Harbor district navigation capability assessment method and system based on digital twinning

Abstract

The invention provides a harbor district navigation capability assessment method and system based on digital twinning, an adaptive modeling framework is constructed by fusing unmanned aerial vehicle oblique photography, laser point cloud data and a BIM parameterized component library, harbor facility details are scanned through laser point cloud, an unmanned aerial vehicle obtains global terrain and building layout, and the harbor district navigation capability assessment method and system based on digital twinning are provided. The two types of data are input into a pre-trained port scene recognition model, standardized modules in a BIM parameterized component library are automatically matched to generate an initial three-dimensional model, and the initial three-dimensional model reserves a parameterized adjustment interface of attributes; the method comprises the following steps: establishing a physical and virtual dynamic calibration and standard adaptation dual mechanism, deploying a multi-dimensional sensor network at an edge node to collect physical entity operation data in real time, eliminating data noise through a Kalman filtering algorithm, comparing cleaned data with a virtual model simulation result, dynamically correcting model physical parameters, and carrying out real-time dynamic calibration. And synchronously formulating a port digital twin modeling uniform interface protocol compatible with a multi-system data format, and defining a model collaborative interaction rule.

Description

Technical Field

[0001] This invention relates to the field of port capacity assessment technology, and more specifically, to a method and system for assessing port navigation capacity based on digital twins. Background Technology

[0002] The current capacity assessment of port areas faces the core challenge of balancing modeling accuracy and comprehensive coverage. Traditional modeling methods often rely on single data sources or manual modeling. The integration of laser point clouds, UAV imagery, and BIM component libraries lacks a unified standard, leading to issues such as incompatible data formats and feature extraction biases, resulting in insufficient accuracy of the initial model. Furthermore, the model's flexibility in adjusting attributes is poor, making it difficult to adapt to the complex and diverse facility types and layout changes within the port area. Simultaneously, the lack of an effective dynamic calibration mechanism between physical entities and virtual models, coupled with significant noise interference in sensor data, results in substantial discrepancies between simulation results and actual operational data. In addition, inconsistent data interaction protocols between different systems create barriers to model collaboration, severely impacting the reliability of the assessment results.

[0003] In terms of computing architecture and optimization mechanisms, traditional port assessment systems often adopt a single computing power deployment model, which makes it difficult to balance the needs of real-time response and high-precision simulation. In core operational scenarios, the latency is too high, while high-precision analysis faces the dilemma of insufficient computing power. In addition, the models lack cross-scale hierarchical adaptation capabilities, cannot dynamically switch accuracy levels according to assessment needs, and have not established a full lifecycle optimization mechanism. It is difficult to continuously iterate model performance through operational data feedback. When faced with scenarios such as port facility upgrades and operational process adjustments, the model has poor adaptability, and the cost of repeated modeling is high, which cannot meet the long-term and dynamic capacity assessment needs of the port area. Summary of the Invention

[0004] In view of the aforementioned problems, and in conjunction with the first aspect of the present invention, embodiments of the present invention provide a method for assessing the navigation capacity of a port area based on digital twins, the method comprising: An adaptive modeling framework is constructed by integrating UAV oblique photography, laser point cloud data and BIM parametric component library. The framework uses laser point cloud scanning to scan port facility details and UAV to acquire global terrain and building layout. The two types of data are input into a pre-trained port scene recognition model, which automatically matches the standardized modules in the BIM parametric component library to generate an initial 3D model. The initial 3D model has reserved parameter adjustment interfaces for its attributes. Establish a dual mechanism for physical and virtual dynamic calibration and standard adaptation. Deploy a multi-dimensional sensor network at the edge node to collect physical entity operation data in real time. Use the Kalman filter algorithm to remove data noise. Compare the cleaned data with the simulation results of the virtual model to dynamically correct the physical parameters of the model. Simultaneously formulate a unified interface protocol for port digital twin modeling that is compatible with multiple system data formats and define the rules for model collaborative interaction. The port area model is divided into a micro-level equipment component layer, a meso-level operation unit layer, and a macro-level global system layer by adopting cross-scale hierarchical heterogeneous modeling and intelligent linkage technology. The micro-level uses a high-fidelity finite element model stored in the cloud and is only called when high-precision analysis scenarios are needed. The meso-level uses a lightweight model that retains operation characteristics and is deployed on edge nodes. The macro-level uses a topology model that focuses on global indicators. The intelligent linkage algorithm automatically switches the level precision according to the evaluation requirements. We construct a lightweight model and edge / cloud collaborative computing architecture, adopt an attention-based model adaptive simplification algorithm, dynamically adjust the accuracy level according to the scenario priority, edge nodes are responsible for real-time sensor data reception, meso- and macro model simulation and simple decision calculation, and utilize low latency characteristics to ensure rapid response, while the cloud focuses on high-fidelity simulation of micro models, historical data mining and model training optimization, and transmits parameters and model update fragments through edge and cloud data incremental synchronization mechanism. By introducing reinforcement learning-driven full lifecycle closed-loop optimization, the simulation results of the virtual model and the port operation data are used as feedback signals to input the reinforcement learning model, continuously optimizing the physical law mapping and statistical prediction algorithm of the model, identifying changes in port facilities through computer vision and automatically triggering model structure updates, establishing a model accuracy evaluation index system, regularly and automatically verifying model performance, and applying historical modeling experience to new scenarios through transfer learning. An adaptive computing power scheduling algorithm is constructed to dynamically allocate computing resources based on the port area's operational load. When massive data processing is triggered in complex scenarios, collaborative computing of edge node clusters is initiated. Simulation tasks are decomposed into parallel sub-tasks and allocated through a load balancing algorithm. A model caching mechanism is adopted to store frequently called models, reducing redundant calculations and data transmission. While ensuring high accuracy in core scenarios, simulation latency is controlled within the threshold required for real-time evaluation.

[0005] Furthermore, embodiments of the present invention also provide a port area capability assessment system based on digital twins, characterized in that it includes: A processor; a machine-readable storage medium for storing machine-executable instructions of the processor; wherein the processor is configured to execute the aforementioned digital twin-based port navigation capability assessment method by executing the machine-executable instructions.

[0006] In another aspect, embodiments of the present invention also provide a computer program product, the computer program product including machine-executable instructions, the machine-executable instructions being stored in a computer-readable storage medium, a processor of a computer device reading the machine-executable instructions from the computer-readable storage medium, the processor executing the machine-executable instructions, causing the computer device to execute the aforementioned port navigation capability assessment method based on digital twins.

[0007] Based on the above, the accuracy and adaptability of the port area's digital twin model were significantly improved through multi-source data fusion modeling, physical virtual dynamic calibration, and cross-scale hierarchical linkage technology. Collaborative modeling using laser point clouds, UAV imagery, and BIM component libraries, combined with intelligent recognition and positioning of pre-trained models, enabled automated and high-precision generation of a full-area 3D model of the port area. Reserved parameterized interfaces ensured flexible model adjustment capabilities. The physical virtual dynamic calibration mechanism, through Kalman filtering for noise reduction and closed-loop correction, ensured a high degree of consistency between the virtual model and the physical entity's operating state. A unified interface protocol broke down data barriers between multiple systems, enabling smooth model collaboration and interaction. Cross-scale hierarchical heterogeneous modeling and intelligent linkage technology allowed for on-demand switching between micro, meso, and macro levels of precision, meeting the high-precision analysis requirements at the component level while ensuring rapid response in core scenarios through lightweight models, thus solving the pain point of traditional modeling where accuracy and efficiency were difficult to balance.

[0008] Leveraging an edge-cloud collaborative computing architecture, reinforcement learning optimization, and scenario-adaptive computing power scheduling, the efficiency, real-time performance, and sustainability of port area capability assessment have been significantly improved. The attention-based model adaptive simplification algorithm achieves dynamic matching between model accuracy and edge node computing power. Edge nodes handle real-time data processing and lightweight simulation, while the cloud focuses on high-fidelity analysis and model training. An incremental synchronization mechanism reduces data transmission losses. Reinforcement learning-driven full lifecycle optimization continuously iterates the model algorithm through operational data feedback, combined with computer vision to automatically identify facility changes, ensuring the model remains adaptable to port area operational needs in the long term. The scenario-adaptive computing power scheduling algorithm, through task decomposition, load balancing, and model caching, achieves efficient resource allocation in complex scenarios, keeping simulation latency within real-time assessment thresholds. This provides scientific and efficient decision support for port area operation optimization, resource scheduling, and risk prediction. Attached Figure Description

[0009] Figure 1 This is a schematic diagram of the execution flow of the port navigation capacity assessment method based on digital twin provided in this embodiment of the invention.

[0010] Figure 2 This is a schematic diagram of exemplary hardware and software components of the port area capability assessment system based on digital twin provided in an embodiment of the present invention. Detailed Implementation

[0011] The present invention will now be described in detail with reference to the accompanying drawings. Figure 1 This is a flowchart illustrating a port navigation capacity assessment method based on digital twins provided in one embodiment of the present invention. The following is a detailed description of this port navigation capacity assessment method based on digital twins.

[0012] Step S110: Integrate UAV oblique photography, laser point cloud data and BIM parametric component library to build an adaptive modeling framework. Use laser point cloud to scan port facility details and UAV to acquire global terrain and building layout. Input the two types of data into the pre-trained port scene recognition model and automatically match the standardized modules in the BIM parametric component library to generate an initial 3D model. The initial 3D model reserves a parametric adjustment interface for the attributes. The system employs a collaborative approach, utilizing ground-based laser scanning equipment deployed at key observation points of port facilities to cover detailed areas. The drones fly along pre-defined routes to acquire comprehensive topographical and building layout images, achieving full data coverage of both local details and the overall layout. Both types of data are transmitted to a processing terminal for initial analysis and preprocessing. The preprocessed fused data is standardized according to the input format of a pre-trained model. A pre-trained port scene recognition model is then used to extract facility features, which are compared with a built-in feature library to automatically identify the facility type. Based on the identification results, the BIM parametric component library's feature index is invoked. Standardized modules are obtained through similarity matching, and parameters are adjusted according to actual dimensions. An initial 3D model is automatically assembled based on spatial relationships, while attribute parameter adjustment interfaces are reserved on the model components to ensure flexible subsequent modifications.

[0013] Step S111: Analyze the port area modeling requirements, clarify the fusion rules of laser point cloud, UAV oblique photography data and BIM parametric component library, define data interaction standards and format conversion protocols, build BIM parametric component library, classify and construct standardized modules according to port facility types, preset editable attribute parameters for each module, and establish a module feature index library of associated facility identification features. The port area modeling requirements were analyzed, clarifying core demands such as facility accuracy, attribute dimensions, and application scenarios. Based on this, fusion rules for laser point clouds, UAV imagery, and the BIM component library were formulated, including data priority determination, conflict handling, and feature association mapping rules. Simultaneously, interaction interface protocols and format conversion standards for the three types of data were defined to ensure smooth data exchange. A BIM parametric component library was built, categorizing port facility types such as loading and unloading equipment, warehousing buildings, and waterway facilities into standardized modules. Editable attribute parameters, including geometric, physical, and operational parameters, were preset for each module. A module feature index library was established, using the morphological, structural, and functional characteristics of port facilities as index items to associate corresponding standardized modules, enabling rapid matching of features and modules and providing support for subsequent automated modeling.

[0014] Step S112: Use ground laser scanning equipment to scan the port facilities in all directions to collect data, plan the drone oblique photography route and set shooting parameters to obtain terrain and building layout image data covering the entire port area, and record shooting auxiliary information simultaneously. Ground-based laser scanning equipment was used to comprehensively collect detailed images of port facilities. Scanning stations were planned according to facility type and size, ensuring overlapping scan areas between adjacent stations to meet subsequent registration requirements. A drone oblique photography flight path was planned, employing a multi-view coverage design to ensure complete capture of terrain and building layout. Reasonable flight altitude, speed, and overlap rate were set, and the drone was launched to perform oblique photography along the flight path, acquiring image data of the entire port area. Key auxiliary information was recorded simultaneously during the acquisition process: the laser scanner recorded station information and equipment operating parameters, while the drone recorded shooting time, flight attitude, geographical location, and weather conditions. This provided a foundation for subsequent data preprocessing and accuracy control, ensuring the usability and relevance of the collected data.

[0015] Step S113: Denoise and register the laser point cloud data, preprocess the UAV oblique photogrammetry image and generate dense point cloud and digital orthophoto map, extract terrain and building information, and perform coordinate transformation and fusion of the processed laser point cloud data and UAV related data to form a three-dimensional data volume in a unified format. Statistical filtering algorithms were used to denoise the laser point cloud data, removing environmental noise points. Then, a feature-based registration method was employed, utilizing overlapping area features from multiple stations to complete point cloud registration, forming a complete detailed point cloud of the facilities. Distortion correction and image enhancement preprocessing were performed on the UAV oblique photogrammetric images. Dense point clouds and digital orthophoto maps were generated through image matching, extracting terrain elevation and building outline information. Based on the port area's unified coordinate system, coordinate transformation was performed on the two types of processed data to ensure they were in the same coordinate system. A point cloud fusion algorithm was then used to remove redundant points and retain valid feature points, forming a unified format 3D data volume. Simultaneously, the extracted terrain and building information was associated with the data volume, enriching the information dimensions to support subsequent recognition and modeling.

