Intelligent ship shore-side collaborative research and development method and system based on group cloud PaaS service

Abstract

The application provides an intelligent ship shore-end collaborative research and development method and system based on a group cloud PaaS service, which relies on network and storage resources provided by the group cloud PaaS service, adopts an MQTT protocol to receive ship end measurement point data and obtain shore end supplementary data, and converges into a multi-source heterogeneous original data set; a two-dimensional data quality monitoring system is constructed to monitor and govern the data quality standards of the generated standardized original data, and output standardized data; a model algorithm file management system and a ship knowledge graph are constructed, the standardized data are associated and bound with model input variables to form a simulation task instance, joint simulation is performed, and the knowledge graph is called to attribute and reason abnormal results; intelligent applications are developed, and running feedback data is input reversely to correct model parameters, trigger incremental learning or update monitoring rules, form an iterative closed loop, can effectively improve the collaborative research and development efficiency, shorten the technical iteration cycle, and provide a quantitative decision basis for green intelligent ship technology research and development.

Description

Technical Field

[0001] This invention relates to the field of green and intelligent ship technology, specifically to an intelligent ship shore-based collaborative R&D method and system based on a group cloud PaaS service. Background Technology

[0002] The research and development of green and intelligent ships faces prominent challenges, including long development cycles, complex processes, numerous stakeholders, and slow feedback on the application effects of green technologies. Early industry focus was on ship-side digital services, ship-to-shore data synchronization, and storage services, failing to fully leverage existing intelligent ship technologies to support technological iteration across all stages of design, construction, equipment, and operation. A unified R&D system was also lacking to guide the safe, efficient, and intelligent development of intelligent ship systems. Furthermore, the failure to break down barriers between multi-source data sources, inefficient model collaboration, and insufficient data sharing security and standardization have led to wasted R&D resources and low collaborative efficiency, hindering the engineering application and industrialization of green and intelligent ship technologies.

[0003] Existing shore-based solutions for ships mainly fall into three categories: independent shore-based data storage systems, single-function data processing tools, and distributed data sharing platforms. Each solution has significant technical shortcomings, as detailed below:

[0004] Independent shore-based data storage systems can only achieve simple reception and storage of ship data, lacking core functions such as data governance and quality monitoring. The data has a low degree of standardization and cannot be directly used for research and development analysis. They have poor compatibility with ship-side transmission protocols, resulting in low data transmission efficiency and insufficient security. Their fixed storage architecture makes it difficult to meet the storage needs of massive amounts of ship data.

[0005] Single-function data processing tools: These tools focus on single aspects such as data cleaning and energy efficiency analysis, resulting in fragmented functions and an inability to achieve a closed-loop process of data aggregation, processing, simulation, and application. They do not support multi-source data fusion processing, and their models and algorithms are closed, making it impossible to collaborate with other R&D tools and leading to low R&D efficiency.

[0006] Distributed data sharing platforms: These platforms use simple file transfer or database integration to achieve data sharing, lacking unified interface standards and security mechanisms; inconsistent data formats can easily lead to data distortion during sharing; the lack of granular permission management and encrypted transmission mechanisms makes it difficult to guarantee data security; and the absence of a data sharing mechanism makes it impossible to define data ownership and usage rights, which can easily lead to disputes of interest.

[0007] Therefore, there is an urgent need for a method and system that can effectively solve the above problems and realize the collaborative R&D of green and intelligent ships on shore. This system can integrate multi-source data, achieve model collaboration, ensure data security and sharing, and form a closed loop for R&D iteration, supporting the rapid iteration and industrialization of green and intelligent ship technology. Summary of the Invention

[0008] To address the problems of weak data governance, insufficient collaboration, poor data sharing, and lack of flexibility in existing green and intelligent ship shore-based collaborative R&D, this invention provides a method for intelligent ship shore-based collaborative R&D based on a group cloud PaaS service. This method integrates ship-side measurement point data with shore-based supplementary data, achieving standardized data governance, multi-disciplinary collaborative simulation, and intelligent application development, forming an iterative optimization closed loop. It features high collaborative efficiency, reliable data quality, and good engineering adaptability. This invention also relates to an intelligent ship shore-based collaborative R&D system based on a group cloud PaaS service.

