An offshore transportation safety analysis method based on AI and ship AIS big data

Abstract

The application discloses a kind of based on AI and ship AIS big data offshore traffic safety analysis method, belong to offshore wind farm and offshore mariculture area siting technical field, specifically include: obtaining the ship AIS data of target sea area generation ship AIS traffic flow analysis diagram;Target sea area is divided into analysis grid unit and the external navigation risk score of each grid unit is calculated;Based on risk score screening and aggregation generate several candidate site area, extract site boundary parameter;Determine the candidate mother port corresponding to each candidate site area, based on electronic chart data planning each mother port to site operation and maintenance route, calculate navigation difficulty coefficient and select minimum value as operation and maintenance path conflict index;The external navigation risk score and operation and maintenance path conflict index are weighted and summed to obtain comprehensive score, select the lowest site area as target site area as score.The application includes external navigation risk and operation and maintenance path safety into unified decision framework, realizes the multidimensional collaborative optimization of wind farm siting.

Description

Technical Field

[0001] This invention relates to the field of offshore wind farm and offshore aquaculture site selection technology, specifically to a method for maritime transportation safety analysis based on AI and ship AIS big data. Background Technology

[0002] The construction of offshore wind farms and offshore aquaculture areas is an important direction for marine energy development. As nearshore wind farm resources become saturated, offshore wind power projects are gradually expanding to deeper waters further offshore. Site selection for offshore wind farms and offshore aquaculture areas is a crucial early-stage step in project development, directly determining their construction costs, operational efficiency, and navigation safety. Currently, site selection for offshore wind farms and offshore aquaculture areas typically relies on geographic information systems and multi-criteria decision-making methods, comprehensively considering multiple factors such as wind resource conditions, water depth and geological conditions, and marine functional zoning, and determining candidate sites through spatial overlay analysis and weighted scoring.

[0003] In navigation safety assessments, existing technologies generally employ Automatic Identification System (AIS) data for traffic flow analysis. Specifically, by collecting historical vessel trajectory data for the target sea area, methods such as kernel density analysis and track strip extraction are used to generate vessel traffic flow density distribution maps, identifying main channel routes, habitual shipping routes, and areas with high fishing vessel density. Based on these analysis results, during site selection, wind farms are located in areas with low vessel traffic density by setting safety distances to prevent the risk of collisions between wind farms and passing vessels. This type of method prioritizes "avoidance" as its core principle, focusing on the impact of wind farm construction on the existing navigation environment.

[0004] However, current technologies only focus on external navigation conflicts between offshore wind farms and offshore aquaculture areas and passing vessels, failing to incorporate post-construction operation and maintenance activities into the site selection assessment system. Offshore wind farms and offshore aquaculture areas have operational cycles of twenty to thirty years, during which maintenance vessels frequently need to navigate from their home ports to the wind farms for operations. The spatial overlap between maintenance routes and main shipping channels and fishing areas directly triggers new navigation safety risks. Due to the lack of quantitative assessment of the safety of maintenance routes, existing technologies may select sites that are far from main shipping channels, but whose operation and maintenance access routes frequently require crossing high-risk waters, resulting in a disconnect between external navigation risks and operation and maintenance route risks in site selection decisions. This fragmented assessment approach exposes the interaction risks between maintenance vessels and passing vessels during the actual operation phase, increasing navigational safety hazards for maintenance vessels. Furthermore, because parameters such as the length of the maintenance route, water depth, and crossing of waterways and fishing areas are not included in the assessment during the site selection phase, the selected sites face problems during the operation phase, such as increased fuel costs and travel time due to excessively long routes, limited vessel passage windows due to shallow water depth, and increased avoidance frequency and navigation risks due to crossing waterways or fishing areas. Ultimately, this affects the overall operational efficiency of the wind farm. Summary of the Invention

[0005] The purpose of this invention is to provide a maritime transportation safety analysis method based on AI and ship AIS big data, addressing the following technical problems:

[0006] Existing technologies only focus on external navigation conflicts between offshore wind farms and offshore aquaculture areas and passing vessels, without incorporating the route safety of maintenance vessels into the site selection assessment. This results in external navigation risks and maintenance route risks being treated separately in the site selection decision-making process, leading to increased navigation safety hazards for maintenance vessels, higher maintenance costs, and reduced operational efficiency at the selected sites during the operation phase.

[0007] The objective of this invention can be achieved through the following technical solutions:

[0008] A maritime transportation safety analysis method based on AI and ship AIS big data includes the following steps:

[0009] S1. Acquire AIS data of ships in the target sea area, and perform trajectory extraction and density analysis on the ship AIS data to generate a ship AIS traffic flow analysis map covering the target sea area.

[0010] S2, the target sea area is divided into several evaluation grid units, the normalized density value of each analysis grid unit is extracted based on the ship AIS traffic flow analysis map, and the external navigation risk score of each analysis grid unit is calculated.

[0011] S3. Based on the external navigation risk score of each evaluation grid unit, several candidate site areas are determined, and the site boundary parameters of each candidate site area are extracted.

[0012] S4. Obtain the surrounding port information corresponding to each candidate site area, and determine several candidate home ports corresponding to each candidate site area. Based on the electronic nautical chart data, plan the operation and maintenance routes from each candidate home port to the corresponding candidate site area, and obtain all operation and maintenance routes corresponding to each candidate site area.

[0013] S5, extract the route feature parameters of each operation and maintenance route, calculate the navigation difficulty coefficient of each operation and maintenance route corresponding to each candidate site area based on the route feature parameters, and select the minimum navigation difficulty coefficient corresponding to each candidate site area as the operation and maintenance path conflict index of the candidate site area.

[0014] S6. Based on the external navigation risk score and operation and maintenance path conflict index of each candidate site area, calculate the comprehensive score of each candidate site area, select the candidate site area with the lowest comprehensive score as the target site area and output it.

[0015] As a further aspect of the present invention: in S1, the specific process for generating the ship AIS traffic flow analysis diagram is as follows:

[0016] Acquire AIS data of ships in the target sea area. The AIS data of ships includes static information and dynamic information of each ship. The static information includes ship type, length and beam. The dynamic information includes ship position, speed, heading and timestamp.

[0017] The ship AIS data is preprocessed to remove abnormal data with missing static information and locations exceeding the target sea area boundary, resulting in preprocessed ship AIS data;

[0018] Based on the preprocessed ship AIS data, the trajectory point sequence of each ship is extracted. The trajectory point sequence is arranged in ascending order by timestamp to form the ship trajectory data of each ship.

[0019] The target sea area is divided into several analysis grid cells. The density of ship trajectory points in each analysis grid cell is calculated using a kernel density estimation algorithm to generate the density value of ship trajectory points in each analysis grid cell. The density value of ship trajectory points in each analysis grid cell is then normalized to obtain the normalized density value of each analysis grid cell.

[0020] Acquire electronic nautical chart data for the target sea area, which includes a channel distribution layer, an anchorage distribution layer, and a navigational obstruction distribution layer; using the electronic nautical chart data as a base map, overlay the normalized density values ​​of each analysis grid cell onto the electronic nautical chart data according to their spatial location to generate a ship AIS traffic flow analysis map.