[0016] Step S114: After processing the pre-processed 3D data volume according to the preset format, input it into the pre-trained port scene recognition model, extract facility features through the model and compare them with the built-in feature library to complete facility type identification and positioning. Step S1141: Parse the original format of the preprocessed 3D data volume, convert it into a unified format according to the input requirements of the pre-trained model, retain information and remove redundant attributes, resample the data to regularize the dataset to a preset dimension to ensure the input dimension is unified, and normalize the resampled data to eliminate feature extraction bias caused by differences in the size of different facilities. The data format parsing module extracts core information such as geometric coordinates and feature points from the 3D data volume, converting it to a unified format according to the input requirements of the pre-trained port scene recognition model, and removing redundant attributes irrelevant to feature extraction. A resampling algorithm is then used to normalize the converted data into a pre-defined dimensional dataset, ensuring consistent data dimensions for the input model. Normalization is performed on the resampled data to eliminate feature extraction biases caused by differences in the dimensions of different port facilities, ensuring the model accurately captures the inherent features of the facilities. The entire process focuses on the core objective of data standardization, using three core operations—format conversion, normalization, and standardization—to provide compliant input data for subsequent model inference.

[0017] Step S1142: Perform lightweight augmentation operations on the standardized point cloud data to improve the robustness of the model to port facility attitude changes and environmental disturbances. If the pre-trained model is a two-dimensional convolutional architecture, convert the three-dimensional point cloud data into a multi-view two-dimensional depth map, retain the depth information and coordinate mapping relationship of each view, and ensure that it can be back-mapped to three-dimensional space in the future. Lightweight augmentation operations such as random rotation, translation, and scaling are performed on the standardized point cloud data to improve the model's robustness to changes in facility posture and environmental disturbances. During the augmentation process, core data features are strictly preserved to avoid distortion. If the pre-trained model is a 2D convolutional architecture, a multi-view projection algorithm is used to convert the 3D point cloud into a multi-view 2D depth map. The mapping relationship between depth information and 3D coordinates from each viewpoint is recorded simultaneously to ensure that the 2D recognition results can be reverse-mapped to 3D space, guaranteeing positioning accuracy. Through two core operations—data augmentation and dimensional adaptation—the input data and model architecture are adapted, improving the applicability and accuracy of model inference.

[0018] Step S1143: Load the deep learning model pre-trained based on the port scene dataset, import the pre-trained weight file, fix the parameters of the first half of the feature extraction network and activate the inference mode of the output layer and feature matching layer, initialize the built-in feature library of the model, which is a set of standard feature vectors of various port facilities, classified and stored according to facility type and an index structure is built to improve the efficiency of feature comparison. Load a pre-trained deep learning model based on a port scenario dataset into the data processing terminal and import the pre-trained weight file. Fix the parameters of the first half of the feature extraction network through the model configuration module, activating only the inference mode of the output layer and feature matching layer to reduce inference computation. Initialize the model's built-in feature library, a classification set of standard feature vectors for various port facilities, establishing a storage directory by facility type and constructing a hash index structure to improve feature comparison and retrieval efficiency. After completing model loading configuration and feature library initialization, the model is in a ready state, capable of receiving standardized input data and performing inference.

[0019] Step S1144: Input the formatted input data into the pre-trained model, and perform spatial alignment, local feature extraction, and global feature fusion through the feature extraction network in sequence to generate a global feature vector that can represent the overall structure and morphological features of the facility. Then, normalize the global feature vector to eliminate scale differences. The formatted input data is fed into the pre-trained model. The feature extraction network first spatially aligns the data to ensure the consistency of features for similar facilities in different locations. Then, a local feature extraction layer captures detailed features of the facilities, and a global feature fusion layer integrates these local features into a global feature vector representing the overall structure and morphology of the facilities. Finally, the global feature vector is normalized to eliminate the impact of scale differences on feature comparison, ensuring the comparability of features for similar facilities of different sizes. Through this coherent process of spatial alignment, local extraction, global fusion, and normalization, a high-quality global feature vector is generated, providing a reliable foundation for subsequent feature matching.

[0020] Step S1145: Input the global feature vector into the feature matching module of the model, retrieve similar feature templates in the built-in feature library through the index structure, calculate the similarity between the feature to be matched and each template feature, set a similarity threshold to filter candidate feature templates, count the number of facility categories corresponding to the candidate templates to determine the preliminary identification results, if there are multiple categories with similarity, extract the local detail features of the area to be identified for secondary comparison to determine the final facility type; The global feature vector is input into the model's feature matching module, which quickly retrieves similar feature templates from the built-in feature library using a hash index structure. A cosine similarity algorithm is used to calculate the similarity between the feature to be matched and each template. A threshold is set to filter candidate templates, and the number of facility categories corresponding to each candidate template is counted. The category with the most similarities is used as the initial identification result. If multiple categories have similarities close to a preset range, local detailed features of the area to be identified are extracted for a secondary comparison. The final facility type is determined through fine-tuning. The entire process, through a hierarchical matching logic of "fast retrieval - preliminary judgment - secondary verification," improves the accuracy and reliability of facility type identification.

[0021] Step S1146: Based on the feature point heatmap output by the model, locate the key feature points of the facility, obtain their coordinates in the model input coordinate system, transform the feature point coordinates to the unified coordinate system of the entire port area through a predefined coordinate mapping matrix, fit the boundary box of the facility according to the feature point coordinates, clarify the precise position and spatial attitude of the facility in the three-dimensional space of the port area, and record the positioning reliability of the facility. Based on the feature point heatmap output from the model, key feature points of the facility are analyzed to obtain their coordinates in the model input coordinate system. Using a predefined coordinate mapping matrix, the feature point coordinates are transformed to a unified coordinate system covering the entire port area. A bounding box fitting algorithm is employed to fit the facility's bounding box based on the feature point coordinates, clarifying the facility's precise position and orientation in 3D space. Simultaneously, the positioning confidence level is calculated and recorded. Through the process of feature point extraction, coordinate transformation, bounding box fitting, and confidence level recording, precise facility positioning is achieved, providing a spatial location basis for subsequent model assembly.

[0022] Step S1147: Combine the prior knowledge of the port area scene to perform consistency verification on the preliminary identification and positioning results, eliminate results that conflict with the logic of the surrounding environment, mark facilities with positioning confidence below the preset threshold as suspicious areas, perform feature extraction and comparison again by supplementing data, and finally output a structured identification and positioning result containing facility type, precise coordinates, bounding box parameters, and identification confidence, and associate the result with the three-dimensional data volume.

[0023] By combining prior knowledge of the port area scenario (such as facility functional layout logic and spatial location constraints), the initial identification and positioning results are validated for consistency, eliminating results that conflict with the surrounding environment (such as loading and unloading equipment identified as storage facilities located within the waterway). Facilities with a positioning confidence level below a preset threshold are marked as suspicious areas, and supplementary laser point cloud or UAV imagery data for these areas are collected, followed by re-performing feature extraction and comparison. The final output is a structured result containing facility type, precise coordinates, bounding box parameters, and identification confidence level, and an association index is established with the 3D data volume to achieve a precise correspondence between the identification results and the original data.

[0024] Step S115: Based on the scene recognition results, call the feature index library of the BIM parametric component library, automatically match the standardized modules through the similarity matching algorithm, and adjust the parameters of the matching modules adaptively in combination with the size information in the three-dimensional data volume. Then, automatically assemble each module according to the spatial position relationship of the facilities to generate an initial three-dimensional model covering the entire port area. Based on the structured identification and positioning results output in step S114, the feature index library of the BIM parametric component library is invoked. A similarity matching algorithm compares the facility features with the feature items in the index library, matching the corresponding standardized modules from the component library. According to the actual size information of the facility in the 3D data volume, the attribute parameters of the matching modules are adaptively adjusted to ensure that the module's geometric dimensions are consistent with the actual facility. Based on the spatial relationship of the facility within the port area, the standardized modules are automatically assembled using a module assembly algorithm. During assembly, the spatial connection relationships between modules are ensured to conform to the actual layout logic, ultimately generating an initial 3D model covering the entire port area, achieving automated and precise assembly of the facilities.

[0025] Step S116: Using parametric programming technology, reserve attribute parameter adjustment interfaces on the initial 3D model components. The interfaces are associated with preset attribute parameters in the BIM component library to verify the accuracy of the initial model. If the deviation exceeds the threshold, the module parameters are fine-tuned in reverse. Finally, a flexibly adjustable initial 3D model of the port area is output.

[0026] Parametric programming technology is employed, reserving attribute parameter adjustment interfaces for each component of the initial 3D model. These interfaces are linked to preset geometric and physical attribute parameters in the BIM component library, allowing direct modification of component parameters. A precision verification module compares the geometric features and spatial positions of the initial model with the preprocessed fused data to verify model accuracy. If the deviation exceeds a preset threshold, the corresponding module parameters are fine-tuned via the parametric interface, and this process is repeated until the accuracy meets the standards. The final output is a flexibly adjustable initial 3D model of the port area, ensuring model accuracy while reserving room for future parameter optimization.

[0027] Step S120: Establish a dual mechanism for physical and virtual dynamic calibration and standard adaptation. Deploy a multi-dimensional sensor network at the edge node to collect physical entity operation data in real time. Use the Kalman filter algorithm to remove data noise. Compare the cleaned data with the simulation results of the virtual model to dynamically correct the physical parameters of the model. Simultaneously formulate a unified interface protocol for port digital twin modeling that is compatible with multiple system data formats and define the rules for model collaborative interaction. A multi-dimensional sensor network is deployed within the coverage area of ​​edge nodes to achieve real-time acquisition of operational data of physical entities in the port area. The collected raw data is input into the Kalman filter algorithm module deployed on the edge nodes. Random noise is removed through iterative algorithm calculations to obtain cleaned and effective data. A data comparison engine is built to correlate and match the cleaned data with the simulation results of the virtual model, calculate the deviation between the two, and if the deviation exceeds a threshold, trigger the virtual model physical parameter correction process to form a dynamic calibration closed loop. At the same time, all systems involved in the port area's digital twin are comprehensively reviewed, the differences in data formats and transmission protocols are analyzed, a unified interface protocol compatible with multiple systems is formulated, and the rules for model collaborative interaction are clarified to achieve the collaborative implementation of the two mechanisms and data interoperability.

[0028] Step S121: Divide the monitoring area according to the distribution of physical entities in the port area, deploy multi-dimensional sensing devices within the coverage area of ​​the edge nodes, determine the installation points of each sensor to fully reflect the operating status of the entities, configure the sensor acquisition parameters, adopt a hybrid transmission mode to transmit the collected physical entity operating data to the local data buffer of the edge nodes in real time, and set up a data transmission verification mechanism. Monitoring areas are divided based on the distribution characteristics of physical entities within the port area to ensure full coverage and no redundancy or overlap. Within the signal coverage area of ​​edge nodes, the installation locations of each sensor are determined based on the monitoring needs of entity operation status, ensuring comprehensive capture of key operational parameters. Core parameters such as sensor acquisition frequency and accuracy are configured, and a hybrid wired + wireless transmission mode is adopted to transmit the collected physical entity operation data to the local data buffer of the edge nodes in real time, avoiding data transmission delays or loss. A data transmission verification mechanism is set up to verify data integrity through checksum comparison. If data anomalies are detected, a retransmission process is triggered to ensure the reliability and availability of the collected data.

[0029] Step S122: Deploy the Kalman filter algorithm module at the edge node, initialize the filter parameters and set the corresponding equation according to the data type of the sensor, input the original sensor data into the filter module, remove random noise through prediction and update iteration process, output the optimal estimate, perform secondary cleaning on the filtered data, detect and remove outliers, supplement missing data, and then perform standardization processing to form a physical entity operation dataset. A Kalman filter algorithm module is deployed at the edge nodes. Filtering parameters are initialized based on the type of sensor data to be processed (e.g., vibration, temperature, operating speed), and corresponding state and observation equations are constructed. The raw sensor data is input into the filtering module, and noise is filtered through a "prediction-update" iterative process, outputting the optimal estimate of the data. A second cleaning process is performed on the filtered data stream, using an outlier detection algorithm to identify and remove data deviating from the normal range, and interpolation to supplement missing data. Finally, the cleaned data is standardized to unify the data volume and format, forming a structured physical entity operation dataset, providing standardized input for subsequent data comparison.

[0030] Step S123: Construct a data comparison engine, set a comparison cycle linked to the data acquisition frequency, associate and match the cleaned physical entity operation data with the corresponding simulation data output by the virtual model according to the preset mapping relationship, use the deviation calculation model to calculate the degree of difference between the two types of data, set the deviation threshold, trigger the model parameter correction process when the deviation exceeds the threshold, and record the comparison information to form a comparison log. A data comparison engine is constructed, and a linked comparison cycle is set according to the data collection frequency to ensure the timeliness and synchronization of data comparison. Based on the preset entity-model mapping relationship, the cleaned physical entity operation data is accurately associated and matched with the corresponding simulation data output by the virtual model to ensure the consistency of the comparison objects of the two types of data. A deviation calculation model is used to calculate the degree of difference between the matched data pairs, and differentiated deviation thresholds are set to adapt to different types of monitoring data. When the calculated deviation value exceeds the corresponding threshold, the virtual model parameter correction process is automatically triggered; at the same time, information such as comparison time, comparison object, and deviation value are recorded in real time to form a complete comparison log, providing a basis for subsequent traceability and optimization.

