Dynamic optimization and visualization decision-making system for port and navigation multimodal transport knowledge graph collaboration

Abstract

The invention provides a dynamic optimization and visual decision-making system for port and navigation multimodal transport knowledge graph collaboration, and solves the problems of high transport mode conversion cost, difficult information sharing and the like. Aiming at the defects that a traditional static data warehouse (such as Hadoop) is insufficient in real-time performance, a general data fusion framework (such as Apache NiFi) is poor in field adaptability and the like, the system innovatively constructs a multi-source heterogeneous database, integrates port production, economic association, infrastructure and freight data, and realizes accurate path planning and personalized logistics scheme recommendation through an intelligent algorithm. The system adopts a React + ECharts + ShadcnUI technology stack to realize knowledge graph visualization, practice shows that the operation efficiency can be remarkably improved, the transportation cost is reduced, and technical support is provided for sustainable development of ports.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of intelligent decision-making, and particularly relates to a port and shipping multimodal transport knowledge graph collaborative dynamic optimization and visual decision-making system. BACKGROUND

[0002] Under the background of rapid development of global trade and logistics system, multimodal transport has become the core mode of modern logistics due to its significant advantages of efficient integration of transportation resources and cost reduction. However, China's multimodal transport still faces many challenges in actual operation: high conversion and connection cost between different transportation modes, difficulty in information interconnection and sharing, and difficulty in path optimization due to dynamic changes in transportation environment (such as extreme weather and traffic congestion). These problems seriously restrict the improvement of transportation efficiency and service quality, and are difficult to meet the growing demand of logistics market.

[0003] Traditional multimodal transport optimization methods have obvious limitations. At the data level, existing researches mostly rely on single transportation mode data analysis, lacking deep fusion of port and shipping multi-source heterogeneous data (such as channel information, railway dispatch, highway freight, and port operation). Although distributed data warehouse technology (such as Hadoop) and general data fusion framework (such as Apache NiFi) have been applied, these technologies are difficult to meet the real-time needs of port and shipping business, and lack of field professional knowledge support, resulting in poor data conflict resolution effect. At the algorithm level, traditional optimization models (such as linear programming and genetic algorithm) usually assume static transportation environment, which cannot adapt to dynamic changes in actual operation, limiting the accuracy and timeliness of path planning. At the application level, existing decision-making systems are mostly designed for professional users, with poor interaction experience, making it difficult for non-professionals to use efficiently, hindering the popularization of technology.

[0004] In recent years, the development of knowledge graph and visualization technology provides a new way to solve the above problems. Knowledge graph can effectively integrate multi-source heterogeneous data and reveal complex relationships through semantic association, but its application in port and shipping multimodal transport is still in the exploratory stage. Although existing visualization tools (such as GIS-based monitoring platforms) can display basic logistics information, they lack intuitive presentation and interactive support for dynamic optimization process. In addition, although some researches attempt to introduce real-time data processing technology (such as stream computing framework Flink) and dynamic multi-objective optimization model, these methods have not formed a complete systematic solution, making it difficult to fully cover the planning, execution and decision-making needs of multimodal transport.

[0005] Therefore, an innovative technical solution is urgently needed to deeply integrate port and shipping multi-modal data, combine adaptive dynamic optimization algorithms with interactive visualization technology, and build an intelligent decision support system. This system needs to have real-time data processing capabilities, dynamic path optimization functions, and user-friendly visualization interfaces to solve the problems of information silos, dynamic environment, and decision complexity in multi-modal transportation, thereby promoting the industry towards intelligent and efficient development. SUMMARY

[0006] To overcome the shortcomings of the prior art, the present application provides a port and shipping multi-modal knowledge graph collaborative dynamic optimization and visualization decision system. In this scheme, a multi-source heterogeneous database is innovatively constructed, covering port production, economic correlation, infrastructure network, and cargo transportation activity data. Data integration, cleaning, and fusion technologies are used to deeply mine data value, and intelligent algorithms are used to achieve accurate path planning. Based on customer diversification needs, a logistics scheme intelligent decision system is developed, which provides personalized logistics solutions for customers through big data analysis and recommendation algorithms. The system uses the React framework to build the front end, combines ECharts to realize knowledge graph visualization, and uses ShadcnUI as the UI library. Practice has proved that the present application significantly improves the operation efficiency of multi-modal transportation, effectively reduces transportation cost and time, enhances the flexibility and accuracy of services, and provides strong technical support for the sustainable development of the port industry.