[0009] The technical solution of the present invention is as follows:

[0010] A collaborative R&D method for intelligent ships on shore based on a group cloud PaaS service, characterized by the following steps:

[0011] Data access and aggregation steps: Relying on the network and storage resources that are dynamically scalable according to business load provided by the Group's cloud PaaS service, the MQTT protocol is used to receive ship-side measurement point data uploaded by at least one green intelligent ship in real time, and at the same time, supplementary shore-side data is obtained; the ship-side measurement point data and shore-side supplementary data are uniformly aggregated to the storage resources of the Group's cloud PaaS service to form a multi-source heterogeneous original dataset.

[0012] Multi-source data standardization and quality governance steps: A unified data channel ID encoding rule is established according to the general international standard for ship data. This rule is then used to standardize the naming and format of the multi-source heterogeneous raw datasets, generating standardized raw data. Based on this multi-source heterogeneous raw dataset, for each ship and each piece of equipment on that ship, the total number of measurement points that should have been received and the number of measurement points that were actually received are calculated. Ship data integrity is calculated based on the actual number of received ship measurement points and the number that should have been received. Similarly, equipment data integrity is calculated based on the actual number of received equipment measurement points and the number that should have been received. A two-dimensional data quality monitoring system is then constructed based on both ship and equipment data integrity. This two-dimensional data quality monitoring system is used to monitor the standardized raw data against data quality standards. For data that does not meet the preset quality standards, data cleaning, data trimming, data synchronization, and data processing are performed sequentially, outputting processed standardized data.

[0013] Collaborative Simulation and Knowledge-Driven Steps: Relying on the elastic computing resources of the Group's cloud PaaS service, a file management system is constructed to store multi-format model algorithm files. These multi-format model algorithm files include multiple ship system simulation models, such as hull fluid model, main engine performance model, exhaust gas treatment model, and navigation control model. A unique file identifier is assigned to each model algorithm file. Fault phenomena, causes, and handling measures extracted from pre-acquired historical fault data are structured using fault tree analysis to construct a ship knowledge graph containing fault nodes, cause nodes, symptom nodes, and their causal relationships. The target model algorithm file is then read from the model algorithm file management system based on the file identifier. According to the physical measurement point location and signal type represented by the unified data channel ID encoding rules, each data field in the standardized data is associated and bound with the input variables of the corresponding physical measurement points in the target model algorithm file to form a simulation task instance. Based on the simulation task instance, collaborative simulation of multiple ship system simulation models is executed, and during the simulation process, the ship knowledge graph is called to perform attribution reasoning on abnormal simulation results, outputting simulation result data.

[0014] The development and iterative closed-loop steps for shore-based intelligent applications are as follows: Based on the standardized data and simulation results, develop intelligent applications including ship energy efficiency analysis, performance prediction, voyage management, and remote operation and maintenance of shipboard equipment; and feed back the feedback data generated during the operation of the intelligent applications to the resource pool of the group's cloud PaaS service, which includes storage resources and elastic computing resources. The feedback data includes model simulation errors, energy efficiency prediction deviations, and data quality monitoring results; use the model simulation errors to correct the model parameters of the model algorithm file; use the energy efficiency prediction deviations to trigger incremental learning or model retraining of the intelligent applications; and use the data quality monitoring results to update the monitoring rules of the dual-dimensional data quality monitoring system or optimize the unified data channel ID encoding rules, thus forming an iterative closed loop of data processing and application optimization.