[0031] Step S124: Start the parameter correction engine based on the deviation calculation results, locate the physical parameters of the virtual model corresponding to the deviation, establish a mapping model between the deviation and the parameter correction amount, calculate the correction amount according to the mapping model, iteratively correct the physical parameters of the virtual model according to the preset correction step size, drive the virtual model to re-simulate and compare again until the deviation is lower than the threshold, forming a closed-loop calibration process, and record the parameter correction information at the same time. Based on the deviation results obtained from data comparison, the parameter correction engine is activated to locate the physical parameters of the virtual model associated with the deviation (such as mechanical coefficients, motion resistance parameters, etc.) through deviation source tracing. A mapping model between deviation values ​​and parameter correction amounts is established, and the precise parameter correction amounts are calculated based on this model. The corresponding physical parameters of the virtual model are iteratively adjusted according to a preset correction step size to avoid model instability caused by sudden parameter changes. After the parameter correction is completed, the virtual model is driven to re-execute the simulation, and the new simulation results are input into the data comparison engine again for comparison with the physical data until the deviation value is lower than a preset threshold, forming a closed-loop calibration process of "comparison-correction-re-comparison". At the same time, detailed information such as the values ​​before and after parameter correction and correction time is recorded.

[0032] Step S125: Organize the various systems involved in the digital twin of the port area, collect data source information, data format, transmission protocol and interaction requirements of each system through interface debugging and document parsing, analyze the characteristics and differences of data of each system, summarize compatibility issues, clarify the requirements that the unified interface protocol needs to cover, and form a protocol requirement specification. A comprehensive review was conducted of all related systems involved in the port area's digital twin system, including equipment operation and maintenance systems, job scheduling systems, and sensor monitoring systems. Interface debugging tools were used to connect with each system, and system technical documentation was analyzed to collect core information such as data source types, data format specifications, transmission protocol types, and interaction requirements. Comparative analysis of the collected data characteristics from multiple systems identified compatibility issues such as data format differences and inconsistent field naming. Based on the analysis results, the compatibility, security, and real-time requirements of the unified interface protocol were clarified, and a structured protocol requirements specification was compiled to provide a clear basis for subsequent protocol design.

[0033] Step S126: Based on the requirements analysis results, design a unified interface protocol framework, adopt a layered architecture with the application layer as the core, define data encoding rules, interface call specifications and message formats, unify data field naming rules and data types, formulate protocol communication specifications and support synchronous / asynchronous transmission modes, design adaptation interfaces for each system, clarify interface parameters, calling methods and error code definitions, and add data encryption transmission and identity authentication mechanisms to ensure transmission security. A unified interface protocol framework was designed based on the protocol requirements specification, adopting a layered architecture with the application layer at its core to ensure the protocol's scalability and adaptability. Unified data encoding rules, interface call specifications, and message formats were defined, and data field naming rules and data types were standardized to achieve unified data format across multiple systems. Communication specifications for the protocol were formulated, supporting both synchronous and asynchronous transmission modes to adapt to different interaction scenarios. Dedicated adaptation interfaces were designed for each related system, clearly defining the interface's input and output parameters, calling methods, and error code definitions to ensure the standardization and traceability of interface calls. Data encryption and authentication mechanisms were incorporated into the protocol, ensuring data transmission security through key verification and access control.

[0034] Step S127: Define model collaboration and interaction rules based on modeling requirements, including data interaction timing rules, data sharing permission rules and conflict resolution rules, formulate multi-model collaboration and linkage process, clarify data interaction triggering conditions, build a verification environment to access simulated data from various systems, test protocol compatibility and rule effectiveness, and optimize protocols and rules based on test results; Based on the digital twin modeling and operation needs of the port area, collaborative interaction rules for models were defined, including the timing requirements for data interaction, the division of data sharing permissions between different modules, and rules for resolving multi-source data conflicts (such as priority arbitration rules). A multi-model collaborative linkage process was developed, clarifying the triggering conditions for data interaction between models (such as triggering by changes in work status, or timed triggering). A simulation verification environment was built, and simulation test data from various related systems was connected to test the compatibility of the unified interface protocol and verify the effectiveness of the collaborative interaction rules. Based on protocol adaptation issues and rule conflict issues discovered during testing, the details of the interface protocol and the collaborative interaction rules were iteratively optimized to ensure smooth collaboration among multiple systems and models.

[0035] Step S128: Integrate the dynamic calibration mechanism and the standard adaptation mechanism into the unified management platform of the edge nodes to realize the closed loop of data flow. Deploy a monitoring module on the platform to monitor the operation status of the two mechanisms in real time and trigger alarms when abnormalities occur. Regularly collect operation data and analyze indicators. Iteratively optimize filtering parameters, deviation thresholds, protocol details and collaboration rules in combination with changes in port operation needs.

[0036] The physical-virtual dynamic calibration mechanism and the standard adaptation mechanism are integrated into a unified management platform for edge nodes, establishing a complete data flow loop encompassing data acquisition, processing, comparison, correction, and interaction. A monitoring module is deployed and runs on the management platform to monitor the operational status of both mechanisms in real time, including data transmission link connectivity, algorithm module status, and protocol interaction success rate. When anomalies are detected, alarms are automatically triggered and pushed to the operations and maintenance terminal. Platform operational data is collected regularly to analyze key indicators such as data processing efficiency, calibration accuracy, and protocol adaptation rate. Based on the dynamic changes in port area operational needs, Kalman filter parameters, deviation thresholds, interface protocol details, and collaborative interaction rules are iteratively optimized to ensure the dual mechanisms continuously adapt to actual operational scenarios.

[0037] Step S130: Using cross-scale hierarchical heterogeneous modeling and intelligent linkage technology, the port area model is divided into a micro-level equipment component layer, a meso-level operation unit layer, and a macro-level global system layer. The micro-level layer uses a high-fidelity finite element model stored in the cloud and is only called when high-precision analysis scenarios are needed. The meso-level layer uses a lightweight model that retains operation characteristics and is deployed on edge nodes. The macro-level layer uses a topology model that focuses on global indicators. The intelligent linkage algorithm automatically switches the level precision according to the evaluation requirements. Based on the hierarchical characteristics of the port area, the digital twin model is divided into a micro-level equipment component layer, a meso-level operational unit layer, and a macro-level global system layer. The micro-level layer uses finite element analysis software to construct a high-fidelity model, which is stored in the cloud and only accessed in high-precision analysis scenarios. The meso-level layer, based on the micro-level model, uses lightweight techniques to remove details while retaining operational characteristics, and is deployed at edge nodes. The macro-level layer employs a topological abstraction method, abstracting meso-level units into nodes and relationships into links to construct a topological model. An intelligent linkage algorithm and linkage decision engine are built, defining a layer-switching trigger condition matrix to automatically switch layer precision based on evaluation needs. Standardized linkage interfaces and data interaction protocols are designed, integrating the three-layer model, linkage algorithm, and interfaces into a unified management platform. Through simulation testing and pilot application iterations and optimizations, accurate layer switching and smooth data synchronization are ensured.

[0038] Step S131: Define the definition criteria and objectives of the micro-level equipment component layer, the meso-level operation unit layer, and the macro-level global system layer; sort out the relationship between the layers; establish entity and parameter mapping tables between the layers; and clarify the elements and accuracy requirements for modeling at each layer. Based on the operational and assessment needs of the port area, the defining criteria for the micro-level equipment component layer, the meso-level operational unit layer, and the macro-level overall system layer are clearly defined. For example, the hierarchical boundaries are divided according to entity granularity and analytical precision requirements. Simultaneously, the core objectives of each layer are determined: the micro-level focuses on component performance analysis, the meso-level focuses on operational process simulation, and the macro-level focuses on overall operational monitoring. The system systematically analyzes the relationships between each layer, establishing entity mapping tables and parameter mapping tables, clarifying the correspondence between entities and the rules for transferring core parameters between different layers. Based on the objectives and relationships of each layer, the core elements that must be covered in the modeling of each layer are determined, and differentiated precision requirements are formulated, providing a clear basis for subsequent layered modeling.

[0039] Step S132: The micro-equipment component layer uses finite element analysis software to construct a high-fidelity finite element model based on the component's three-dimensional geometric data and design parameters. Key parts are refined and non-critical structures are appropriately simplified. The meso-level operation unit layer uses lightweight model technology to remove micro-details and retain operation features based on the outer contour features and operation process of the micro-level model. The macro-level global system layer uses topological abstraction method to abstract meso-level operation units into topological nodes and the relationships between operation units into topological links. Node and link-type topological models are constructed and parameters and indicators are associated. Based on the 3D geometric data and design parameters of equipment components, a high-fidelity finite element model is constructed using finite element analysis software. Detailed modeling is performed on key components such as bearings and gears, while non-critical structures are appropriately simplified to balance accuracy and computational efficiency. Meso-level operational unit modeling: The outer contour features of the micro-level model are extracted. Combined with key nodes in the operational process, lightweight modeling techniques are used to remove component-level micro-details, retaining core operational features such as operational trajectories and collaborative relationships, generating a lightweight model. Macro-level system-wide modeling: Using topological abstraction methods, meso-level operational units (such as quay crane operational units and yard units) are abstracted into topological nodes. The material transfer and signal interaction relationships between units are abstracted into topological links, constructing a node-link topological model. Operational indicators, resource utilization, and other parameters are associated with corresponding nodes and links.

[0040] Step S133: Build a cloud and edge node collaborative storage architecture, classify and store the micro-level high-fidelity finite element model in the cloud distributed database and establish a model index library, and call it only when high-precision analysis scenarios. Deploy the meso-level lightweight model according to the job unit type on the local storage of the corresponding edge node, and use a redundant backup mechanism to ensure real-time calling. Deploy the macro-level topology model on both the cloud and edge nodes. A collaborative storage architecture between the cloud and edge nodes is constructed, employing a layered storage strategy of "high-precision storage in the cloud and real-time access at the edge." High-fidelity finite element models at the microscopic level are categorized by device type and stored in a distributed cloud database. A model index is established for rapid retrieval, and these models are only accessed during high-precision analysis scenarios such as component fault analysis. Lightweight models at the mesoscopic level are deployed according to work unit type (e.g., loading / unloading units, scheduling units) in the local storage modules of corresponding edge nodes, employing a redundant backup mechanism to prevent data loss and ensure real-time access. Macroscopic topology models are deployed simultaneously in the cloud and edge nodes. The cloud version is used for comprehensive analysis, while the edge node version is used for rapid response to local monitoring needs, achieving collaborative adaptation between the cloud and edge.

[0041] Step S134: Construct a linkage decision engine, define a matrix of hierarchical switching trigger conditions including assessment requirement type, data accuracy requirements, response latency requirements, and abnormal event types, construct an intelligent linkage algorithm based on rule engine and machine learning algorithm, process explicit trigger conditions and optimize hierarchical switching strategy, and design a smooth transition mechanism for hierarchical switching. Build a linkage decision-making engine, sort out key factors such as evaluation requirement types, data accuracy requirements, response latency requirements, and abnormal event types, establish a hierarchical switching trigger condition matrix, and clarify the specific conditions for triggering hierarchical switching in different scenarios. Build an intelligent linkage algorithm based on a rule engine and machine learning algorithms. The rule engine processes clear trigger conditions (such as high-precision analysis requirements triggering the invocation of the micro layer), and the machine learning algorithm optimizes the hierarchical switching strategy to improve the rationality and efficiency of switching. Design a smooth transition mechanism for hierarchical switching to maintain the continuity of data transmission and the consistency of model states during the hierarchical switching process, and avoid simulation interruptions or data breaks during the switching process.

[0042] Step S135: Design a standardized linkage interface and corresponding invocation method for the three-layer model, adapt to hierarchical switching requests with different real-time requirements, define a unified data interaction protocol, clarify the data transmission format and data fields, establish a verification mechanism and a fault tolerance mechanism for data interaction, and deploy a data gateway at the edge node; For the linkage requirements of the three-layer model, design a standardized linkage interface, clarify the invocation method of the interface, and adapt to hierarchical switching requests with different real-time requirements. For example, synchronous invocation is used for high-real-time requests, and asynchronous invocation is used for non-real-time requests. Define a unified data interaction protocol, standardize data encoding rules, transmission formats, and core data fields to ensure data interoperability between different hierarchical models. Establish a verification mechanism for data interaction to ensure the validity of transmitted data through data integrity verification and format verification; set up a fault tolerance mechanism to automatically trigger a retry or degradation strategy when data transmission is abnormal. Deploy a data gateway at the edge node to forward, filter, and adapt the interaction data between layers to ensure the security and efficiency of data transmission.

[0043] Step S136: Integrate the three-layer heterogeneous model, intelligent linkage algorithm, and linkage interface into the unified management platform of the port digital twin, build a simulation test environment, simulate different evaluation requirement scenarios to test the triggering accuracy, response latency, and data synchronization accuracy of hierarchical switching, collect test metrics and analyze the optimization direction, and conduct pilot applications in the actual port scenario, and iteratively optimize the hierarchical division standard, model accuracy, and linkage logic in combination with real operation data.

[0044] The three-layer heterogeneous model, intelligent linkage algorithm, and standardized linkage interface are integrated into the port area's unified digital twin management platform to achieve collaborative linkage between modules. A simulated test environment is built to recreate typical port area assessment scenarios (such as daily monitoring, fault analysis, and full-area scheduling), testing the trigger accuracy, response latency, and data synchronization precision of level switching. Key indicators are collected during the testing process, and optimization directions such as the rationality of level division and the effectiveness of linkage logic are analyzed. Representative pilot areas are selected for on-site deployment, connecting to the on-site sensor network and operating system to obtain real operational data. Based on pilot application data and on-site feedback, the level division standards, model accuracy parameters, and linkage trigger logic are iteratively optimized to improve the system's adaptability to real-world scenarios.