[0007] The technical problem of the present application is solved by the following technical solution: a port and shipping multi-modal knowledge graph collaborative dynamic optimization and visualization decision system, comprising the following steps:

[0008] S1: Construct a multi-source heterogeneous database, integrate port production data, economic correlation data, infrastructure network data, and cargo transportation activity data, form a data resource pool covering the whole process of multi-modal transportation, use data integration, cleaning, and fusion technologies to standardize the processing of multi-source heterogeneous data, solve the problem of inconsistent data format, standard, and transmission protocol, and improve the integrity and consistency of the data;

[0009] S2: Design a visualization decision support system with a three-layer architecture to realize multi-view interactive analysis, support users to view multi-dimensional information such as transportation path, cost distribution, and time efficiency prediction in the same interface, and provide a drag-and-drop interactive interface to reduce the technical threshold;

[0010] Data layer: Integrate multi-source heterogeneous data and use distributed computing to realize real-time processing;

[0011] Algorithm layer: Use a dynamic multi-objective optimization model to generate the optimal solution;

[0012] Presentation layer: Front-end interface is built through React framework, knowledge graph visualization is realized by combining ECharts, and user experience is optimized by using ShadcnUI component library;

[0013] S3: A four-layer architecture is adopted to build the port and shipping multimodal transport knowledge graph:

[0014] The data layer integrates multi-source heterogeneous data based on the ontology model, defines core entities such as freight orders, transportation nodes, transportation resources, and events, etc.

[0015] The extraction layer combines rule-based and attention mechanism-based bidirectional LSTM models to realize entity recognition and relationship extraction.

[0016] The fusion layer solves data conflicts through credibility weighting and spatiotemporal consistency verification (Kalman filtering);

[0017] The reasoning layer uses the SPAROL engine for monitoring and analyzing real-time data streams.

[0018] S4: Data visualization analysis is realized, which can visually display freight order data, path selection and associated relationships in the freight transportation process, and the path status between the starting point and the ending point.

[0019] Further: The construction steps of the multi-source heterogeneous database are as follows:

[0020] S1-1: Distributed ETL (Extract-Transform-Load) is used to realize the collection and cleaning of multi-source heterogeneous data, and a data quality control method combining rule verification and machine learning is used to ensure data integrity and consistency;

[0021] S1-2: Based on the hierarchical storage architecture, the database system is designed, the logical layer uses the spatiotemporal data model to realize the correlation of multi-modal data, and the physical layer relies on the Hadoop HDFS distributed file system to realize efficient storage and access of massive data;

[0022] S1-3: Multi-dimensional data standardization processing is implemented, including data format conversion, semantic alignment, and conflict resolution, to ensure the accuracy of data fusion;

[0023] S1-4: Through the directed acyclic graph (DAG) task scheduling mechanism, the parallelization of data processing flow is realized, combined with the Kafka message queue to build a real-time data transmission channel, forming a full-process data resource pool covering port production, economic association, infrastructure, and transportation activities, providing data support for subsequent path optimization and intelligent decision-making;

[0024] S1-5: The standardized spatio-temporal data model unifies the modeling of the multimodal transport full-process data. The consistency and interoperability of the data are ensured by defining the core field system. The model records the time stamp accurate to the millisecond in the ISO 8601 international standard format, uses the WGS-84 global coordinate system to represent the geographic spatial position information, and introduces the industry general transport state coding system. The three core fields together constitute the basic framework of the transport data;

[0025] S1-6: The multi-level data checking mechanism ensures the data quality. The field integrity check verifies the required fields at the syntax checking level, and the format checking is performed through the regular expression matching. The business rule verification and the value range checking of the numerical value range are performed at the logic checking level, thereby constructing a comprehensive data quality control system.

[0026]

[0027] S1-8: The conversion of the three data formats of EDI, XML and JSON is supported.

[0028] Further, the construction steps of the visual decision support system are as follows:

[0029] S2-1: The data layer is constructed based on a distributed architecture, and multi-source heterogeneous data such as port operation data, ship AIS data and highway and railway transport data are integrated. The system uses a hybrid storage scheme of HBase+HDFS to realize data management and storage.

[0030] S2-2: The algorithm layer adopts a multi-objective optimization model, and the transport time, cost and carbon emission are the core optimization indexes. The dynamic weight adjustment mechanism is used to realize the differentiated demand response.

[0031] The general form of the objective function of the multi-objective optimization problem is as follows:

[0032] Minimize F(x) = [f1(x), f2(x), f3(x)] (1)

[0033] Wherein: f1(x) is the transport time, f2(x) is the transport cost, and f3(x) is the carbon emission; x is the decision variable (such as path selection, transport mode, etc.).

[0034] S2-3: The dynamic weight adjustment mechanism converts the multi-objective into a single objective through the weighted sum method (Weighted Sum Method), and the dynamic weight is adjusted by the user (0% to 100%): ​

[0035]

[0036] where w1, w2, w3: user-defined weights (w1+w2+w3=1). T1, T2, T3 are normalization factors to eliminate dimensional differences, and dynamic weights can be adjusted in real time according to user needs;

[0037] S2-5: The performance layer adopts modular design, and builds a three-view linkage interface with transportation network topology graph, real-time state monitoring panel and path planning result display area as the core. Data association mechanism is established among the three views;

[0038] S2-6: The view association mechanism is built based on Redux state management framework, and the transportation path graph, cost distribution graph and time efficiency prediction are realized through unified data flow management Figure Three Real-time linkage and data synchronization of large core view;

[0039] S2-7: The interactive function realizes intuitive drag operation components through React-Draggable library, so that users can complete complex query construction through simple dragging.

