Port vertical large model system based on general large language model and construction method

Abstract

The invention relates to the technical field of port vertical large model systems, in particular to a port vertical large model system based on a general large language model and a construction method. Comprising the following modules: a data acquisition module, a pre-training corpus construction module, an incremental pre-training module, a fine adjustment module, a port adaptation and optimization module, a knowledge base management module, a reasoning and man-machine interaction module and a system monitoring and optimization module. The port vertical large model has the beneficial effects that the port vertical large model can dynamically adapt to emergencies in port operation through incremental pre-training, a fine tuning technology and a real-time knowledge base updating mechanism, and multi-source data is integrated and analyzed to form a whole-process business simulation and optimization scheme. The modularized knowledge management mechanism enables the port management system to have a real-time decision support function, and improves the overall efficiency and safety of the port in the aspects of equipment maintenance, cargo scheduling, environment monitoring and the like.

Description

Technical Field

[0001] The present invention relates to the technical field of large vertical model systems of ports, and in particular to a large vertical model system of ports based on a universal large language model and a construction method thereof. Background Art

[0002] As an important hub of global trade, ports play a key role in international cargo transportation and logistics management. With the acceleration of globalization and the continuous expansion of cargo flow, efficient operation of ports has become the basis for maintaining a stable supply chain. Traditional port management mainly relies on manual operation and experience accumulation, but this method can no longer meet the complex and changing needs of port operations today. In order to improve operational efficiency and reduce human errors, port intelligence and automation technologies have emerged. The combination of the Internet of Things, big data analysis and artificial intelligence technology provides a new solution for port operations. Among them, intelligent systems based on large models can quickly adapt to industry changes, understand complex environments, and provide intelligent decision support.

[0003] Traditional general-purpose large language models (such as GPT, BERT, etc.) are mainly composed of four modules: data preprocessing, model training, reasoning, and human-computer interaction. Although it performs well in natural language processing, it still has limitations in applications in specific industries. At the same time, due to the lack of in-depth understanding of the industry field, these models find it difficult to accurately understand professional terms, handle complex operating procedures, and details such as equipment monitoring. Therefore, although these models perform well in general scenarios, their versatility becomes a limitation when facing specific industry needs. Specifically, there are the following defects:

[0004] (1) Inability to conduct adequate testing: Existing methods for building large industry knowledge models based on large language models usually rely on existing data and manual debugging for model optimization, and lack equipment in real scenarios (such as unmanned container trucks, bridge cranes, etc.) for actual operation testing. This makes it impossible to simulate and verify complex operation processes.

[0005] (2) Insufficient real-time and adaptability: Existing methods often require comprehensive training or fine-tuning of large language models to adapt to new industry needs or changes. This is not only time-consuming and labor-intensive, but also makes the model slow to respond to rapidly changing scenarios, making it difficult to achieve real-time intelligent decision-making and response. The lack of rapid knowledge updates and model adaptability will make the system at a loss when faced with emergencies or environmental changes, thereby reducing the practicality and reliability of the system.

[0006] (3) High resource consumption: In existing technologies, updating the knowledge base usually requires retraining or fine-tuning the entire large language model, resulting in a significant increase in computing resources and time costs. This high resource consumption approach is difficult to maintain, especially in application scenarios involving multiple industries or requiring frequent knowledge updates. At the same time, the collection, processing, and annotation of large-scale data is a huge workload, which not only increases labor costs, but also limits the deployment and promotion of models. Summary of the invention

[0007] In order to overcome the technical defects in the prior art and effectively solve the technical problems in the background technology, the present invention provides the following technical solutions:

[0008] The port vertical large model system based on the general large language model includes the following modules: data acquisition module, pre-training corpus construction module, incremental pre-training module, fine-tuning module, port adaptation and optimization module, knowledge base management module, reasoning and human-computer interaction module, system monitoring and optimization module;

[0009] The functions of the data acquisition module include multi-source data fusion, real-time data update, and data cleaning and preprocessing. The data of the data acquisition module includes text data, image data, video data, and structured data;

[0010] The functions of the pre-training corpus construction module include corpus collection and organization, corpus annotation and classification, and data storage and management;

[0011] The functions of the incremental pre-training module include model initialization, industry knowledge enhancement, performance evaluation and optimization;

[0012] The functions of the fine-tuning module include fine-tuning data preparation, business process simulation and fine-tuning result verification;

[0013] The functions of the port adaptation and optimization module include personalized learning of port operations, dynamic optimization and adjustment, cross-port adaptability, continuous optimization of operating processes and equipment operations, and intelligent prediction and decision support;

[0014] The functions of the knowledge base management module include knowledge base creation and maintenance, knowledge retrieval and query, and real-time knowledge update;

[0015] The functions of the reasoning and human-computer interaction module include intelligent reasoning, real-time data support and feedback, multi-round dialogue support, personalized user experience and integration of business processes and user feedback;

[0016] The functions of the system monitoring and optimization module include operation status monitoring and self-feedback optimization mechanism.

[0017] As a preferred technical solution of the present invention, the fine-tuning data preparation includes integrating safety data, scheduling efficiency data and actual scenarios.

[0018] As a preferred technical solution of the present invention, the business process simulation includes the following steps:

[0019] Step 1: Establish business process model and data integration, including flowchart drawing, key node definition, data collection and model initialization;

[0020] Step 2: Security scenario and equipment operation simulation, including security scenario simulation and equipment operation and task allocation simulation;

[0021] Step 3: Scheduling strategy optimization and model decision-making, including scheduling strategy optimization and model response and automatic decision-making;

[0022] Step 4: Verification, optimization and real-time application, including simulation result analysis and feedback optimization, real-time application and continuous monitoring.

[0023] As a preferred technical solution of the present invention, the verification of the fine-tuning results includes model performance evaluation and result comparison, field testing and feedback collection, and continuous optimization and adaptive adjustment.

[0024] The port vertical large model system and construction method based on the general large language model are characterized by comprising the following steps:

[0025] S1: First, the data acquisition module is used to obtain key data such as the operating status and environmental monitoring of various types of equipment and sensors in the port in real time, covering parameters such as operation progress, equipment performance indicators and environmental dynamics, providing high-quality and comprehensive data support for model training;

[0026] S2: Then, a pre-trained corpus building module is used to build a rich corpus using multi-source data, covering the work processes and equipment operations of different ports, ensuring the wide applicability of the model;

[0027] S3: Through the incremental pre-training module, new data is continuously learned and model parameters are dynamically adjusted to ensure the model's latest adaptability to port business and adaptability to different port business scenarios, enhance its generalization ability to changing scenarios, and thus provide more accurate prediction and decision support;

[0028] S4: At the same time, the fine-tuning module can be used to fine-tune the model according to the needs of specific ports to improve its accuracy in specific scenarios;

[0029] S5: Analyze the characteristics of different ports through the port adaptation and optimization module, and intelligently adjust the model to adapt to the equipment and operation processes of each port, so as to achieve efficient management;

[0030] This module can dynamically adjust model parameters to customize the scene. In addition to self-adaptation, it also has the following functions:

[0031] 1. Self-learning and innovation capabilities: For example, if port 1 uses device A, port 2 uses device A, but port 3 uses device B, even if it has no knowledge of device B, it can self-learn based on information related to the features of the device and provide reasonable suggestions on the new device;

[0032] 2. Personalized learning and optimization for different ports: Real-time optimization of equipment scheduling and operation processes to match different ports. That is, a model that has been optimized at port A does not need to be trained from scratch if there is the same business scenario. At the same time, timely optimization is carried out based on the results of data feedback to improve efficiency. In addition, the module can make personalized adjustment measures according to different ports based on the differences in equipment scheduling, yard management and other processes, and provide measures that are more in line with the situation of the port.

[0033] 3. Predictive analysis: Analyze the historical data, real-time status and future operation plan of the equipment, and the module can provide intelligent prediction;

[0034] S6: Dynamically maintain a knowledge base that records best practices and operating specifications through the knowledge base management module, providing decision support and intelligent reasoning basis;

[0035] S7: Through the reasoning and human-computer interaction module, natural language processing and logical reasoning algorithms are integrated to achieve multi-round information interaction and dynamic reasoning capabilities based on context, accurately capture user needs, support the layered decomposition and in-depth analysis of complex problems, and generate optimization solutions based on port business scenarios, providing strong support for intelligent decision-making;

[0036] S8: Monitor the operating status of the entire system through the system monitoring and optimization module, regularly evaluate the model performance, and ensure the continuous and efficient operation of the system;

[0037] S9: Through the coordinated operation of the eight modules S1-S8, a port vertical large model system based on a general large language model can be built, which can significantly improve the intelligent management level and operating efficiency of the port and is conducive to the digital transformation of the port industry.