[0021] As a further aspect of the present invention: the specific calculation process for the external navigation risk score in S2 is as follows:

[0022] Obtain the ship AIS data corresponding to each analysis grid cell, and extract the normalized density value of each analysis grid cell and the number of trajectory points corresponding to each ship type in each analysis grid cell from the ship AIS data. The ship types include merchant ships, fishing ships, cargo ships, passenger ships and engineering ships.

[0023] Based on the ratio of the number of trajectory points corresponding to each ship type in each analysis grid cell to the total number of trajectory points of all ship types in that analysis grid cell, the composition ratio of trajectory points of each ship type in each analysis grid cell is calculated.

[0024] Multiply the normalized density value of each analysis grid cell by the proportion of trajectory points for each ship type to obtain the partial density value of each ship type in each analysis grid cell;

[0025] Obtain a preset set of ship risk weight coefficients, which includes risk weight coefficients corresponding to each ship type;

[0026] Multiply the component density value of each ship type within each analysis grid cell by the risk weight coefficient of the corresponding ship type to obtain the component risk value of each ship type; sum the component risk values ​​of all ship types within the analysis grid cell to obtain the external navigation risk score of each analysis grid cell.

[0027] As a further aspect of the present invention: the specific process for determining several candidate site areas in step S3 is as follows:

[0028] Obtain the external navigation risk score for each analysis grid cell, and mark the analysis grid cells with external navigation risk scores lower than the preset risk threshold as candidate grid cells;

[0029] Connectivity aggregation is performed on each candidate grid cell, adjacent candidate grid cells are merged into connected regions, several candidate site areas are generated, and the boundary coordinate sequence of each candidate site area is extracted as the site boundary parameter of each candidate site area.

[0030] As a further aspect of the present invention: in step S4, the specific planning process for the operation and maintenance routes from each candidate homeport to the corresponding candidate site area is as follows:

[0031] Obtain the site boundary parameters and corresponding candidate home ports for each candidate site area. The site boundary parameters include the boundary coordinate sequence of each candidate site area, and the candidate home ports include port location coordinates. Obtain electronic nautical chart data for the target sea area. The electronic nautical chart data includes a water depth distribution layer, a navigational obstruction distribution layer, and a restricted navigation zone distribution layer.

[0032] The geometric center point of each candidate site area is determined by the boundary coordinate sequence of each candidate site area as the end point of the route, and the port location coordinates of the candidate home port corresponding to each candidate site area are used as the starting point of the route. Based on the electronic nautical chart data, the shortest navigation path is planned that meets the constraints of water depth not less than a preset water depth threshold and avoiding the distribution layers of navigation obstacles and restricted areas, and the operation and maintenance route from each candidate home port to the corresponding candidate site area is generated.

[0033] As a further aspect of the present invention: in step S5, the specific extraction process of the route feature parameters of each maintenance route is as follows:

[0034] Obtain the operation and maintenance routes from each candidate homeport to the corresponding candidate site area. The operation and maintenance routes include a sequence of waypoints arranged in the order of navigation, and each waypoint includes location coordinates.

[0035] Calculate the Euclidean distance between adjacent path points sequentially along the maintenance route, and sum the distance values ​​between each adjacent path point to obtain the route length;

[0036] Extract the water depth values ​​of each path point on the maintenance route, and select the minimum water depth value among all path point values ​​to obtain the minimum water depth of the route;

[0037] Obtain the channel distribution layer of the target sea area, perform spatial overlay analysis on the operation and maintenance route and the channel distribution layer, count the number of intersections between the operation and maintenance route and the channel distribution layer, and obtain the number of times the route crosses the channel.

[0038] Acquire fishing area distribution data for the target sea area, perform spatial overlay analysis on the operation and maintenance route and the fishing area distribution data, count the number of intersections between the operation and maintenance route and the fishing area distribution data, and obtain the number of times the route crosses the fishing area.

[0039] The route length, minimum water depth, number of times the route crosses a waterway, and number of times the route crosses a fishing area are used as route characteristic parameters for operation and maintenance routes.

[0040] As a further aspect of the present invention: the specific calculation process of the operation and maintenance path conflict index in S5 is as follows:

[0041] Obtain all operation and maintenance routes corresponding to each candidate site area, and extract the route feature parameters of each operation and maintenance route. The route feature parameters include route length, minimum water depth of the route, number of times the route crosses the waterway, and number of times the route crosses the fishing area.

[0042] For each maintenance route, the reciprocal of the minimum water depth of the route is taken as the water depth hazard value; the total route length, the total water depth hazard value, the total number of times the route crosses the waterway, and the total number of times the route crosses the fishing area are calculated for all maintenance routes.

[0043] Obtain a preset set of route feature weight coefficients, which includes route length weight coefficient, water depth hazard value weight coefficient, number of times the route crosses a waterway weight coefficient, and number of times the route crosses a fishing area weight coefficient.

[0044] For each maintenance route, the navigation difficulty coefficient is obtained by summing the following: the ratio of route length to the total route length multiplied by the route length weighting coefficient; the ratio of water depth hazard value to the total water depth hazard value multiplied by the water depth hazard value weighting coefficient; the ratio of the number of times the route crosses a waterway to the total number of times the route crosses a waterway multiplied by the number of times the route crosses a waterway multiplied by the number of times the route crosses a fishing area to the total number of times the route crosses a fishing area multiplied by the ...

[0045] As a further aspect of the present invention: in step S6, the specific process for generating the target site area is as follows:

[0046] Obtain the external air traffic risk score and operation and maintenance path conflict index for each candidate site area. For each candidate site area, the external air traffic risk score and operation and maintenance path conflict index are weighted and summed to obtain the comprehensive score of each candidate site area. Select the candidate site area with the lowest comprehensive score as the target site area.

[0047] The beneficial effects of this invention are:

[0048] 1) This invention first generates a traffic flow analysis map covering the target sea area through ship AIS data extraction and density analysis. The sea area is then divided into analysis grid units. Combining the proportion of trajectory points of different ship types within each grid unit with preset risk weight coefficients, the external navigation risk score of each grid unit is calculated. Candidate site areas with lower external navigation risks are selected to ensure basic navigation safety for wind farms and passing ships. Furthermore, an operation and maintenance dimension is introduced. For each candidate site area, available candidate home ports are identified in the surrounding area. Based on electronic nautical chart data, operation and maintenance routes from each home port to the site are planned. By extracting characteristic parameters such as route length, minimum water depth, number of channel crossings, and number of fishing area crossings, the navigation difficulty coefficient of each route is calculated, and the minimum value is taken as the operation and maintenance path conflict index for that site. Finally, the external navigation risk score and the operation and maintenance path conflict index are weighted and summed, and the lowest comprehensive score is used as the selection criterion for the target site. Through the above-mentioned hierarchical screening and multi-indicator fusion mechanism, this invention integrates external navigation risks and operation and maintenance path risks into the same decision-making framework, so that the final selected site is not only far away from the main channel and densely populated fishing areas, but also has convenient and safe operation and maintenance access channels, avoiding the problem of sacrificing the safety of another dimension due to single-dimensional optimization, and achieving synergistic optimization and balanced control of the two types of risks.