[0045] Step S1361: Analyze the technical architecture characteristics of the three-layer heterogeneous model, intelligent linkage algorithm, and linkage interface; clarify the input and output parameters and operating environment of each module; encapsulate each module based on the microservice architecture, split it into independent microservices; define the communication specifications and registration mechanism of each microservice; build a unified management platform framework for the port area digital twin; integrate the data access layer, model management layer, algorithm engine layer, interface gateway layer, and visualization display layer; and clarify the functions of each layer. This paper analyzes the technical architecture characteristics of the three-layer heterogeneous model, intelligent linkage algorithm, and linkage interface, clarifying the input and output parameters, runtime dependencies, and resource requirements of each module. Based on a microservice architecture, each module is encapsulated, separating functions such as micro-model management, meso-model simulation, macro-model analysis, and linkage decision-making into independent microservices, defining the communication specifications and service registration mechanisms for each microservice. A unified management platform framework for the port area's digital twin is built, integrating the data access layer, model management layer, algorithm engine layer, interface gateway layer, and visualization layer. The core functions of each layer are clearly defined; for example, the data access layer is responsible for multi-source data aggregation, and the model management layer is responsible for unified model scheduling and lifecycle management, ensuring the scalability and maintainability of the platform architecture.

[0046] Step S1362: Deploy the encapsulated three-layer heterogeneous model microservices according to the corresponding nodes, establish model indexes and call links through the model management layer to achieve unified scheduling, integrate the intelligent linkage algorithm into the algorithm engine layer, configure the interaction links between the algorithm and other layers, integrate the standardized linkage interface into the interface gateway layer, configure interface routing rules and access control policies, conduct module joint debugging tests, verify the synergy of model calls, algorithm decisions, and interface communication, and fix various synergy problems that occur during the joint debugging process; The encapsulated three-tiered heterogeneous model microservices are deployed to the cloud and edge nodes according to storage deployment requirements. A unified model index and call chain are established through the model management layer to achieve unified scheduling of models at different levels. The intelligent linkage algorithm is integrated into the platform's algorithm engine layer, and the interaction links between the algorithm and the model management layer and the data access layer are configured to ensure that the algorithm can obtain scene requirements and data status in real time and output level switching instructions. The standardized linkage interface is integrated into the interface gateway layer, and interface routing rules and permission control policies are configured to achieve unified management and control of interface calls. Module integration testing is carried out to verify the responsiveness of model calls, the accuracy of algorithm decisions, and the stability of interface communication. Collaboration issues (such as data transmission latency and instruction execution deviations) discovered during integration testing are repaired and optimized.

[0047] Step S1363: Build a cloud-edge collaborative hardware architecture consistent with the real port area, construct a hybrid transmission network to simulate the real data transmission environment, build the corresponding software environment for cloud and edge nodes, install testing tools and log collection system, construct test datasets based on real port area operation data, generate simulated data covering different evaluation scenarios, and label the expected results and core requirements corresponding to the data. A cloud-edge collaborative hardware architecture consistent with that of a real port area was constructed, deploying corresponding servers, edge node devices, and network equipment to build a hybrid wired + wireless transmission network to simulate a real data transmission environment. Corresponding operating systems, database software, simulation engines, and testing tools were installed on both the cloud and edge nodes, and a log collection system was deployed to record all operational data during the testing process. Based on operational data from the real port area, test datasets covering different evaluation scenarios were constructed, generating simulated data including normal operation, fault anomalies, and peak load scenarios. The expected results and core test requirements for each test data point were labeled, providing standardized input for subsequent testing.

[0048] Step S1364: Identify typical assessment needs scenarios in the port area and clarify the testing objectives for each scenario. Design test cases for each scenario and clarify the test steps, input data and expected outputs. We identified typical assessment scenarios for the port area, including daily operation monitoring, equipment fault diagnosis, operational efficiency optimization, and global resource scheduling. We clarified the core testing objectives for each scenario; for example, in the fault diagnosis scenario, the focus was on testing the accuracy and response speed of micro-level calls. Detailed test cases were designed for each scenario, specifying the test steps, input scenario data, core functions to be verified, and expected output results. For instance, in the global resource scheduling scenario, the input resource distribution data was specified, the test step was triggering a switch from the macro to the meso layer, and the expected output was a successful switch with no data synchronization deviation. This resulted in a complete test case set covering the testing requirements for both independent operation in a single scenario and concurrent operation in multiple scenarios.

[0049] Step S1365: Execute single-scenario independent tests and multi-scenario concurrent tests in the order of test cases. During the test, record the response time of level switching and data synchronization deviation through the test tool, record the trigger results and the running status of each module through the log collection system, mark test abnormal scenarios and record relevant information. Following the order of the test case set, single-scenario independent testing and multi-scenario concurrent testing were conducted sequentially. During the testing process, professional testing tools were used to record key indicators such as response time and data synchronization deviation during level switching in real time; a log collection system was used to record information such as trigger condition matching results, the running status of each module, and interface call logs. For abnormal scenarios such as false triggering of level switching, data synchronization failure, and excessive response latency that occurred during the testing process, detailed information such as the time of occurrence, triggering conditions, and module running status was recorded to provide complete data support for subsequent problem analysis.

[0050] Step S1366: Organize test data and statistically analyze test indicators, set qualified thresholds for indicators, compare test results with thresholds to filter out substandard indicators and corresponding scenarios, analyze the reasons for substandard indicators, identify optimization directions, and formulate targeted optimization plans. We organized and analyzed the collected metrics and log information during the testing process, focusing on core test metrics such as level switching accuracy, response latency, and data synchronization precision. Based on preset acceptable thresholds, we identified non-compliant metrics and their corresponding test scenarios, such as level switching response latency exceeding the threshold in a certain scenario. For the non-compliant metrics, we conducted in-depth analysis of the root causes, such as excessive response latency potentially stemming from excessively long decision-making time in the linkage algorithm or low interface communication efficiency. We then pinpointed specific optimization directions and developed targeted optimization solutions, such as optimizing algorithm logic and improving interface transmission efficiency.

[0051] Step S1367: Carry out optimization work according to the optimization plan. After the optimization is completed, reuse the original test cases to carry out regression testing to verify whether the indicators meet the standards and the impact of the optimization plan on other modules. Repeat the optimization and regression testing process until the indicators meet the requirements. Based on the optimization plan, targeted optimization work is carried out. For example, to address the issue of excessively long algorithm decision-making time, the feature extraction logic and decision rules of the linkage algorithm are optimized; to address the issue of interface communication efficiency, the data transmission protocol is optimized and the amount of transmitted data is compressed. After optimization, the original test cases are reused to conduct regression testing to verify whether the optimized indicators have reached the qualified thresholds, and to check whether the optimization plan has a negative impact on the functionality of other modules. If there are still unmet indicators or new problems in the regression testing, the optimization and regression testing process is repeated until all core test indicators meet the requirements.

[0052] Step S1368: Select a representative pilot area that covers the entities corresponding to the three-layer model and has a complete operation process, conduct on-site research on the pilot area, adjust the platform parameters and interface configuration to adapt to the actual environment, deploy a unified management platform in the pilot area, connect to the real sensor network and operation system on site, complete the cloud and edge node configuration and build a visual monitoring interface. Representative pilot areas covering the micro, meso, and macro levels of the model, along with their corresponding entities and complete operational processes, were selected for on-site investigation. This involved analyzing the equipment distribution, operational processes, and sensor deployment in these areas. Based on the investigation results, the parameter configuration and interface adaptation rules of the unified management platform were adjusted to ensure the platform's compatibility with the actual environment of the pilot areas. The unified management platform was then deployed in the pilot areas, connecting to the on-site sensor networks, PLC systems, and TOS systems. Network and permission configurations for cloud and edge nodes were completed, and a visual monitoring interface was built to achieve unified display and control of the three-layer model in the pilot areas.

[0053] Step S1369: Start the trial operation of the pilot area platform, monitor the platform's operating status in real time and record the actual operating data, collect feedback from on-site staff on the platform, and record the actual problems that occur during the trial operation; The pilot area platform was launched for trial operation. The platform monitoring module monitored the three-layer model's operational status, layer switching, and data transmission link connectivity in real time, synchronously recording actual operational data, including layer switching frequency, response latency, and resource utilization of each module. A feedback collection mechanism was established to gather feedback from on-site maintenance and management personnel regarding the platform's ease of use, functional completeness, and data accuracy. Detailed records were kept of actual problems encountered during the trial operation, such as model accuracy deviations, unreasonable linkage logic, and interface communication anomalies.

[0054] Step S13610: Comprehensively analyze the trial operation data and feedback, evaluate the platform's adaptability and practicality, optimize the platform based on the problems found during the trial operation and the actual situation on site, and verify it again in the pilot area after optimization.

[0055] By comprehensively analyzing pilot operation data and on-site feedback, the platform's adaptability and practicality to actual port area scenarios were evaluated, and optimization directions for aspects such as hierarchical division, model accuracy, and linkage logic were identified. For issues discovered during the pilot operation, optimization plans were adjusted based on actual on-site operational needs. For example, the rules for retaining operational features in the mid-level model were optimized according to actual work processes, and the triggering conditions for hierarchical switching were optimized based on maintenance feedback. After optimization, a second pilot operation was launched in the pilot area to verify whether the issues were resolved and to ensure that the platform's performance and functionality met actual operational requirements.

[0056] Step S140: Construct a lightweight model and edge / cloud collaborative computing architecture. Adopt an attention-based model adaptive simplification algorithm and dynamically adjust the accuracy level according to the scene priority. Edge nodes are responsible for real-time sensor data reception, meso- and macro-level model simulation and simple decision calculation. Low latency characteristics ensure fast response. The cloud focuses on high-fidelity simulation of micro-level models, historical data mining and model training optimization. Parameters and model update fragments are transmitted through edge and cloud data incremental synchronization mechanism. A lightweight model and edge / cloud collaborative computing architecture are constructed, specifically implemented as follows: An attention-based adaptive simplification algorithm framework is built. This framework extracts core features of the port area scene and calculates attention weights, dynamically adjusting the model's accuracy level based on scene priorities. The functional boundaries between edge nodes and the cloud are clearly defined. Edge nodes deploy real-time data receiving modules responsible for meso- and macro-level model simulation and simple decision calculations, leveraging low latency to ensure rapid response in operational scenarios. The cloud deploys high-performance computing modules focusing on high-fidelity micro-level model simulation, historical data mining, and model training optimization. An edge-cloud incremental data synchronization mechanism is designed, transmitting only model update fragments and core parameters to reduce data transmission volume. After integrating the algorithm and architecture, the adaptability of model accuracy adjustments under different scenarios is verified. The data synchronization strategy between the edge and cloud is optimized to ensure that the lightweight model adapts to the computing power of edge nodes while meeting accuracy requirements.

[0057] Step S141: Review the requirements for digital twin modeling and evaluation in the port area, clarify the functional boundaries of edge and cloud collaboration, divide the priority levels of scenarios and match the corresponding model accuracy strategies, design the overall framework of edge real-time processing and cloud deep computing collaborative architecture, and clarify the data flow and module interaction relationships. To address the digital twin modeling and evaluation needs of the port area, a combined survey approach of interviews and questionnaires was employed to collect feedback from relevant stakeholders regarding modeling accuracy, evaluation scenarios, real-time response, and computing power constraints. A requirements list was compiled and prioritized, resulting in a specification document. The requirements were broken down into quantifiable indicators. Based on the real-time nature and computing power requirements of these indicators, the task scope was defined: edge nodes were responsible for real-time processing and local caching, while the cloud was responsible for high-computing power and non-real-time processing. The points of collaboration between the two were clarified. Based on the operational characteristics of the port area, three levels of scenario priority were established, and differentiated model accuracy strategies were developed for different priorities, establishing a mapping relationship between scenarios, accuracy, and simplification strategies. A four-level collaborative architecture—perception layer, edge layer, cloud layer, and application layer—was constructed. The modules, functions, and deployment requirements of each layer were clarified, data and interaction links were analyzed, and rules for task allocation and data synchronization were defined. Finally, the architecture's adaptability was verified against the requirements specification document and adjusted to fully meet the requirements.

[0058] Step S1411: Use a combined survey approach to obtain the demands of all relevant parties in the port area, clarify the accuracy and consistency requirements at the modeling level and the scenario requirements at the evaluation level, and simultaneously collect real-time response, computing power, bandwidth, and compatibility constraints, sort them into a list of requirements and complete the priority ranking, and output the requirement specification document. A combined research approach, integrating interviews, on-site surveys, and document analysis, was employed to engage with relevant stakeholders in port operations, maintenance, and technology to obtain accuracy and consistency requirements at the modeling level, as well as scenario requirements at the evaluation level. Constraints such as real-time response latency, computing resource limits, network bandwidth, and system compatibility were simultaneously collected. These requirements and constraints were categorized and organized to form a structured requirements list. Prioritizing the list according to the principle of "core requirements first, key constraints first," the requirements hierarchy was clearly defined, clarifying mandatory, expected, and optional requirements. The final output was a requirements specification document containing requirement descriptions, priorities, constraints, and acceptance criteria, providing a basis for subsequent architecture design.

[0059] Step S1412: Decompose the sorted requirements into specific indicators for modeling and evaluation, and clarify the quantitative standards and definitions of each indicator; The streamlined modeling and evaluation requirements were broken down into quantifiable metrics, covering core dimensions such as model accuracy, simulation response speed, data synchronization efficiency, and computing power utilization. For each metric, clear quantitative standards and definitions were established. For example, model accuracy was defined as the range of geometric / physical parameter deviations between the model and the physical entity; simulation response speed was defined as the time interval from receiving the evaluation request to outputting the result; and computing power utilization was defined as the upper limit of the proportion of edge / cloud computing resources used during model runtime. A metric dictionary was established, labeling the dimension, calculation method, and associated scenarios of each metric, ensuring that all requirements were transformed into verifiable and implementable quantifiable metrics, providing a quantitative basis for subsequent edge-cloud function partitioning.