[0015] Preferably, after the collaborative simulation and knowledge-driven steps, a data sharing step is also included:

[0016] The standardized data and simulation results are shared externally using RESTful API and MQTT modes. The RESTful API mode combines token authentication, role-based access control, and call monitoring to ensure that only authorized users can access standardized data and simulation reports for specific ships or models. The MQTT mode establishes TOPIC naming rules, access authorization policies, quality of service usage principles, and data packet size limits, and pushes standardized data streams and key early warning information from simulation results in real time. Simultaneously, authentication interfaces, customized data sharing interfaces, standardized data channel interfaces, standardized time-series data interfaces, and customized data push interfaces are configured, and AES encrypted transmission and firewall isolation are used to ensure data security.

[0017] Preferably, in the multi-source data standardization and quality governance steps, the specific construction method of the dual-dimensional data quality monitoring system is as follows: pre-setting ship-level integrity thresholds and equipment-level integrity thresholds; when the ship data integrity of a certain ship is lower than the ship-level integrity threshold, the ship is marked as a high-risk object for data loss; when the equipment data integrity of a certain device is lower than the equipment-level integrity threshold, the device is marked as a priority object for data anomaly investigation; and the execution priority of subsequent data cleaning, data trimming, data synchronization, and data processing is dynamically adjusted based on the marking results.

[0018] Preferably, in the collaborative simulation and knowledge-driven step, the attribution reasoning of the ship knowledge graph specifically includes: during the joint simulation of multiple ship system simulation models, real-time monitoring of the intermediate output results of each simulation model; when any intermediate output result exceeds a preset safety threshold, marking the intermediate output result exceeding the preset safety threshold as an abnormal simulation result, and extracting the fault phenomenon corresponding to the abnormal simulation result as a query condition; based on the query condition, matching fault nodes with the same or similar fault phenomena as the query condition in the ship knowledge graph; tracing back the causal relationship of the matched fault nodes to obtain a list of candidate fault causes leading to the abnormal simulation result and a confidence score for each candidate fault cause; and using the list of candidate fault causes and the confidence score as simulation result data.

[0019] Preferably, in the collaborative simulation and knowledge-driven step, the specific implementation method for associating and binding each data field in the standardized data with the input variables of the corresponding physical measurement points in the target model algorithm file is as follows: The target model algorithm file is read from the model algorithm file management system according to the file identifier and a loading operation is performed; after the loading operation, a mapping configuration table is established between the unified data channel ID code and the input variables in the target model algorithm file, wherein the mapping configuration table uses the unified data channel ID code as the key and the memory address of the input variable in the target model algorithm file as the value; according to the physical measurement point location and signal type represented by the unified data channel ID code rule, each data field in the standardized data is written to the memory address of the corresponding input variable in the target model algorithm file according to the mapping configuration table, thereby completing the association and binding, and forming a simulation task instance.

[0020] Preferably, in the closed-loop process of the shore-based intelligent application development and iteration, the incremental learning method is as follows: using historical data of the period when the energy efficiency prediction deviation exceeds the preset deviation threshold as incremental samples, the weight parameters of the ship energy efficiency analysis model in the intelligent application are updated using an online gradient descent algorithm, without reloading the full historical training data.

[0021] The specific method of model retraining is as follows: when the cumulative duration of the energy efficiency prediction deviation exceeding the preset deviation threshold reaches the preset retraining trigger condition, the full amount of historical training data and new data are reloaded to perform complete training on the energy efficiency analysis model in the intelligent application and update all model parameters.

[0022] Preferably, in the closed-loop step of the shore-based intelligent application development and iteration, the calculation method of the model simulation error is as follows: select the simulation result data and the standardized data within the same time window, and calculate the root mean square error of the two; when the root mean square error exceeds the preset error threshold for N consecutive time windows, it is determined that the model parameters need to be corrected, where N is a positive integer greater than or equal to 3.

[0023] Preferably, in the data access and aggregation step, the ship-side measurement data includes main engine speed, fuel consumption, exhaust emission parameters, navigation attitude and equipment operating temperature, and the shore-side supplementary data includes AIS data, meteorological data, historical model data, algorithm parameter data and fault knowledge base data.