[0049] 2) After planning the operation and maintenance routes from each candidate homeport to the candidate site area, this invention extracts route length from the distance dimension to reflect sailing time and fuel costs, minimum water depth from the physical safety dimension to reflect grounding risk, and the number of times crossing waterways and fishing areas from the interaction and conflict dimension to reflect the probability of conflict with passing vessels. These multi-dimensional feature parameters comprehensively characterize the navigation difficulty of the operation and maintenance routes. Based on this, by normalizing and weighting the parameters of all operation and maintenance routes, the navigation difficulty coefficient of each route is calculated, making the operation and maintenance routes from different homeports to the same site objectively comparable. This method provides a scientific basis for operation and maintenance safety in site selection decisions, enabling decision-makers to choose the optimal operation and maintenance entry and exit scheme among multiple candidate homeports.

[0050] 3) This invention controls operational risks from the outset by evaluating maintenance activities at the site selection stage. By selecting the minimum navigation difficulty coefficient for each candidate site area as the maintenance path conflict index, this index comprehensively reflects the navigation risk level of entering and exiting the site from the optimal home port. By incorporating the maintenance path conflict index and external navigation risk score into a unified comprehensive scoring system, this invention can effectively identify and avoid problems such as increased fuel costs and travel time due to excessively long maintenance routes, limited vessel passage windows due to shallow water, and increased avoidance frequency and collision risk due to frequent crossings of waterways or fishing areas. Selecting the site with the lowest comprehensive score means that the site has the optimal maintenance entry and exit route while meeting external navigation safety requirements, thereby reducing navigation risks for maintenance vessels at the source, reducing maintenance costs, improving maintenance efficiency and reliability, and laying a solid foundation for the long-term safe and stable operation of wind farms. Attached Figure Description

[0051] The invention will now be further described with reference to the accompanying drawings.

[0052] Figure 1 This is a schematic diagram of a maritime transportation safety analysis method based on AI and ship AIS big data according to the present invention. Detailed Implementation

[0053] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0054] Please see Figure 1 As shown, this invention is a maritime transportation safety analysis method based on AI and ship AIS big data, including the following steps:

[0055] S1. Acquire AIS data of ships in the target sea area, and perform trajectory extraction and density analysis on the ship AIS data to generate a ship AIS traffic flow analysis map covering the target sea area.

[0056] When acquiring AIS data for vessels in the target sea area, this data originates from the shipborne Automatic Identification System, recording information continuously broadcast by each vessel during its voyage. This includes static parameters such as the position coordinates, speed, heading, vessel type, length, and beam of a bulk carrier at a given moment. Because the raw data may contain invalid records—such as outliers whose position coordinates significantly exceed the sea area boundaries or missing crucial information about the vessel type—preprocessing is necessary to remove unusable data, ensuring that every record used subsequently is complete and reliable. Next, all trajectory points of each vessel are concatenated chronologically according to its unique identifier to form its navigation trajectory. For example, the trajectory points of a container ship from entering to leaving the sea area are arranged chronologically to reconstruct its complete route. Finally, the target sea area is divided into many identical square grids of a fixed size, each grid corresponding to a small area within the sea area. The density value of each grid cell is calculated using a kernel density estimation algorithm. The principle is that each trajectory point acts like a light source, contributing to the surrounding grid cells within a certain range. The closer the point, the greater the contribution; the farther away, the smaller the contribution. The contributions of multiple trajectory points to the same grid cell are superimposed to obtain the density value of that grid cell. Higher density indicates more frequent ship activity at that location. Since the total number of ships varies across different areas, directly comparing the original density values ​​is unfair. Therefore, the density values ​​of all grid cells are normalized, compressing the values ​​to a uniform range, allowing for direct comparison between different areas. Simultaneously, electronic nautical chart data for the target sea area is acquired. This data, presented in layers, includes information on channel distribution, anchorage locations, and obstructions such as the main channel centerline, anchorage boundaries, and reef areas. Finally, using the electronic nautical chart as a base map, the calculated normalized density values ​​of each grid cell are superimposed onto the chart according to their actual locations, forming a comprehensive map that simultaneously displays the intensity of ship activity and navigational environmental factors—the ship AIS traffic flow analysis map.

[0057] By combining data on waterways and anchorages from electronic nautical charts, density distribution and navigational environmental elements are aligned on the same coordinate system, providing precise spatial location data for subsequent risk assessments. Subsequent site selection no longer relies on vague, empirical judgments but can directly avoid high-risk areas such as main channels and anchorage perimeters based on objective density data, ensuring basic navigational safety for wind farms and passing vessels from the outset. For the entire scheme, the traffic flow analysis map generated in this step serves as the data source for all subsequent external navigational risk calculations, ensuring that the ultimately selected target site has basic guarantees in terms of external navigational safety.

[0058] S2, the target sea area is divided into several evaluation grid units, the normalized density value of each analysis grid unit is extracted based on the ship AIS traffic flow analysis map, and the external navigation risk score of each analysis grid unit is calculated.

[0059] The system acquires AIS data for each analysis grid cell, extracts the normalized density value and the number of trajectory points for each ship type within that grid cell, including merchant ships, fishing vessels, cargo ships, passenger ships, and engineering vessels. Based on the ratio of the number of trajectory points for each ship type within that grid cell to the total number of trajectory points for all ship types within that grid cell, the system calculates the proportion of trajectory points for each ship type within that grid cell. This proportion reflects the relative percentage of different ship types within that grid cell; for example, in a certain grid cell, merchant ship trajectory points may account for 60%, fishing vessel trajectory points for 30%, and other types for a combined 10%.

[0060] The normalized density value of each analysis grid cell is multiplied by the proportion of trajectory points for each ship type to obtain the component density value of each ship type within each analysis grid cell. For example, if the normalized density value of a grid cell is 0.5, and the proportion of merchant ships is 0.6, then the component density value of merchant ships within that grid cell is 0.3. A preset set of ship risk weight coefficients is obtained. This set of ship risk weight coefficients includes risk weight coefficients corresponding to each ship type. These coefficients are preset based on factors such as ship size, cargo characteristics, and maneuverability. For example, large merchant ships are given higher weights due to their large inertia and poor avoidance capabilities, while fishing vessels are given medium weights due to their highly random behavior. The component density value of each ship type within each analysis grid cell is multiplied by the corresponding risk weight coefficient to obtain the component risk value for each ship type. The component risk values ​​of all ship types within the analysis grid cell are summed to obtain the external navigation risk score for each analysis grid cell. This score comprehensively reflects the impact of ship traffic flow within the grid cell on the external navigation risk of wind farm site selection.

[0061] S3. Based on the external navigation risk score of each evaluation grid unit, several candidate site areas are determined, and the site boundary parameters of each candidate site area are extracted.

[0062] The external navigation risk score of each analysis grid cell is obtained, and analysis grid cells with external navigation risk scores lower than a preset risk threshold are marked as candidate grid cells. The preset risk threshold can be set according to the project's safety requirements. For example, the threshold can be set to 0.05, meaning that grids with risk scores lower than 0.05 are considered to be within the acceptable navigation safety zone.

