A wharf risk assessment and management method based on artificial intelligence

Abstract

The application discloses a wharf risk assessment and management method based on artificial intelligence, and particularly relates to the field of wharf risk assessment and management, and is used for solving the problems that the existing wharf is difficult to quantize the influence of delay on the supply chain network level and cannot intervene in the supply chain risk in advance. The application builds a comprehensive data platform of wharf operation and supply chain cargo linkage, analyzes and predicts the future delivery window bottleneck time point based on the ship arrival density, operation carrying capacity and stacking structure, retrieves the historical supply chain interruption events and extracts the corresponding cargo and node information, tracks the conduction path of the interruption event in the supply chain network, quantizes the influence of the bottleneck on the downstream node and establishes a supply chain risk influence model, generates a supply chain risk influence graph in combination with the future bottleneck prediction, and finally builds a composite risk portrait and implements targeted risk intervention on the wharf and the supply chain node. The application realizes intelligent prediction and management of the wharf and the supply chain whole-chain risk.

Description

An Artificial Intelligence-Based Method for Terminal Risk Assessment and Management Technical Field

[0001] This invention relates to the field of port risk assessment and management technology, and more specifically, to an artificial intelligence-based port risk assessment and management method. Background Technology

[0002] In the modern port environment deeply integrated with the global supply chain, terminals are no longer simply logistics nodes undertaking loading and unloading tasks, but rather core hubs at critical junctures in the supply chain. Their operational efficiency directly determines the pace of cargo flow and the stability of the downstream network. With the combined effects of factors such as dense berthing of large ships, shortened container turnover cycles, concurrent operations by multiple cargo owners, and compressed delivery windows, the frequency of bottlenecks at terminals during peak hours, such as insufficient operational capacity, complex storage structures, and congested container pick-up routes, has increased significantly. Any bottleneck will generate a chain reaction effect in the supply chain network, preventing manufacturers from obtaining key components in a timely manner, causing rapid depletion of inventory in distribution centers, and even leading to the failure of cross-border e-commerce delivery windows, resulting in order defaults and high cost losses.

[0003] Traditional port risk control relies primarily on manual experience or single-point indicator judgments, lacking the ability to comprehensively analyze multiple factors such as ship arrival density, yard structure, cargo value density, and delivery time windows. Furthermore, it cannot track the propagation path of historical delays within the supply chain network, making it difficult to identify systemic risks caused by localized operational bottlenecks in a timely manner. In the context of highly interconnected global supply chains, a single port operation delay often spreads along the supply network to multiple countries and regions, creating a multiplier effect of delivery delays. This leads to multiple cascading consequences, including inventory shortages, rescheduling of transportation plans, and losses in end-sales revenue at upstream and downstream nodes.

[0004] As the application of artificial intelligence technology in port digitalization continues to deepen, how to use supply chain disruption events to infer vulnerable links in the supply chain, form a visual map of supply chain risks, and implement precise intervention on potentially high-risk nodes has become a key technical problem that urgently needs to be solved in current terminal operations and supply chain management. Summary of the Invention

[0005] In order to overcome the above-mentioned deficiencies of the prior art, embodiments of the present invention provide an artificial intelligence-based terminal risk assessment and management method to solve the problems mentioned in the background art.

[0006] To achieve the above objectives, the present invention provides the following technical solution:

[0007] An artificial intelligence-based method for port risk assessment and management includes the following steps:

[0008] S1. Obtain the flow status of containers at the terminal, extract the value density and delivery window of the supply chain goods, and establish a comprehensive data platform linking terminal operations and supply chain goods.

[0009] S2. Based on the ship arrival density and current terminal operation capacity, combined with the container stacking structure analysis, the container lifting efficiency is analyzed to predict the bottleneck time points of future supply chain cargo delivery window operations.

[0010] S3. Retrieve supply chain network disruption events caused by port operation bottlenecks from the historical data of the integrated data platform, and extract the corresponding cargo information and affected supply chain network nodes.

[0011] S4. Track the transmission path and amplification effect of supply chain disruption events in the supply chain network. Based on the correlation between port operation bottlenecks and supply chain network disruptions, quantify the impact of different operation bottleneck data on the downstream distribution network and generate a supply chain risk impact model.

[0012] S5. Based on the predicted future supply chain goods delivery window bottleneck time points, combined with the supply chain risk impact model, output the supply chain risk impact map;

[0013] S6. Based on the supply chain risk impact map, construct a composite risk profile that includes supply chain-related risks and economic loss risks, and conduct targeted risk intervention management for port operations and supply chain nodes.

[0014] As a further aspect of the present invention, in step S1, obtaining the flow status of containers at the terminal, extracting the value density and delivery window of the supply chain goods, and establishing a comprehensive data platform linking terminal operations and supply chain goods specifically includes:

[0015] Collect real-time location information and status change records of containers at the terminal, including the coordinates of container positions in the yard and the sequence of movement trajectory points;

[0016] Obtain basic cargo data from the cargo list submitted by the supply chain, including the type, scale, value density, and planned delivery time window of the cargo. The value density is determined based on the value of a unit of cargo, and the loss rate of value density over time is also marked.