[0040] Further: the port and shipping multimodal transport knowledge graph function has the following specific steps:

[0041] S3-1: Ontology similarity calculation of data layer

[0042] Sim(e1, e2) = a Jaccard(Attr(e1), Attr(e2)) + b Cosine(RelEmbed(e1), RelEmbed(e2)) (3)

[0043] Where Attr(e) is the attribute set of the entity, RelEmbed(e) is the relationship embedding vector based on TransE, a = 0.6, b = 0.4 are the attribute and relationship weights;

[0044] S3-2: Time stamp alignment of space-time data standardization of data layer:

[0045]

[0046] Where Δt is the window size;

[0047] S3-3: Coordinate conversion of space-time data standardization of data layer:

[0048] (x', y') = GIS_Transform(x, y, CRS source , WGS84) (5)

[0049] S3-4: Bidirectional LSTM output of entity recognition of extraction layer (BiLSTM-Attention model):

[0050] h t = [LSTM forward (x t ,h t-1 );LSTM backward (x t ,h t+1 )] (6)

[0051] where: t is the current time step, x t is the input feature of the current time step, h t-1 is the hidden state of the forward LSTM layer at t-1 time step, h t+1 is the hidden state of the backward LSTM layer at t+1 time step, h t is the spliced hidden state of the bidirectional LSTM at time step, used for subsequent entity recognition;

[0052] S3-5: Attention weight of entity recognition of extraction layer (BiLSTM-Attention model):

[0053]

[0054] where α t is the attention weight of the current time step t, indicating the importance of this position in entity recognition. v, W h , W x are trainable parameter matrices used to calculate attention scores;

[0055] S3-6: CRF decoding of entity recognition of extraction layer (BiLSTM-Attention model):

[0056]

[0057] where X is the input sequence, y is the predicted label sequence, T is the sequence length, ψ emit (y t ,h t ) is the emission score, indicating the probability of hidden state h t generating label y t , ψ trans (y t ,y t-1 ) is the transition score, indicating the probability of label y t-1 transiting to y t , y' is all possible label sequences, used for normalization calculation;

[0058] S3-7: Relation classification in relation extraction (BiLSTM-Attention model) of the extraction layer:

[0059] P(r|e1,e2) = Softmax(W r · [ATT(e1); ATT(e2)]) (9)

[0060] Where ATT(e) =∑ t∈span(e) α t h t , e1, e2 are two entities, r: the type of relationship between entities, span(e) is the time step range of entity e in the text, W r : parameter matrix for relation classification, used to map aggregated entity features to relation probability;

[0061] S3-8: State prediction of the fusion layer Kalman filter:

[0062]

[0063] Where F k is the state transition matrix; State vector at time step k-1, Prior state prediction at time step k, F k : state transition matrix, describing the change of state over time, P k-1 : Covariance matrix at time step k-1, indicating state uncertainty, Prior covariance matrix at time step k, Q k : Process noise covariance matrix, describing model error;

[0064] S3-9: Normalization factor of the inference layer (SPARQL engine):

[0065]

[0066] S3-10: Scatter plot layout is one of the important ways for the system to realize knowledge graph visualization, which converts abstract node relationships into intuitive two-dimensional space distribution through carefully designed algorithms;

[0067] S3-11: Guided layout is based on the principle of energy minimization, defining the energy function:

[0068] E =∑ {i≠j} (k rep / ||x i -x j || α )+∑ {(i,j)∈E} k spr (||x i -x j||-l ij ) 2 (12)

[0069] where α = 2 is the repulsive decay factor, l ij is the edge ideal length, k rep is the repulsive constant, controlling the strength of repulsion between nodes, k spr is the attractive constant, controlling the strength of stretching of edges, E is the energy function, balancing the repulsion between nodes and the attraction of edges, making the layout stable;

[0070] S3-12: Node position iterative update formula:

[0071]

[0072] where t is the iteration number, is the coordinate of node i in the tth iteration, η: learning rate, controlling the moving step size of each iteration, is the gradient of the energy function at , guiding the node to move in the direction of energy reduction, β is the damping coefficient, introducing the momentum term to avoid oscillation and accelerate convergence;

[0073] S3-13: Visual encoding uses multi-dimensional mapping:

[0074]

[0075] where (x, y) is the node coordinate, calculated by the layout algorithm. deg(v) is the degree of node v, s max , s min the maximum and minimum values of node size, type(v) is the node type, |T| is the total number of node types, used to map the type to the hue dimension of HSL color, σ type is the preset shape according to the node type;

[0076] Further: the port and shipping multimodal transport knowledge graph function further comprises the following steps:

[0077] S3-14: The hierarchical graph analyzes the type information of all nodes, constructs a tree-like hierarchy, and the root node represents the entire graph, the first layer of child nodes represents different node type categories, and the second layer is each specific node instance. In terms of spatial allocation, the system uses an improved treemap algorithm to recursively divide the canvas space into multiple rectangular regions according to the node quantity proportion, and the size of each rectangle corresponds to the importance sum of the nodes it contains;