[0063] Connectivity aggregation is performed on each candidate grid cell. An eight-neighbor connectivity algorithm is used to merge adjacent candidate grid cells into connected regions, each of which constitutes a candidate site area. Connected regions with areas smaller than the minimum site area requirement are removed, and those meeting the area requirement are retained as the final candidate site areas. The boundary coordinate sequence of each candidate site area is extracted as the site boundary parameter for each candidate site area. This boundary parameter describes the spatial extent of the candidate site area and provides a spatial reference for subsequent operation and maintenance route planning.

[0064] S4. Obtain the surrounding port information corresponding to each candidate site area, and determine several candidate home ports corresponding to each candidate site area. Based on the electronic nautical chart data, plan the operation and maintenance routes from each candidate home port to the corresponding candidate site area, and obtain all operation and maintenance routes corresponding to each candidate site area.

[0065] Obtain the site boundary parameters and corresponding candidate home ports for each candidate site area. The site boundary parameters include the boundary coordinate sequence of each candidate site area, and the candidate home ports include port location coordinates. The candidate home ports are determined based on the survey results of surrounding ports, selecting ports that meet the requirements of the operation and maintenance vessel type in terms of water depth, berth capacity, and logistical support. Obtain electronic nautical chart data for the target sea area. The electronic nautical chart data includes a water depth distribution layer, a navigational obstruction distribution layer, and a restricted navigation zone distribution layer.

[0066] The geometric center point of each candidate site area is determined using the boundary coordinate sequence of each candidate site area, serving as the endpoint of the route. The port location coordinates of the corresponding candidate home port are used as the starting point of the route. Based on the electronic nautical chart data, the shortest navigation path is planned that satisfies the constraint that the water depth is not less than a preset water depth threshold and avoids the obstruction distribution layer and the restricted navigation zone distribution layer, thus generating maintenance routes from each candidate home port to the corresponding candidate site area. For example, if a candidate site area corresponds to candidate home ports A, B, and C, three maintenance routes are planned from port A to the site, port B to the site, and port C to the site, respectively. This planning process ensures that the generated routes are physically safe and feasible, providing a foundation for subsequent route feature extraction.

[0067] S5, extract the route feature parameters of each operation and maintenance route, calculate the navigation difficulty coefficient of each operation and maintenance route corresponding to each candidate site area based on the route feature parameters, and select the minimum navigation difficulty coefficient corresponding to each candidate site area as the operation and maintenance path conflict index of the candidate site area.

[0068] The operation and maintenance routes from each candidate homeport to the corresponding candidate site area are obtained. These routes consist of a sequence of waypoints arranged in chronological order, with each waypoint containing its location coordinates. The Euclidean distance between adjacent waypoints is calculated sequentially along the route, and the distances between adjacent waypoints are summed to obtain the route length. This parameter reflects the navigation distance and time cost. The water depth values ​​of each waypoint along the operation and maintenance route are extracted, and the minimum water depth is selected to obtain the minimum water depth of the route. This parameter reflects the risk of ship grounding; the shallower the water, the higher the risk. A channel distribution layer of the target sea area is obtained. The operation and maintenance routes are spatially overlaid with the channel distribution layer, and the number of intersections between the operation and maintenance routes and the channel distribution layer is counted to obtain the number of times the route crosses a channel. This parameter reflects the probability of conflict with vessels passing through the main channel. Fishing area distribution data of the target sea area is obtained. The operation and maintenance routes are spatially overlaid with the fishing area distribution data, and the number of intersections between the operation and maintenance routes and the fishing area distribution data is counted to obtain the number of times the route crosses a fishing area. This parameter reflects the probability of conflict with fishing vessel operations.

[0069] Then, the navigation difficulty coefficient and operation and maintenance path conflict index are calculated. All operation and maintenance routes corresponding to each candidate site area are obtained, and the route characteristic parameters of each route are extracted. For each operation and maintenance route, the reciprocal of the minimum water depth is used as the water depth hazard value, so that the physical law that shallower water corresponds to higher hazard values ​​is accurately reflected in the quantitative indicators. The total route length, total water depth hazard value, total number of times the route crosses a waterway, and total number of times the route crosses a fishing area are statistically analyzed for all operation and maintenance routes. A preset set of route characteristic weight coefficients is obtained, which includes weight coefficients for route length, water depth hazard value, number of times the route crosses a waterway, and number of times the route crosses a fishing area. These weight coefficients can be adjusted according to the project's emphasis on cost, safety, and conflict.

[0070] For each maintenance route, the navigation difficulty coefficient is obtained by summing the following: the ratio of route length to the total number of route lengths multiplied by a route length weighting coefficient; the ratio of water depth hazard value to the total number of water depth hazard values ​​multiplied by a water depth hazard value weighting coefficient; the ratio of the number of times the route crosses a waterway to the total number of times the route crosses a waterway to the total number of times the route crosses a waterway to the total number of times the route crosses a fishing area ...

[0071] For each candidate site area, the minimum navigation difficulty coefficient among all corresponding operation and maintenance routes is selected as the operation and maintenance path conflict index for that candidate site area. This index represents the lowest navigation difficulty corresponding to entering and exiting from the optimal home port of that candidate site area, reflecting the best accessibility level of the site in terms of operation and maintenance path safety.

[0072] S6. Based on the external navigation risk score and operation and maintenance path conflict index of each candidate site area, calculate the comprehensive score of each candidate site area, select the candidate site area with the lowest comprehensive score as the target site area and output it.

[0073] The external navigation risk score and operation and maintenance path conflict index of each candidate site area are obtained. For each candidate site area, the external navigation risk score and operation and maintenance path conflict index are weighted and summed to obtain the comprehensive score of each candidate site area. The weighting coefficient can be set according to the relative importance of navigation safety and operation and maintenance safety of the project. For example, if the project pays more attention to the navigation safety of maintenance vessels during the operation phase, the weight of the operation and maintenance path conflict index can be appropriately increased. The candidate site area with the lowest comprehensive score is selected as the target site area. This site achieves overall optimization in both the external navigation risk and operation and maintenance path safety dimensions, avoiding the problem of losing control of the risk in one dimension due to optimization of one dimension.

[0074] It is worth noting that this invention expands wind farm site selection from the traditional single-dimensional assessment of external navigation risks to a collaborative assessment of both external navigation risks and the safety of operation and maintenance routes. Its core lies in recognizing the differences in the selection of home ports for operation and maintenance (O&M) services corresponding to different candidate site areas. Furthermore, the O&M routes from different home ports to the site vary in length, water depth, number of channel crossings, and number of fishing area crossings. These differences directly determine the navigation safety and costs of O&M vessels during the operational phase. Traditional site selection methods only focus on external navigation conflicts between wind farms and passing vessels, failing to incorporate the safety of O&M routes into the assessment system. This results in selected sites, although far from main shipping channels, requiring frequent crossings of high-risk waters for O&M access, leading to increased navigation safety hazards and higher costs for O&M vessels during the operational phase. This invention incorporates two types of risks into the decision-making framework by constructing a comprehensive scoring mechanism that combines external navigation risk scoring with operation and maintenance path conflict index. This ensures that the final selected target site not only guarantees the navigation safety of passing vessels but also has the optimal operation and maintenance access route. This reduces the navigation risks and costs of operation and maintenance vessels from the source and realizes the scientific and forward-looking nature of wind farm site selection decisions.