[0017] Based on the correspondence between container numbers and cargo bill of lading numbers, the operational data of the terminal and the basic data of the supply chain cargo are integrated into a unified time-scaled system to build a comprehensive data platform that provides a collaborative view of terminal operations and supply chain cargo.

[0018] As a further aspect of the present invention, in S2, based on the ship arrival density and current terminal operation capacity, and combined with the container stacking structure analysis to determine container retrieval efficiency, the prediction of future supply chain cargo delivery window bottleneck time points specifically includes:

[0019] Based on the ship arrival timetable and berth allocation plan in the integrated data platform, calculate the spatiotemporal distribution density of ships in the future period;

[0020] Assess the operational capacity saturation for future periods based on the current number of available quay cranes and their average operational efficiency.

[0021] Analyze the specific distribution location of containers in the yard and the complexity of the container retrieval path, and calculate the expected retrieval time by combining historical container retrieval operation data;

[0022] By performing multi-dimensional superposition and predictive analysis of ship spatiotemporal density, operational capacity saturation, and expected container handling time, key operational bottleneck time points that may affect the delivery time of goods in the supply chain can be identified.

[0023] As a further aspect of the present invention, in step S3, retrieving supply chain network disruption events caused by port operation bottlenecks from the historical data of the integrated data platform and extracting the corresponding cargo information and affected supply chain network nodes specifically includes:

[0024] Retrieve historical supply chain disruption data recorded in the integrated data platform, and filter out cargo delivery delay events caused by port operation bottlenecks. The port operation bottleneck data includes the spatiotemporal density of ships, the saturation of operational capacity, and the complexity of container pick-up routes during the corresponding time period of the event.

[0025] For each selected delay event, extract the basic data of the cargo corresponding to the affected containers and the planned delivery delay duration within the time period of the event;

[0026] Based on the upstream and downstream enterprise information associated with the bill of lading number, the affected supply chain network nodes can be located, including specific manufacturing plants, distribution centers, and retail outlets.

[0027] By matching and associating operational bottleneck data with corresponding supply chain node information, a set of historical interruption event cases can be established.

[0028] As a further aspect of the present invention, step S4, specifically tracking the transmission path and amplification effect of supply chain disruption events in the supply chain network, includes:

[0029] Construct a directed graph representation of the supply chain network topology, where nodes represent supply chain entities, edges represent logistics relationships, and edge weights reflect the strength of logistics dependence.

[0030] Based on a set of historical supply chain disruption cases, a graph traversal algorithm is used to search for affected upstream and downstream nodes layer by layer along the directed edge direction, starting from the initial affected node. The impact on each node is calculated based on the product of cargo size, value density loss rate, and the difference in delivery delay time between upstream and downstream nodes.

[0031] By comparing the ratio of the influence of upstream nodes to the influence of downstream nodes, and removing the inventory buffer weight of each node, the amplification effect coefficient of the influence of each node in the interruption event is obtained.

[0032] As a further aspect of the present invention, in step S4, based on the correlation between port operation bottlenecks and supply chain network disruptions, the degree of impact of different operation bottleneck data on the downstream distribution network is quantified, and a supply chain risk impact model is generated, specifically including:

[0033] The random forest algorithm is used to train a supply chain risk impact model that links operational bottlenecks with supply chain disruptions. The data on port operational bottlenecks and cargo data from the historical supply chain disruption event case set are used as input features, and the amplification effect coefficient of the supply chain node impact is used as the prediction target.

[0034] Deploy the trained supply chain risk impact model as an application service.

[0035] As a further aspect of the present invention, in step S5, the output of the supply chain risk impact map, based on the predicted future supply chain goods delivery window bottleneck time points and combined with the supply chain risk impact model, specifically includes:

[0036] Input the predicted future operational bottleneck time points corresponding to the terminal operational bottleneck data and cargo basic data into the trained supply chain risk impact model, and output the prediction results of the supply chain node impact amplification effect coefficient of each node in the supply chain network.

[0037] Based on the prediction results, the risk labels of nodes in the supply chain network topology are updated, and a dynamic layout is used to generate a visualized supply chain risk impact map with time-series characteristics.

[0038] As a further aspect of the present invention, in step S6, constructing a composite risk profile based on the supply chain risk impact map, which includes supply chain-related risks and economic loss risks, and conducting targeted risk intervention management for port operations and supply chain nodes specifically includes:

[0039] Based on the cargo scale and value density of the initial nodes in the supply chain network corresponding to the supply chain risk impact map, and combined with the amplification effect coefficient of each node's impact, an economic loss risk index for each node is generated. Based on the magnitude of the economic loss risk index, targeted risk intervention management is carried out on port operations and supply chain nodes.

[0040] The technical effects and advantages of the artificial intelligence-based terminal risk assessment and management method of this invention are as follows:

[0041] By constructing a comprehensive data platform that deeply integrates terminal operation data and supply chain cargo information, this invention overcomes the limitations of traditional terminal risk assessments that only focus on on-site operational load. It enables end-to-end correlation analysis of key factors such as cargo value density, delivery window, and storage structure, allowing for the early identification of future high-risk operational bottlenecks and quantification of their impact on the downstream supply network. This transforms terminal risk control from a reactive response to a proactive prediction approach. Furthermore, by tracing the path of historical supply chain disruptions and modeling their amplification effects, this invention reveals how local bottlenecks propagate within the complex topology of the supply chain, identifying hidden sensitive nodes and vulnerable links. This information is then used to train a supply chain risk impact model, enabling risk analysis to be learnable and iteratively optimized. The resulting supply chain risk impact map presents node risk levels and potential economic losses in a time-series manner, providing precise intervention criteria for terminal operations, shipping schedule organization, yard scheduling, and supply chain enterprises. This achieves a management upgrade from "bottleneck discovery" to "bottlenecks prevention."