[0078] S3-15: Hierarchical space division is based on the weighted treemap algorithm, satisfying:

[0079]

[0080] where v is the current node, parent(v) is the parent node of v, sib(v) is the sibling node set of v, A(v) is the display area of node v, which is proportional to its weight w(v). w(v) is the node weight, which usually represents the node importance, and∑w(sib(v)) is the sum of sibling node weights, which is used to proportionally distribute the parent node area;

[0081] S3-16: The rectangular optimization goal is to minimize the aspect ratio deviation:

[0082]

[0083] where w i , h i is the width and height of the i-th rectangle;

[0084] S3-17: Color coding:

[0085] d(d) = RGB(r base ×(1-γ) d ,g base ×(1-γ) d ,b base ×(1-γ) d ) (17)

[0086] where d is the node level depth, r base , g base b base are the base RGB color values, and γ is the decay factor (0<γ<1) that controls the degree of darkening of the color with increasing level depth;

[0087] S3-18: Label visibility:

[0088]

[0089] where A(v) is the node display area, A min is the minimum area threshold for label visibility, l(v) is the node label text length, and k is the proportional constant that controls the allowed proportion of label length to area;

[0090] S3-19: In the configuration of scatter plots, the system establishes a complete visual coding system, which converts data attributes into visual elements through carefully designed mapping rules. The position of the node in the scatter plot is determined by the (x, y) coordinates calculated by the layout algorithm, and the size of the node is proportional to its importance.

[0091] S3-20: The node size is calculated by a linear mapping function to ensure that important nodes can obtain more significant visual representation:

[0092]

[0093] wherein importance is a node importance indicator (such as transport volume, connection times), is a value range of the node size;

[0094] S3-21: The configuration of the hierarchical graph adopts different visual strategies, highlighting the hierarchical features of the data. The system uses nested rectangles to represent nodes at different levels, with the size of the rectangle being proportional to the number of child nodes it contains. Through a recursive spatial partitioning algorithm, the system ensures that each level is given a reasonable display space, enhancing the visualization effect;

[0095] S3-22: The color coding scheme is associated with the level depth, using a linear gradient to make the high-level area color lighter, and as the level deepens, the color gradually deepens, forming a clear visual hierarchy. The label rendering uses intelligent adaptive technology, with the system calculating the available space for each rectangle in real time, and deciding to display the full label, abbreviated label, or only the icon based on the space size. When the space is insufficient, the system will automatically enable the function of hovering to display the full information, to improve the readability of the information.

[0096] Further: the path visualization function comprises the following steps:

[0097] S4-1: The system overview module provides a comprehensive overview interface to help users monitor and analyze the overall status of the system in real time. The statistical data on the interface shows the number of current tasks, allowing users to quickly understand the distribution of tasks. The chart section uses pie charts and line graphs to visually display the status distribution and time trend of tasks, making it easier for users to identify potential problems and adjust strategies in a timely manner. The "latest task" list displays information about recently created tasks, making it easy for users to quickly view and handle the latest work items.

[0098] S4-2: The freight order management module is used to display and manage task information in the system. Users can view detailed data of tasks through an intuitive table interface, including order number, cargo type, empty container, and other key attributes. This module provides a task overview, allowing users to quickly monitor task status. The table clearly lists detailed information for each task, supporting editing, deletion, or viewing task details, and improving user experience.

[0099] S4-3: The freight graph visualization builds a multi-dimensional knowledge network to visually display the path selection and relationship in the freight transportation process. Based on advanced visualization algorithms, the system converts complex transportation networks into interactive graphical displays, allowing users to clearly identify key information such as topological connections between different transportation paths, transfer nodes, and path attributes.

[0100] S4-4: The path state analysis module allows users to input the starting node and target node to calculate the optimal path between them. After the user selects the nodes in the input box above the interface, the system will calculate and display the path between them in real time. In the path display area, users can see the specific nodes and connection relationships of the path, and intuitively understand the changes and structure of the path.

[0101] S4-5: The path state optimization module focuses on maximum path selection. The system will compare the path lengths according to the user's settings, update the path display in real time, and finally select the shortest path calculated.

[0102] S4-6: The path optimization analysis module provides in-depth path analysis. Users can select specific starting and ending points, and the system will display the path connecting the two nodes and its related attributes. The system selects the optimal path based on length, required time, risk assessment, etc.

[0103] Compared with related prior art, the application has the following main technical advantages:

[0104] Improved generalization ability: Through innovative multi-modal data fusion architecture and adaptive feature extraction mechanism, the system can effectively handle various complex transportation scenarios, greatly improving the adaptability and robustness of the model in different port environments and transportation modes.

[0105] Dynamic prediction capability: The system not only accurately identifies the status of each link in multimodal transport, but also predicts the dynamic trend of the transportation network in real time, which is crucial for risk warning and emergency dispatch in actual business;

[0106] Improved decision-making accuracy: By introducing a spatiotemporal perception deep optimization algorithm and a domain knowledge enhanced reasoning engine, the system's scheme accuracy in multi-objective path planning is improved by more than 35% compared to traditional methods.