[0060] Step S1413: Based on the real-time and computing power requirements of the demand indicators, define the task scope of edge nodes and the cloud, clarify the real-time processing and local caching tasks of edge nodes, the high computing power and non-real-time processing tasks of the cloud, and determine the collaborative connection points and data interaction content between the two. Based on the real-time nature and computing power requirements of the demand indicators, the task boundaries between edge nodes and the cloud are defined: edge nodes handle low-latency, low-computing-power tasks such as real-time sensor data reception, parsing, and local caching; meso / macro model simulation; and simple job decision calculations; while the cloud handles non-real-time, high-computing-power tasks such as high-fidelity micro-model simulation, massive historical data mining, and model training and optimization. The collaborative connection points between the two are clearly defined, such as edge nodes uploading abnormal data and computing load warnings to the cloud, and the cloud sending model update packages and optimized decision rules to edge nodes. An interactive data list is compiled, clarifying data types, transmission trigger conditions, and format requirements to ensure seamless integration between edge and cloud tasks.

[0061] Step S1414: Combining the characteristics of port operations and the priority of needs, the scenarios are divided into three priority levels, and the coverage and focus of each level of scenarios are clarified. Based on the characteristics and priority of port operations, scenarios are divided into three priority levels: Level 1 priority covers core operational scenarios such as ship berthing and container loading / unloading, focusing on key operational equipment and real-time operational processes; Level 2 priority covers routine operational scenarios such as yard storage management and auxiliary facility monitoring, focusing on the overall operational status of operational units; Level 3 priority covers non-core scenarios such as idle area inspections and routine equipment maintenance, focusing on the overall resource utilization status. The coverage scope of each level of scenario is clearly defined, and the criteria for scenario switching are defined. For example, an alarm from operational equipment triggers a Level 1 scenario, and routine inspections trigger a Level 3 scenario, thus forming a scenario priority classification standard.

[0062] Step S1415: Develop differentiated model accuracy strategies for different priority scenarios, establish a mapping relationship between scenario priority and model accuracy and simplification strategies, and clarify the computational requirements for each scenario model. A differentiated model accuracy strategy was developed for the three priority levels of scenarios: Level 1 scenarios adopted high-precision models, retaining the detailed features and complete physical parameters of core equipment; Level 2 scenarios adopted standard-precision models, eliminating non-critical details and retaining the core features of the operational process; Level 3 scenarios adopted basic-precision models, retaining only the macroscopic topology and the correlation between key indicators. A mapping table was established between scenario priority, model accuracy, and simplification strategies, clarifying the computational requirements for the number of triangles, parameter dimensions, and computational step size in each scenario, ensuring that the model complexity matches the importance of the scenario and the computing power of edge nodes, avoiding wasted computing power or insufficient accuracy.

[0063] Step S1416: Construct a four-level collaborative architecture consisting of a perception layer, edge layer, cloud layer, and application layer using a layered approach, and clarify the constituent modules, functions, and hardware and software deployment requirements of each layer; A four-tiered collaborative architecture is constructed using a layered approach: The perception layer deploys multi-dimensional sensors and data acquisition terminals, responsible for physical entity data collection, with hardware adapted to the complex port environment and software supporting multi-protocol data access; the edge layer deploys edge servers and gateway devices, responsible for real-time data processing and lightweight model operation, equipped with a lightweight operating system and real-time computing framework; the cloud layer deploys a distributed server cluster, responsible for high-fidelity simulation and data mining, configured with large-capacity storage and high-performance computing components; and the application layer deploys a visualization platform and evaluation terminals, responsible for demand distribution and result display. The hardware and software deployment requirements for each layer are clearly defined, such as edge nodes needing to meet industrial-grade protection and millisecond-level response, and the cloud needing to meet distributed storage and elastic computing power expansion requirements.

[0064] Step S1417: Analyze the data links between each level and module of the architecture, clarify the transmission content and method of each data flow, and form a closed-loop data link; This document outlines the data links between different layers and modules of the architecture. It clarifies the transmission of raw sensor data from the perception layer to the edge layer, incremental and anomaly data from the edge layer to the cloud layer, model update packages and optimization parameters from the cloud layer to the edge layer, and simulation results and evaluation reports from the edge / cloud layer to the application layer. It defines the transmission methods for each data flow, such as using MQTT for real-time data, HTTP for batch data, and fragmented transmission for large files. A closed-loop data flow diagram is drawn, marking all nodes in the data generation, transmission, processing, and feedback process to ensure uninterrupted data links and efficient and controllable data interaction at each layer.

[0065] Step S1418: Clarify the interaction links between modules within the architecture, define rules for task allocation, data synchronization, fault switching, and permission management, and write module interaction specifications; Clearly define the interaction links between modules within the architecture and define core interaction rules: task allocation rules follow the principle of "proximity processing and computing power adaptation"; data synchronization rules distinguish between timed synchronization and event-triggered synchronization; fault switching rules clarify the cloud takeover process in case of edge node failure; and permission management rules divide module operation permissions for different roles. Analyze module interaction scenarios, including normal operation, model updates, and emergency fault handling, and write module interaction manuals that detail the module call order, input / output parameters, and exception handling methods in each scenario to ensure that inter-module interactions are regulated and reduce the probability of collaborative failures.

[0066] Step S1419: Verify the functional coverage and indicator satisfaction of the collaborative architecture against the requirements specification, and adjust the architecture design for adaptability issues until the architecture fully adapts to the requirements.

[0067] Against the requirements specification, verify the functional coverage and metric fulfillment of the collaborative architecture item by item. For example, verify whether the edge layer meets the millisecond-level response time for core scenarios and whether the cloud layer meets the high-fidelity simulation computing power requirements. For compatibility issues discovered during verification, such as insufficient edge node computing power or excessive data synchronization latency in a certain scenario, adjust the architecture design, such as optimizing edge node computing power configuration and adjusting data synchronization triggering conditions. Repeat the verification-adjustment process until all functional modules of the architecture cover the requirements list and all quantitative metrics meet the preset standards, forming the final edge-cloud collaborative architecture design scheme.

[0068] Step S142: Construct the algorithm framework to determine input and output elements, design a scene feature attention extraction module, calculate feature attention weights, develop dynamic precision adjustment logic, adopt a differentiated simplification strategy based on scene priority and attention weights, and embed a model complexity monitoring module. An adaptive simplification algorithm framework was constructed, with the algorithm goal of adapting to edge computing power and achieving scenario-based accuracy adjustment. The framework was divided into seven modules, including input parsing, feature attention extraction, and weight calculation, with defined module boundaries and interaction specifications. Input elements were identified as model data and scene priorities, while output elements included lightweight model files. Data parsing and encapsulation specifications were established. A port area scene feature system was constructed, a feature extraction network was built and trained and optimized, a multi-dimensional weight calculation model was designed, and weight parameters were dynamically calibrated. A precision adjustment rule base was established, a decision engine was developed and combined with edge computing power to fine-tune precision, differentiated simplification strategies were formulated, and a complexity monitoring module was embedded to form a closed loop. After module integration, testing was conducted, and parameters were optimized until performance met the standards.

[0069] Step S1421: Define the algorithm goal as achieving scene-adaptive precision adjustment and lightweighting of the port area digital twin model, adapting to the computing power capacity of edge nodes, and building a modular algorithm framework based on logic, which is divided into an input parsing module, a scene feature attention extraction module, a weight calculation module, a dynamic precision adjustment module, a differentiated simplification execution module, a complexity monitoring and feedback module, and an output encapsulation module. Define the functional boundaries and data interaction specifications of each module, and clarify the data flow form, transmission protocol and triggering conditions between modules. The core objective of the algorithm is clearly defined as achieving scene-adaptive precision adjustment and lightweighting of the port area's digital twin model, adapting to the computing power capabilities of edge nodes. Based on this objective, a modular algorithm framework is built, divided into seven modules: input parsing, scene feature attention extraction, weight calculation, dynamic precision adjustment, differentiated simplification execution, complexity monitoring and feedback, and output encapsulation. The functional boundaries of each module are defined; for example, the input parsing module is responsible for data format conversion, and the attention extraction module is responsible for core feature recognition. The data interaction specifications between modules are clearly defined, including data flow formats, transmission protocols, and triggering conditions. For example, complexity monitoring results trigger the dynamic precision adjustment module to recalculate, ensuring the algorithm framework is modular and scalable.

[0070] Step S1422: Input elements include original 3D model data, scene priority labels, real-time evaluation requirement parameters, and edge node computing power status data. Define parsing rules, coding specifications, and transmission interfaces for various input data. Output elements include adaptive simplified lightweight model files, model accuracy configuration reports, and complexity monitoring result reports. Define the encapsulation format of output data and the interface specifications for downstream modules. The algorithm's input elements are clearly defined, including raw 3D model data, scene priority labels, real-time evaluation requirements parameters, and edge node computing power status data. Parsing rules are established for various input data types, such as a unified model data format and standardized scene priority encoding. Input data transmission interfaces are designed to support both local edge node reading and cloud-based delivery. The output elements are defined as an adaptively simplified lightweight model file, a model accuracy configuration report, and a complexity monitoring result report. Output data encapsulation formats are defined, such as a lightweight format for model files and a structured JSON format for reports. Interface specifications for downstream modules (model management, simulation engine) are designed to ensure that output data can be directly used by downstream modules.

[0071] Step S1423: Construct a port area scene feature system, classify and sort geometric features, operational features, and physical features, establish a feature classification dictionary and attribute description system, build a feature extraction network architecture, develop a feature post-processing module, structure and organize the extracted feature vectors to generate feature maps, and train and optimize the network through labeled datasets to ensure that the feature recognition and positioning accuracy meets the standards. A port area scene feature system was constructed, classifying and organizing geometric features (equipment outlines, structural dimensions), operational features (operational trajectories, collaborative relationships), and physical features (mechanical parameters, operating status). A feature classification dictionary and attribute description system were established. A deep learning-based feature extraction network architecture was built, including a feature input layer, convolutional layers, pooling layers, and an output layer. A feature post-processing module was developed to structure and generate feature maps from the extracted feature vectors. The network was trained using a labeled port area scene dataset. By adjusting network parameters and adding regularization layers to optimize the model, the feature recognition and localization accuracy were verified until a preset standard was reached, ensuring accurate extraction of core features under different scenarios.

[0072] Step S1424: Determine the inherent importance of features, scenario priority, and real-time evaluation requirements as the core dimensions for weight calculation, design a multi-dimensional weighted fusion weight calculation model, normalize the initial weights, develop a dynamic weight calibration mechanism, update weight parameters based on historical feedback data, and build a weight visualization module. The inherent importance of features, scenario priority, and real-time evaluation requirements are identified as the core dimensions for weight calculation. A multi-dimensional weighted fusion model is designed, and the initial weight proportion of each dimension is determined using the analytic hierarchy process (AHP). The calculated initial weights are then normalized to ensure that the sum of the weights is 1. A dynamic weight calibration mechanism is developed, which updates the weight parameters regularly based on the model simplification effect and evaluation feedback data from historical scenarios, improving the adaptability of the weights to real-world scenarios. A weight visualization module is built to display the weight values ​​of each dimension and the weight adjustment process in chart form, facilitating algorithm debugging and effect evaluation, and ensuring that the attention weights accurately reflect the importance of the core features of the scenario.

[0073] Step S1425: Construct a precision adjustment rule base based on scene priority and attention weight, establish a three-dimensional mapping relationship, define the technical parameters corresponding to each precision level, develop a dynamic precision adjustment decision engine, fine-tune the precision level in combination with the computing power status of edge nodes, write precision adjustment logic code and reserve rule update interface, and add a logic verification module. A precision adjustment rule base is constructed based on scene priority and attention weight, establishing a three-dimensional mapping relationship of "scene priority - attention weight - precision level." Technical parameters corresponding to each precision level are defined, such as high precision level retaining over 90% of core features and basic precision level retaining 60% of core features. A dynamic precision adjustment decision engine is developed. The engine receives scene priority, attention weight, and edge node computing power status data, outputs precision level adjustment instructions, and fine-tunes the precision level based on computing power status to avoid edge node overload. Precision adjustment logic code is written, reserving rule update interfaces for subsequent iterations, and a logic verification module is added to verify the rationality of the adjustment instructions and prevent precision adjustment anomalies.

[0074] Step S1426: For different scenarios, priorities, attention weights and accuracy levels, formulate a differentiated simplification strategy system, develop a simplified algorithm component library, build a simplified execution scheduling module, call the corresponding algorithm component to perform simplification operations according to the accuracy adjustment instruction and record the simplification log; A differentiated simplification strategy system is developed for different scenarios with varying priorities, attention weights, and accuracy levels. For example, in Level 1 scenarios, key physical parameters are retained, while in Level 3 scenarios, detailed geometric features are removed. A simplification algorithm component library is developed, including components for geometric simplification, parameter dimensionality reduction, and mesh optimization. A simplification execution scheduling module is constructed to automatically match and call the corresponding component to perform simplification operations based on accuracy adjustment instructions. Simplification logs are recorded during the simplification process, including information such as simplification strategy type, parameter adjustment values, and changes in model complexity. This facilitates tracking the simplification process, analyzing its effects, and ensuring that model simplification operations in different scenarios are standardized and traceable.