[0042] This invention can effectively reduce the risk of delays in key components, reduce downstream inventory disruptions and order defaults, improve supply chain continuity and efficiency, and ultimately achieve overall risk convergence and minimize economic losses in port operations and supply chain networks, demonstrating significant engineering and commercial value. Attached Figure Description

[0043] Figure 1 is a schematic diagram of a port risk assessment and management method based on artificial intelligence according to the present invention. Detailed Implementation

[0044] 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 a part of the embodiments of the present invention, and not all of them. 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. Embodiment 1

[0045] Figure 1 illustrates a terminal risk assessment and management method based on artificial intelligence according to the present invention, which includes the following steps:

[0046] S1. Obtain the flow status of containers at the terminal, extract the value density and delivery window of the supply chain goods, and establish a comprehensive data platform linking terminal operations and supply chain goods.

[0047] S2. Based on the ship arrival density and current terminal operation capacity, combined with the container stacking structure analysis, the container lifting efficiency is analyzed to predict the bottleneck time points of future supply chain cargo delivery window operations.

[0048] S3. Retrieve supply chain network disruption events caused by port operation bottlenecks from the historical data of the integrated data platform, and extract the corresponding cargo information and affected supply chain network nodes.

[0049] S4. Track the transmission path and amplification effect of supply chain disruption events in the supply chain network. Based on the correlation between port operation bottlenecks and supply chain network disruptions, quantify the impact of different operation bottleneck data on the downstream distribution network and generate a supply chain risk impact model.

[0050] S5. Based on the predicted future supply chain goods delivery window bottleneck time points, combined with the supply chain risk impact model, output the supply chain risk impact map;

[0051] S6. Based on the supply chain risk impact map, construct a composite risk profile that includes supply chain-related risks and economic loss risks, and conduct targeted risk intervention management for port operations and supply chain nodes.

[0052] In S1, the flow status of containers at the terminal is obtained, the value density and delivery window of the supply chain goods are extracted, and a comprehensive data platform linking terminal operations and supply chain goods is established.

[0053] Real-time location identification and status recording are performed on all containers in the terminal, using a data collection method consistent with on-site production processes to ensure continuous, complete, and meaningful spatiotemporal data. Specifically, after each container is dropped or moved, its final location is recorded in the yard bay coordinate table. This coordinate table includes the container's location within the yard, its row and column number, and its corresponding stacking height, clearly identifying the static spatial position of each container. Furthermore, during loading, unloading, moving, and lifting operations, on-site equipment automatically records the start and end positions of each action, as well as the time of occurrence. By organizing the sequence of these actions, a sequence of trajectory points can be obtained showing the movement of containers between different areas of the terminal. This sequence consists of several nodes with specific timestamps, area numbers, and spatial coordinates, comprehensively describing the dynamic process of a container moving from one processing station to another. In this embodiment, all container coordinates and trajectory point sequences are merged in chronological order to form a container dynamic spatial dataset covering the entire operation cycle, enabling subsequent steps to accurately grasp the flow process, dwell area, movement path, and time distribution characteristics of each container.

[0054] The basic attributes of goods corresponding to each container are obtained by using the cargo manifests submitted by supply chain companies, and all goods are broken down into data item by item based on the manifest content. Specifically, the cargo manifest is usually provided by the shipping company during customs declaration or advance declaration, and includes information such as the category, batch, packaging form, and container quantity of goods. This example classifies and codes the types of goods to make different types of goods searchable in subsequent processes. For cargo size, the quantity or volume information recorded in the manifest is used, and multiple batches loaded into the same container are summarized to establish a stable association with the container's unique number. For value density, this example uses the declared value information submitted by the supply chain companies, dividing the total value of the goods by the quantity to obtain the unit value of the goods, thus obtaining a clear quantification of value density. Since some goods are sensitive to delivery time, the supply chain companies indicate the percentage of value loss due to time delays. This percentage is not a fixed percentage and is provided by the cargo owner according to their industry characteristics; for example, perishable goods will have a higher loss percentage, while general cargo will have almost no loss. Planned delivery time windows are recorded, provided by the supply chain companies, including the earliest and latest receiving times.

[0055] This embodiment establishes a correspondence between the dynamic locations of the aforementioned containers and the basic cargo data to form a collaborative view that can be used for supply chain linkage analysis. To achieve this, this embodiment first uses the container number as the primary key and associates it with the bill of lading number recorded in the cargo manifest. Since one bill of lading number may correspond to multiple containers, the contents of the manifest are traversed to categorize all containers related to the same bill of lading number, ensuring that cargo information is consistently mapped to specific containers. Subsequently, the real-time location records of containers within the terminal, the time of status changes, and the sequence of trajectory points are merged with the corresponding cargo type, value density, and delivery time window. Because the time systems of different data sources may differ, the time base of the terminal operation scheduling system is selected as the unified time standard, and all time fields from the cargo owner's manifest are converted to the same time base, thereby ensuring that different data have a consistent time reference after merging. After the unified time standard is established, all information is organized into continuous records according to the container flow sequence, so that the complete process of each container from arrival at the port, unloading, relocation, pickup to departure is presented synchronously with its cargo information. The comprehensive data platform ultimately constructed in this embodiment uses containers as the core index unit. By associating locations, operational actions, and cargo attributes, it integrates the terminal operation view and the supply chain view into a unified data framework that can be accessed.