[0107] High real-time performance: Based on a streaming computing framework and distributed architecture design, the system can maintain complex functions while still achieving sub-second response, meeting the high timeliness requirements of port operations. Actual measurements show that its processing speed is more than 5 times that of traditional systems. BRIEF DESCRIPTION OF DRAWINGS

[0108] Figure 1 is the model framework structure provided by the present application;

[0109] Figure 2 is the system overview module diagram provided by the present application;

[0110] Figure 3 is the freight order management module diagram provided by the present application;

[0111] Figure 4 is a freight map visualization module diagram provided by the present application;

[0112] Figure 5 is a path state analysis module diagram provided by the present application;

[0113] Figure 6 is a path state optimization module diagram provided by the present application;

[0114] Figure 7 is a path state analysis module diagram provided by the present application; DETAILED DESCRIPTION

[0115] The technical solutions of the present application will be described in detail below in combination with the drawings and examples, but the embodiments of the present application are not limited thereto. Those skilled in the art can make appropriate adjustments and modifications without departing from the core idea of the present application.

[0116] Example 1: Implementation of port and shipping multimodal transport intelligent decision system

[0117] 1. Dataset and experimental environment

[0118] This embodiment uses the Lianyungang port multimodal transport logistics dataset provided by Jiangsu Blue Star Ball Technology Co., Ltd., which contains more than 5000 complete transportation records, and the data fields include:

[0119] Basic information: order number, bill of lading number, cargo type

[0120] Space-time data: starting point coordinates, destination coordinates, timestamp

[0121] Transportation attributes: empty container status, transportation mode, cost details

[0122] Environmental parameters: weather conditions, traffic congestion index

[0123] Experimental hardware configuration:

[0124] Processor: Intel Xeon Gold 6248R (3.0GHz)

[0125] Memory: 128GB DDR4 ECC

[0126] Graphics card: NVIDIA Tesla T4 (16GB GDDR6)

[0127] Storage: 2TB NVMe SSD array

[0128] Software environment:

[0129] Operating system: Ubuntu 20.04LTS

[0130] Development framework: React 18.2+TypeScript 4.9

[0131] Visualization library: ECharts 5.4+ECharts GL 2.0

[0132] Backend service: Spring Boot 3.1+Apache Flink 1.17

[0133] 2. Overall system architecture implementation

[0134] The present application optimizes the system from multiple aspects such as data integration, system design, knowledge graph construction, and visual analysis. First, a multi-source heterogeneous database is constructed to integrate port production data, economic correlation data, infrastructure network data, and cargo transportation activity data, forming a data resource pool covering the entire process of multimodal transport. Through the use of data integration, cleaning, and fusion technology, multi-source heterogeneous data is standardized, solving the problem of inconsistent data format, standard, and transmission protocol, thereby improving the integrity and consistency of the data.

[0135] On this basis, a visual decision support system is designed, which realizes multi-view interactive analysis through a three-layer architecture. The system supports users to view multi-dimensional information such as transportation path, cost distribution, and time efficiency prediction in the same interface, and provides a drag-and-drop interactive interface to reduce the technical threshold. The specific architecture design is as follows: the data layer integrates multi-source heterogeneous data and realizes real-time processing through distributed computing; the algorithm layer uses a dynamic multi-objective optimization model to generate the optimal solution; the presentation layer builds the front-end interface through the React framework, realizes knowledge graph visualization combined with ECharts, and uses the ShadcnUI component library to optimize user experience.

[0136] Further, a four-layer architecture is used to construct the port-harbor multimodal transport knowledge graph. The data layer integrates multi-source heterogeneous data based on the ontology model, defines core entities such as cargo order, transportation node, transportation resource, and event; the extraction layer uses a bidirectional LSTM model combined with rules and attention mechanism to realize entity recognition and relationship extraction; the fusion layer solves data conflicts through credibility weighting and spatiotemporal consistency verification (Kalman filtering); the reasoning layer uses a SPARQL engine for real-time data stream monitoring and analysis.

[0137] Finally, through the realization of data visualization analysis, the cargo order data, path selection and associated relationships in the cargo transportation process, and the path status between the starting point and the ending point can be visualized. This series of measures will comprehensively improve the intelligent management level of port-harbor multimodal transport and provide strong support for decision-making.

[0138] In summary, aiming at the problems in the development of port-harbor multimodal transport in China, the research proposes a solution based on port-harbor multi-source heterogeneous big data fusion, realizes intelligent optimization of multimodal transport path and intelligent decision of logistics scheme by constructing multi-source heterogeneous database and fusing multi-source data, and forms a complete technical system from data acquisition and preprocessing, standardized processing, synchronization mechanism, task management and resource optimization to cluster management and fault-tolerant mechanism in the port-harbor multi-source heterogeneous big data fusion technology.