[0038] Compared with the prior art, the present invention has the following beneficial effects:

[0039] (1) Knowledge adaptability and updating efficiency: Through incremental pre-training, real-time knowledge base updating, and modular knowledge management, rapid iteration and precise application of port professional knowledge are achieved, enabling the system to respond promptly to new situations and new demands in port operations, effectively solving the problem of slow knowledge updating and insufficient adaptability of existing port intelligent systems in a rapidly changing business environment.

[0040] (2) Multi-source data integration and decision support: By integrating multi-dimensional data such as equipment sensors, TOS systems, and environmental monitoring, combined with the reasoning capabilities of large language models, comprehensive and accurate decision support and optimization solutions are provided for the complex operation processes of ports. By proposing multi-source data integration and full-process simulation methods, the problem that traditional systems are difficult to effectively integrate and utilize multiple data sources in ports is effectively solved.

[0041] (3) System performance and resource efficiency: By introducing system monitoring and self-optimization mechanisms, real-time adjustment of model performance and dynamic optimization of resource usage can be achieved, significantly improving the operating efficiency and reliability of the system in the actual port environment, while reducing maintenance costs and resource consumption, thereby effectively solving the performance bottlenecks and resource consumption problems faced by port intelligent systems in large-scale applications.

[0042] (4) Enhanced knowledge dynamic update and real-time response capabilities: The system integrates incremental pre-training, real-time knowledge base update and multimodal data fusion technology to achieve dynamic update and efficient response of port domain knowledge. Compared with traditional static knowledge systems, it can adapt to changes in the business environment in real time, quickly identify and respond to new equipment, new processes or emergencies, and provide continuous and accurate knowledge support for port operations.

[0043] (5) Scenario Adaptation and Multi-Port Optimization Capabilities: The system supports the differentiated needs of different ports in terms of business scale, equipment type and operation process through cross-scenario knowledge migration, adaptive parameter adjustment and personalized optimization strategies. It can not only be quickly deployed in new ports, but also dynamically adjust the optimization plan based on real-time feedback, achieving efficient adaptation from a single port to multiple ports, significantly improving the application scope and universality of the system.

[0044] (6) Prediction-driven decision support and risk management: The system integrates historical equipment operation data, real-time status and future operation plans, and builds a prediction-driven decision support system through machine learning models and knowledge reasoning technology. It can accurately predict potential risks such as equipment failures and operational bottlenecks, and provide preventive maintenance and optimization strategy recommendations to help managers achieve forward-looking risk management and optimal resource allocation in a complex and dynamic environment.

[0045] (7) System collaborative optimization and intelligent operation upgrade: Through modular design and knowledge sharing mechanism, the system realizes cross-module and cross-port collaborative optimization, forming a unified intelligent operation management framework. Multi-module collaborative operation runs through the entire port management process, from data collection to knowledge reasoning, from scenario adaptation to real-time optimization. Compared with traditional single-point optimization systems, this system is significantly ahead in intelligence, systematization and scalability, providing systematic support for the digital transformation of the port industry.

[0046] Specifically, system collaborative optimization and intelligent operation upgrades are achievable under the background of this invention patent.

[0047] 1. Modular design and multi-source data integration: The patent document describes that the system consists of multiple modules, including multi-dimensional data collection of ports, heterogeneous data fusion, entity relationship extraction, efficiency evaluation, and adaptive knowledge graph fusion. These modules work together to form an overall optimization capability. This modular design facilitates cross-module collaborative optimization.

[0048] 2. Knowledge reasoning and optimization supported by large models: The system supports knowledge reasoning and decision optimization modules through large language models, providing full process management from real-time data to deep reasoning, from data collection to contextual knowledge modeling. This feature enables the system to respond flexibly to diverse business scenarios and adapt to the rapidly changing port operation environment, thereby achieving intelligent operation upgrades.

[0049] 3. Adaptive update and conflict resolution: The patent clearly states that the system can adaptively update and resolve conflicts in the knowledge graph to ensure data consistency and real-time knowledge. These functions enable the knowledge graph to remain effective and efficient when data is constantly updated, which is conducive to the intelligent collaborative management of the entire system.

[0050] In summary, the modules and technical features in the patent document support the system's ability to theoretically achieve collaborative optimization and intelligent operation upgrades. BRIEF DESCRIPTION OF THE DRAWINGS

[0051] Figure 1 This is a flow chart of the operating system integration architecture of the present invention.

[0052] Figure 2 It is a data collection and analysis flow chart of the present invention.

[0053] Figure 3 It is a flow chart of the port equipment and operation process control of the present invention.

[0054] Figure 4 It is a flow chart of data classification and information flow of the present invention.

[0055] Figure 5The system deployment architecture flow chart of the present invention.

[0056] Figure 6 This is a flow chart of intelligent reasoning and human-computer interaction of the present invention.

[0057] Figure 7 It is a flow chart of the port vertical model system of the present invention. DETAILED DESCRIPTION

[0058] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0059] See also Figure 1-7 The present invention provides a technical solution: a port vertical large model system and construction method based on a general large language model.

[0060] 1. The system consists of:

[0061] This port big prediction model aims to improve the intelligent management level of port operations. It consists of eight main modules: data acquisition module, pre-training corpus construction module, incremental pre-training module, fine-tuning module, port adaptation and optimization module, knowledge base management module, reasoning and human-computer interaction module, and system monitoring and optimization module.

[0062] First, the data acquisition module is responsible for collecting data from various equipment and sensors in the port in real time, including operation progress and environmental status. These data provide the basis for model training. Subsequently, the pre-training corpus construction module uses multi-source data to build a rich corpus covering the operation processes and equipment operations of different ports, ensuring that the model has wide applicability.

[0063] To keep the model updated and adaptable, the incremental pre-training module continuously learns new data and dynamically adjusts model parameters to ensure that it is consistent with the ever-changing port environment. At the same time, the fine-tuning module focuses on fine-tuning the model according to the needs of specific ports to improve its accuracy in specific scenarios.

[0064] The port adaptation and optimization module focuses on analyzing the characteristics of different ports and intelligently adjusting the model to adapt to the equipment and operation processes of each port, thereby achieving efficient management. The knowledge base management module dynamically maintains a knowledge base that records best practices and operating specifications, providing decision support and intelligent reasoning basis.

[0065] In addition, the reasoning and human-computer interaction module uses natural language processing technology to interact with port staff, provide real-time suggestions and decision support, and enhance human-computer collaboration. The system monitoring and optimization module monitors the operating status of the entire system, regularly evaluates model performance, and ensures that the system continues to operate efficiently.

[0066] Through the coordinated operation of these eight modules, this port big prediction model will significantly improve the port's intelligence level and operational efficiency, and provide strong support for the digital transformation of the port industry ( Figure 1 The system structure of the method of one embodiment of the present invention is an interactive operating system integrated architecture flow chart).

[0067] Specifically, the functions of the eight modules are as follows:

[0068] 1. Data acquisition module

[0069] This module is designed to collect real-time information from various data sources in port operations, including ship location, cargo status, equipment operating parameters, port environment monitoring data, etc. Through integration with port systems, equipment and sensors, the data collection module ensures the real-time and accuracy of data, providing high-quality training data for subsequent port large models.

[0070] (1) The module specifically includes the following functions:

[0071] a) Multi-source data fusion: Integrate data from multiple sources such as equipment monitoring systems and environmental perception systems to ensure the comprehensiveness and consistency of data.

[0072] b) Real-time data update: Dynamically monitor the status of port equipment, ships and logistics operations, and continuously update data sources to adapt to changes in daily port operations.

[0073] c) Data cleaning and preprocessing: Clean, denoise and standardize the collected data to ensure that the data meets the input requirements of the large language model and improve data utilization efficiency.

[0074] (2) Data source:

[0075] a) Text data:

[0076] i. Data source: TOS (Terminal Operating System) and ECS (Equipment Control System).

[0077] ii. Data types: job reports, equipment logs, dispatch instructions, container tracking records, etc.

[0078] iii. Collection equipment: server log system, database system, etc.

[0079] b) Image data:

[0080] i. Data source: CCTV (video surveillance system), AGV (automatic guided vehicle) camera.

[0081] ii. Data types: containers, cargo loading and unloading conditions, key images during equipment operation.

[0082] iii. Collection equipment: port surveillance cameras and built-in visual sensors of the equipment.

[0083] c) Video data:

[0084] i. Data source: TCS (Terminal Control System) and port monitoring system.