[0056] In S2, based on the density of ships arriving at the port and the current terminal operation capacity, combined with the container stacking structure analysis, the container lifting efficiency is predicted to identify the bottleneck time points for future supply chain cargo delivery windows.

[0057] By retrieving vessel arrival schedules and berth allocation plans for future periods from a comprehensive data platform, the arrival order and space occupancy of vessels at different time points can be obtained. The vessel arrival schedule records the estimated arrival time at the anchorage outside the port, the estimated entry time, and the planned berthing time for each vessel. This data is submitted in advance by shipping companies and verified by the terminal scheduling department. The berth allocation plan specifies the berth number, berthing length, berthing time interval, and expected departure time for each vessel, ensuring a complete record of berth occupancy for future periods. The future timeline is scanned hourly in a time-series manner. At each scanned time point, the number of vessels in berth status is counted, and the berth intervals occupied by these vessels are recorded. This method allows for the determination of the occupancy level of the terminal waters for berthing operations at that time point. Subsequently, the number of vessels, vessel lengths, and berth distribution at each time point are combined to form a record of vessel spatiotemporal occupancy for future periods. To obtain continuous trend changes, the number of ships in each berth interval is statistically analyzed segment by segment using a sliding time interval, so that the ship arrival density presents a continuous density curve on the time axis, and can clearly reflect the specific location and duration of the arrival peak in different time periods.

[0058] When assessing the terminal's operational capacity for future periods, the number of currently available quay cranes is determined, and the operational efficiency of each crane is summarized. The number of quay cranes, provided by the equipment management department, includes those currently in operation, on standby, and under maintenance but expected to be restored, thus clarifying the available equipment status for a specific future period. Operational efficiency is statistically analyzed based on the historical loading and unloading efficiency of similar vessels; for example, a quay crane typically achieves a certain volume of work per hour when handling a particular type of vessel, serving as a reference value. Subsequently, the distribution of vessels in the projected berthing schedule for the future period is matched with the number of available quay cranes. The number of quay cranes that can be allocated to each vessel during different operating periods is clearly defined, and available equipment is matched according to the berth arrangement order, providing a quantifiable equipment basis for each vessel's loading and unloading capacity. Based on this, the total operational capacity of available quay cranes is calculated for each time period and compared with the planned workload of each vessel in the future period to calculate saturation. If, within a certain time period, the planned workload approaches or exceeds the total operational capacity that the quay cranes can provide, that time period is marked as an operational capacity saturation period.

[0059] To obtain the expected pickup time for each cargo, the specific distribution location of each container to be picked up within the yard is determined, including its location within the yard, bay row and column number, and stack height. The yard location determines how container loading equipment needs to enter the yard, from which direction it needs to approach the container, and whether there are other stacked containers obstructing the view, thus directly affecting the complexity of the pickup path. By analyzing the yard structure and container locations, the composition of the pickup path is identified, such as straight-line pickup, detouring around the yard, and cross-area movement, and these paths are categorized into different levels based on their topological complexity. Subsequently, historical pickup operation data is retrieved, and pickup operation records with similar yard locations, similar stack heights, and similar path structures are filtered. The operation stages such as equipment travel time, stack height reduction time, and container loading time in these historical records are statistically analyzed to provide a clear time reference for future pickup operations. For containers at the same location, the corresponding path entry time, the dwell time within the yard, and the time required for loading and leaving the yard are statistically analyzed. The sum of these three times is used as the expected pickup time for that container.

[0060] By overlaying three types of data—ship arrival density, operational capacity saturation, and expected container pick-up time—a comprehensive predictive map of future terminal operational pressure is generated. During the overlay analysis, the three types of data are aligned according to the same time interval along a unified time axis, ensuring that each time interval contains corresponding records of arrival density, equipment capacity status, and expected container pick-up time. Subsequently, the three types of data are combined and analyzed time-by-time. If the ship arrival density is high during a certain time period, and the corresponding operational capacity saturation is also strained, while the expected container pick-up time for a large number of goods is long, then that time period will experience superimposed pressure, exhibiting clear operational bottleneck characteristics. The specific judgment threshold is set based on the actual operational capacity of the terminal. By continuously sliding and statistically analyzing the changing trend of superimposed pressure on the time axis, the periods with the longest duration, highest intensity, and greatest impact on cargo pick-up behavior are identified among all periods of significantly increased pressure and marked as critical operational bottleneck time points.

[0061] In step S3, historical data from the integrated data platform is used to retrieve supply chain network disruption events caused by port operation bottlenecks, and to extract corresponding cargo information and affected supply chain network nodes.