[0075] In a preferred embodiment of the present invention, the specific process of generating the ship AIS traffic flow analysis diagram in step S1 is as follows:

[0076] First, data from the Automatic Identification System (AIS) of the target sea area is acquired. This data records the static and dynamic information continuously broadcast by each vessel during its navigation. Static information, such as vessel type, length, and beam, is used to distinguish the physical attributes of different vessels, while dynamic information, such as vessel position, speed, heading, and timestamp, is used to reconstruct the vessel's navigation trajectory. Since the raw data may contain anomalies such as missing static information, abrupt changes in dynamic information, or position coordinates exceeding the target sea area boundary, these anomalies can affect the accuracy of subsequent analysis. Therefore, preprocessing is necessary to remove such data, ensuring that every piece of data used for analysis is complete and valid. Based on the preprocessed data, all trajectory points of each vessel are arranged in ascending order of timestamp according to its identifier, forming a trajectory point sequence for that vessel. This sequence completely records the vessel's complete navigation path within the target sea area from entry to exit. The target sea area is then divided into several identical analysis grid cells according to a preset grid size. This division aims to discretize the continuous sea area space, allowing the spatial distribution of vessel trajectory points to be analyzed systematically. Statistical analysis was conducted using a reference frame. A kernel density estimation algorithm was employed to calculate the density of ship trajectory points within each analysis grid cell. This algorithm works by contributing density values ​​to surrounding grid cells based on distance decay, with closer points contributing more and farther points contributing less. Ultimately, the density value of each grid cell reflects the degree of aggregation of ship trajectory points around that location, i.e., the frequency of ship activity. Due to significant differences in the number of ships in different sea areas, directly using the original density values ​​would be difficult due to dimensional differences. Therefore, the density values ​​of each grid cell were normalized, compressing them into a uniform range to ensure comparability between different regions. Simultaneously, electronic nautical chart data for the target sea area was acquired. This data, stored in layers, contains geographic information such as channel distribution, anchorage distribution, and obstruction distribution—essential components of the navigation environment. Finally, using the electronic nautical chart data as a base map, the calculated normalized density values ​​of each analysis grid cell were accurately overlaid onto the electronic nautical chart data according to their spatial location, generating a ship AIS traffic flow analysis map.

[0077] By dividing the area into grid cells and calculating the density of each cell, the intensity of ship activity in space can be accurately expressed numerically, avoiding the subjectivity of making general qualitative judgments about large sea areas. Normalization eliminates the dimensional effects caused by differences in the total number of ships in different areas, allowing for comparison of navigation density at different locations on the same scale. By overlaying electronic nautical chart layers containing channels, anchorages, and navigational obstructions, ship activity information is integrated with geographic environmental information, providing a spatially aligned data foundation for subsequent assessments of external navigation risks. For the entire wind farm site selection scheme, this traffic flow analysis map is the fundamental input for all subsequent external navigation risk assessments. Its accuracy and detail directly determine the reliability of candidate site selection. Only by accurately identifying areas of dense and sparse ship activity through this map can the selected candidate sites be guaranteed basic safety in terms of external navigation.

[0078] In another preferred embodiment of the present invention, the specific calculation process of the external air traffic risk score in step S2 is as follows:

[0079] For each pre-defined analysis grid cell, the normalized density value is first extracted from the ship AIS data. This value reflects the overall density of ship trajectory points within the grid cell, and normalization ensures comparability of densities between different grid cells. Simultaneously, the ship types of all trajectory points within the grid cell are identified from the AIS data, categorized by type (merchant ship, fishing vessel, cargo ship, passenger ship, engineering vessel, etc.). For example, a grid cell might contain forty merchant ship trajectory points, twenty fishing vessel trajectory points, and ten other types, thus revealing the frequency of each ship type within the grid cell. Based on this, the ratio of trajectory points of each ship type to the total number of trajectory points in the grid cell is calculated, yielding the proportion of each ship type within the grid cell. For instance, merchant ship trajectory points might account for four-tenths, and fishing vessel trajectory points for two-tenths. This ratio directly reflects the relative activity intensity of different ship types within the grid cell; a higher number of trajectory points for a particular ship type indicates more frequent activity at that location and a greater impact on the navigation environment. Subsequently, the normalized density value of the grid cell is multiplied by the proportion of trajectory points for each ship type to obtain the component density value for each ship type within that grid cell. For example, if the normalized density value is 0.5 and the proportion of merchant ships is 0.4, then the component density value for merchant ships is 0.2. This value considers both the overall density level and the contribution of different ship types, because the activity intensity of different ship types at the same location has different impacts on safety. Next, a preset set of ship risk weight coefficients is obtained. This set assigns a weight value to each ship type. The weight is set according to the ship's physical characteristics and behavioral features. For example, large merchant ships are given higher weights due to their large inertia, difficulty in maneuvering, and high risk factor for carrying cargo; fishing boats are given medium weights due to their highly random and unpredictable behavior; and small cargo ships have relatively low weights. These weights reflect the differentiated risk contributions of different ship types to navigation safety. Multiplying the sub-density value of each vessel type by the corresponding risk weight coefficient yields the sub-risk value for each vessel type within the grid cell. For example, multiplying the merchant ship sub-density value by the merchant ship weight yields the merchant ship risk contribution value. This product reflects the cumulative effect of "higher density and greater weight, higher risk." Finally, the sub-risk values ​​of all vessel types within the grid cell are summed to obtain the external navigation risk score for that grid cell. This comprehensive score integrates the activity intensity and risk characteristics of different vessel types into a single value, thereby quantifying the overall external navigation risk level of the grid cell caused by vessel traffic flow.

[0080] By differentiating vessel types and assigning them varying weights, risk assessment no longer treats all types of vessels equally. Instead, it accurately reflects the fundamental differences between large merchant ships and small fishing vessels in terms of collision energy, avoidance capabilities, and emergency response, thus preventing misjudgments due to neglecting individual vessel differences. By multiplying the normalized density value by the component proportion, a refined decomposition from overall density to type-specific density is achieved, ensuring that subsequent weighted summation is based on this scientific decomposition. The resulting external navigation risk score, presented in grid cells, provides a spatially continuous and comparable quantitative risk map for wind farm site selection. This score directly serves the subsequent candidate site selection process: only grid cells with risk scores below a preset threshold are eligible to be aggregated into candidate site areas, ensuring that the initially selected sites have basic guarantees in terms of external navigation safety.

[0081] In another preferred embodiment of the present invention, the specific process of determining a plurality of candidate site areas in step S3 is as follows:

[0082] After obtaining the external navigation risk score for each analysis grid cell, the risk score of each grid cell is compared with a pre-set risk threshold. This threshold is determined based on the project's acceptance of navigation safety; for example, only grid cells with risk scores not exceeding a certain value are considered suitable as candidate areas for wind farm site selection. Grid cells with risk scores below this threshold are marked as candidate grid cells, meaning that the cell meets the basic requirements for external navigation safety. Subsequently, connected component aggregation is performed on these candidate grid cells. This involves checking whether each candidate grid cell and its surrounding adjacent grid cells (including vertical, horizontal, and diagonal directions) are also marked as candidate grid cells. If they are adjacent, they belong to the same connected region. By traversing all candidate grid cells, interconnected cells are merged into a single connected region. For example, in a certain sea area, there may be a continuous area composed of dozens of adjacent candidate grid cells, while another area composed of a dozen or so adjacent candidate grid cells is identified as another independent connected region. Thus, each connected region corresponds to a candidate site area. Finally, extract the outer boundary of each connected region, that is, find all the candidate grid cells that constitute the outermost part of the region, and record the coordinates of the center points of these grid cells in clockwise or counterclockwise order to form a closed boundary coordinate sequence. This sequence is the site boundary parameter of the candidate site area, which accurately describes the spatial range of the site.

