIS. First, the points of the ship's geographical coordinate information are spatially joined with the surfaces of all port docking ranges to extract AIS data whose geographical locations are within the port docking space ranges. Second, calculate the continuous stay time t of the ship at the port, and determine whether the continuous stay time t is greater than the stay time threshold t α . Third, if the stay time is greater than the time threshold, calculate the average speed and the maximum speed v max of the ship during the stay time. If the average speed is less than the average speed threshold v β , and the maximum speed v max is less than the maximum speed threshold v γ , then it is determined that the ship has successfully called at the port, and the ship port call identification is completed.

[0070] Step 3: Establish a port network topology graph

[0071] Based on the port call information identified in Step 2, establish a port network topology graph. Using all the ports called at by ships as nodes and the ship routes between two ports as edges, the port network topology graph G is established. G=(V, E) is a directed graph. When there is a ship traveling from port A to port B, there is a route connection from port A to port B, but there is no route connection from port B to port A. The graph G contains a set of points V and a set of edges E, where:

[0072]

[0073] In the formula, N is the number of all ports in the port network, and vi is the i-th port;

[0074]

[0075] In the formula, e ij is the edge from the i-th port to the j-th port.

[0076] Step 4: Construct a port feature matrix

[0077] In this step, based on the port network topology graph already established in Step 3, generate the port feature matrix X. The port feature matrix X of the port is an N*Q matrix, where Q represents the feature dimension of the matrix. The features of the port are divided into three parts, including the historical ship traffic flow of the port, the features of the port in the network topology structure, and the common features of the port network.

[0078] The historical ship traffic flow of the port consists of the ship traffic flows in m historical time periods of the port corresponding to the previous prediction time period.

[0079] The characteristics of ports in the network topology are calculated based on the indicators of complex network theory. Existing research has proven that the shipping network constructed from AIS data is a complex network. Therefore, the indicators in complex networks can be used to reflect the characteristics of ports in the network topology and quantify the degree of association between a port and other ports in the network. The selected indicators include out-degree, in-degree, weighted in-degree, weighted out-degree, betweenness centrality, and port clustering coefficient. The indicator formulas are as follows:

[0080]

[0081] In the formula, U i is the in-degree of the i-th port, and x ij indicates whether there is a connection from the i-th port to the j-th port. If connected, it is 1; otherwise, it is 0.

[0082]

[0083] In the formula, S i is the out-degree of the i-th port.

[0084]

[0085] In the formula, UW i is the weighted in-degree of the i-th port, and w ij represents the number of ships from the i-th port to the j-th port, that is, the weight of the edge.

[0086]

[0087] In the formula, SW i is the weighted out-degree of the i-th port.

[0088]

[0089] In the formula, B i is the betweenness centrality of the i-th port, and g jk represents the number of the shortest paths from the j-th port to the k-th port, and g ji (i) represents the number of the shortest paths from the j-th port to the k-th port passing through the i-th port.

[0090] The clustering coefficient of a port reflects the aggregation situation among ports. In a directed graph, assume that port i is unidirectionally or bidirectionally connected to k i ports. Then, there are at most k i edges among these k i (k i -1) edges. Thus, the formula for the clustering coefficient of a port is as follows:

[0091]

[0092] Wherein, C i is the clustering coefficient of the i-th port, and r i is k i the actual number of edges existing between ports.

[0093] The common features of the port network can also be calculated using complex network metrics. Incorporating the common features enables the feature matrix to contain information about the overall port network. The selected metrics include the average shortest path, network efficiency, and network clustering coefficient, and the metric formulas are as follows:

[0094]

[0095] Wherein, D is the average shortest path of the port network, and d ij represents the shortest path from the i-th port to the j-th port.

[0096]

[0097] Wherein, E is the network efficiency of the port network.

[0098]

[0099] Wherein, C is the network clustering coefficient of the port network.

[0100] Thus, the historical ship traffic flow of the port in m dimensions, the features of the port in the network topology in 6 dimensions, and the common features of the port network in 3 dimensions are obtained. Combining these features yields the N*Q-dimensional feature matrix X, where Q = m + 9.