[0062] Historical operational records provided by the integrated data platform are used to retrieve past supply chain disruption events. This platform records all container flow anomalies, container pickup delays, delivery timeouts, and actual receiving delays reported by supply chain network nodes during terminal operations. To clarify the correlation between events and terminal operational bottlenecks, all historical disruption events are sorted by occurrence time, and the terminal operational status on the day of the event and in adjacent time periods is retrieved for each event. Operational status information includes the ship arrival density, the operational capacity saturation of quay crane equipment, and the complexity of container pickup routes within the yard. The records show that certain time periods simultaneously exhibited high ship berthing, equipment shortages, and concentration of deep-positioned containers in the yard. These situations are considered bottleneck characteristics and compared with the time periods of actual delay events. If the occurrence time of a delay event overlaps with the bottleneck period, and the delay duration is consistent with the bottleneck duration, then the event is identified as a supply chain disruption event caused by a terminal operational bottleneck. Subsequently, all records meeting the above criteria were screened, removing delays caused by weather, port controls, or external transportation, retaining only events highly relevant to operational bottlenecks to ensure accuracy. This resulted in a set of historical delivery delay events caused by terminal operational bottlenecks, each event containing a clear time of occurrence, delay duration, number of containers involved, and relevant bottleneck data at the time of the event.

[0063] Each selected delay event was analyzed step-by-step, examining the affected containers and their corresponding cargo information. First, the container numbers in the event records were used to query the corresponding basic cargo data in the integrated data platform, including cargo type, cargo value density, cargo size, and planned delivery window. The actual delay duration in each event is directly related to the planned delivery window; therefore, the planned delivery delay duration for each container was quantified and recorded to clearly define the impact of delays on cargo value and supply chain progress during subsequent modeling. After extracting cargo information, upstream and downstream enterprise information was further retrieved based on the cargo's bill of lading number. The bill of lading number corresponds to the shipping company, carrier, and receiving company in the integrated data platform. By reviewing the records, the corresponding supply chain network nodes for the cargo could be identified, including manufacturing plants, regional distribution centers, or retail outlets. For cargo simultaneously allocated to multiple downstream destinations, their corresponding nodes were recorded item by item, making the scope of the event's impact clearly visible. Subsequently, the operational bottleneck data during the event period, including ship arrival density, operational capacity saturation, and container pickup route complexity, were correlated with the corresponding supply chain nodes for each cargo. This ensured that each delay event had a complete bottleneck cause, cargo attributes, node location, and delay duration. After processing, all events were constructed into a unified case set, with each case including the bottleneck background, involved containers, supply chain node locations, and delay characteristics.

[0064] In S4, the transmission path and amplification effect of supply chain disruption events in the supply chain network are tracked.

[0065] When constructing the supply chain network topology, the first step is to accurately identify all entities involved in the supply chain operation, including raw material suppliers, component suppliers, manufacturing plants, warehousing centers, regional distribution centers, and final delivery nodes, and to clarify the business positioning and functional attributes of these entities. Then, based on the direction of logistics flow between entities, the actual logistics flow paths are converted into directional connectivity relationships, forming the directed connection structure of the supply chain network. In constructing these connections, stable logistics links are extracted from historical transportation paths, order delivery paths, and fixed replenishment routes, and edges are established according to the direction of logistics flow. The weights of the edges are quantified using preset methods to reflect the intensity of logistics dependence, such as using the average number of shipment batches, the size of a single batch of materials, the frequency of long-term cooperation, or the difficulty of logistics substitution between nodes as the basis for weight quantification. Finally, a complete, stable, and accurately reflecting logistics dependence is constructed through the combination of nodes and edges.

[0066] Representative historical disruption events were selected as samples. These events needed to have a clearly defined initial affected node, verifiable records of cascading impacts on the logistics chain, and complete upstream and downstream delivery delay data. For each historical event, the initial affected node was first identified as the starting point for the search. Based on the directed supply chain graph structure constructed in the previous steps, a layer-by-layer diffusion traversal process was performed according to the logistics transmission direction. Specifically, the search proceeded sequentially from the starting node along its downstream nodes. Visited nodes then continued to extend further downstream, ensuring the impact propagation path fully covered all affected links. For business nodes with bidirectional logistics interaction, a supplementary traversal of upstream nodes along their reverse supply direction was also performed to prevent overlooking diffusion links caused by the supply side. During the layer-by-layer traversal, a combination factor of the cargo size, value density loss rate, and delivery delay duration difference was extracted for each visited node in the event. The cargo size was quantified using the node's typical business capacity in the current period, the value density loss rate was obtained from the company's historical loss records, and the delivery delay duration difference was calculated based on the delay records in the event. The three factors are multiplied to generate the node's influence value. This method clearly reflects the combined effect of scale, value, and time loss, and ensures that the influence of each node can be quantified during the traversal process.