[0083] By setting risk thresholds and marking eligible grid cells, areas with acceptable external navigation risks can be quickly selected from the entire target sea area, avoiding subsequent analysis at the same depth in all sea areas and significantly improving site selection efficiency. Connectivity aggregation merges scattered candidate grid cells into continuous regions because wind farms require concentrated, contiguous sea areas for development; isolated individual grid cells have no practical development value, and only sufficiently large continuous areas can serve as candidate sites. Extracting the site boundary coordinate sequence provides necessary spatial geometric information for subsequent steps, such as determining the geometric center point of the site when planning operation and maintenance routes, and determining the site coverage area when assessing external navigation risks.

[0084] In another preferred embodiment of the present invention, the specific planning process for the operation and maintenance routes from each candidate homeport to the corresponding candidate site area in step S4 is as follows:

[0085] After obtaining the site boundary parameters for each candidate site area, these parameters record the spatial extent of the candidate site area in the form of a boundary coordinate sequence. The geometric center point of the site area is obtained by averaging the coordinates of these boundary points. This geometric center point is used as the endpoint of the navigation route because the maintenance vessel ultimately needs to reach the interior of the site area for operations, and the geometric center point represents the core location of the site area, ensuring that the planned navigation route reaches the interior of the site area rather than just its edge. Simultaneously, the port location coordinates of the candidate home ports corresponding to each candidate site area are obtained, and these coordinates are used as the starting point of the navigation route, since the maintenance vessel departs from the home port, and the entrance / exit position of the home port is the starting point of the voyage. Electronic nautical chart data of the target sea area is obtained. This data includes a water depth distribution layer, a navigational obstruction distribution layer, and a prohibited navigation zone distribution layer. The water depth distribution layer records the water depth values ​​at various locations in the sea area; the navigational obstruction distribution layer marks the locations of obstacles such as reefs, shipwrecks, and aquaculture areas; and the prohibited navigation zone distribution layer marks areas where entry is prohibited, such as military zones and protected areas. Using the starting and ending points of the route as the two endpoints, a navigation path from the starting point to the ending point is searched on the electronic nautical chart. Two constraints must be met during the search: first, the water depth at all points along the path must be no less than a preset water depth threshold, for example, twice the ship's draft, to ensure the ship will not run aground; second, the path must avoid areas marked by the obstruction distribution layer and the restricted navigation zone distribution layer, as these areas contain physical obstacles or legal prohibitions that prevent ships from passing through. Under these two constraints, the shortest path is selected from all feasible paths as the maintenance route, because a shorter distance means shorter sailing time, less fuel consumption, and less time the ship is exposed to risks. Finally, maintenance routes from each candidate homeport to the corresponding candidate site area are generated. This route is a continuous path from the starting point to the ending point, consisting of a series of waypoints arranged in sailing order.

[0086] The site's geometric center is chosen as the route endpoint because the maintenance vessel's operational range covers the entire site area. Selecting the center point represents the vessel's target location upon arrival, providing a unified reference benchmark for the planned route. The port's location coordinates are used as the route starting point because the home port is the maintenance vessel's departure base; starting from the actual departure location aligns with real-world operational scenarios. Electronic nautical chart data, including water depth, navigational obstructions, and restricted areas, are incorporated as constraints for route planning because vessel navigation must adhere to physical environmental and legal limitations. Insufficient water depth leads to grounding, navigational obstructions cause collisions, and restricted areas prevent entry. Only routes that meet these constraints are truly feasible and safe. The shortest route among feasible routes that satisfy these constraints is selected because maintenance activities require frequent round trips. Shorter routes result in shorter travel times, lower fuel consumption, and less risk exposure time for the vessel en route, making it the optimal choice from both economic and safety perspectives. By taking into account the geographical location of the candidate site area and the available home port resources in the surrounding area, multiple feasible, safe and compliant operation and maintenance routes were generated for each candidate site area. These routes will have their feature parameters extracted and quantitatively evaluated in subsequent steps, so that the safety of the operation and maintenance path can be used as a quantifiable basis for site selection decision-making. This has achieved a leap from a single-dimensional assessment of "external navigation safety" to a dual-dimensional collaborative optimization of "external navigation safety and operation and maintenance path safety".

[0087] In another preferred embodiment of the present invention, the specific extraction process of the route feature parameters of each maintenance route in step S5 is as follows:

[0088] After obtaining the operational routes from each candidate homeport to the corresponding candidate site area, the route consists of a series of waypoints arranged in voyage order. Each waypoint contains its latitude and longitude coordinates. These waypoints are generated by a path planning algorithm, and their continuous arrangement forms a complete curve, representing the actual navigation trajectory of the ship from the homeport to the geometric center of the site area. Along this operational route, the Euclidean distance between any two adjacent waypoints is calculated sequentially, i.e., the straight-line distance is calculated based on the coordinates of the two points. Since the distances between adjacent waypoints are straight-line segments, the total length of the entire route is obtained by summing up all the distances between all adjacent points. This route length reflects the total distance the ship needs to travel from the homeport to the wind farm. For each waypoint on the maintenance route, the water depth value is extracted from the water depth distribution data of the electronic nautical chart. Since each waypoint corresponds to a specific geographic coordinate, and the electronic nautical chart records the water depth information for that coordinate, the water depth values ​​of all waypoints along the entire route are compared, and the smallest value is selected as the minimum water depth of the route. This represents the shallowest water area the vessel may encounter along the entire route; the shallower the water, the higher the risk of grounding. A channel distribution layer for the target sea area is obtained. This layer marks the centerline or boundary of the main channel in the form of geometric lines. The planned maintenance route is spatially overlaid with the channel distribution layer to determine whether each segment of the route intersects with the lines in the channel layer. The number of all intersection points is counted to obtain the number of times the route crosses the channel. This value reflects the number of times the maintenance vessel needs to cross the main channel during navigation; each crossing means a potential conflict with passing vessels in the main channel. The data on the distribution of fishing areas in the target sea area is obtained. This data marks the areas where fishing vessels frequently operate in the form of polygonal regions. The operation and maintenance routes are spatially overlaid with the data on the distribution of fishing areas to determine whether each segment of the route enters the interior of the fishing area polygon. The number of times the route enters the fishing area is counted to obtain the number of times the route crosses the fishing area. This value reflects the number of times the operation and maintenance vessels need to pass through the fishing area. Fishing vessels usually operate at low speed and have frequently changing courses within the fishing area. Crossing the fishing area increases the risk of collision.