[0085] ii. Data type: Real-time video stream, including monitoring videos of loading and unloading operations and equipment operation videos.

[0086] iii. Collection equipment: CCTV cameras, drone video surveillance systems.

[0087] d) Structured data:

[0088] i. Data sources: TOS, TCS, external documents (such as shipping company's electronic data interchange files, EDI system).

[0089] ii. Data types: ship arrival schedule, container location data, equipment operating status, sensor data (temperature, humidity, equipment status, etc.).

[0090] iii. Collection equipment: sensor equipment, database, logistics system, etc.

[0091] 2. Pre-training corpus construction module

[0092] This module focuses on building a specific industry corpus covering port business to enhance the large language model's ability to understand and process port domain knowledge. By collecting, organizing and annotating text data such as port-related documents, operation manuals, case studies, etc., a comprehensive industry corpus is established to provide basic data support for model pre-training. The specific functions of this module include:

[0093] (1) Data collection and organization

[0094] a) Data sources: The collected data comes from a variety of different sources, such as port operating procedures, equipment maintenance manuals, industry standards, and historical operation data. Depending on the characteristics of the industry, text data may include operating manuals, expert case studies, etc., to ensure that all aspects of port operations are covered.

[0095] b) Treatment method

[0096] i. Text data: Use natural language processing (NLP) technology to pre-process the original documents, including removing irrelevant information and cleaning up duplicate data. This is to ensure that the collected corpus is of high quality and relevance. Based on industry-specific vocabulary, a dedicated vocabulary is built to enhance the model's understanding of port field terminology.

[0097] ii. Image and video data: Images are labeled with equipment and container types through deep learning technology (such as convolutional neural network CNN), and video data are labeled with operation processes through action recognition models (such as 3DCNN) to ensure that key operation steps are accurately classified.

[0098] (2) Corpus annotation and classification

[0099] a) Marking method

[0100] i. Text data: With the participation of experts, fine-grained annotation of text corpus is performed, such as annotation of operation steps, equipment names, work areas (such as shore, yard, road), etc. Classification is based on different equipment (horizontal transportation such as internal container trucks and external container trucks; vertical transportation such as reach stackers and forklifts) and operation processes (shore, yard operations, etc.).

[0101] ii. Image and video data: Images are labeled with equipment and container types through deep learning technology (such as convolutional neural network CNN), and video data are labeled with operation processes through action recognition models (such as 3DCNN) to ensure that key operation steps are accurately classified.

[0102] b) Annotation goal: Ensure that the corpus covers the multi-level knowledge system of port operations, and build a multi-level knowledge system of port operations, covering equipment types, operation processes, operating specifications, etc. The combined annotation of text, image and video data will help the model fully understand the complexity of port operations, and help the systematic classification of the corpus, and improve the model's ability to handle port-specific processes.

[0103] (3) Data storage and management: The corpus is stored in a structured manner to support version control, dynamic updating and expansion of the corpus, ensuring that the data always keeps pace with the latest developments in port business and providing the most timely and accurate knowledge support for the model.

[0104] a) Structured storage: All processed and annotated data (including text, images, and videos) will be stored in a structured manner and managed by equipment type, operation process, time period, etc. The corpus supports version control and can be updated dynamically to ensure that the data is always synchronized with the latest developments of the port.

[0105] b) Knowledge base management: Build a corpus system that supports real-time query and call. The corpus provides efficient support for the incremental learning, fine-tuning and reasoning modules of the model. Through the dynamic update mechanism, ensure the timeliness and accuracy of the model.

[0106] 3. Incremental pre-training module

[0107] This module aims to incrementally pre-train the general large language model by gradually introducing port-specific data, so that the model can gradually enhance its understanding of port business based on the original general knowledge. This module improves the adaptability and update efficiency of the model by dynamically introducing new corpus to make small adjustments without retraining the entire model. Specific functions include:

[0108] (1) Model initialization: Based on the existing general large language models (such as LLaMA, Wenxinyiyan, etc.), the model is incrementally pre-trained using the pre-trained corpus. This step improves the model's ability to understand specific industry contexts.

[0109] (2) Industry knowledge enhancement: Collect specific corpus from multiple channels such as port industry-related documents, manuals, and operating procedures. Use natural language processing technology to clean and annotate the corpus to ensure data quality and enable the model to master industry-specific terminology and operating procedures. Optimize the model's parameters during training so that it can better understand and generate industry-related content.

[0110] (3) Performance evaluation and optimization: Use standardized evaluation indicators to monitor model performance. After each incremental training, use the test set to evaluate the model performance to ensure the accuracy and generalization ability of the model after incremental pre-training. Based on the evaluation results, further optimize the training process and parameter settings.

[0111] 4. Fine-tune the module

[0112] This module is responsible for customizing and fine-tuning the pre-trained model in specific port business scenarios, so that the model can handle more complex industry tasks and provide accurate business support. Fine-tuning is based on actual port operation data to ensure that the model can meet specific business needs.

[0113] (1) Fine-tune data preparation

[0114] a) Integrate safety data: During the fine-tuning data preparation stage, all data related to equipment operation safety in the port are collected, including equipment failure records, operating specifications in the safety manual, and historical accident cases. By annotating each key process node with data, the big model can truly understand the standardized process of port operation tasks, enable it to identify potential safety hazards, and optimize safety-related decision-making capabilities.

[0115] b) Scheduling efficiency data: A complete scheduling efficiency data set is established by retrieving equipment scheduling logs, real-time task allocation data, etc. The data set will cover aspects such as equipment utilization, task execution time, and path planning, ensuring that the model can fully learn how to optimize equipment scheduling and improve overall operating efficiency during the fine-tuning stage.

[0116] c) Actual scenarios: For example, during crane operation, fault records and equipment load data are collected and combined with the equipment usage recorded by the dispatching system, and incorporated into the fine-tuning data to ensure that the model is fully trained in terms of safety assurance and efficient dispatching.

[0117] (2). Business process simulation

[0118] Step 1: Establish business process model and data integration

[0119] Flowchart drawing and key node definition:

[0120] a) Flowchart drawing: Use flowchart software (such as Visio or Lucidchart) to draw a detailed flowchart of port operations, including loading and unloading, storage, transportation, scheduling, etc. The flowchart should clearly indicate the name of each link, the time required and the equipment involved.

[0121] b) Critical node identification: Mark the critical nodes in the process, especially the safety nodes, such as equipment failure detection points, personnel operation specification execution points, work instruction confirmation points, etc. These nodes are the focus of future simulations to help identify potential risks.

[0122] Data collection and model initialization

[0123] a) Historical data collection: extract historical operation data from the port management system, covering operation duration, equipment status (such as failure, maintenance records), accident records, etc. At the same time, the impact of environmental factors (such as weather, tide) on operations should also be considered.

[0124] b) Data integration: Integrate the collected safety data with the scheduling efficiency data to build a comprehensive data set. Ensure that the data set contains information of various dimensions so that the model can fully understand the operating environment and identify efficiency and safety hazards during the training process.

[0125] c) Model initialization: Use the cleaned and preprocessed data to initialize the business process model to ensure that the model accurately reflects the operational characteristics and risk points of the port, providing a basis for subsequent simulations.

[0126] Step 2: Security scenario and equipment operation simulation

[0127] Security scenario simulation

[0128] a) Construct emergency scenarios: Set up a variety of emergency events in the simulation environment, such as bad weather (strong winds, heavy rains), equipment failures (power failures, mechanical failures), etc. These scenarios should be designed based on historical events to ensure their realism and applicability.

[0129] b) Triggering safety events: Introduce randomly triggered safety events such as over-temperature, overload, and equipment shutdown in the simulation, record the model's reaction time and response strategy, and ensure that the simulation covers various potential risks.

[0130] c) Response strategy learning: Use historical data to fine-tune the model so that it learns automatic response strategies when facing emergencies and triggers corresponding safety measures, such as reducing operating speed, suspending operations, or initiating emergency procedures to ensure operational safety.

[0131] Equipment operation and task allocation simulation

[0132] a) Refine the operation process: For each equipment, simulate the specific operation process and task allocation, including the working mode and efficiency standard of cranes, container trucks, conveyors and other equipment. Ensure that each equipment can operate in its best working condition, thereby improving operation efficiency.

[0133] b) Safety monitoring mechanism integration: embed real-time monitoring functions into the model, collect equipment operation data, and monitor changes in equipment status and operating environment in real time. For example, monitor indicators such as temperature, load, and working hours to ensure that the equipment operates within a safe range.