[0024] And / or, in the multi-source data standardization and quality governance steps, the unified data channel ID encoding rule includes rule version number, model version number, functional device identifier, attribute label and remarks information; the unified naming and format standardization processing includes original signal list inspection, data model construction, functional device matching, attribute label selection, secondary function matching, signal uniqueness inspection, cross-ship naming consistency inspection and encoding generation and binding; the data cleaning includes removing outliers and filling missing values; the data pruning includes removing redundant fields; the data synchronization includes unifying timestamps; and the data processing includes format conversion and unit unification.

[0025] A smart ship shore-based collaborative R&D system based on a group cloud PaaS service is characterized by comprising, in sequence, a data access and aggregation module, a multi-source data standardization and quality governance module, a collaborative simulation and knowledge-driven module, and a shore-based intelligent application development and iteration closed-loop module.

[0026] The data access and aggregation module relies on the network and storage resources that are dynamically scalable according to the business load provided by the Group's cloud PaaS service. It uses the MQTT protocol to receive ship-side measurement data uploaded by at least one green intelligent ship in real time, and at the same time obtains shore-side supplementary data. The ship-side measurement data and shore-side supplementary data are uniformly aggregated to the storage resources of the Group's cloud PaaS service to form a multi-source heterogeneous original dataset.

[0027] The multi-source data standardization and quality governance module formulates unified data channel ID encoding rules according to the general international standards for ship data. It then uses these rules to uniformly name and format the multi-source heterogeneous raw datasets, generating standardized raw data. Based on these multi-source heterogeneous raw datasets, it calculates the total number of measurement points that should be received and the actual number of measurement points received for each ship and each piece of equipment on that ship. It calculates the ship data integrity based on the actual number of received measurement points and the number that should be received, and calculates the equipment data integrity based on the actual number of received measurement points and the number that should be received. A two-dimensional data quality monitoring system is then constructed based on these two dimensions. This system monitors the standardized raw data for quality standards, and for data that does not meet the preset quality standards, it sequentially performs data cleaning, data trimming, data synchronization, and data processing, outputting the processed standardized data.

[0028] The collaborative simulation and knowledge-driven module, relying on the elastic computing resources of the group's cloud PaaS service, constructs a file management system that stores multi-format model algorithm files. These multi-format model algorithm files include multiple ship system simulation models, such as a hull fluid model, main engine performance model, exhaust gas treatment model, and navigation control model. Each model algorithm file is assigned a unique file identifier. Fault phenomena, causes, and handling measures extracted from pre-acquired historical fault data are structured using fault tree analysis to construct a ship knowledge graph containing fault nodes, cause nodes, symptom nodes, and their causal relationships. The target model algorithm file is then read from the model algorithm file management system based on the file identifier. According to the physical measurement point location and signal type represented by the unified data channel ID encoding rules, each data field in the standardized data is associated and bound with the input variables of the corresponding physical measurement points in the target model algorithm file, forming a simulation task instance. Based on the simulation task instance, multiple ship system simulation models are collaboratively simulated, and during the simulation process, the ship knowledge graph is called to perform attribution reasoning on abnormal simulation results, outputting simulation result data.

[0029] The shore-based intelligent application development and iteration closed-loop module, based on the standardized data and simulation results, develops intelligent applications including ship energy efficiency analysis, performance prediction, voyage management, and remote operation and maintenance of shipboard equipment. The module then feeds back the feedback data generated during the operation of these intelligent applications to a resource pool within the group's cloud PaaS service, which includes storage and elastic computing resources. This feedback data includes model simulation errors, energy efficiency prediction deviations, and data quality monitoring results. The module uses the model simulation errors to correct the model parameters in the model algorithm file, the energy efficiency prediction deviations to trigger incremental learning or model retraining in the intelligent applications, and the data quality monitoring results to update the monitoring rules of the dual-dimensional data quality monitoring system or optimize the unified data channel ID encoding rules, thus forming an iterative closed loop for data processing and application optimization.