[0089] In another preferred embodiment of the present invention, the specific calculation process of the operation and maintenance path conflict index in step S5 is as follows:

[0090] After obtaining all operational routes corresponding to each candidate site area, four characteristic parameters were extracted from each operational route: route length, minimum water depth, number of channel crossings, and number of fishing area crossings. For each operational route, the reciprocal of its minimum water depth was used to obtain the water depth hazard value. This is because the shallower the water depth, the higher the risk of ship grounding; taking the reciprocal results in a larger reciprocal for a shallower water depth, thus transforming the physical law of "the shallower the water depth, the higher the risk" into a positive indicator of higher risk for a larger value, facilitating subsequent unified calculations. Next, the total route length of all operational routes was calculated, i.e., the lengths of all routes were added together; the total water depth hazard values ​​of all operational routes were calculated, i.e., the total number of channel crossings of all operational routes were calculated, i.e., the total number of channel crossings of all operational routes were calculated, i.e., the total number of fishing area crossings of all operational routes were calculated, i.e., the total number of fishing area crossings of all operational routes were calculated, i.e., the total number of fishing area crossings of all operational routes were calculated. These four sums represent the overall performance of all operational routes in terms of length, water depth hazard, channel crossing, and fishing area crossing dimensions. A pre-defined set of route feature weight coefficients is obtained, containing four weight coefficients corresponding to route length, water depth hazard value, number of channel crossings, and number of fishing area crossings. These weight values ​​reflect the project's emphasis on each indicator; for example, if the project prioritizes navigation costs, the weight of route length can be increased, and if navigation safety is prioritized, the weight of water depth hazard value can be increased. For each maintenance route, calculate the ratio of its route length to the total length of all routes. This ratio reflects the relative length of the route compared to all routes; a larger ratio indicates that the route is longer than the overall average. Calculate the ratio of its water depth hazard value to the total water depth hazard value. This ratio reflects the relative water depth hazard of the route compared to all routes; a larger ratio indicates that the route is shallower and riskier than the overall average. Calculate the ratio of the number of times the route crosses a channel to the total number of times the route crosses a channel. This ratio reflects the relative frequency of the route crossing a channel compared to all routes; a larger ratio indicates that the route needs to cross the main channel more frequently than the overall average. Calculate the ratio of the number of times the route crosses fishing areas to the total number of times the route crosses fishing areas. This ratio reflects the relative frequency of the route crossing fishing areas compared to all routes; a larger ratio indicates that the route needs to cross fishing areas more frequently than the overall average. Multiply these four ratios by their respective weighting coefficients and sum them to obtain the navigation difficulty coefficient of the maintenance route. This coefficient comprehensively considers the route's relative performance in four dimensions: length, water depth, channel crossing, and fishing area crossing. The larger the coefficient, the higher the overall navigation difficulty of the route compared to all other routes.

[0091] By converting the minimum water depth of a route into a water depth hazard value, the physical law that shallower water carries higher risk can be positively expressed in mathematical calculations, avoiding calculation biases caused by inconsistent numerical directions. By summing all the data for all routes and calculating the ratio of a single route to the total, relative comparisons between different routes on the same dimension are achieved, ensuring comparability of indicators on a unified scale and avoiding unfair comparisons due to differences in the number of routes or numerical dimensions. Introducing weighting coefficients allows projects to flexibly adjust the emphasis on different indicators according to actual needs. For example, in areas rich in fishery resources, the weight of the number of times fishing areas are crossed can be increased; in areas with limited water depth, the weight of the water depth hazard value can be increased, enhancing the adaptability and flexibility of the method. The final calculated navigation difficulty coefficient integrates multi-dimensional characteristics into a single value, allowing direct comparisons of operation routes from different home ports to the same site. A smaller coefficient indicates a lower overall navigation difficulty for the route. For each candidate site area, the route with the lowest navigation difficulty coefficient among all corresponding operation and maintenance routes is selected as the representative of that site. This minimum value reflects the optimal operation and maintenance path level that can be achieved when selecting the optimal home port for entry and exit from all available home ports of that site, i.e., the operation and maintenance path conflict index. This index directly serves the subsequent comprehensive scoring stage of S6, enabling different candidate site areas to be compared under a unified operation and maintenance path safety index, thereby selecting the target site with the best overall performance in both external navigation risk and operation and maintenance path safety dimensions.

[0092] In another preferred embodiment of the present invention, the specific process of generating the target site area in step S6 is as follows:

[0093] After obtaining the external navigation risk score and operation and maintenance path conflict index for each candidate site area, these two values ​​have already been calculated in the previous steps for each candidate site area. The external navigation risk score reflects the degree of impact of vessel traffic flow in the area where the site is located on external navigation safety, while the operation and maintenance path conflict index reflects the overall navigation difficulty of the operation and maintenance route when entering and exiting from the optimal home port available for the site. These two values ​​are then weighted and summed. First, a weight coefficient is assigned to the external navigation risk score, and another weight coefficient is assigned to the operation and maintenance path conflict index. The sum of the two weight coefficients is one. Then, the external navigation risk score is multiplied by its weight coefficient, and the operation and maintenance path conflict index is multiplied by its weight coefficient. Finally, the two products are added together to obtain the comprehensive score for the candidate site area. The value of the weight coefficient can be flexibly adjusted according to the project's emphasis on different dimensions. For example, if the navigation environment in the project area is complex and there is a high density of external vessels, the weight of the external navigation risk score can be appropriately increased. If the project is far from shore and is sensitive to operation and maintenance costs, the weight of the operation and maintenance path conflict index can be appropriately increased. For multiple candidate site areas, calculate their respective comprehensive scores according to the above method, then compare all the comprehensive scores and select the candidate site area with the smallest value as the target site area. This is because the smaller the comprehensive score, the better the site's overall performance in terms of external navigation risk and operational route safety. In other words, it can meet the navigation safety requirements of passing ships and has good operational access conditions.

[0094] The foregoing has provided a detailed description of one embodiment of the present invention, but this description is merely a preferred embodiment and should not be construed as limiting the scope of the invention. All equivalent variations and modifications made within the scope of the claims of this invention should still fall within the patent coverage of this invention.

Claims

1. A maritime transportation safety analysis method based on AI and ship AIS big data, characterized in that, Includes the following steps: S1. Acquire AIS data of ships in the target sea area, and perform trajectory extraction and density analysis on the ship AIS data to generate a ship AIS traffic flow analysis map covering the target sea area. S2, the target sea area is divided into several analysis grid units, the normalized density value of each analysis grid unit is extracted based on the ship AIS traffic flow analysis map, and the external navigation risk score of each analysis grid unit is calculated. S3. Based on the external navigation risk score of each analysis grid cell, several candidate site areas are determined, and the site boundary parameters of each candidate site area are extracted. S4. Obtain the surrounding port information corresponding to each candidate site area, and determine several candidate home ports corresponding to each candidate site area. Based on the electronic nautical chart data, plan the operation and maintenance routes from each candidate home port to the corresponding candidate site area, and obtain all operation and maintenance routes corresponding to each candidate site area. S5, extract the route feature parameters of each operation and maintenance route, calculate the navigation difficulty coefficient of each operation and maintenance route corresponding to each candidate site area based on the route feature parameters, and select the minimum navigation difficulty coefficient corresponding to each candidate site area as the operation and maintenance path conflict index of the candidate site area. S6. Based on the external navigation risk score and operation and maintenance path conflict index of each candidate site area, calculate the comprehensive score of each candidate site area, select the candidate site area with the lowest comprehensive score as the target site area and output it.