[0075] Step S1427: Define the model complexity evaluation index system, clarify the calculation method and collection frequency of each index, develop the complexity monitoring data collection module, set the index thresholds under different scenarios, design the monitoring and feedback closed-loop logic, link the monitoring module with the dynamic accuracy adjustment module, and develop the complexity monitoring visualization module. Define a model complexity evaluation index system, including indicators such as the number of triangular faces, parameter dimensions, computation time, and computing power utilization. Clarify the calculation methods and collection frequency for each indicator, such as collecting computing power utilization every 10 seconds. Develop a complexity monitoring data acquisition module to collect complexity index data in real time during model operation, setting thresholds for indicators under different scenarios, such as an 80% threshold for edge node computing power utilization. Design a monitoring and feedback closed-loop logic; when an indicator exceeds a threshold, trigger a dynamic precision adjustment module to recalculate the precision level. Develop a complexity monitoring visualization module to display real-time trend changes in indicators, facilitating algorithm optimization and operational monitoring.

[0076] Step S1428: Complete the integration and connection of each module and open up the data interface, carry out single module functional testing and end-to-end full-process joint debugging testing, verify the module function and the stability of the whole process operation, and optimize the module parameters and interaction logic. Complete the integration and docking of all algorithm modules, establish data interfaces between modules, and ensure smooth data flow in input parsing, feature extraction, and precision adjustment. Conduct single-module functional testing to verify whether the functionality of each module meets the design requirements, such as testing whether the attention extraction module can accurately identify core features. Conduct end-to-end full-process integration testing to simulate different input scenarios and verify the stability and accuracy of the algorithm throughout the entire process. For issues such as module interaction stuttering and precision adjustment deviations discovered during testing, optimize module parameters and interaction logic until all algorithm modules function normally and the entire process runs stably.

[0077] Step S1429: Construct a test dataset covering all scenario types in the port area, conduct batch verification of algorithm performance, statistically analyze performance indicators, analyze the root causes of unsatisfactory test cases, optimize algorithm parameters, iterate and verify until all performance indicators meet the requirements, and generate an algorithm performance verification report.

[0078] A test dataset covering all scenarios within the port area was constructed, including model data and scenario parameters for core operations, routine operations, and non-core operations. Based on this dataset, batch verification of algorithm performance was conducted, statistically analyzing performance metrics such as model simplification efficiency, accuracy retention rate, and edge computing power utilization. For test cases that failed to meet the standards, the root causes were analyzed, such as weight calculation deviations or unreasonable simplification strategies, and algorithm parameters were optimized, such as adjusting the weight dimension ratio and updating simplification rules. The batch verification and optimization process was repeated until all performance metrics met the preset requirements, generating an algorithm performance verification report that clearly defines the algorithm's performance and applicability in different scenarios.

[0079] Step S143: For the meso- and macro-level models, perform lightweight processing based on the adaptive simplification algorithm to generate multi-precision level model versions and establish a precision and scene priority mapping table. Retain core features and remove redundant details according to the hierarchical characteristics. Perform performance testing on the lightweight model and backtrack to optimize the simplification parameters. For meso- and macro-level models, lightweight processing is performed based on an adaptive simplification algorithm: meso-level models retain core operational features of work units while removing component-level micro-details; macro-level models retain the overall topology and key indicator relationships while removing non-core operational unit details. Multi-precision model versions are generated, corresponding to different scenario priorities, and a precision-scenario priority mapping table is established. Performance testing is conducted on the lightweight models to verify whether the model's running efficiency and accuracy deviation meet the requirements. If the test results are unsatisfactory, the simplification algorithm parameters, such as the feature retention ratio and mesh simplification threshold, are adjusted backtrackingly until a balance is achieved between the model's lightweighting effect and accuracy requirements.

[0080] Step S144: Select suitable edge node hardware and configure relevant components, deploy a lightweight operating system and the software framework and middleware required for real-time processing, configure the communication link between the edge node and the sensor, enable the local data caching mechanism, deploy the model management module, preload the corresponding accuracy model and establish a fast call index; Select edge node hardware suitable for the port area's industrial environment, including edge servers and industrial gateways, and configure them with high-temperature resistant and interference-resistant protective components. Deploy a lightweight operating system, install software components such as a real-time computing framework and message middleware, configure communication links between edge nodes and sensors, support multi-protocol data access, and enable a local data caching mechanism to cache recent sensor data and model files. Deploy a model management module, preload lightweight mesoscopic and macroscopic models of different accuracy levels, establish a fast model retrieval index, and associate the index with scene priority to ensure that edge nodes can quickly retrieve the corresponding accuracy model according to the scene, meeting real-time response requirements.

[0081] Step S145: Deploy the cloud server cluster and configure storage and computing components, install the operating system, distributed database, high-fidelity simulation software and machine learning framework, build the cloud model training and optimization module, deploy the cloud and edge collaborative management platform and clarify the overall coordination function; Deploy a cloud server cluster, configuring large-capacity distributed storage components and high-performance computing components to meet the needs of high-fidelity simulation and massive data processing. Install server operating systems, distributed databases, high-fidelity finite element simulation software, and machine learning frameworks to build a cloud-based model training and optimization module. This module includes sub-modules for data preprocessing, model training, and performance evaluation, supporting iterative optimization of digital twin models based on historical data. Deploy a cloud-edge collaborative management platform, clearly defining the platform's overall functions, including edge node computing power monitoring, model version management, task scheduling, and data synchronization control, to achieve unified management of the edge-cloud collaborative process.

[0082] Step S146: Define the task allocation rules between edge nodes and the cloud, design a dynamic scheduling engine, monitor the computing load and task queue status of edge nodes in real time, and formulate strategies for task migration and update instruction issuance. Define edge-cloud task allocation rules, adhering to the principle of "real-time tasks processed at the edge, high-computing-power tasks processed in the cloud," and clarify the task allocation logic under different scenarios and computing power states. Design a dynamic scheduling engine that monitors the computing power load and task queue length of edge nodes in real time. When the load of an edge node exceeds a threshold, a task migration strategy is triggered to migrate some non-core tasks to the cloud. When the cloud issues model update instructions, an instruction issuance strategy is formulated to prioritize the issuance of lightweight model update packages to high-load edge nodes. Develop a monitoring interface for the scheduling engine to visually display the task allocation status and node load, facilitating manual intervention and strategy optimization.

[0083] Step S147: Define the incremental synchronization data type between the edge and the cloud, design a synchronization triggering mechanism using a dual-mode approach of timed synchronization and event-triggered synchronization, process the synchronization data using data compression and differential coding technology, deploy synchronization verification and fault tolerance modules, and formulate data conflict handling principles. Define incremental synchronization data types between the edge and cloud, including model update fragments, core parameters, anomaly data, and evaluation results, avoiding full data transmission. Design a dual-mode synchronization system: scheduled synchronization transmits regular parameters at preset intervals, while event-triggered synchronization immediately transmits relevant data when model updates or job anomalies occur. Employ data compression and differential coding techniques to process synchronization data, reducing the amount of data transmitted and improving synchronization efficiency. Deploy synchronization verification and fault tolerance modules to verify the integrity and accuracy of synchronized data, triggering a retry mechanism when synchronization fails. Establish data conflict handling principles; for example, in the event of a conflict between local data on edge nodes and cloud data, the data with the latest timestamp will prevail.

[0084] Step S148: Integrate each module into a unified management platform, conduct joint debugging tests, verify the smoothness of business processes, cloud high-fidelity simulation processes and edge and cloud incremental synchronization processes, test response and synchronization performance under different scenarios, and fix various problems that occur during the joint debugging process; The lightweight model module, edge-cloud task scheduling module, and incremental synchronization module were integrated into the port area's unified digital twin management platform, establishing seamless data interfaces and interaction links between the modules. Joint testing was conducted to verify the smoothness of core business processes (scene recognition - model simplification - edge simulation), cloud-based high-fidelity simulation processes (data upload - model invocation - simulation calculation), and edge-cloud incremental synchronization processes. Response performance and data synchronization accuracy were tested under different scenarios, such as edge response latency in core scenarios and the success rate of model update package synchronization. Issues such as interface incompatibility, synchronization packet loss, and simulation stuttering encountered during joint testing were addressed by fixing module vulnerabilities and optimizing interaction logic to ensure smooth operation throughout the entire process.

[0085] Step S149: Deploy a collaborative architecture in the pilot area of ​​the port area, access real sensor data and operational scenarios to conduct trial operation, monitor the operational status of edge nodes, cloud and data synchronization in real time, collect trial operation data and feedback, analyze the balance between model accuracy and computing efficiency, and iteratively optimize algorithm strategies, computing power configuration and synchronization mechanism.

[0086] An edge-cloud collaborative computing architecture was deployed in the pilot area of ​​the port area. Hardware installation and software configuration of edge nodes and cloud servers were completed, and real-world sensor data and operational scenario data from the pilot area were integrated. Trial operation of the architecture was initiated, and operational metrics such as edge node computing load, cloud simulation status, and data synchronization success rate were monitored in real time through a unified management platform. Feedback from on-site staff on model accuracy and response speed was collected. Trial operation data was analyzed to evaluate the balance between model accuracy and computational efficiency. For issues such as synchronization latency and insufficient model accuracy, algorithm strategies were iteratively optimized, edge / cloud computing power configurations were adjusted, and incremental synchronization mechanisms were improved until the architecture adapted to the actual operational needs of the pilot area.

[0087] Step S150: Introduce reinforcement learning-driven full lifecycle closed-loop optimization, use virtual model simulation results and port operation data as feedback signals to input reinforcement learning model, continuously optimize the physical law mapping and statistical prediction algorithm of model, identify changes in port facilities through computer vision and automatically trigger model structure updates, establish a model accuracy evaluation index system, regularly and automatically verify model performance, and apply historical modeling experience to new scenarios through transfer learning. This process involves clarifying the objectives, defining the responsibilities and interactions of the participants in the virtual model and reinforcement learning model, constructing a closed-loop lifecycle, and specifying the triggering conditions and data flow requirements for each stage. It also includes building a reinforcement learning model architecture adapted to the port area scenario, selecting training algorithms, defining the state space and action space, and designing a reward function aimed at minimizing bias. The feedback signal sources are defined as simulation data and real operational data; extraction rules are established, and a preprocessing workflow is designed to generate a standardized feedback signal dataset. A full lifecycle data acquisition module is deployed, a distributed data integration platform is built, a data association mechanism is established, and data storage strategies are set. After model pre-training, an online training mechanism is designed, with input feedback signals and output optimization instructions. An execution module is built to adjust parameters, and step size control is set to ensure stability. After optimization, simulation verification is initiated, an evaluation system and thresholds are constructed, and deployment is performed if the thresholds are met; otherwise, a new round of optimization is triggered, establishing an application tracking mechanism to form a closed loop. A dynamic model adaptation mechanism is designed, a training monitoring module is established, operation and maintenance specifications are formulated, and full lifecycle optimization logs are recorded. An evaluation index system is constructed to regularly evaluate the optimization effect, specifically optimize model parameters and rules, expand the dataset to improve generalization ability, and achieve continuous iteration.

[0088] Step S160: Construct a scenario-adaptive computing power scheduling algorithm, dynamically allocate computing resources based on port area operation load, start edge node cluster collaborative computing when massive data processing is triggered in complex scenarios, decompose simulation tasks into parallel sub-tasks and allocate them through load balancing algorithm, adopt model caching mechanism to store high-frequency calling models, reduce redundant calculations and data transmission, and control simulation latency within the threshold required for real-time evaluation while ensuring high accuracy of core scenarios.

[0089] Through field research and historical data mining, typical operational scenarios in the port area were categorized, the computing power requirements of each scenario were clarified, computing power scheduling evaluation indicators were defined, and trigger thresholds for complex scenarios were set to determine whether to initiate edge cluster collaborative computing. A four-level algorithm architecture of perception, decision-making, execution, and feedback was built, clarifying the functions of each level, defining the interaction interfaces and data flow specifications between levels, and using lightweight message queues to achieve low-latency data transmission. Multi-dimensional data acquisition units were deployed to collect operational load, edge node status, and task attribute data in real time, dynamically adjusting the acquisition frequency, constructing a scenario recognition model based on machine learning algorithms, and developing a load prediction submodule based on time-series prediction algorithms. A multi-scenario computing power scheduling strategy library was formulated, matching differentiated scheduling strategies for different scenario types and load levels, designing a scheduling decision engine, integrating multi-source data, solving for the optimal scheduling scheme through multi-objective optimization algorithms, and constructing a complex scenario trigger mechanism. Decomposition rules based on task characteristics were designed. The algorithm is decomposed, outputting a list of subtasks containing key information and verifying parallelism. A subtask priority ranking mechanism is constructed to ensure that core subtasks are executed first. An improved load balancing algorithm adapted to the port area scenario is selected, and node resource performance weights are introduced to calculate node load values ​​and allocate subtasks. A load balancing adjustment mechanism is developed, and cluster load balancing is ensured through subtask migration. A computing power resource pool management module is constructed to integrate edge cluster resources. Dynamic resource allocation logic based on job load is designed to achieve accurate resource allocation and idle resource reclamation. A resource elastic scaling mechanism is deployed to adapt to changes in cluster load. A scheduling execution monitoring module is deployed to track task execution and resource usage status. A scheduling effect evaluation submodule is developed, and a feedback optimization mechanism is constructed to iteratively optimize scheduling parameters and models. All modules are integrated into a unified computing power scheduling system, relevant interfaces are opened, a full-scenario test dataset is constructed to conduct functional and performance tests, and algorithm parameters and interaction logic are optimized until the requirements are met.