[0134] c) Exception handling mechanism: Once an equipment abnormality is detected (such as excessive load, abnormal temperature), the model can respond quickly, automatically adjust the operation plan or trigger an alarm to notify relevant operators to avoid safety accidents.

[0135] Step 3: Scheduling strategy optimization and model decision making

[0136] Scheduling strategy optimization

[0137] a) Real-time data analysis: Real-time analysis of the equipment status, location and task priority, and dynamic optimization of scheduling strategies. For example, according to the current job tasks and equipment idleness, the equipment's working order and task allocation are adjusted to maximize resource utilization and reduce job waiting time.

[0138] b) Application of path optimization algorithm: Use path planning algorithms (such as Dijkstra algorithm and A algorithm) to find the optimal scheduling path for equipment to reduce the equipment's movement time and idling and improve work efficiency.

[0139] c) Operation process collaboration: simulate collaborative operations between devices, consider dependencies and resource sharing between operations, optimize the overall operation flow, reduce resource conflicts and time delays, and improve overall efficiency.

[0140] Model responses and automated decision making

[0141] a) Real-time monitoring and analysis: The model continuously monitors device status and environmental changes, and analyzes possible safety hazards. For example, it tracks key indicators such as temperature, pressure, and load in real time to ensure that all devices operate within a safe range.

[0142] b) Preset threshold monitoring: Set safety thresholds and monitor the operating indicators of the equipment. Once an indicator exceeds the preset safety threshold, the model will automatically trigger an alarm and start the emergency response mechanism to reduce the probability of accidents.

[0143] c) Automatic decision-making and execution: When the model finds safety hazards, it can respond quickly and automatically adjust the work process, suspend operations or notify relevant personnel to ensure work safety.

[0144] Step 4: Verification, optimization and real-time application

[0145] Simulation results analysis and feedback optimization

[0146] a) Result evaluation: Conduct in-depth analysis of the data recorded during the simulation process to evaluate the model's response speed, decision-making effectiveness, and implementation of safety measures. Through comprehensive data analysis, identify problems in the simulation and make targeted improvements.

[0147] b) Bottleneck identification: Identify bottleneck links in the simulation and analyze issues such as response time when a security incident occurs and information transmission delay in the decision-making chain in order to carry out targeted optimization.

[0148] c) Model optimization: Based on the feedback results, continuously adjust and optimize the model's parameters and decision logic to improve the model's adaptability and reliability in actual operations.

[0149] Real-time application and continuous monitoring:

[0150] a) Model deployment: Apply the optimized model to real port operations, monitor operation progress, equipment status and safety risks in real time, and ensure the effectiveness of the model in actual applications.

[0151] b) Continuous monitoring and feedback mechanism: Establish a monitoring and feedback mechanism to regularly collect actual operation data and evaluate the performance of the model to ensure that it adapts to the ever-changing port environment.

[0152] c) Dynamic adjustment: In actual operation, the model can identify safety hazards or scheduling efficiency issues in real time, and adjust strategies in time to optimize operating processes to ensure safe and efficient port operations.

[0153] (3) Verification of fine-tuning results

[0154] Model performance evaluation and result comparison

[0155] a) Validation set testing: Use a dedicated validation set to perform performance tests on the fine-tuned model and evaluate its performance in different operating scenarios. This includes testing the model’s response speed and decision-making accuracy in emergencies such as equipment failure and bad weather.

[0156] b) Comparative analysis: Compare the simulation results before and after fine-tuning, focusing on key indicators such as response time to security incidents, equipment scheduling efficiency, and job completion time. Through quantitative analysis, verify whether the fine-tuning has achieved the expected security and efficiency improvements.

[0157] c) Application of statistical methods: Use statistical analysis tools (such as T-test, analysis of variance) to compare results and conduct significance tests to ensure that the observed improvements are statistically significant and avoid misjudgment due to accidental factors.

[0158] Field testing and feedback collection:

[0159] a) Pilot implementation: Select specific areas or equipment in actual port operations for small-scale pilot testing to observe how the fine-tuned model performs in a real environment. This can help identify problems that the model may encounter in complex environments.

[0160] b) Real-time monitoring and data collection: During the pilot period, the operation of the model was monitored in real time, and the operation progress, equipment status, and frequency of safety incidents were recorded to ensure a comprehensive understanding of the actual performance of the model.

[0161] c) Operator feedback: Collect feedback and suggestions from front-line operators to understand the advantages and disadvantages of the model in actual application. These feedbacks will provide valuable data support for subsequent optimization.

[0162] Continuous optimization and adaptive adjustment

[0163] a) Dynamic monitoring mechanism: Establish a continuous monitoring mechanism to track the difference between the model’s performance and expected results in real time, ensuring that the model can quickly adapt to the changing port environment and operational needs.

[0164] b) Regular review and iteration: Develop a regular review plan and perform iterative optimization based on monitoring results and operator feedback to ensure that the adaptability and reliability of the model in actual operations are continuously improved.

[0165] c) Data-driven improvements: Utilize the collected historical operational data and emerging risk information to adjust and optimize the model’s parameters and decision-making logic, maintain the model’s foresight and effectiveness, and further improve the port’s safety and operational efficiency.

[0166] 5. Port Adaptation and Optimization Module

[0167] This module enables the model to dynamically adapt to the specific business needs of different ports through continuous learning and optimization of port business scenarios, equipment operations and work processes. This module mainly targets the diversity of port business, ensuring that the model has flexible adaptability in different port environments and can be optimized in a targeted manner. The following are the specific functions and effects of this module:

[0168] (1) Personalized learning of port operations

[0169] a) By analyzing and learning the historical data and real-time operation data of the ports, the specific needs and patterns of different ports in terms of operation processes and equipment operation are identified. For example, different ports may have differences in equipment scheduling, yard management, and loading and unloading processes. The module can dynamically adjust the model according to these personalized needs.

[0170] b) Example: The equipment scheduling task of one port prioritizes the operation of container trucks, while another port may prioritize the scheduling of forklifts. The Adaptation and Optimization module will learn these different priorities and provide the best operation suggestions for different port scenarios.

[0171] (2) Dynamic optimization and adjustment

[0172] a) The model is continuously optimized by real-time monitoring of the equipment operation status and operation process of the port. As the port's business needs change (such as equipment failure, weather changes, etc.), the module can dynamically adjust the model parameters, optimize the equipment scheduling and operation process to ensure efficient operation of the port. For example, when a port faces a sudden equipment failure, the module will reschedule according to the status of other equipment, give priority to the available equipment to take over the task of the failed equipment, and avoid operation delays.

[0173] (3) Cross-port adaptability

[0174] a) Through in-depth learning of business scenarios, equipment characteristics and operating modes of multiple ports, the module has the ability to adapt across ports. Even if different ports have different equipment configurations and operating processes, the model can still adapt to these changes and provide optimized scheduling and management solutions. For example, a model that has been optimized in Port A can also adapt to the operating needs of Port B without having to be trained completely from scratch. By comparing and adapting to the operating modes of the two, the model can flexibly respond to the operational needs of multiple ports.

[0175] (4) Continuous optimization of work processes and equipment operations

[0176] a) The module not only focuses on optimizing equipment scheduling, but also continuously optimizes equipment operation specifications and operating procedures based on feedback data from the actual port operation process. Through the real-time feedback mechanism, the system can identify bottlenecks or inefficient links in the current operation process, and then optimize these processes to improve the overall operation efficiency of the port. For example, if a container truck is found to be inefficient when driving on the road at a certain port, the module will optimize the route planning and driving scheduling by analyzing the driving route, traffic conditions and other data of the container truck, and reduce congestion and empty vehicle driving time.

[0177] (5) Intelligent prediction and decision support

[0178] a) By analyzing the historical data, real-time status and future operation plans of the equipment, the module can provide intelligent prediction and decision support. This function is particularly suitable for predicting the maintenance needs of equipment and the changing trend of equipment load. For example, during the operation of port equipment, the module will predict the maintenance cycle and potential failure risks of the equipment based on the load status and historical usage of the equipment, provide early warning and arrange equipment maintenance plans to avoid operation interruptions caused by equipment failures.

[0179] 6. Knowledge base management module

[0180] This module is responsible for building and managing the knowledge base in the port field, and combines it with model reasoning capabilities to provide intelligent knowledge services. This module supports real-time updates and modular management, and can dynamically insert or remove specific knowledge modules to ensure that the knowledge base always keeps pace with the latest developments in the industry. Specific functions include:

[0181] (1) Knowledge base creation and maintenance

[0182] The knowledge base can dynamically manage the data of horizontal and vertical transportation equipment, including the operating specifications of internal and external container trucks, reach stackers and forklifts, ensuring efficient maintenance and real-time updating of equipment information.