[0030] Preferably, after the collaborative simulation and knowledge-driven module, a data sharing module is also included:

[0031] The standardized data and simulation results are shared externally using RESTful API and MQTT modes. The RESTful API mode combines token authentication, role-based access control, and call monitoring to ensure that only authorized users can access standardized data and simulation reports for specific ships or models. The MQTT mode establishes TOPIC naming rules, access authorization policies, quality of service usage principles, and data packet size limits, and pushes standardized data streams and key early warning information from simulation results in real time. Simultaneously, authentication interfaces, customized data sharing interfaces, standardized data channel interfaces, standardized time-series data interfaces, and customized data push interfaces are configured, and AES encrypted transmission and firewall isolation are used to ensure data security.

[0032] The beneficial effects of this invention are as follows:

[0033] This invention provides a method for collaborative R&D of intelligent ships and shore-based systems based on the Group's Cloud PaaS service. Through a standardized data system, flexible architecture design, collaborative simulation mechanism, and secure sharing technology, it achieves full-process collaborative R&D on shore. First, relying on the network and storage resources provided by the Group's Cloud PaaS service, it uses the MQTT protocol to receive ship-side measurement point data in real time and obtain supplementary shore-side data, aggregating them into a multi-source heterogeneous raw dataset. This data access mechanism breaks down the information silos of ship-shore data separation in traditional R&D. By establishing a unified data aggregation channel, it solves the problems of single data sources and delayed updates in green intelligent ship R&D, providing a highly timely data foundation for subsequent full-link R&D analysis. The storage resources of the Group's Cloud PaaS service can dynamically scale elastically with business load, solving the technical defects of traditional fixed storage architecture, such as insufficient storage space during peak access of massive ship time-series measurement point data and idle and wasted storage resources during low-load business phases. Storage capacity can be dynamically allocated according to the number of ships accessing and the frequency of data reporting, ensuring stable and persistent storage of massive ship-shore heterogeneous data while significantly improving cloud resource utilization and reducing the infrastructure deployment and maintenance costs of ship collaborative R&D.

[0034] Then, based on internationally accepted standards for ship data, a unified data channel ID coding rule was established to standardize the naming and formatting of multi-source heterogeneous raw datasets, generating standardized raw data. This achieved standardized representation of multi-source heterogeneous data, effectively solving data silos and comprehension biases, significantly improving data integrity, and providing a unified semantic foundation for subsequent end-to-end data processing, while also adapting to the collaborative needs of different ship types and equipment. Building upon this, a dual-dimensional data quality monitoring system based on "ship-level integrity" and "equipment-level integrity" was constructed. This system dynamically marks "high-risk data missing objects" and "priority data anomaly detection objects" using preset thresholds, guiding subsequent data cleaning, trimming, synchronization, and processing. This refined governance mechanism transforms traditional "passive" data correction into a proactive "prevention-correction" closed loop, significantly improving data credibility and usability. Simultaneously, this specific quality monitoring mechanism, through integrity calculations at both the ship-level and equipment-level dimensions, achieves refined data quality control from macro to micro levels. In particular, by cleaning up outliers, filling in missing values, and synchronizing data to unify timestamps, the numerical accuracy and stability of subsequent simulation calculations and model training are fundamentally guaranteed.

[0035] Furthermore, this invention leverages the elastic computing resources of the group's cloud PaaS service to construct a file management system for storing multi-format model algorithm files. Each model algorithm file is assigned a unique file identifier. Fault phenomena, causes, and handling measures extracted from historical fault data are structured using fault tree analysis to construct a ship knowledge graph containing fault nodes, cause nodes, symptom nodes, and their causal relationships. This specific architecture, combining model algorithm file management with knowledge graph technology, achieves unified management and knowledge reuse of R&D assets. Then, the target model algorithm file is read based on the file identifier and loaded. According to the physical measurement point location and signal type represented by the unified data channel ID encoding rules, each data field in the standardized data is precisely associated and bound with the corresponding input variables of the physical measurement points in the target model algorithm file (establishing a mapping between ID encoding and memory addresses), forming a simulation task instance. This binding mechanism achieves automated connection between standardized data and the simulation model without manual intervention, significantly improving the efficiency of collaborative simulation. Based on simulation task instances, multiple ship system simulation models (hull fluid, main engine performance, etc.) are executed in a collaborative simulation. During the simulation process, the ship knowledge graph is called to perform attribution reasoning on abnormal simulation results. This allows anomalies that occur during the simulation to be diagnosed in a timely manner and the root cause to be traced. This effectively reduces the fault diagnosis time from several hours to minutes, significantly improves the efficiency and accuracy of R&D testing, greatly reduces repetitive work, shortens the R&D cycle, and provides a quantitative basis for model improvement.