2. The maritime transportation safety analysis method based on AI and ship AIS big data as described in claim 1, characterized in that, In S1, the specific process for generating the ship AIS traffic flow analysis diagram is as follows: Acquire AIS data of ships in the target sea area. The AIS data of ships includes static information and dynamic information of each ship. The static information includes ship type, length and beam. The dynamic information includes ship position, speed, heading and timestamp. The ship AIS data is preprocessed to remove abnormal data with missing static information and locations exceeding the target sea area boundary, resulting in preprocessed ship AIS data; Based on the preprocessed ship AIS data, the trajectory point sequence of each ship is extracted. The trajectory point sequence is arranged in ascending order by timestamp to form the ship trajectory data of each ship. The target sea area is divided into several analysis grid cells. The density of ship trajectory points in each analysis grid cell is calculated using a kernel density estimation algorithm to generate the density value of ship trajectory points in each analysis grid cell. The density value of ship trajectory points in each analysis grid cell is then normalized to obtain the normalized density value of each analysis grid cell. Acquire electronic nautical chart data for the target sea area, which includes a channel distribution layer, an anchorage distribution layer, and a navigational obstruction distribution layer; using the electronic nautical chart data as a base map, overlay the normalized density values ​​of each analysis grid cell onto the electronic nautical chart data according to their spatial location to generate a ship AIS traffic flow analysis map.

3. The maritime transportation safety analysis method based on AI and ship AIS big data as described in claim 2, characterized in that, In S2, the specific calculation process for the external air traffic risk score is as follows: Obtain the ship AIS data corresponding to each analysis grid cell, and extract the normalized density value of each analysis grid cell and the number of trajectory points corresponding to each ship type in each analysis grid cell from the ship AIS data. The ship types include fishing boats, cargo ships, passenger ships and engineering vessels. Based on the ratio of the number of trajectory points corresponding to each ship type in each analysis grid cell to the total number of trajectory points of all ship types in that analysis grid cell, the composition ratio of trajectory points of each ship type in each analysis grid cell is calculated. Multiply the normalized density value of each analysis grid cell by the proportion of trajectory points for each ship type to obtain the partial density value of each ship type in each analysis grid cell; Obtain a preset set of ship risk weight coefficients, which includes risk weight coefficients corresponding to each ship type; Multiply the component density value of each ship type within each analysis grid cell by the risk weight coefficient of the corresponding ship type to obtain the component risk value of each ship type; The external navigation risk score for each analysis grid cell is obtained by summing the risk values ​​of all ship types within the analysis grid cell.

4. The maritime transportation safety analysis method based on AI and ship AIS big data as described in claim 1, characterized in that, In step S3, the specific process for determining several candidate site areas is as follows: Obtain the external navigation risk score for each analysis grid cell, and mark the analysis grid cells with external navigation risk scores lower than the preset risk threshold as candidate grid cells; Connectivity aggregation is performed on each candidate grid cell, adjacent candidate grid cells are merged into connected regions, several candidate site areas are generated, and the boundary coordinate sequence of each candidate site area is extracted as the site boundary parameter of each candidate site area.

5. The maritime transportation safety analysis method based on AI and ship AIS big data as described in claim 1, characterized in that, In S4, the specific planning process for the operation and maintenance routes from each candidate homeport to the corresponding candidate site area is as follows: Obtain the site boundary parameters and corresponding candidate home ports for each candidate site area. The site boundary parameters include the boundary coordinate sequence of each candidate site area, and the candidate home ports include port location coordinates. Obtain electronic nautical chart data for the target sea area. The electronic nautical chart data includes a water depth distribution layer, a navigational obstruction distribution layer, and a restricted navigation zone distribution layer. The geometric center point of each candidate site area is determined by the boundary coordinate sequence of each candidate site area as the end point of the route, and the port location coordinates of the candidate home port corresponding to each candidate site area are used as the starting point of the route. Based on the electronic nautical chart data, the shortest navigation path is planned that meets the constraints of water depth not less than a preset water depth threshold and avoiding the distribution layers of navigation obstacles and restricted areas, and the operation and maintenance route from each candidate home port to the corresponding candidate site area is generated.

6. The maritime transportation safety analysis method based on AI and ship AIS big data as described in claim 1, characterized in that, In S5, the specific extraction process of route feature parameters for each maintenance route is as follows: Obtain the operation and maintenance routes from each candidate homeport to the corresponding candidate site area. The operation and maintenance routes include a sequence of waypoints arranged in the order of navigation, and each waypoint includes location coordinates. Calculate the Euclidean distance between adjacent path points sequentially along the maintenance route, and sum the distance values ​​between each adjacent path point to obtain the route length; Extract the water depth values ​​of each path point on the maintenance route, and select the minimum water depth value among all path point values ​​to obtain the minimum water depth of the route; Obtain the channel distribution layer of the target sea area, perform spatial overlay analysis on the operation and maintenance route and the channel distribution layer, count the number of intersections between the operation and maintenance route and the channel distribution layer, and obtain the number of times the route crosses the channel. Acquire fishing area distribution data for the target sea area, perform spatial overlay analysis on the operation and maintenance route and the fishing area distribution data, count the number of intersections between the operation and maintenance route and the fishing area distribution data, and obtain the number of times the route crosses the fishing area. The route length, minimum water depth, number of times the route crosses a waterway, and number of times the route crosses a fishing area are used as route characteristic parameters for operation and maintenance routes.

7. The maritime transportation safety analysis method based on AI and ship AIS big data as described in claim 1, characterized in that, In S5, the specific calculation process of the operation and maintenance path conflict index is as follows: Obtain all operation and maintenance routes corresponding to each candidate site area, and extract the route feature parameters of each operation and maintenance route. The route feature parameters include route length, minimum water depth of the route, number of times the route crosses the waterway, and number of times the route crosses the fishing area. For each maintenance route, the reciprocal of the minimum water depth of the route is taken as the water depth hazard value; the total route length, the total water depth hazard value, the total number of times the route crosses the waterway, and the total number of times the route crosses the fishing area are calculated for all maintenance routes. Obtain a preset set of route feature weight coefficients, which includes route length weight coefficient, water depth hazard value weight coefficient, number of times the route crosses a waterway weight coefficient, and number of times the route crosses a fishing area weight coefficient. For each maintenance route, the navigation difficulty coefficient is obtained by summing the following: the ratio of route length to the total route length multiplied by the route length weighting coefficient; the ratio of water depth hazard value to the total water depth hazard value multiplied by the water depth hazard value weighting coefficient; the ratio of the number of times the route crosses a waterway to the total number of times the route crosses a waterway multiplied by the number of times the route crosses a waterway multiplied by the number of times the route crosses a fishing area to the total number of times the route crosses a fishing area multiplied by the ...

8. The maritime transportation safety analysis method based on AI and ship AIS big data as described in claim 1, characterized in that, In step S6, the specific process for generating the target site area is as follows: Obtain the external air traffic risk score and operation and maintenance path conflict index for each candidate site area. For each candidate site area, the external air traffic risk score and operation and maintenance path conflict index are weighted and summed to obtain the comprehensive score of each candidate site area. Select the candidate site area with the lowest comprehensive score as the target site area.

Images