[0183] a) Management of port equipment diversity: The knowledge base will list in detail the operating specifications and maintenance guidelines for various equipment, divided into two categories:

[0184] i. Horizontal transport equipment: including internal container trucks and external container trucks, which are responsible for the transfer of goods between the terminal and the yard. The knowledge base will record the best operating procedures, daily inspection standards and troubleshooting solutions for each type of equipment.

[0185] ii. Longitudinal transport equipment: including flow machines (reach stackers, forklifts), which are mainly used for vertical loading, unloading and stacking of goods. The knowledge base will provide the equipment's operating performance parameters, maintenance cycles and troubleshooting guides.

[0186] Through modular management, equipment data can be quickly updated or replaced, ensuring that port operators always have access to the latest equipment usage information.

[0187] b) Comprehensive support for operational processes: The knowledge base will cover standard operating procedures (SOPs) and emergency plans for operational processes such as shore operations, yard management, and road driving.

[0188] i. Shoreside operations: Detailed records of container loading and unloading sequences, equipment scheduling, and best practices for human-machine collaboration.

[0189] ii. Yard management: Efficient stacking and picking strategies for container trucks and forklifts to ensure maximum space utilization.

[0190] iii. Road driving: Provide vehicle driving safety specifications and traffic rules to ensure safe and efficient transportation.

[0191] Through this module, users can dynamically adjust the job processes in the knowledge base according to real-time business needs and respond to business changes without retraining the model.

[0192] (2) Knowledge retrieval and query

[0193] By combining the knowledge base with model reasoning capabilities, the system can provide operators with personalized operational suggestions to help optimize equipment scheduling and respond to emergencies.

[0194] a) Query of port operation process: Through the knowledge retrieval function, users can query port equipment and operation processes in natural language. For example, users can query "how to dispatch container trucks under severe weather conditions", and the system will provide best practices and historical data support from the knowledge base. At the same time, its query results not only include operation steps, but also relevant precautions and previous successful cases, which can enhance users' decision-making confidence.

[0195] b) Intelligent suggestions: The knowledge base will provide personalized operation suggestions and equipment scheduling optimization solutions based on the current operation status or equipment conditions, combined with the reasoning module. This is especially useful for quickly querying and obtaining response solutions when equipment fails or operation process bottlenecks occur.

[0196] i. Equipment failure: When the system detects an abnormal equipment status (such as overtemperature or overload), the knowledge base will automatically recommend corresponding emergency handling procedures and alternative equipment usage plans.

[0197] ii. Operation process bottleneck: When a bottleneck is encountered in the operation process, the system can quickly provide adjustment suggestions to optimize equipment scheduling and improve operation efficiency.

[0198] (3) Real-time knowledge update

[0199] The knowledge base management module supports real-time updates based on the latest port operation data, ensuring that it always reflects actual operations.

[0200] a) Equipment and process updates: With the introduction of new equipment (such as the latest model of forklift or flow machine), the knowledge base can automatically or manually insert new equipment information and operating specifications to ensure that the content is synchronized with the update of port equipment. This process includes:

[0201] i. Collect operating manuals, maintenance guides, and usage data for new equipment.

[0202] ii. Update equipment performance parameters and recommended operating procedures to ensure they meet the latest port requirements.

[0203] b) Best Practice Updates: In port operations, real-time updated best practices (such as the latest dispatching rules for internal and external container trucks) will continue to enter the knowledge base to help operators obtain the best operation strategy at any time. The module supports timely updates based on industry trends, equipment technology changes and other factors.

[0204] 7. Reasoning and human-computer interaction module

[0205] This module mainly provides reasoning capabilities and human-computer interaction functions based on the port large model, enabling the model to understand the user's natural language needs in real time and provide intelligent feedback based on industry knowledge. The module supports multi-round dialogues and reasoning in complex business scenarios, helping port operators to handle daily work and respond to emergencies more efficiently. Specific functions include:

[0206] (1) Intelligent Reasoning

[0207] a) Port knowledge big model reasoning guidance: The module loads the port knowledge big model, integrates relevant knowledge and real-time business data in the port field, and performs intelligent reasoning and decision support. This process can help users handle complex business applications, such as equipment scheduling, operation process scheduling, etc., to ensure that all operational decisions are based on current best practices and real-time data.

[0208] b) Business detailed reasoning: For different types of business applications, the system can classify them according to preset business rules and standards to help users determine the best operation steps. For example, when a user asks how to optimize the use of the yard during peak hours, the system can integrate historical data and current yard status to provide detailed yard management suggestions.

[0209] (2) Real-time data support and feedback:

[0210] When dealing with user issues, the system can call up real-time port operation data (such as equipment operating status, task progress, etc.), and combine historical best practices to give more accurate suggestions. For example, when a user faces equipment failure, the system will infer the possible cause of the failure and provide a processing flow, such as switching to backup equipment or temporarily adjusting the task allocation.

[0211] (3) Multi-round dialogue support

[0212] a) Context management and deepening of business consultation: The module supports multiple rounds of dialogue with users to ensure the consistency of context, so that users can gradually deepen business consultation in continuous dialogue. For example, the user may initially ask how to dispatch internal collection trucks, and then further ask how to optimize the dispatch strategy under specific weather conditions. The system can remember the previous context and provide more complete suggestions.

[0213] b) In-depth business scenario simulation: Users can simulate actual business scenarios through the system, input specific business problems, and the system will guide users step by step to obtain the best solution to the problem. This function helps users conduct experimental operations in complex business scenarios and reduce the risk of errors in real operations.

[0214] (4) Personalized user experience

[0215] a) Intelligent customization based on user roles: The system can adjust the answer style and content according to the user's role and input requirements. For example, for front-line operators, the system may provide detailed steps and operation prompts; while for managers, it focuses more on the status of the equipment and the optimization of overall operating efficiency. This personalized feedback not only improves the user experience, but also helps different user groups complete their tasks more effectively.

[0216] b) Personalized recommendations and improvements: By continuously analyzing user feedback and operating habits, the system can continuously learn and adapt to user behavior and provide recommendations that are more in line with user preferences. For example, if a user frequently inquires about the operating steps of a certain device, the system will prioritize providing relevant information in future interactions, reducing the user's query time.

[0217] (5) Integration of business processes and user feedback:

[0218] The module can continuously absorb user feedback in actual operations, adjust the system's knowledge reasoning rules, and continuously optimize and upgrade the system. For example, by analyzing user complaints and question records, the module can identify and improve common business process bottlenecks, thereby improving the system's operational effectiveness and the user's overall experience.

[0219] a) Decision support and operation support: The system provides feasible solutions based on the demands or questions input by the user and combines the reasoning model to support business decisions and operations. For example, when the user asks how to optimize the yard layout, the system can analyze real-time data flow, equipment utilization and other factors to propose specific implementation plans to support the implementation at the operational level.

[0220] 8. System monitoring and optimization module

[0221] This module is responsible for real-time monitoring of the model's operating status, ensuring the system's stability and performance under high load, and dynamically optimizing model parameters based on user feedback and system logs to improve the overall efficiency of the system. Specific functions include:

[0222] (1) Operation status monitoring: Real-time monitoring of system performance indicators, such as response time, accuracy, user satisfaction, etc., to ensure that the model runs efficiently in actual business.

[0223] (2) Self-feedback optimization mechanism: Dynamically adjust model parameters and reasoning strategies through user feedback and system log data to continuously optimize system performance.

[0224] 2. System deployment:

[0225] The deployment of the port vertical big model adopts a layered architecture to ensure performance, security and flexibility. First, a data collection layer is set up in the internal network to connect port systems such as TOS, LMS, and IoT devices to collect real-time data. The data processing layer uses an ETL server cluster and data warehouse to clean, integrate and store data.

[0226] Model training and inference layers are deployed in high-performance computing clusters, including GPU server clusters for incremental training and fine-tuning, distributed storage systems to manage large-scale data, model version control to ensure stability, and inference acceleration engines to optimize real-time response.

[0227] The middle-tier network deploys the application service layer, including API gateway, load balancer, application server cluster and knowledge base management system, to process user requests and provide intelligent services. Finally, a secure external access interface is set up to allow authorized users to access securely.

[0228] 3. Workflow:

[0229] 1. Data acquisition module:

[0230] (1) Historical data collection

[0231] a) Equipment history document information

[0232] Collect information such as equipment operation time, maintenance records, failure rates, control parameter settings, etc. This historical data can be used to analyze equipment performance, develop maintenance plans, and predict equipment life.