[0036] Building upon this foundation, this invention develops intelligent applications based on processed standardized data and simulation results, encompassing ship energy efficiency analysis, performance prediction, voyage management, and remote equipment operation and maintenance, covering the entire spectrum of green and intelligent ship R&D and operation. More importantly, this invention establishes a full-link iterative closed loop: feedback data generated during the operation of the intelligent applications (including model simulation errors, energy efficiency prediction deviations, and data quality monitoring results) is fed back to the resource pool within the group's cloud PaaS service, which includes storage and elastic computing resources. Model simulation errors are used to correct model parameters in the model algorithm file; energy efficiency prediction deviations are used to trigger incremental learning or model retraining in the intelligent applications; and data quality monitoring results are used to update the monitoring rules of the dual-dimensional data quality monitoring system or optimize data channel ID encoding rules. This iterative closed-loop mechanism enables the model to continuously adapt and optimize as the ship's condition evolves and the market environment changes, breaking the inefficient "development-deployment-waiting for feedback-redevelopment" model of traditional R&D and achieving a continuously self-evolving collaborative R&D system.

[0037] The core of this invention lies in constructing a complete collaborative R&D system encompassing "multi-source data aggregation and access + standardized data quality governance + knowledge-driven collaborative simulation + intelligent application development + iterative optimization closed loop." This system thoroughly overcomes the limitations of existing independent shore-based data storage systems, single-function data processing tools, and decentralized data sharing platforms in handling collaborative R&D of green and intelligent ships, achieving a high-efficiency, low-obstacle collaborative development model throughout the entire process. Specifically, it includes the following aspects:

[0038] 1) Breaking down barriers between multiple data sources: This invention receives ship-side measurement data in real time and obtains supplementary shore-side data through the MQTT protocol, and aggregates multiple data sources into a multi-source heterogeneous original dataset, realizing the unified aggregation of ship-side and shore-side data. This provides a complete data foundation for subsequent standardized governance and collaborative simulation, and solves the problem of data dispersion and difficulty in integration in traditional solutions.

[0039] 2) Constructing a standardized data governance system: This invention formulates unified data channel ID coding rules according to the general international standards for ship data, performs unified naming and format standardization processing on multi-source heterogeneous raw datasets, and conducts data quality monitoring and cleaning processing by constructing a "ship + equipment" dual-dimensional data quality monitoring system. This standardized governance system ensures the consistency, integrity and availability of data, quantifies the integrity and credibility of data, and ensures the high quality of input data by dynamically adjusting the priority of data cleaning and processing, providing reliable data support for subsequent high-precision simulation.

[0040] 3) Achieving Knowledge-Driven Collaborative Simulation: This invention constructs a model algorithm file management system, assigning a unique file identifier to each model algorithm file, realizing version control, access management, and full lifecycle traceability of model assets. Simultaneously, it constructs a ship knowledge graph, using fault tree analysis to structure and store fault phenomena, causes, and remedial measures. Furthermore, during simulation, the knowledge graph is invoked for attribution reasoning of abnormal results, achieving deep integration of simulation and knowledge and improving the intelligence level of collaborative simulation. Unlike traditional single-function tools or black-box models, this invention innovatively utilizes causal relationships in the knowledge graph for anomaly attribution reasoning. When anomalies occur in the simulation (such as a decline in main engine performance), the system can automatically associate potential causes such as "fuel filter blockage" and "fuel injector wear" and provide confidence levels, achieving a technological leap from "phenomenon description" to "root cause localization," significantly reducing reliance on expert experience.