[0067] The calculated impact of each node is compared according to its upstream and downstream structure. Based on the directional relationship of nodes within the supply chain, upstream nodes are categorized as supply nodes, and downstream nodes as demand nodes. The impact of each category is then extracted and ranked. When calculating the amplification effect, the ratio of upstream to downstream impact for each node is used as the evaluation basis. This ratio reflects the difference between a node's ability to disrupt upstream dependencies and its ability to transmit disruptions to downstream delivery links during supply chain disruptions. Subsequently, to avoid structural masking effects from inventory buffer comparisons, nodes are excluded based on their own inventory buffer weight. The inventory buffer weight is obtained by quantifying factors such as the node's average inventory coverage days, emergency replenishment frequency, and material substitution degree, quantitatively describing the node's ability to offset time delays during disruption events. The exclusion method involves reducing the aforementioned impact ratio by the inventory buffer weight, resulting in nodes with larger inventory buffers having lower amplification effect coefficients, thus allowing the amplification effect to more accurately reflect the node's true disruption propagation capability. The final impact amplification coefficient can quantitatively represent the accelerating diffusion effect of a node on the entire supply chain during an interruption event.

[0068] In S4, based on the correlation between port operation bottlenecks and supply chain network disruptions, the impact of different operation bottleneck data on the downstream distribution network is quantified, and a supply chain risk impact model is generated.

[0069] A systematic compilation of historical supply chain disruption case studies was conducted to identify the bottleneck data in port operations involved in each disruption, such as ship arrival density, operational capacity saturation, and container handling route complexity, ensuring that all event information was stored in a structured manner. Simultaneously, the corresponding cargo data was integrated to ensure that input features covered the complete dimensions from operational to cargo flow characteristics. After data processing, the input features underwent a unified format conversion, using labeling, numericalization, and normalization to ensure consistent input formats for different feature types entering the training phase. Subsequently, training and validation samples were separated using a strict chronological partitioning method to prevent information leakage and make the training results more closely resemble real-world supply chain scenarios. During model training, the parameters of the random forest algorithm were explicitly set, including setting the number of random trees to no less than 300 to enhance model stability, the maximum tree depth to approximately 15 layers to balance expressive and generalization capabilities, and the minimum number of split samples per node to 5 to ensure reliable splitting processes for each tree and avoid bias towards extreme samples. During node splitting, the random forest evaluates all available feature splitting points and selects the feature and threshold combination that maximizes the purity of the split dataset, enabling the system to gradually form a decision path consistent with the actual supply chain impact propagation pattern. As the number of decision trees increases, the model gradually forms a prediction mechanism that integrates voting from multiple tree structures. After model training, a validation set is used to evaluate model performance. The evaluation includes the deviation level of the predicted impact amplification effect, the contribution ranking of bottleneck features and cargo basic data within the model, and the overall stability index of the random forest. If the evaluation results meet the set performance requirements, the model training is considered complete; if not, training is re-executed by adjusting parameters such as the number or depth of trees until the model performance meets expectations. The finally trained model has a stable impact prediction capability and can accurately reflect the amplification effect of different operational bottleneck combinations in the supply chain network.

[0070] The supply chain risk impact model is deployed as an application service for actual business use. The deployment process first involves serializing the trained model so that it can be stored as an independent file and loaded through a standard service interface. The serialized model is then deployed to the risk assessment application's runtime environment and configured with unified input and output interfaces, including the terminal operation bottleneck field and cargo basic field at the input end, and the predicted value of the supply chain node impact amplification effect coefficient at the output end.

[0071] In S5, based on the predicted future supply chain goods delivery window bottleneck time points, and combined with the supply chain risk impact model, a supply chain risk impact map is output.

[0072] The identified future operational bottleneck time points are structurally integrated with the corresponding terminal operational bottleneck data. Simultaneously, basic cargo data related to these time points is also integrated. To ensure model input consistency, the above data undergoes field matching and format normalization, ensuring the input structure fully corresponds to the feature requirements of the trained supply chain risk impact model. The processed data is then imported into the risk impact model, which automatically invokes its embedded node impact transmission rules, node sensitivity correlations, and network structure feature learning results to extrapolate the diffusion path of the input bottleneck data within the supply chain network. During the model extrapolation, the flow structure of the network topology, historical inter-node linkages, typical blocking propagation links, and node resilience are comprehensively calculated, ultimately outputting the predicted impact amplification effect coefficients for each node in the supply chain network.

[0073] After obtaining the predicted impact amplification coefficients for each node, these results are compared node-by-node with the original topology of the supply chain network. Based on node location, functional attributes, and the relative magnitude of the predicted impact amplification coefficients, the risk classification labels of nodes in the network are updated. This update process clarifies the risk level classification criteria according to a preset risk label setting method, ensuring that nodes with different risk levels remain clearly distinguishable in subsequent presentations. After completing the risk label revision, the updated node information is input into a dynamic layout drawing. By introducing temporal information, the risk changes of nodes at different time points are continuously presented visually. During the layout generation process, an automatic adjustment method based on the strength of node relationships is used to place key nodes in visual centers or prominent positions. At the same time, the directionality and distance of the connecting lines are adjusted according to the flow relationships between nodes, so that the propagation path of supply chain risks can be intuitively displayed in the map. The final visualized supply chain risk impact map has clear temporal characteristics, showing the trend of node risk changes over time, the transmission process of risk from bottleneck nodes to other nodes, and the dynamic form of the risk diffusion link.

[0074] In S6, a composite risk profile containing supply chain-related risks and economic loss risks is constructed based on the supply chain risk impact map, and targeted risk intervention management is carried out on port operations and supply chain nodes.