[0139] The above embodiments only express one embodiment of the present application, and the description is more specific and detailed, but it cannot be understood as a limitation on the scope of the patent. It should be noted that for ordinary skilled persons in the art, without departing from the concept of the present application, a number of modifications and improvements can be made, which are within the scope of protection of the present application.

Claims

1. A dynamic optimization and visualization decision-making system based on the knowledge graph of port and shipping multimodal transport, characterized by: The steps include: S1: Build a multi-source heterogeneous database to integrate port production data, economic data, infrastructure network data, and cargo transportation activity data to form a data resource pool covering the entire multimodal transport process. Use data integration, cleaning, and fusion technologies to standardize multi-source heterogeneous data, resolve issues with inconsistent data formats, standards, and transmission protocols, and improve data integrity and consistency. S2: Design a visual decision support system with a three-tier architecture to implement multi-view linkage analysis. This system allows users to view multi-dimensional information such as transportation routes, cost distribution, and timeliness forecasts on a single interface. It also provides a drag-and-drop interactive interface to lower the barrier to entry for technology users. Data layer: Integrate multi-source heterogeneous data and achieve real-time processing with the help of distributed computing; Algorithm layer: Use dynamic multi-objective optimization model to generate the optimal solution; Presentation layer: The front-end interface is built using the React framework, combined with ECharts to achieve knowledge graph visualization, and the ShadcnUI component library is used to optimize the user experience; S3: A four-layer architecture is used to build a knowledge graph of port and shipping multimodal transport: The data layer integrates multi-source heterogeneous data based on the ontology model and defines core entities such as freight orders, transportation nodes, transportation resources and events; The extraction layer combines rules and a bidirectional LSTM model with an attention mechanism to achieve entity recognition and relationship extraction; The fusion layer resolves data conflicts through credibility weighting and spatiotemporal consistency verification (Kalman filtering); The reasoning layer uses the SPAROL engine to monitor and analyze real-time data streams; S4: Realizes data visualization analysis, which can visualize freight order data, path selection and correlation relationships during cargo transportation, and path status between the starting point and the end point.

2. The dynamic optimization and visualization decision-making system for port and shipping multimodal transport knowledge graph collaboration according to claim 1 is characterized in that: The steps for constructing the multi-source heterogeneous database are as follows: S1-1: Use distributed ETL (Extract-Transform-Load) to collect and clean multi-source heterogeneous data. Ensure data integrity and consistency through a data quality control method that combines rule verification with machine learning. S1-2: Design a database system based on a tiered storage architecture. The logical layer uses a spatiotemporal data model to implement multimodal data association, and the physical layer relies on the Hadoop HDFS distributed file system to achieve efficient storage and access of massive data. S1-3: Implement multi-dimensional data standardization, including data format conversion, semantic alignment, and conflict resolution, to ensure the accuracy of data fusion; S1-4: Parallelize the data processing process through a directed acyclic graph (DAG) task scheduling mechanism. Combined with Kafka message queues, a real-time data transmission channel is established to form a full-process data resource pool covering port production, economic connections, infrastructure, and transportation activities, providing data support for subsequent route optimization and intelligent decision-making. S1-5: A standardized spatiotemporal data model unifies the data of the entire multimodal transport process. By defining a core field system, data consistency and interoperability are ensured. The model records millisecond-accurate timestamps in the ISO 8601 international standard format, uses the WGS-84 global coordinate system to represent geospatial location information, and introduces the industry-wide transport status coding system. These three core fields together form the basic framework of transport data. S1-6: A multi-level data validation mechanism ensures data quality. At the syntax validation level, field integrity checks are implemented to verify required fields. Regular expression matching is also used for format verification. At the logic validation level, business rule verification and value range checks are performed to establish a comprehensive data quality control system. S1-7: A multi-dimensional data alignment method enables heterogeneous data fusion. The GIS system performs unified spatial coordinate conversion to ensure geographic location consistency. A sliding window mechanism is used to address timestamp deviations for precise time alignment. Semantic matching based on predefined transportation domain ontologies eliminates business concept ambiguity. S1-8: Supports conversion of three data formats: EDI, XML, and JSON.

3. The dynamic optimization and visualization decision-making system for port and shipping multimodal transport knowledge graph collaboration according to claim 1 is characterized in that: The steps for constructing the visual decision support system are as follows: S2-1: The data layer is built on a distributed architecture, integrating multi-source heterogeneous data such as port operation data, ship AIS data, and road and rail transportation data. The system uses an HBase+HDFS hybrid storage solution to achieve data management and storage; S2-2: The algorithm layer adopts a multi-objective optimization model, with transportation time, cost, and carbon emissions as core optimization indicators, and achieves differentiated demand response through a dynamic weight adjustment mechanism; The general form of the objective function for modeling multi-objective optimization problems is: MinimizeF(x)=[f1(x),f2(x),f3(x)] (1) Where: f1(x) transportation time, f2(x) transportation cost, f3(x) carbon emissions; x decision variables (such as route selection, transportation mode, etc.); S2-3: Dynamic weight adjustment mechanism converts multiple objectives into a single objective through the weighted sum method. The dynamic weight is adjusted by the user (0% to 100%): Where w1, w2, w3: user-defined weights (w1 + w2 + w3 = 1), T1, T2, T3 are normalization factors used to eliminate dimensional differences, and dynamic weights can be adjusted in real time according to user needs; S2-5: The presentation layer adopts a modular design and builds a three-view linkage interface centered on the transportation network topology map, the real-time status monitoring panel, and the path planning results display area. A data association mechanism is established between the three views. S2-6: The view association mechanism is built on the Redux state management framework. Through unified data flow management, it enables real-time linkage and data synchronization of the three core views: the transportation route map, the cost distribution map, and the time efficiency forecast map. S2-7: The interactive function implements intuitive drag-and-drop operation components through the React-Draggable library, allowing users to complete the construction of complex queries by simply dragging and dropping.