[0233] b) Business historical data

[0234] i. Extract business data such as historical loading and unloading tasks, ship arrival information, container lists and berth allocation from the TOS system to simulate and optimize the operation process.

[0235] ii. Import historical data of goods flow from the logistics management system (LMS) for trend analysis and training of prediction models.

[0236] iii. Historical AIS data: Analyze the historical entry and exit of ships to optimize future berth allocation and entry and exit management.

[0237] (2) Real-time data collection

[0238] a) Real-time equipment status data:

[0239] Collect the real-time status of production equipment from equipment sensors or control systems, such as current operating status, operating parameters, workload, etc., to ensure that the model can instantly reflect the current working conditions of the equipment.

[0240] b) Environmental data:

[0241] Environmental sensors are used to collect real-time environmental conditions of the port, including temperature, humidity, wind speed, weather conditions, etc. At the same time, third-party APIs are integrated to obtain real-time meteorological information and forecast data to support emergency response and operation adjustments.

[0242] c) Real-time business data:

[0243] i. Extract real-time operation tasks from the TOS system, including loading and unloading task progress, ship arrival time, container loading and unloading status, etc.

[0244] ii. Automatic Identification System (AIS): collects the real-time position, speed, heading and other information of the ship to support the planning of port ship entry and exit and berth scheduling.

[0245] iii. Obtain the loading and unloading status, transportation route and delivery time of goods from the logistics management system (LMS) to ensure real-time tracking and optimization of the flow of goods.

[0246] (3) Equipment data collection

[0247] a) Device sensor data:

[0248] Collect equipment operating status, operating parameters, maintenance requirements, etc. from equipment sensors to ensure the integrity of equipment data for real-time monitoring and predictive maintenance.

[0249] b) Terminal Equipment Management System (EMS):

[0250] Monitor the operating status, utilization and maintenance requirements of equipment such as quay cranes, gantry cranes, and automated guided vehicles to ensure that the equipment is operating in the best condition. EMS data can be used to simulate equipment scheduling and maintenance plans, optimize equipment utilization and reduce downtime.

[0251] (4) Truck data collection

[0252] a) Internal and external container cards:

[0253] Collect data on the operating status, load conditions, and transportation routes of internal and external container trucks. This data helps optimize the dispatch of container trucks, ensure efficient transportation of goods, and avoid idling of container trucks and road congestion.

[0254] (5) Data processing and storage

[0255] a) Data cleaning and centralized storage:

[0256] After preliminary cleaning and preprocessing, the collected data are stored in a centralized data warehouse to ensure the integrity and consistency of the data and provide support for subsequent corpus construction.

[0257] 2. Pre-training corpus construction module

[0258] (1) Data cleaning and standardization

[0259] a) Cleaning step: Process the missing values, duplicate values, and outliers of the collected data to ensure data integrity and consistency.

[0260] i. Missing data completion: Use historical data or predictive models to complete key missing data, such as equipment status or scheduling instructions.

[0261] ii. Outlier processing: Set upper and lower limits to eliminate or correct abnormal data, such as extreme wind speeds in meteorological data and abnormal peaks in traffic flow.

[0262] b) Data standardization: Unified coding and formatting of multi-source data to ensure that all data conform to the large model input standard. For example, ship position data is standardized into a global unified coordinate system, and timestamps are uniformly expressed in UTC time.

[0263] (2) Business classification and hierarchical processing

[0264] a) Data classification: Data is divided into different business scenario categories according to port business needs and operating equipment characteristics. Mainly including:

[0265] i. Equipment operation and maintenance data: covering equipment status, fault records, maintenance plans, etc.

[0266] ii. Dispatch data: including ship dispatch, cargo handling, equipment dispatch instructions, etc.

[0267] iii. Environmental data: including external influencing factors such as meteorological conditions, traffic flow, and emergencies.

[0268] b) Layered processing: Data is processed in layers according to business scenarios, time periods, device types, and other dimensions to ensure that the model has sufficient detailed data support when processing different tasks.

[0269] (3) Data semantic conversion and annotation

[0270] a) Data semantic conversion: Convert the processed business data into a semantic format that can be understood by the large language model to ensure that the model can correctly parse the professional terms and operating instructions in port business.

[0271] b) Data annotation: annotate historical data, including safety incidents, dispatch instructions, equipment failure causes, etc. The purpose of annotation is to help the model learn different tasks and their results in port operations.

[0272] 3. Incremental pre-training module

[0273] (1) Incremental corpus update

[0274] a) Update the business data in the corpus based on the processed incremental data so that the model can receive the latest port operation information in real time.

[0275] i. Real-time guarantee: Ensure that incremental data can be quickly transferred to the corpus so that the model can make the latest judgments based on the new data.

[0276] b) New business data knowledge point annotation, for example:

[0277] i. Added device scheduling logic: When device configuration is changed or added, the scheduling logic is updated.

[0278] ii. Emergency event handling strategies: By adding new emergency event data, emergency handling strategies are annotated to help the model learn diverse methods to deal with complex events.

[0279] (2) Start model training and adjust parameters

[0280] a) Model training parameter adjustment: According to the changes in incremental data, adjust the model's training parameters, especially the learning rate, optimization algorithm, etc., to ensure that the model has the best learning effect on the new data. For example:

[0281] i. Setting the learning rate: Set the learning rate based on the importance and quantity of the data to balance the impact of incremental data on the model and avoid excessive interference with historical data.

[0282] b) Model weight update: During the training process of incremental data, the model will adjust the weights based on the new data, especially in the fields of equipment scheduling and emergency handling, to ensure that the model can understand the best handling methods in new scenarios.

[0283] (3) Model verification and evaluation

[0284] a) Model validation set selection: Select key task scenarios in incremental data, establish a model validation set, and evaluate the performance of the model for newly added equipment, scheduling strategies, or abnormal events.

[0285] b) Evaluation criteria setting: The accuracy, response speed and stability of the model are evaluated through incremental data. The evaluation criteria include:

[0286] i. The model’s adaptability to new task scenarios.

[0287] ii. Accuracy of new equipment scheduling.

[0288] iii. The degree of optimization of emergency response time and response strategies.

[0289] c) Verification result feedback: If the model performance is not ideal, return to adjust the model training parameters or re-label the data until the predetermined performance standards are met.

[0290] 4. Fine-tune the module

[0291] (1) Define the mission objectives

[0292] a) Identify the mission objectives that the model needs to optimize in port operations, such as ship scheduling, cargo handling, equipment maintenance, or emergency response.

[0293] b) Set clear performance indicators for each target task, including:

[0294] Accuracy: The accuracy of the model in judging the business process.

[0295] Response time: How fast the model makes decisions in real-time scenarios.

[0296] Resource utilization: the effectiveness of equipment and manpower scheduling.

[0297] (2) Freeze the general model layer

[0298] a) Freeze the base layers related to general language processing in the large model to ensure that the basic language capabilities are not modified.

[0299] b) Only the model layers related to port-specific tasks are trained so that they can focus on performance optimization in specific scenarios and avoid overfitting or affecting the generality of the model.

[0300] (3) Adjusting task-specific layer parameters

[0301] a) Adjust parameters of the layers in the model that handle port-specific tasks:

[0302] Device scheduling layer: Optimize device scheduling logic to ensure efficient allocation of devices and resources and reduce idle time.

[0303] Ship scheduling layer: adjust the ship scheduling algorithm, optimize the ship's entry and exit time, and improve port operation efficiency.

[0304] Emergency management layer: Improve the model's response speed in emergency situations and enhance its ability to respond to emergencies (such as equipment failure and severe weather).

[0305] b) Adjust the learning rate and optimization algorithm to ensure that the model can converge quickly under limited data sets and tasks, thereby improving training efficiency.

[0306] (4) Model evaluation and verification

[0307] a) Use actual port business scenarios to validate the model, including scheduling optimization, equipment management, emergency response and other tasks, and evaluate its performance in actual business.

[0308] b) Evaluation criteria:

[0309] i. Business accuracy: Evaluate the accuracy of the model in task processing in the port business scenario and whether it meets business requirements.

[0310] ii. Real-time evaluation: Test the response time of the model in real-time operations to ensure that it can quickly respond to emergencies and business changes.

[0311] c) Performance feedback: If the model performance does not meet expectations, return to adjust the model parameters (such as learning rate, model layer weights, etc.) and retrain until the expected goal is achieved.

[0312] (5) Model deployment

[0313] a) After verification and evaluation, the fine-tuned model is deployed into the actual port business system.