[0090] Based on the same inventive concept, please refer to Figure 2 This diagram illustrates a schematic block diagram of a digital twin-based port capacity assessment system 100, provided in an embodiment of this application, for executing the aforementioned digital twin-based port navigation capacity assessment method. The digital twin-based port capacity assessment system 100 may include a communication unit 110, a machine-readable storage medium 120, and a processor 130. Alternatively, the machine-readable storage medium 120 may be integrated into the processor 130 and can communicate and interact with external systems through the communication unit 110. The machine-readable storage medium 120 stores machine-executable instructions for executing the scheme of this application, and the processor 130 executes the machine-executable instructions stored in the machine-readable storage medium 120 to implement the digital twin-based port navigation capacity assessment method provided in the aforementioned method embodiment.

[0091] It should be noted that, in order to simplify the description of the present invention and thus help to understand one or more embodiments of the invention, multiple features may sometimes be grouped into one embodiment, drawing or description thereof in the foregoing description of the embodiments of the present invention.

Claims

1. A method for assessing port navigation capacity based on digital twins, characterized in that, The method includes: An adaptive modeling framework is constructed by integrating UAV oblique photography, laser point cloud data and BIM parametric component library. The framework uses laser point cloud scanning to scan port facility details and UAV to acquire global terrain and building layout. The two types of data are input into a pre-trained port scene recognition model, which automatically matches the standardized modules in the BIM parametric component library to generate an initial 3D model. The initial 3D model has reserved parameter adjustment interfaces for its attributes. Establish a dual mechanism for physical and virtual dynamic calibration and standard adaptation. Deploy a multi-dimensional sensor network at the edge node to collect physical entity operation data in real time. Use the Kalman filter algorithm to remove data noise. Compare the cleaned data with the simulation results of the virtual model to dynamically correct the physical parameters of the model. Simultaneously formulate a unified interface protocol for port digital twin modeling that is compatible with multiple system data formats and define the rules for model collaborative interaction. The port area model is divided into a micro-level equipment component layer, a meso-level operation unit layer, and a macro-level global system layer by adopting cross-scale hierarchical heterogeneous modeling and intelligent linkage technology. The micro-level uses a high-fidelity finite element model stored in the cloud and is only called when high-precision analysis scenarios are needed. The meso-level uses a lightweight model that retains operation characteristics and is deployed on edge nodes. The macro-level uses a topology model that focuses on global indicators. The intelligent linkage algorithm automatically switches the level precision according to the evaluation requirements. We construct a lightweight model and edge / cloud collaborative computing architecture, adopt an attention-based model adaptive simplification algorithm, dynamically adjust the accuracy level according to the scenario priority, edge nodes are responsible for real-time sensor data reception, meso- and macro model simulation and simple decision calculation, and utilize low latency characteristics to ensure rapid response, while the cloud focuses on high-fidelity simulation of micro models, historical data mining and model training optimization, and transmits parameters and model update fragments through edge and cloud data incremental synchronization mechanism. By introducing reinforcement learning-driven full lifecycle closed-loop optimization, the simulation results of the virtual model and the port operation data are used as feedback signals to input the reinforcement learning model, continuously optimizing the physical law mapping and statistical prediction algorithm of the model, identifying changes in port facilities through computer vision and automatically triggering model structure updates, establishing a model accuracy evaluation index system, regularly and automatically verifying model performance, and applying historical modeling experience to new scenarios through transfer learning. An adaptive computing power scheduling algorithm is constructed to dynamically allocate computing resources based on the port area's operational load. When massive data processing is triggered in complex scenarios, collaborative computing of edge node clusters is initiated. Simulation tasks are decomposed into parallel sub-tasks and allocated through a load balancing algorithm. A model caching mechanism is adopted to store frequently called models, reducing redundant calculations and data transmission. While ensuring high accuracy in core scenarios, simulation latency is controlled within the threshold required for real-time evaluation.

2. The port navigation capacity assessment method based on digital twins according to claim 1, characterized in that, The adaptive modeling framework, which integrates UAV oblique photography, laser point cloud data, and a BIM parametric component library, utilizes laser point cloud scanning to capture port facility details and UAV data to obtain global terrain and building layout. These two types of data are input into a pre-trained port scene recognition model, which automatically matches standardized modules from the BIM parametric component library to generate an initial 3D model. The initial 3D model includes a reserved interface for parametric adjustment of attributes, including: The port area modeling requirements were analyzed, the integration rules of laser point cloud, UAV oblique photography data and BIM parametric component library were clarified, and data interaction standards and format conversion protocols were defined. A BIM parametric component library was built, standardized modules were constructed according to port facility types, and editable attribute parameters were preset for each module. At the same time, a module feature index library with associated facility identification features was established. Ground-based laser scanning equipment was used to conduct a comprehensive scan of the port facilities to collect data. The drone oblique photography route was planned and shooting parameters were set to obtain topographic and building layout image data covering the entire port area, while simultaneously recording shooting auxiliary information. The laser point cloud data is denoised and registered. The UAV oblique photogrammetric images are preprocessed to generate dense point cloud and digital orthophoto maps. Terrain and building information are extracted. The processed laser point cloud data and UAV related data are coordinate transformed and fused to form a three-dimensional data volume in a unified format. After processing the pre-processed 3D data volume according to the preset format, it is input into the pre-trained port scene recognition model. The model extracts facility features and compares them with the built-in feature library to complete facility type identification and location. Based on the scene recognition results, the feature index library of the BIM parametric component library is called. The standardized modules are automatically matched through the similarity matching algorithm. The parameters of the matched modules are adaptively adjusted by combining the size information in the 3D data volume. Then, the modules are automatically assembled according to the spatial position relationship of the facilities to generate an initial 3D model covering the entire port area. Parametric programming technology is used to reserve attribute parameter adjustment interfaces on the initial 3D model components. The interfaces are associated with preset attribute parameters in the BIM component library to verify the accuracy of the initial model. If the deviation exceeds the threshold, the module parameters are fine-tuned in reverse, and the final output is a flexibly adjustable initial 3D model of the port area.

3. The port navigation capacity assessment method based on digital twins according to claim 2, characterized in that, The process involves processing the pre-processed 3D data volume according to a preset format, inputting it into a pre-trained port scene recognition model, extracting facility features through the model and comparing them with a built-in feature library to complete facility type identification and localization, including: The original format of the preprocessed 3D data volume is parsed, converted into a unified format according to the input requirements of the pre-trained model, information is retained and redundant attributes are removed, the data is resampled to regularize the dataset to a preset dimension to ensure the uniformity of the input dimension, and the resampled data is normalized to eliminate the feature extraction bias caused by the difference in the size of different facilities. Lightweight augmentation operations are performed on the standardized point cloud data to improve the robustness of the model to port facility attitude changes and environmental disturbances. If the pre-trained model is a two-dimensional convolutional architecture, the three-dimensional point cloud data is converted into a multi-view two-dimensional depth map, preserving the depth information and coordinate mapping relationship of each view, so as to ensure that it can be back-mapped to three-dimensional space later. Load the deep learning model pre-trained based on the port scene dataset, import the pre-trained weight file, fix the parameters of the first half of the feature extraction network and activate the inference mode of the output layer and feature matching layer, initialize the built-in feature library of the model, which is a set of standard feature vectors of various port facilities, and store them according to facility type and build an index structure to improve the efficiency of feature comparison. The formatted input data is fed into the pre-trained model, and spatial alignment, local feature extraction, and global feature fusion are performed sequentially through the feature extraction network to generate a global feature vector that can represent the overall structure and morphological features of the facility. The global feature vector is then normalized to eliminate scale differences. The global feature vector is input into the feature matching module of the model. Similar feature templates in the built-in feature library are retrieved through the index structure. The similarity between the feature to be matched and each template feature is calculated. A similarity threshold is set to filter candidate feature templates. The number of facility categories corresponding to the candidate templates is counted to determine the preliminary identification results. If multiple categories have similarity, local detail features of the area to be identified are extracted for secondary comparison to determine the final facility type. Based on the feature point heatmap output by the model, the key feature points of the facility are located, and their coordinates in the model input coordinate system are obtained. Through a predefined coordinate mapping matrix, the feature point coordinates are transformed to the unified coordinate system of the entire port area. The boundary box of the facility is fitted according to the feature point coordinates to clarify the precise position and spatial attitude of the facility in the three-dimensional space of the port area, and the positioning reliability of the facility is recorded at the same time. By combining prior knowledge of the port area scenario, the initial identification and positioning results are verified for consistency. Results that conflict with the logic of the surrounding environment are eliminated. Facilities with positioning confidence scores below a preset threshold are marked as suspicious areas. Feature extraction and comparison are performed again by supplementing data. Finally, a structured identification and positioning result containing facility type, precise coordinates, bounding box parameters, and identification confidence score is output, and the result is associated with the three-dimensional data volume.

4. The port navigation capacity assessment method based on digital twins according to claim 1, characterized in that, The establishment of a dual mechanism for physical and virtual dynamic calibration and standard adaptation involves deploying a multi-dimensional sensor network at edge nodes to collect real-time operational data of physical entities. Kalman filtering is used to remove data noise, and the cleaned data is compared with the simulation results of the virtual model to dynamically correct the model's physical parameters. Simultaneously, a unified interface protocol for port digital twin modeling compatible with multiple system data formats is developed, defining rules for model collaborative interaction, including: The monitoring area is divided according to the distribution of physical entities in the port area. Multi-dimensional sensing devices are deployed within the coverage area of ​​the edge nodes. The installation points of each sensor are determined to fully reflect the operating status of the entities. Sensor acquisition parameters are configured. The collected physical entity operating data is transmitted to the local data buffer of the edge nodes in real time using a hybrid transmission mode. A data transmission verification mechanism is also set up. Deploy the Kalman filter algorithm module at the edge node, initialize the filter parameters and set the corresponding equation according to the data type of the sensor, input the raw sensor data into the filter module, remove random noise through prediction and update iteration process, output the optimal estimate, perform secondary cleaning on the filtered data, detect and remove outliers, supplement missing data, and then perform standardization processing to form a physical entity operation dataset. A data comparison engine is built, and a comparison cycle linked to the data acquisition frequency is set. The cleaned physical entity operation data is associated and matched with the corresponding simulation data output by the virtual model according to the preset mapping relationship. The deviation calculation model is used to calculate the degree of difference between the two types of data, and a deviation threshold is set. When the deviation exceeds the threshold, the model parameter correction process is triggered, and the comparison information is recorded to form a comparison log. Based on the deviation calculation results, the parameter correction engine is started, the physical parameters of the virtual model corresponding to the deviation are located, a mapping model between the deviation and the parameter correction amount is established, the correction amount is calculated according to the mapping model, the physical parameters of the virtual model are iteratively corrected according to the preset correction step size, the virtual model is driven to re-simulate and compare again until the deviation is lower than the threshold, forming a closed-loop calibration process, and the parameter correction information is recorded at the same time. The various systems involved in the digital twin of the port area were sorted out. Data source information, data format, transmission protocol and interaction requirements of each system were collected through interface debugging and document parsing. The characteristics and differences of data of each system were analyzed, compatibility issues were summarized, the requirements to be covered by the unified interface protocol were clarified, and a protocol requirement specification was formed. Based on the requirements analysis results, a unified interface protocol framework was designed. A layered architecture was adopted with the application layer as the core. Data encoding rules, interface call specifications and message formats were defined. Data field naming rules and data types were unified. Protocol communication specifications were formulated and synchronous / asynchronous transmission modes were supported. Adaptation interfaces for each system were designed. Interface parameters, call methods and error code definitions were clarified. Data encryption transmission and identity authentication mechanisms were added to ensure transmission security. Define model collaboration and interaction rules based on modeling requirements, including data interaction timing rules, data sharing permission rules, and conflict resolution rules; formulate multi-model collaboration and linkage processes; clarify data interaction triggering conditions; build a verification environment to access simulated data from various systems; test protocol compatibility and rule effectiveness; and optimize protocols and rules based on test results. The dynamic calibration mechanism and the standard adaptation mechanism are integrated into the unified management platform of the edge nodes to realize the closed loop of data flow. A monitoring module is deployed on the platform to monitor the operation status of the two mechanisms in real time and trigger alarms when abnormalities occur. Operation data is collected and indicators are analyzed regularly. Filtering parameters, deviation thresholds, protocol details and collaboration rules are iteratively optimized in combination with changes in port operation needs.