4. The dynamic optimization and visualization decision-making system for realizing the coordination of port and shipping multimodal transport knowledge graph according to claim 1 is characterized in that: The specific steps of the port and shipping multimodal transport knowledge graph function are as follows: S3-1: Data layer ontology similarity calculation Sim(e1,e2)=α·Jaccard(Attr(e1),Attr(e2))+β·Cosine(RelEmbed(e1),RelEmbed(e2)) (3) Where Attr(e) is the attribute set of the entity, RelEmbed(e) is the relation embedding vector based on TransE, α=0.6, β=0.4 are the attribute and relation weights; S3-2: Timestamp alignment of data layer spatiotemporal data standardization: Where Δt is the window size; S3-3: Coordinate transformation of data layer spatiotemporal data standardization: (x′,y′)=GIS_Transform(x,y,CRS source ,WGS84) (5) S3-4: Extraction layer (BiLSTM-Attention model) entity recognition bidirectional LSTM output: h t =[LSTM forward (x t ,h t-1 );LSTM backward (x t ,h t+1 )] (6) Where: t is the current time step, x t is the input feature of the current time step, h t-1 is the hidden state of the forward LSTM layer at time step t-1, h t+1 is the hidden state of the backward LSTM layer at time step t+1, h t It is the concatenated hidden state of the bidirectional LSTM at the time step, which is used for subsequent entity recognition; S3-5: Extraction layer (BiLSTM-Attention model) entity recognition attention weight: where α t is the attention weight of the current time step t, indicating the importance of the position in entity recognition, v, W h 、W x is a trainable parameter matrix used to calculate the attention score; S3-6: Extraction layer (BiLSTM-Attention model) entity recognition CRF decoding: Where X is the input sequence, y is the predicted label sequence, T is the sequence length, ψ emit (y t ,h t ) is the emission fraction, indicating the hidden state h t Generate label y t The probability of trans (y t ,y t-1 ) is the transfer score, indicating the label y t-1 Transfer to y t The probability of , y ′ is all possible label sequences, used for normalization calculation; S3-7: Relation classification in extraction layer (BiLSTM-Attention model): P(r|e1,e2)=Softmax(W r ·[ATT(e1);ATT(e2)]) (9) Where ATT(e)=∑ t∈span(e) α t h t , e1, e2 are two entities, r: the relationship type between entities, span(e) is the time step range of entity e in the text, W r : The parameter matrix of relation classification, used to map aggregated entity features to relation probabilities; S3-8: State prediction of Kalman filter in fusion layer: Among them F k is the state transfer matrix; The state vector at k-1 time steps, The prior state prediction for k time steps, F k : State transfer matrix, describing the change of state over time, P k-1 : The covariance matrix of k-1 time steps, representing the state uncertainty, The prior covariance matrix for k time steps, Q k : process noise covariance matrix, describing the model error; S3-9: Reasoning layer (SPARQL engine) normalization factor: S3-10: Scatter plot layout is one of the important ways for the system to realize knowledge graph visualization. It transforms abstract node relationships into intuitive two-dimensional spatial distribution through a carefully designed algorithm. S3-11: Guided layout is based on the principle of energy minimization and defines the energy function: E=S {i≠j} (k rep / ||x i -x j || α )+S {(i,j)∈E} k spr (||x i -x j ||-l ij ) 2 (12) Where α = 2 is the repulsive force attenuation factor, l ij is the ideal length of the edge, k rep is the repulsive force constant, which controls the repulsion strength between nodes, k spr is the gravitational constant, which controls the tensile strength of the edge, and E is the energy function, which balances the repulsion between nodes and the gravitational force of the edge to make the layout stable; S3-12: Node position iterative update formula: Where t is the number of iterations, is the coordinate of node i at the tth iteration, η: learning rate, which controls the moving step size of each iteration, The energy function is The gradient at , guides the node to move in the direction of lower energy, β is the damping coefficient, the momentum term is introduced to avoid oscillation and accelerate convergence; S3-13: Visual encoding uses multi-dimensional mapping: Where (x, y) is the node coordinate, calculated by the layout algorithm, deg(v) is the degree of node v, s max 、s min The maximum and minimum values ​​of the node size, type(v) is the node type, |T| is the total number of node types, used to map the type to the hue dimension of HSL color, σ type The shape is preset according to the node type.