[0041] 4) Forming an intelligent application and iterative closed loop: This invention develops various intelligent applications based on standardized data and simulation results, and feeds back the feedback data generated by the application operation to the resource pool of the group's cloud PaaS service for correcting model parameters, triggering incremental learning, updating monitoring rules, or optimizing coding rules. This iterative closed-loop mechanism enables this R&D solution to have the ability to continuously learn and self-evolve, adapting to the needs of ship R&D scenarios under different ship types and sea conditions, significantly improving the accuracy and timeliness of predictions, greatly shortening the technology iteration cycle, and improving collaborative R&D efficiency.

[0042] In summary, this invention organically integrates the five mechanisms mentioned above—multi-source data aggregation and access, standardized data quality governance, knowledge-driven collaborative simulation, intelligent application development, and iterative optimization closed loop—to form a complete collaborative R&D technology system. This multi-mechanism collaborative architecture not only significantly improves the collaborative efficiency and data quality of green and intelligent ship R&D but also possesses excellent adaptability, enabling it to meet the new R&D needs of ammonia-fueled ships, intelligent energy efficiency systems, and other fields, thus promoting the high-end, intelligent, and green transformation of the shipbuilding industry.

[0043] Furthermore, this invention employs a dual transmission mode of Restful API and MQTT to achieve bidirectional sharing and interaction of standardized data and simulation results, accommodating both batch structured data retrieval and real-time streaming data push scenarios, and adapting to different types of data interaction needs in collaborative ship development. The Restful API transmission link introduces token authentication, role-based access control, and call monitoring mechanisms to precisely constrain the data access scope of different R&D entities. Only authorized users can view business data and simulation reports corresponding to specific ships and models, effectively avoiding security risks such as unauthorized data access and arbitrary leakage. The MQTT link, through standardized TOPIC naming rules, access authorization policies, service quality application principles, and data packet size limits, ensures low-latency and stable delivery of massive measurement point data streams and simulation early warning information, and can also securely receive externally issued simulation scheduling instructions, achieving bidirectional and controllable interaction between upstream and downstream businesses. By unifying the configuration of authentication, data sharing, time-series transmission, and message push through standardized interfaces, it can quickly connect to multiple R&D business systems, reducing the development costs of cross-entity system integration. Coupled with dual protection measures of AES encrypted transmission and firewall isolation, a security protection system is built from the transmission link and network boundary levels, comprehensively ensuring the transmission security of ship-shore multi-source R&D data during bidirectional sharing and interaction. This solves the technical drawbacks of traditional ship R&D data sharing methods, such as single access control, lack of interaction security, and cumbersome multi-party system integration, achieving efficient collaborative data interaction while building a solid data security protection barrier.

[0044] This invention also relates to an intelligent ship shore-based collaborative R&D system based on a group cloud PaaS service. This system corresponds to the aforementioned intelligent ship shore-based collaborative R&D method based on a group cloud PaaS service. It can be understood as a system that implements the aforementioned intelligent ship shore-based collaborative R&D method based on a group cloud PaaS service. It includes a data access and aggregation module, a multi-source data standardization and quality governance module, a collaborative simulation and knowledge-driven module, and a shore-based intelligent application development and iterative closed-loop module, connected sequentially. These modules work collaboratively, resolving the multi-source data barrier problem by aggregating ship-side measurement point data and shore-based supplementary data. Through standardization governance and data quality monitoring, it solves the problems of low data standardization and uncontrollable quality. Through model algorithm file management and knowledge graph construction, it achieves model asset management and knowledge-driven collaborative simulation. Through intelligent application development and feedback data reverse input, it forms an iterative optimization closed loop. The modules are interconnected and progressively advance, constructing a full-link collaborative R&D system from data access to iterative optimization. Attached Figure Description

[0045] Figure 1 This is a flowchart of the intelligent ship shore-based collaborative R&D method based on the group's cloud PaaS service.