[0101] Step Five: Predict the port ship traffic flow based on the graph convolutional neural network model

[0102] In this step, the established prediction model is a two-layer graph convolutional neural network, and the port feature matrix obtained in Step Four is used as the input of the model for prediction. The propagation rule of each layer of the model is as follows:

[0103]

[0104] Wherein, H (1) is the port feature of the first layer, σ is the activation function, is the concatenated adjacency matrix, represents the degree matrix of, and W (1) is the parameter matrix of the first layer.

[0105]

[0106] Wherein, A is the adjacency matrix of graph G, and I is the N*N-dimensional identity matrix.

[0107] Through the propagation rule, the graph convolutional neural network can aggregate the features of adjacent ports together, thus taking the spatial dependence between ports into account in the prediction of ship traffic flow. In the first layer of the model, the input matrix is the port feature matrix obtained in Step 3, i.e., H (0) = X. After the first layer of the graph convolutional neural network finishes the calculation, the information of the port has been updated by the information of the first-order adjacent ports and is input into the second layer of the graph convolutional neural network. The second layer performs the same calculation. At this time, the port obtains the information of the second-order adjacent ports. Finally, the softmax function is used to output the prediction result. Using a two-layer graph convolutional neural network can enable the corresponding port to obtain the features of the second-order adjacent ports that are not directly connected but indirectly connected. At the same time, it can avoid diluting the information weight of important associated ports by the relatively weak third-order and higher-order adjacent ports. Based on this, the output result of the model is as follows:

[0108]

[0109] where Z is the N*1 feature vector output by the model, and both softmax and ReLU are activation functions.

[0110] Step 6: Prediction error test

[0111] After obtaining the predicted port ship traffic flow of the model in Step 5, an error test is carried out on the prediction result to verify the prediction effect. The metrics used are the mean absolute error MAE and the mean absolute percentage error MAPE. The metric formulas are as follows:

[0112]

[0113]

[0114] where is the predicted value of the ship traffic flow of the i-th port, and y i is the actual value of the ship traffic flow of the i-th port.

[0115] As described above, only the preferred specific embodiments of the present invention are provided, but the protection scope of the present invention is not limited thereto. Any person skilled in the art within the technical scope disclosed by the present invention, according to the technical solution and inventive concept of the present invention, making equivalent substitutions or changes, shall be covered by the protection scope of the present invention.

Claims

1. A method for predicting the traffic flow of port ships considering spatial dependence, characterized in that The following steps are involved: Step 1: AIS data acquisition and preprocessing; Step 2: Ship port identification; Step 3: Establish port network topology; Step 4: Construct port feature matrix; Step 5: Predict port ship traffic flow based on the graph convolutional neural network model; Step 6: Forecast error test; In step 4, based on the port network topology map established in step 3, the port feature matrix X is generated. The port feature matrix X is an N*Q matrix, where Q represents the feature dimension of the matrix. The port features are divided into three parts, including the port's historical ship traffic flow, the port's features in the network topology, and the common features of the port network. In step five, the prediction model established is a two-layer graph convolutional neural network.

2. The port ship traffic flow prediction method considering spatial dependence according to claim 1, wherein The AIS data described in step 1 is obtained from the actual voyage of the ship. The original AIS data is parsed by the AIS decoding algorithm, and then the required AIS data fields are extracted, including the ship name, ship geographic coordinates, time, and speed. The data is then preprocessed, including data cleaning, data outlier removal, and data completion.

3. A method for predicting the traffic flow of port ships considering spatial dependence according to claim 2, characterized in that, The ship port identification described in step 2 refers to the ship port identification based on the ship's geographic coordinates, time and speed information. The specific steps are: The first step is to spatially connect the points of the ship's geographic coordinate information with the surfaces of all port berthing ranges, and extract the AIS data whose geographical location is within the port berthing space range; Step 2: Calculate the continuous stay time t of the ship in the port and determine whether the continuous stay time t is greater than the stay time threshold t α ; Step 3: If the residence time is greater than the time threshold, calculate the average speed of the ship during the residence time and the maximum speed v max . If the average speed is less than the average speed threshold v β , and the maximum speed v max is less than the maximum speed threshold v γ , it is determined that the ship has successfully berthed at the port, and the ship's port call identification is completed.