5. The port navigation capacity assessment method based on digital twins according to claim 1, characterized in that, The proposed method employs cross-scale hierarchical heterogeneous modeling and intelligent linkage technology, dividing the port area model into a micro-level equipment component layer, a meso-level operational unit layer, and a macro-level global system layer. The micro-level layer uses a high-fidelity finite element model stored in the cloud, only accessed during high-precision analysis scenarios. The meso-level layer uses a lightweight model that retains operational characteristics, deployed on edge nodes. The macro-level layer uses a topology model focusing on global indicators. An intelligent linkage algorithm automatically switches the layer precision based on evaluation requirements, including: Define the criteria and objectives for each of the micro-level equipment component layer, meso-level operation unit layer, and macro-level global system layer; sort out the relationship between each layer; establish entity and parameter mapping tables between layers; and clarify the elements and accuracy requirements for modeling at each layer. The micro-equipment component layer is based on the three-dimensional geometric data and design parameters of the components. A high-fidelity finite element model is constructed using finite element analysis software. Key parts are refined and non-critical structures are appropriately simplified. The meso-level operation unit layer is based on the outer contour features and operation process of the micro-level model. A lightweight model is generated by removing micro-details while retaining operation features. The macro-level global system layer uses a topology abstraction method to abstract meso-level operation units into topology nodes and the relationships between operation units into topology links. A node and link-type topology model is constructed and parameters and indicators are associated. A collaborative storage architecture between the cloud and edge nodes is established. The high-fidelity finite element models of the micro-level are classified and stored in the distributed database in the cloud and a model index library is established. They are only called when high-precision analysis scenarios are required. The lightweight models of the meso-level are deployed in the local storage of the corresponding edge nodes according to the type of work unit. A redundant backup mechanism is adopted to ensure real-time access. The topology models of the macro-level are deployed in both the cloud and edge nodes. Construct a linkage decision engine, define a matrix of hierarchical switching trigger conditions including assessment demand type, data accuracy requirements, response latency requirements, and abnormal event types, build an intelligent linkage algorithm based on rule engine and machine learning algorithm, handle explicit trigger conditions and optimize hierarchical switching strategy, and design a smooth transition mechanism for hierarchical switching. Design standardized linkage interfaces and corresponding calling methods for the three-layer model, adapt to layer switching requests with different real-time requirements, define a unified data interaction protocol, clarify data transmission format and data fields, establish a data interaction verification mechanism and fault tolerance mechanism, and deploy data gateways at edge nodes. The three-layer heterogeneous model, intelligent linkage algorithm, and linkage interface are integrated into the port area's digital twin unified management platform. A simulation test environment is built to simulate different assessment needs scenarios to test the accuracy of level switching triggers, response latency, and data synchronization accuracy. Test indicators are collected and optimization directions are analyzed. Pilot applications are carried out in actual port area scenarios, and the level division standards, model accuracy, and linkage logic are iteratively optimized based on real operational data.

6. The port navigation capacity assessment method based on digital twins according to claim 5, characterized in that, The process involves integrating a three-layer heterogeneous model, intelligent linkage algorithm, and linkage interface into a unified digital twin management platform for the port area. A simulated testing environment is built to test the accuracy of level switching triggers, response latency, and data synchronization precision under different evaluation scenarios. Test indicators are collected and optimization directions are analyzed. Pilot applications are then conducted in actual port area scenarios, including: The technical architecture characteristics of the three-layer heterogeneous model, intelligent linkage algorithm, and linkage interface are analyzed. The input and output parameters and operating environment of each module are clarified. Based on the microservice architecture, each module is encapsulated and split into independent microservices. The communication specifications and registration mechanism of each microservice are defined. The unified management platform framework of the port area digital twin is built, integrating the data access layer, model management layer, algorithm engine layer, interface gateway layer, and visualization display layer, and clarifying the functions of each layer. The encapsulated three-layer heterogeneous model microservices are deployed according to the corresponding nodes. Model indexes and call links are established through the model management layer to achieve unified scheduling. Intelligent linkage algorithms are integrated into the algorithm engine layer. The interaction links between the algorithm and other layers are configured. Standardized linkage interfaces are integrated into the interface gateway layer. Interface routing rules and access control policies are configured. Module joint debugging tests are carried out to verify the synergy of model calls, algorithm decisions, and interface communication. Various synergy issues that occur during the joint debugging process are fixed. Build a cloud-edge collaborative hardware architecture consistent with the real port area, construct a hybrid transmission network to simulate the real data transmission environment, build the corresponding software environment for cloud and edge nodes, install testing tools and log collection system, construct test datasets based on real port area operation data, generate simulated data covering different evaluation scenarios, and label the expected results and core requirements corresponding to the data. We identified typical assessment needs scenarios in the port area and clarified the testing objectives for each scenario. For each scenario, we designed test cases and defined the testing steps, input data, and expected outputs. Single-scenario independent tests and multi-scenario concurrent tests are executed in the order of test cases. During the test, the response time of level switching and data synchronization deviation are recorded by the test tools, and the trigger results and the running status of each module are recorded by the log collection system. Test abnormal scenarios are marked and relevant information is recorded. Organize test data and statistically analyze test indicators, set qualified thresholds for indicators, compare test results with thresholds to filter out non-compliant indicators and corresponding scenarios, analyze the reasons for non-compliance, identify optimization directions, and formulate targeted optimization plans. Optimize according to the optimization plan. After optimization, reuse the original test cases to conduct regression testing to verify whether the indicators meet the standards and the impact of the optimization plan on other modules. Repeat the optimization and regression testing process until the indicators meet the requirements. Select representative pilot areas that cover the entities corresponding to the three-layer model and have complete work processes, conduct on-site surveys of the pilot areas, adjust platform parameters and interface configurations to adapt to the actual environment, deploy a unified management platform in the pilot areas, connect to the real sensor network and operation system on site, complete the configuration of cloud and edge nodes and build a visual monitoring interface; The pilot area platform was launched for trial operation, the platform's operating status was monitored in real time and actual operating data was recorded, feedback from on-site staff on the platform was collected, and actual problems encountered during the trial operation were recorded; By comprehensively analyzing the trial operation data and feedback, we evaluated the platform's adaptability and practicality. Based on the problems discovered during the trial operation and the actual situation on site, we optimized the platform and then re-verified it in the pilot area.

7. The port navigation capacity assessment method based on digital twins according to claim 1, characterized in that, The lightweight model and edge / cloud collaborative computing architecture employs an attention-based adaptive simplification algorithm that dynamically adjusts the accuracy level according to scene priority. Edge nodes are responsible for real-time sensor data reception, meso- and macro-level model simulation, and simple decision calculations, leveraging low latency to ensure rapid response. The cloud focuses on high-fidelity simulation of micro-level models, historical data mining, and model training optimization. Parameters and model update segments are transmitted through an edge / cloud data incremental synchronization mechanism, including: The requirements for digital twin modeling and evaluation in the port area were sorted out, the functional boundaries of edge and cloud collaboration were clarified, the priority levels of scenarios were divided and corresponding model accuracy strategies were matched, the overall framework of the collaborative architecture of real-time edge processing and deep cloud computing was designed, and the data flow and module interaction relationship were clarified. The algorithm framework is constructed to determine the input and output elements, a scene feature attention extraction module is designed, feature attention weights are calculated, dynamic precision adjustment logic is developed, a differentiated simplification strategy is adopted according to scene priority and attention weights, and a model complexity monitoring module is embedded. For meso- and macro-level models, lightweight processing is performed based on an adaptive simplification algorithm to generate multi-precision level model versions and establish a precision and scene priority mapping table. Core features are retained and redundant details are removed according to hierarchical characteristics. The performance of the lightweight model is tested and the simplification parameters are backtracked and optimized. Select suitable edge node hardware and configure relevant components, deploy a lightweight operating system and the software framework and middleware required for real-time processing, configure the communication link between edge nodes and sensors, enable the local data caching mechanism, deploy the model management module, preload the corresponding accuracy model and establish a fast call index; Deploy a cloud server cluster and configure storage and computing components, install operating systems, distributed databases, high-fidelity simulation software and machine learning frameworks, build a cloud model training and optimization module, deploy a cloud-edge collaborative management platform and clarify its overall coordination functions; Define the task allocation rules between edge nodes and the cloud, design a dynamic scheduling engine, monitor the computing load and task queue status of edge nodes in real time, and formulate strategies for task migration and update instruction issuance. Define incremental synchronization data types for edge and cloud, design a synchronization triggering mechanism using a dual-mode approach of timed synchronization and event-triggered synchronization, process synchronization data using data compression and differential coding techniques, deploy synchronization verification and fault tolerance modules, and formulate data conflict handling principles. Integrate all modules into a unified management platform, conduct joint debugging and testing, verify the smoothness of business processes, cloud high-fidelity simulation processes and edge and cloud incremental synchronization processes, test response and synchronization performance in different scenarios, and fix various problems that occur during the joint debugging process. In the pilot area of ​​the port, a collaborative architecture is deployed to access real sensor data and operational scenarios for trial operation. The operational status of edge nodes, cloud and data synchronization is monitored in real time, trial operation data and feedback are collected, the balance between model accuracy and computing efficiency is analyzed, and algorithm strategies, computing power configuration and synchronization mechanism are iteratively optimized.

8. The port navigation capacity assessment method based on digital twins according to claim 7, characterized in that, The document outlines the requirements for digital twin modeling and evaluation in the port area, clarifies the functional boundaries of edge and cloud collaboration, classifies scenario priority levels and matches corresponding model accuracy strategies, designs the overall framework of a collaborative architecture for real-time edge processing and deep cloud computing, and clarifies data flow and module interaction relationships, including: We adopted a combined survey approach to obtain the demands of all relevant parties in the port area, clarified the accuracy and consistency requirements at the modeling level and the scenario requirements at the assessment level, and simultaneously collected real-time response, computing power, bandwidth, and compatibility constraints. We then sorted out the requirements into a list, prioritized them, and output a requirements specification document. The identified requirements are broken down into specific indicators for modeling and evaluation, and the quantitative standards and definitions of each indicator are clarified. Based on the real-time requirements and computing power needs of the demand indicators, the task scope of edge nodes and the cloud is defined, the real-time processing and local caching tasks of edge nodes and the high computing power and non-real-time processing tasks of the cloud are clarified, and the collaborative connection points and data interaction content between the two are determined. Based on the characteristics of port operations and the priority of needs, the scenarios are divided into three priority levels, and the coverage and focus of each level of scenarios are clearly defined. Develop differentiated model accuracy strategies for different priority scenarios, establish a mapping relationship between scenario priority and model accuracy and simplification strategies, and clarify the computational requirements for each scenario model. A four-tier collaborative architecture consisting of a perception layer, edge layer, cloud layer, and application layer is constructed using a layered approach, clearly defining the components, functions, and hardware and software deployment requirements of each layer. Analyze the data links between different levels and modules of the architecture, clarify the transmission content and methods of each data flow, and form a closed-loop data link; Define the interaction links between modules within the architecture, define rules for task allocation, data synchronization, fault switching, and permission management, and write module interaction specifications; Verify the functional coverage and metric satisfaction of the collaborative architecture against the requirements specification, and adjust the architecture design to address adaptability issues until the architecture fully adapts to the requirements.

9. The port navigation capacity assessment method based on digital twins according to claim 7, characterized in that, The constructed algorithm framework determines input and output elements, designs a scene feature attention extraction module, calculates feature attention weights, develops dynamic precision adjustment logic, adopts a differentiated simplification strategy based on scene priority and attention weights, and embeds a model complexity monitoring module, including: The algorithm aims to achieve scene-adaptive precision adjustment and lightweighting of the digital twin model for the port area, adapting to the computing power capacity of edge nodes. Based on logic, a modular algorithm framework is built, divided into an input parsing module, a scene feature attention extraction module, a weight calculation module, a dynamic precision adjustment module, a differentiated simplified execution module, a complexity monitoring and feedback module, and an output encapsulation module. The functional boundaries and data interaction specifications of each module are defined, and the data flow forms, transmission protocols, and triggering conditions between modules are clarified. Input elements include original 3D model data, scene priority labels, real-time evaluation requirements parameters, and edge node computing power status data. The parsing rules, coding specifications, and transmission interfaces for various input data are defined. Output elements include adaptively simplified lightweight model files, model accuracy configuration reports, and complexity monitoring result reports. The encapsulation format of output data and the interface specifications for downstream modules are defined. Construct a port area scene feature system, classify and sort geometric features, operational features, and physical features, establish a feature classification dictionary and attribute description system, build a feature extraction network architecture, develop a feature post-processing module, structure and organize the extracted feature vectors to generate feature maps, and train and optimize the network through labeled datasets to ensure that the feature recognition and positioning accuracy meets the standards. The inherent importance of features, scenario priority, and real-time evaluation requirements are identified as the core dimensions for weight calculation. A multi-dimensional weighted fusion weight calculation model is designed, the initial weights are normalized, a dynamic weight calibration mechanism is developed, the weight parameters are updated based on historical feedback data, and a weight visualization module is built. A precision adjustment rule base is built based on scene priority and attention weight. A three-dimensional mapping relationship is established, the technical parameters corresponding to each precision level are defined, a dynamic precision adjustment decision engine is developed, the precision level is fine-tuned in combination with the computing power status of edge nodes, precision adjustment logic code is written and a rule update interface is reserved, and a logic verification module is added. For different scenarios, priorities, attention weights and accuracy levels, we formulate a differentiated simplification strategy system, develop a simplified algorithm component library, build a simplified execution scheduling module, call the corresponding algorithm component to perform simplified operations according to the accuracy adjustment instruction and record the simplification log; Define a model complexity evaluation index system, clarify the calculation method and collection frequency of each index, develop a complexity monitoring data collection module, set index thresholds under different scenarios, design monitoring and feedback closed-loop logic, link the monitoring module with the dynamic accuracy adjustment module, and develop a complexity monitoring visualization module. Complete the integration and connection of each module and open up data interfaces; conduct single-module functional testing and end-to-end full-process joint debugging testing; verify the module functions and the stability of the entire process operation; and optimize module parameters and interaction logic. Construct a test dataset covering all scenarios in the port area, conduct batch verification of algorithm performance, statistically analyze performance indicators, analyze the root causes of unsatisfactory test cases, optimize algorithm parameters, iterate and verify until all performance indicators meet the requirements, and generate an algorithm performance verification report.

10. A port area capability assessment system based on digital twins, characterized in that: include: processor; A machine-readable storage medium for storing machine-executable instructions of the processor; The processor is configured to execute the port navigation capacity assessment method based on any one of claims 1 to 9 by executing the machine-executable instructions.

Images