[0075] The initial node cargo size and value density recorded in the graph are read according to a predetermined evaluation order, arranged based on the node's position in the supply chain transmission path, ensuring the integrity and consistency of data use. Next, after reading the cargo size and value density of each node, the cargo size, value density, and corresponding value density loss rate are quantified and integrated using the node influence amplification effect coefficient stored in the graph as a standard. During the integration process, a linear superposition method is used, multiplying the cargo size by the value density proportionally and summing the results. This summation value is then amplified step-by-step by the influence amplification effect coefficient to form a loss risk indicator for each node in the risk transmission chain. This indicator is then normalized to ensure cross-node comparability across the entire graph. The normalization method uses a fixed-range linear mapping, mapping the highest loss risk indicator to a set upper limit risk level and the lowest loss risk indicator to a set lower limit risk level, ensuring that the economic loss risk indicators of all nodes exhibit a uniform scale. The normalized values ​​are then used as the node's economic loss risk indicators.

[0076] After the economic loss risk indicators for all nodes are generated, targeted risk intervention management processes are implemented based on the magnitude of the risk indicators. First, an intervention priority sequence is established according to the economic loss risk indicators of nodes from high to low. Nodes with high risk indicators are listed as core intervention targets, nodes with medium risk indicators are listed as regular intervention targets, and nodes with low risk indicators are listed as monitoring targets. This sequence serves as the basis for the subsequent intervention execution order. Next, enhanced management measures are implemented for nodes listed as core intervention targets. These measures include restrictive adjustments to the corresponding terminal operation processes and strengthening cross-departmental collaborative review between nodes, enabling high-risk nodes to reduce potential loss exposure in a short period. For nodes listed as regular intervention targets, based on the characteristic of medium-level risk indicators, management measures focused on process optimization are implemented. These include improving the time control accuracy of the unloading process, increasing the verification intensity of the cargo transfer process, and adjusting the connection rhythm between nodes, thereby reducing risk spread while maintaining operational stability. For nodes under monitoring, a routine monitoring mechanism is established based on their relatively low risk indicators. Periodic checks and updates to the spectral data are used to determine if their risk has changed. If an upward trend in the indicators is detected, the node is immediately moved to a higher-level intervention target sequence. Finally, the aforementioned targeted intervention management content is integrated into the port operation management system, allowing economic loss risk indicators to directly serve as the basis for operational control, thereby forming a sustainable, iterative, and real-time responsive risk management mechanism for terminal operations and supply chain nodes.

[0077] The above formulas are all dimensionless calculations. The formulas are derived from software simulations based on a large amount of collected data to obtain the most recent real-world results. The preset parameters and thresholds in the formulas are set by those skilled in the art according to the actual situation.

[0078] The above embodiments can be implemented, in whole or in part, by software, hardware, firmware, or any other combination thereof. When implemented using software, the above embodiments can be implemented, in whole or in part, as a computer program product. The computer program product includes one or more computer instructions or computer programs. When the computer instructions or computer programs are loaded or executed on a computer, all or part of the processes or functions described in the embodiments of this application are generated. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions can be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another. For example, the computer instructions can be transmitted from one website, computer, server, or data center to another website, computer, server, or data center via wired (e.g., infrared, wireless, microwave, etc.) means. The computer-readable storage medium can be any available medium that a computer can access or a data storage device such as a server or data center that includes one or more sets of available media. The available medium can be a magnetic medium (e.g., floppy disk, hard disk, magnetic tape), an optical medium (e.g., DVD), or a semiconductor medium. The semiconductor medium can be a solid-state drive.

[0079] Those skilled in the art will recognize that the modules and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0080] Those skilled in the art will understand that, for the sake of convenience and brevity, the specific working processes of the systems, devices, and modules described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here.

[0081] In the several embodiments provided in this application, it should be understood that the disclosed systems, apparatuses, and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of modules is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple modules or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces; the indirect coupling or communication connection between apparatuses or modules may be electrical, mechanical, or other forms.

[0082] The modules described as separate components may or may not be physically separate. The components shown as modules may or may not be physical modules; they may be located in one place or distributed across multiple network modules. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs.

[0083] In addition, the functional modules in the various embodiments of this application can be integrated into one processing module, or each module can exist physically separately, or two or more modules can be integrated into one module.

[0084] If the aforementioned functions are implemented as software functional modules and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or a portion of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0085] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