5. The dynamic optimization and visualization decision-making system for realizing the coordination of port and shipping multimodal transport knowledge graph according to claim 4 is characterized in that: The port and shipping multimodal transport knowledge graph function also includes the following steps: S3-14: The hierarchical graph analyzes the type information of all nodes and constructs a tree-like hierarchy. The root node represents the entire graph, the first-level child nodes represent different node type categories, and the second-level child nodes represent specific node instances. For spatial allocation, the system uses an improved treemap algorithm to recursively divide the canvas space into multiple rectangular areas proportional to the number of nodes. The size of each rectangle corresponds to the sum of the importance of the nodes it contains. S3-15: Hierarchical space partitioning is based on the weighted treemap algorithm, which satisfies: Where v is the current node, parent(v) is the parent node of v, sib(v) is the set of sibling nodes of v, A(v) is the display area of ​​node v, which is proportional to its weight w(v), w(v) is the node weight, which usually indicates the importance of the node, ∑w(sib(v)) is the sum of the weights of the sibling nodes, which is used to distribute the area of ​​the parent node proportionally; S3-16: The goal of rectangle optimization is to minimize the aspect ratio deviation: where w i , h i is the width and height of the i-th rectangle; S3-17: Color coding: c(d)=RGB(r base ×(1-γ) d ,g base ×(1-γ) d ,b base ×(1-γ) d ) (17) Where d is the node level depth, r base 、g base b base is the base RGB color value, and γ is the attenuation factor (0<γ<1), which controls how dark the color becomes as the level deepens; S3-18: Label visibility: Where A(v) is the node display area, A min is the minimum area threshold for label visibility, l(v) is the length of the node label text, and k is a proportional constant that controls the allowable ratio of label length to area; S3-19: In the configuration of scatter plots, the system has established a complete visual encoding system, which converts data attributes into visual elements through carefully designed mapping rules. The position of nodes in the scatter plot is determined by the (x, y) coordinates calculated by the layout algorithm, and the size of the node is proportional to its importance. S3-20: Node sizes are calculated using a linear mapping function to ensure that important nodes have more prominent visual representation: Among them, importance is the node importance index (such as transportation volume, number of connections), is the value range of node size; S3-21: The hierarchical graph configuration uses a different visual strategy to highlight the hierarchical structure of the data. The system uses nested rectangles to represent nodes at different levels. The size of the rectangles is proportional to the number of child nodes they contain. Through a recursive space segmentation algorithm, the system ensures that each level has a reasonable display space, thereby enhancing the visualization effect. S3-22: The color coding scheme is associated with the depth of the hierarchy, using a linear gradient to make the high-level areas lighter in color. As the hierarchy deepens, the color gradually darkens, thus forming a clear visual hierarchy. Label rendering uses intelligent adaptive technology. The system calculates the available space of each rectangle in real time and decides to display the full label, abbreviated label, or only the icon based on the size of the space. When there is insufficient space, the system automatically enables the function of floating the full information to improve the readability of the information.

6. The dynamic optimization and visualization decision-making system for realizing the coordination of port and shipping multimodal transport knowledge graph according to claim 1 is characterized in that: The path visualization function includes the following steps: S4-1: The System Overview module provides a comprehensive overview interface, helping users monitor and analyze the overall system status in real time. The statistics above display the number of current tasks, allowing users to quickly understand the distribution of tasks. The chart section uses pie charts and line graphs to intuitively display the status distribution and time trends of tasks, making it easier for users to identify potential problems and adjust strategies in a timely manner. The "Latest Tasks" list displays information about recently created tasks, allowing users to quickly view and process the latest work items. S4-2: The Freight Order Management module displays and manages task information in the system. Users can view detailed task data through an intuitive table interface, including key attributes such as order number, cargo type, and empty container. This module provides a task overview, allowing users to quickly monitor task status. The table clearly lists detailed information for each task and supports operations such as editing, deleting, and viewing task details, improving the user experience. S4-3: Freight Map Visualization builds a multi-dimensional knowledge network to intuitively display the path selection and association relationships during the freight transportation process. Based on advanced visualization algorithms, the system transforms complex transportation networks into interactive graphical displays, allowing users to clearly identify key information such as the topological connections between different transportation routes, transit nodes, and path attributes; S4-4: The path status analysis module allows users to enter a starting node and a destination node to calculate the optimal path between them. After the user selects a node in the input box at the top of the interface, the system calculates and displays the path between the two in real time. In the path display area, users can see the specific nodes and connection relationships of the path, intuitively understanding the changes and structure of the path; S4-5: The path state optimization module focuses on maximum path selection. The system compares path lengths based on user settings, updates the path display in real time, and ultimately selects the shortest calculated path. S4-6: The path optimization analysis module provides in-depth path analysis. Users can select a specific starting point and end point. The system will display the path connecting these two nodes and its related attributes. The system selects the optimal path based on length, required time, risk assessment, etc.

Images