[0314] b) The model is connected with core systems such as the port dispatching system, equipment management system, and traffic monitoring system to ensure that the model can provide intelligent dispatching and decision-making support in real time.

[0315] c) Push the fine-tuned model to the actual operation platform, such as equipment scheduling, ship management, emergency plan execution and other key links, conduct on-site testing and observe its performance.

[0316] (6) Continuous monitoring and feedback

[0317] a) Real-time monitoring of the model’s performance in actual port operations, with a focus on monitoring its responsiveness and optimization effects in different operational scenarios.

[0318] b) When business processes or scenarios change, timely record feedback to determine whether further fine-tuning or retraining of the model is needed to ensure that the model is continuously optimized and adapted to the dynamic needs of the port.

[0319] c) The model performance monitoring system will continuously record key indicators of model performance and automatically generate optimization suggestion reports for technical personnel to refer to, ensuring the continued stable operation of the model.

[0320] 5. Port Adaptation and Optimization Module

[0321] The port adaptation and optimization module aims to enable the port model to continuously optimize its performance and adaptability according to the port's business needs and environmental changes, thereby improving the port's operating efficiency and ability to cope with complex situations. The following is the specific workflow of this module:

[0322] Step 1: Data monitoring and analysis

[0323] Real-time data collection and input:

[0324] a) Continuously collect real-time data from various data sources such as port equipment, environmental sensors, TOS system, etc., including equipment operating status, operation progress, environmental changes, etc.

[0325] b) After the data is input, the system will compare it with historical data to find the deviation between the current state and the expected model.

[0326] Analysis of key business indicators:

[0327] a) Analyze the core key indicators of the port (such as equipment utilization, operation time, ship turnaround time, truck dispatch efficiency, etc.) and monitor their changing trends.

[0328] b) Evaluate each indicator and identify abnormal situations that affect port operation efficiency or safety.

[0329] Step 2: Model Adaptive Adjustment

[0330] Identify adaptation needs:

[0331] a) Based on the analysis results of real-time data, determine whether there is a need to adjust the model. For example, if it is detected that the efficiency of container truck scheduling has decreased, the model will identify whether the container truck scheduling strategy needs to be adjusted.

[0332] b) Evaluate the impact of changes in external conditions (such as weather and equipment load changes) on model reasoning and decision-making, and determine the direction of adjustment.

[0333] Adaptive Optimization Strategy Development:

[0334] a) Develop adaptive optimization strategies based on the identified adaptation needs and the best practices in the knowledge base.

[0335] b) For example, in the case of adverse weather conditions, optimize the equipment operation speed and scheduling method, and adjust the task allocation of equipment to ensure safety.

[0336] Optimization algorithm application:

[0337] a) Use optimization algorithms (such as genetic algorithms, reinforcement learning, etc.) to simulate and optimize existing operation plans to find the optimal solution for port operations. For example, simulate the impact of different scheduling plans on container loading and unloading efficiency to find the most suitable solution for the current environment.

[0338] b) Compare and analyze the optimized strategy with the original strategy to ensure that the new solution can effectively improve efficiency or reduce risks in actual operations.

[0339] Step 3: Implementation and Feedback Loop

[0340] Optimization plan implementation:

[0341] a) Gradually implement the optimized operation strategy in the port operation process. According to the optimization results of the model, adjust the operation, task allocation and scheduling plan of specific equipment.

[0342] b) During the implementation process, real-time monitoring of feedback data from equipment and business processes will be conducted to ensure that adjustments can achieve the expected improvement effects.

[0343] Implementation effect evaluation:

[0344] a) Evaluate the effectiveness of the optimization strategy by monitoring key indicators after implementation (such as equipment efficiency, operating time, failure rate, etc.). If the indicators are not significantly improved after optimization, re-optimize and adjust.

[0345] b) For the improvements that have been achieved, the system will store the optimization experience in the knowledge base as experience support for similar scenarios in the future.

[0346] Step 4: Self-learning and continuous optimization

[0347] Model updating and self-learning:

[0348] a) The model continuously updates parameters and improves adaptability by self-learning the data in the optimization process. For example, the model improves its prediction and decision-making capabilities for port equipment scheduling by learning from the records of multiple operation optimizations.

[0349] b) The self-learning process incorporates new operational experience, equipment status changes and other data into the model to ensure that the port model has the latest operational knowledge and adaptability.

[0350] Feedback mechanism and continuous improvement:

[0351] a) Establish a continuous feedback mechanism to integrate user operation feedback, system monitoring results and optimization implementation effects into the model for further optimization.

[0352] b) Regularly review the optimized operation process to identify possible areas for improvement. Ensure that the port model continues to improve its adaptability to the port and respond to changes in the port environment and business needs.

[0353] 6. Knowledge base management module

[0354] (1) Knowledge base construction

[0355] a) Knowledge classification and organization: Classify and organize the knowledge information in port business, including: port business processes, equipment management manuals, etc.

[0356] b) Data conversion and structuring: Convert unstructured knowledge data (such as documents and manuals) into structured knowledge points to facilitate model calls.

[0357] (2) Knowledge base update mechanism

[0358] a) Real-time data input: Dynamically update new knowledge in daily port operations, such as equipment maintenance records, scheduling adjustments, etc.

[0359] b) Incremental update: The knowledge base supports an incremental update mechanism to reduce the processing and storage of duplicate data and achieve efficient updates.

[0360] (3) Knowledge base management and optimization

[0361] a) Knowledge base structure optimization: According to different business scenarios (such as equipment maintenance and ship scheduling), the knowledge base structure is adjusted to improve the knowledge call efficiency of the model.

[0362] b) Delete outdated knowledge: Automatically detect and delete outdated or no longer applicable knowledge points to ensure the accuracy of model reasoning.

[0363] (4) Knowledge retrieval and call

[0364] Semantic search optimization: Supports natural language queries and intelligent semantic retrieval, and the model can call the most relevant knowledge based on the context.

[0365] (5) Hierarchical calling: According to the different requirements of the task, the knowledge base calls the knowledge of the relevant levels as needed to ensure the accuracy of task processing.

[0366] (6) Knowledge feedback and optimization

[0367] a) Knowledge feedback mechanism: User feedback information will be used to optimize relevant content in the knowledge base and improve the accuracy of knowledge points.

[0368] b) Knowledge base expansion: User feedback can also drive the expansion of the knowledge base, ensuring that the coverage of the knowledge base content continues to grow.

[0369] (7) Integration of knowledge base and model reasoning

[0370] a) Dynamic knowledge call: During the model reasoning process, relevant knowledge points are called in real time for decision support based on current business needs.

[0371] b) Interaction between knowledge base and reasoning process: During the reasoning process, the knowledge base management module and the reasoning module cooperate with each other to dynamically supplement and optimize the knowledge requirements in the reasoning process.

[0372] (8) Security management and authority control

[0373] a) Hierarchical authority management: Control the access and call of the knowledge base to ensure that different users can only access the knowledge points within their authorized scope.

[0374] b) Data security: Sensitive data in the knowledge base is stored and transmitted in encrypted form to ensure that the data is securely protected during the call and update process.

[0375] 7. Reasoning and human-computer interaction module

[0376] (1) User request input

[0377] a) Input form: Users input questions or instructions through natural language, and the system parses them into task requirements that the model can handle.

[0378] b) Request processing: The system parses the input content and extracts key information, such as query content, device status, task requirements, etc., to ensure that the request can be accurately passed to the model inference module.

[0379] (2) Initiation of the reasoning process

[0380] a) Model reasoning: Based on the user's input, the reasoning module calls relevant knowledge points from the knowledge base for logical reasoning, and generates answers or operational suggestions in combination with real-time business data. The reasoning results may involve specific business scenarios such as port equipment scheduling, cargo management, and emergency response.

[0381] b) Multi-level reasoning: The reasoning module supports multi-level reasoning, first dealing with basic problems, such as basic device status query, and then deepening into complex reasoning tasks, such as scheduling plans for responding to emergencies.

[0382] (3) Interaction result generation

[0383] a) Natural language generation: The inference results are converted into natural language answers that users can understand through the natural language generation module (NLG). For example, when a user inquires about the status of a device, the system converts the inference results into intuitive answers such as "The current device is under maintenance and is expected to resume operation in 2 hours."

[0384] b) Data display and visualization: For reasoning results involving real-time data, equipment status or complex tasks, the system provides visual display (such as charts, progress bars, maps, etc.) to enable users to understand the situation more intuitively.