[0086] In conclusion, the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for port risk assessment and management based on artificial intelligence, characterized in that, The process includes the following steps: S1. Obtain the flow status of containers at the terminal, extract the value density and delivery window of the supply chain goods, and establish a comprehensive data platform linking terminal operations and supply chain goods; S2. Based on the ship arrival density and current terminal operation capacity, analyze container retrieval efficiency in conjunction with container stacking structure, and predict the bottleneck time points for future supply chain goods delivery windows; S3. Retrieve supply chain network disruption events caused by terminal operation bottlenecks from the historical data of the comprehensive data platform, and extract the corresponding goods information and affected supply chain network nodes; S4. Track the transmission path and amplification effect of supply chain disruption events in the supply chain network, and determine the relationship between terminal operation bottlenecks and supply chain network disruptions. The steps are as follows: S5, based on the predicted future delivery window of the supply chain and the bottleneck time point, combined with the supply chain risk impact model, output the risk impact map of the supply chain; S6, based on the supply chain risk impact map, construct a composite risk profile including supply chain associated risks and economic loss risks, and conduct targeted risk intervention management for terminal operations and supply chain nodes; In S3, in the historical data of the integrated data platform, retrieve supply chain network interruption events caused by terminal operation bottlenecks, and extract the corresponding cargo information and affected supply chain network nodes, specifically including: retrieving data from the integrated data platform. The system records historical supply chain disruption data, filters out cargo delivery delays caused by port operation bottlenecks, and includes vessel spatial and temporal density, operational capacity saturation, and container pick-up route complexity for the corresponding time period. For each filtered delay event, it extracts basic data and planned delivery delay duration for the cargo corresponding to the affected containers during the event's occurrence period. Based on the upstream and downstream enterprise information associated with the cargo bill of lading number, it locates the affected supply chain network nodes, including specific manufacturing plants, distribution centers, and retail terminals. It matches and associates the operational bottleneck data with the corresponding supply chain node information to establish a set of historical disruption event cases. In step S4, it tracks the supply chain... The transmission path and amplification effect of disruption events in the supply chain network specifically include: constructing a directed graph representation of the supply chain network topology, where nodes represent supply chain entities, edges represent logistics relationships, and edge weights reflect the strength of logistics dependence; based on a set of historical supply chain disruption event cases, a graph traversal algorithm is used to search for affected upstream and downstream nodes layer by layer along the directed edge direction, starting from the initial affected node; the impact degree on each node is calculated based on the product of cargo size, value density loss rate, and the difference in delivery delay time between upstream and downstream nodes; by comparing the ratio of the impact degree of upstream nodes to the impact degree of downstream nodes, and removing the inventory buffer weight of each node, the amplification effect coefficient of the impact degree of each node in the disruption event is obtained.

2. The method for port risk assessment and management based on artificial intelligence according to claim 1, characterized in that, In step S1, acquiring the flow status of containers at the terminal, extracting the value density and delivery window of the supply chain goods, and establishing a comprehensive data platform linking terminal operations and supply chain goods specifically includes: collecting real-time location information and status change records of terminal containers, including yard bay coordinates and movement trajectory point sequences; obtaining basic data of supply chain goods from the goods list submitted by the supply chain, including the type, scale, value density, and planned delivery time window of the goods, wherein the value density is determined based on the value of a unit of goods, and the loss rate of value density over time is also marked; and integrating the terminal operation data and the basic data of supply chain goods into a unified time-scaled system based on the correspondence between container numbers and bills of lading numbers, constructing a comprehensive data platform that provides a collaborative view of terminal operations and supply chain goods.

3. The method for port risk assessment and management based on artificial intelligence according to claim 1, characterized in that, In S2, based on the ship arrival density and current terminal operation capacity, and combined with the container stacking structure analysis to predict container retrieval efficiency, the prediction of future supply chain cargo delivery window bottleneck time points specifically includes: calculating the spatiotemporal distribution density of ships in the future period based on the ship arrival timetable and berth allocation plan in the integrated data platform; assessing the operational capacity saturation in the future period based on the number of currently available quay cranes and their average operating efficiency; analyzing the specific distribution location of containers in the yard and the complexity of the container retrieval path, and calculating the expected container retrieval time based on historical container retrieval operation data; and performing multi-dimensional superposition prediction analysis of ship spatiotemporal density, operational capacity saturation, and expected container retrieval time to identify key operational bottleneck time points that may affect the supply chain cargo delivery time.

4. The method for port risk assessment and management based on artificial intelligence according to claim 1, characterized in that, In step S4, based on the correlation between port operation bottlenecks and supply chain network disruptions, the impact of different operational bottleneck data on the downstream distribution network is quantified, and a supply chain risk impact model is generated. Specifically, this includes: using a random forest algorithm to train a supply chain risk impact model that correlates operational bottlenecks with supply chain disruptions; using port operation bottleneck data and cargo basic data from the historical supply chain disruption event case set as input features; and using the supply chain node impact amplification effect coefficient as the prediction target; and deploying the trained supply chain risk impact model as an application service.

5. The method for port risk assessment and management based on artificial intelligence according to claim 1, characterized in that, In step S5, based on the predicted future supply chain cargo delivery window bottleneck time points and combined with the supply chain risk impact model, the output of the supply chain risk impact map specifically includes: inputting the terminal operation bottleneck data and cargo basic data corresponding to the predicted future operation bottleneck time points into the trained supply chain risk impact model, and outputting the prediction results of the supply chain node impact amplification effect coefficient of each node in the supply chain network; updating the risk labels of nodes in the supply chain network topology according to the prediction results, and using dynamic layout to generate a visualized supply chain risk impact map with time-series characteristics.

6. The method for port risk assessment and management based on artificial intelligence according to claim 1, characterized in that, In S6, a composite risk profile containing supply chain-related risks and economic loss risks is constructed based on the supply chain risk impact map. Targeted risk intervention management for port operations and supply chain nodes specifically includes: generating economic loss risk indicators for nodes based on the cargo scale and value density of the initial nodes in the supply chain network corresponding to the supply chain risk impact map, combined with the amplification effect coefficient of each node's influence; and conducting targeted risk intervention management for port operations and supply chain nodes based on the magnitude of the economic loss risk indicators.

Images