4. A method for predicting the traffic flow of port ships considering spatial dependence according to claim 3, characterized in that The specific steps for establishing the port network topology diagram described in step 3 are as follows: Based on the port information identified in step 2, a port network topology graph is established; all ports that ships have called at are nodes, and the ship routes connecting the two ports are edges, that is, a port network topology graph G is established; G = (V, E) is a directed graph. When a ship travels from port A to port B, there is a route connecting port A to port B, but there is no route connecting port B to port A; the graph G contains a set of points V and a set of edges E, where: Where N is the number of all ports in the port network, and vi i is the i-th port; where e ij is the edge from the i-th port to the j-th port.

5. A method for predicting the traffic flow of port ships considering spatial dependence according to claim 4, characterized in that, The specific steps of constructing the port feature matrix described in step 4 are as follows: The historical ship traffic flow of the port is composed of the ship traffic flow of the m historical time periods of the port before the corresponding prediction time period; The characteristics of ports in the network topology are calculated based on the indicators of complex network theory. The selected indicators include out-degree, in-degree, weighted in-degree, weighted out-degree, betweenness centrality and port clustering coefficient. The indicator formula is as follows: where U i is the in-degree of the i-th port, and x ij indicates whether there is a connection from the i-th port to the j-th port. If there is a connection, it is 1; otherwise, it is 0. where S i is the out-degree of the i-th port; where, UW i is the weighted in-degree of the \(i\)-th port, and \(w\) ij represents the number of ships departing from the \(i\)-th port to the \(j\)-th port, that is, the weight of the edge; where SW i is the weighted out-degree of the i-th port; where B i is the betweenness centrality of the i-th port, and g jk represents the number of the existing shortest paths from the j-th port to the k-th port, and g ji (i) represents the number of the shortest paths from the j-th port to the k-th port passing through the i-th port; The clustering coefficient of ports reflects the clustering situation among ports. In a directed graph, assuming that port i is connected to k i ports either unidirectionally or bidirectionally, then among these k i ports, there are at most k i (k i - 1) edges. Thus, the formula for the clustering coefficient of ports is as follows: Where C i is the clustering coefficient of the i-th port, and r i is the actual number of edges existing between k i ports; The common features of the port network are also calculated using complex network indicators. The addition of common features makes the feature matrix contain the information of the entire port network. The selected indicators include the average shortest path, network efficiency and network clustering coefficient. The indicator formula is as follows: Where D is the average shortest path of the port network, and d ij represents the shortest path from the i-th port to the j-th port; Where, E is the network efficiency of the port network; Where C is the network clustering coefficient of the port network; Thus, we obtain the m-dimensional historical ship traffic flow of the port, the 6-dimensional characteristics of the port in the network topology, and the 3-dimensional common characteristics of the port network; combining these characteristics gives us the N*Q-dimensional feature matrix X, where Q=m+9.

6. The port ship traffic flow prediction method considering spatial dependence according to claim 5, characterized in that The specific steps of predicting port ship traffic flow based on the graph convolutional neural network model described in step 5 are as follows: Use the port feature matrix obtained in Step 4 as the input of the model for prediction; the propagation rules of each layer of the prediction model are as follows: where H (l) is the port feature of the l-th layer, σ is the activation function, is the adjacency matrix after connection, denotes the degree matrix of, W (l) is the parameter matrix of the l-th layer; In the formula, A is the adjacency matrix of graph G, and I is the N*N-dimensional identity matrix; Through the propagation rule, the graph convolutional neural network aggregates the features of adjacent ports, thereby incorporating the spatial dependence between ports into the consideration of ship traffic flow prediction; in the first layer of the model, the input matrix is the port feature matrix obtained in step three, i.e., H (0) = X; when the calculation of the first-layer graph convolutional neural network is completed, the information of the port has been updated by the information of the first-order adjacent ports and is input into the second-layer graph convolutional neural network. The second layer completes the same calculation. At this time, the port obtains the information of the second-order adjacent ports, and finally the softmax function is used to output the prediction result; based on this, the output result of the model is as follows: In the formula, Z is the N*1 feature vector output by the model, and both softmax and ReLU are activation functions.

7. A method for predicting the traffic flow of port ships considering spatial dependence according to claim 6, characterized in that, The prediction error test described in Step 6 is specifically as follows: After obtaining the predicted port ship traffic flow of the model in Step 5, conduct an error test on the prediction results to verify the prediction effect. The indicators used are the mean absolute error MAE and the mean absolute percentage error MAPE; the indicator formulas are as follows: In the formula, is the predicted value of the ship traffic flow at the i-th port, and y i is the actual value of the ship traffic flow at the i-th port.

Images