[0385] (4) User feedback and result correction

[0386] a) User feedback mechanism: Users can provide feedback on the model's reasoning results, marking whether they are satisfied or providing further supplementary information. The system will adjust the next reasoning strategy based on the feedback.

[0387] b) Correction of inference results: If the model inference results do not meet user expectations, the system can automatically adjust the model parameters based on feedback, or submit the feedback information to the knowledge base management module for knowledge point optimization.

[0388] (5) Continuous dialogue and task tracking

[0389] a) Dialogue Management: The module supports multi-round dialogue. Users can continue to ask new questions based on the results of the previous dialogue. The system will track the context to ensure the consistency of the dialogue.

[0390] b) Task tracking and reminders: For more complex tasks (such as equipment scheduling or emergency response), the system can continuously track the progress of the task in the background, and push the task completion status and related reminders to the user in real time through the interactive interface.

[0391] 8. System monitoring and optimization module

[0392] (1) Real-time monitoring

[0393] a) System performance monitoring: Real-time monitoring of the performance indicators of each module of the system, including model inference time, task response time, system load, memory usage, etc., to ensure the stability of the system.

[0394] b) Business data monitoring: Real-time monitoring of various indicators of port operations, such as ship scheduling, equipment utilization rate, etc., to determine whether the model performs normally in actual business scenarios and whether it needs to be optimized.

[0395] c) Abnormal event monitoring: The system will automatically detect sudden abnormalities, such as equipment failure, model reasoning errors, system response delays, etc., and promptly notify the operation and maintenance personnel or automatically start the fault handling procedure.

[0396] (2) Automated optimization mechanism

[0397] a) Model optimization mechanism: The system dynamically adjusts the model parameters and learning rate based on the quality of the inference results and user feedback to ensure that the model can adapt to the changing needs of port operations.

[0398] b) Incremental training: When the system detects new business patterns or frequent emergencies, the system will automatically call the incremental training module to fine-tune the model and improve its adaptability.

[0399] c) Task scheduling optimization: The system automatically optimizes task scheduling plans based on real-time business data. For example, when it detects that the utilization rate of port equipment is uneven, the system will automatically optimize equipment scheduling to ensure efficient use of resources.

[0400] (3) Exception handling and alarm

[0401] a) Anomaly detection and alarm: When the system detects model reasoning errors, task timeouts, or equipment failures, it will issue an alarm notification to prompt relevant personnel to handle the situation. For more serious anomalies, the system will also initiate an emergency response procedure.

[0402] b) Automatic recovery mechanism: The system has a built-in fault recovery mechanism. When an anomaly is detected, the system will automatically try to fix the problem and record the repair process in a log for further analysis and optimization.

[0403] (4) System health analysis

[0404] a) Health indicators: The system generates a system health report by analyzing model performance, business indicators and user feedback. The health report includes the model's reasoning accuracy, response time, error rate and the load of each module of the system.

[0405] b) Trend analysis: The system predicts performance bottlenecks or failures that may occur in the future through trend analysis of historical data, and performs optimization or upgrades in advance to prevent business interruptions.

[0406] (5) Feedback and continuous optimization

[0407] a) Feedback data collection: By collecting user feedback and business data, the system will automatically analyze which business scenarios or tasks need further optimization. For example, if the response time of some reasoning tasks is too long, the system will adjust the reasoning algorithm or hardware resource allocation according to the situation.

[0408] Automated Updates: Regularly and automatically detect new data or model performance issues, update the knowledge base with incremental data, and fine-tune model parameters to ensure that the system remains efficient and stable when processing port operations.

[0409] Specifically, during use, the following points can be used to achieve the system's collaborative optimization and intelligent operation upgrade capabilities:

[0410] 1. Modular design and multi-source data integration: The system consists of multiple modules, including port multi-dimensional data collection, heterogeneous data fusion, entity relationship extraction, efficiency evaluation, and adaptive knowledge graph fusion. These modules work together to form an overall optimization capability. This modular design facilitates cross-module collaborative optimization.

[0411] 2. Knowledge reasoning and optimization supported by large models: The system supports knowledge reasoning and decision optimization modules through large language models, providing full process management from real-time data to deep reasoning, from data collection to contextual knowledge modeling. This feature enables the system to respond flexibly to diverse business scenarios and adapt to the rapidly changing port operation environment, thereby achieving intelligent operation upgrades.

[0412] 3. Adaptive update and conflict resolution: The system can adaptively update and resolve conflicts in the knowledge graph to ensure data consistency and real-time knowledge. These functions enable the knowledge graph to remain effective and efficient when data is constantly updated, which is conducive to the intelligent collaborative management of the entire system.

[0413] Although embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions and variations may be made to the embodiments without departing from the principles and spirit of the present invention, and that the scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. A port vertical large model system based on a general large language model, characterized by: It includes the following modules: data collection module, pre-training corpus construction module, incremental pre-training module, fine-tuning module, port adaptation and optimization module, knowledge base management module, reasoning and human-computer interaction module, and system monitoring and optimization module; The functions of the data acquisition module include multi-source data fusion, real-time data update, and data cleaning and preprocessing. The data of the data acquisition module includes text data, image data, video data, and structured data; The functions of the pre-training corpus construction module include corpus collection and organization, corpus annotation and classification, and data storage and management; The functions of the incremental pre-training module include model initialization, industry knowledge enhancement, performance evaluation and optimization; The functions of the fine-tuning module include fine-tuning data preparation, business process simulation and fine-tuning result verification; The functions of the port adaptation and optimization module include personalized learning of port operations, dynamic optimization and adjustment, cross-port adaptability, continuous optimization of operating processes and equipment operations, and intelligent prediction and decision support; The functions of the knowledge base management module include knowledge base creation and maintenance, knowledge retrieval and query, and real-time knowledge update; The functions of the reasoning and human-computer interaction module include intelligent reasoning, real-time data support and feedback, multi-round dialogue support, personalized user experience and integration of business processes and user feedback; The functions of the system monitoring and optimization module include operation status monitoring and self-feedback optimization mechanism.

2. The port vertical large model system based on the general large language model according to claim 1 is characterized by: The fine-tuning data preparation includes integrating safety data, scheduling efficiency data and actual scenarios.

3. The port vertical large model system based on the general large language model according to claim 1 is characterized by: The business process simulation includes the following steps: Step 1: Establish business process model and data integration, including flowchart drawing, key node definition, data collection and model initialization; Step 2: Security scenario and equipment operation simulation, including security scenario simulation and equipment operation and task allocation simulation; Step 3: Scheduling strategy optimization and model decision-making, including scheduling strategy optimization and model response and automatic decision-making; Step 4: Verification, optimization and real-time application, including simulation result analysis and feedback optimization, real-time application and continuous monitoring.

4. The port vertical large model system based on the general large language model according to claim 1 is characterized by: The verification of the fine-tuning results includes model performance evaluation and result comparison, field testing and feedback collection, and continuous optimization and adaptive adjustment.

5. The method for constructing a port vertical large model system based on a general large language model according to any one of claims 1 to 4, characterized in that: The following steps are involved: S1: First, the data acquisition module is used to obtain the operating status and environmental monitoring data of the equipment and sensors in the port in real time, including operation progress, equipment performance indicators and environmental dynamic parameters, to provide high-quality and comprehensive data support for model training; S2: Then, a pre-trained corpus building module is used to build a rich corpus using multi-source data, covering the work processes and equipment operations of different ports, ensuring the wide applicability of the model; S3: Through the incremental pre-training module, new data is continuously learned and model parameters are dynamically adjusted to ensure the model's latest adaptability to port business and adaptability to different port business scenarios, enhance its generalization ability to changing scenarios, and thus provide more accurate prediction and decision support; S4: At the same time, the fine-tuning module can be used to fine-tune the model according to the needs of specific ports to improve its accuracy in specific scenarios; S5: Analyze the characteristics of different ports through the port adaptation and optimization module, and intelligently adjust the model to adapt to the equipment and operation processes of each port, so as to achieve efficient management; S6: Dynamically maintain a knowledge base that records best practices and operating specifications through the knowledge base management module, providing decision support and intelligent reasoning basis; S7: Through the reasoning and human-computer interaction module, natural language processing technology is used to interact with port staff, provide real-time suggestions and decision support, and enhance human-computer collaboration; S8: Monitor the operating status of the entire system through the system monitoring and optimization module, regularly evaluate the model performance, and ensure the continuous and efficient operation of the system; S9: Through the coordinated operation of the eight modules S1-S8, a port vertical large model system based on a general large language model can be built, which can significantly improve the intelligent management level and operating efficiency of the port and is conducive to the digital transformation of the port industry.

Images