[0046] Figure 2 This is a diagram of the intelligent ship shore-based collaborative R&D system architecture based on the group's cloud PaaS service. Detailed Implementation

[0047] The present invention will now be described with reference to the accompanying drawings.

[0048] This invention relates to a collaborative R&D method for intelligent ships based on a group cloud PaaS service. Due to problems such as multi-source data barriers, inefficient model collaboration, insufficient sharing security, and long technology iteration cycles in the R&D process of green intelligent ships, this invention combines four steps: data access and aggregation, multi-source data standardization and quality governance, collaborative simulation and knowledge-driven approach, and intelligent application development and iterative closed-loop. It analyzes and processes ship-side measurement point data and shore-side supplementary data to obtain standardized data and simulation results, forming an iterative optimization closed loop to achieve collaborative R&D of green intelligent ships. This method can effectively improve collaborative R&D efficiency, shorten the technology iteration cycle, and provide a quantitative decision-making basis for the R&D of green intelligent ship technologies. The flowchart of this method is as follows: Figure 1 As shown, the steps are as follows:

[0049] I. Data Access and Aggregation Steps: Relying on the network and dynamically scalable storage resources provided by the Group's Cloud PaaS service, the system uses the MQTT protocol to receive real-time data from ship-side measurement points uploaded by at least one green and intelligent vessel, while simultaneously acquiring supplementary shore-side data. The ship-side measurement point data includes engine speed, fuel consumption, exhaust emission parameters, navigation attitude, and equipment operating temperature, while the shore-side supplementary data includes AIS data and meteorological data. The ship-side measurement point data and shore-side supplementary data are then aggregated to the storage resources of the Group's Cloud PaaS service to form a multi-source heterogeneous raw dataset.

[0050] This step aims to unify and aggregate multi-source data scattered across shipboard and shore-based environments, laying a data foundation for subsequent standardized governance. Specifically, firstly, relying on the network provided by the group's cloud PaaS service and storage resources that dynamically scale with business load, data from shipboard measurement points uploaded by at least one green intelligent vessel (shipboard) is received in real-time via the MQTT protocol, while supplementary shore-based data is also acquired. The shipboard measurement point data includes main engine speed, fuel consumption, exhaust emission parameters, navigation attitude, and equipment operating temperature. The shore-based supplementary data includes AIS data, meteorological data (such as wind speed, wind direction, wave height, and ocean currents), historical model data (such as historical versions of hull fluid models and main engine performance models), algorithm parameter data (such as energy efficiency analysis algorithm parameters and fuel consumption prediction algorithm parameters), and fault knowledge base data (such as fault phenomena, fault causes, and handling measures). Then, the shipboard measurement point data and shore-based supplementary data are aggregated into a multi-source heterogeneous raw dataset.

[0051] II. Multi-Source Data Standardization and Quality Governance Steps: A unified data channel ID encoding rule is established according to international standards for ship data. This rule is then used to standardize the naming and formatting of the multi-source heterogeneous raw datasets, generating standardized raw data. Based on this dataset, for each ship and each piece of equipment on board, the total number of measurement points that should have been received and the actual number of measurement points received are calculated. Ship data integrity is calculated based on the actual number of received measurement points and the number that should have been received. Similarly, equipment data integrity is calculated based on the actual number of received measurement points and the number that should have been received. A two-dimensional data quality monitoring system is then constructed based on both ship and equipment data integrity. This system monitors the standardized raw data against data quality standards. For data that does not meet the preset quality standards, data cleaning, data trimming, data synchronization, and data processing are performed sequentially, outputting the processed standardized data.

[0052] This step aims to break down barriers between multi-source data, unify data standards, and ensure the correctness, timeliness, accuracy, completeness, relevance, and accessibility of data through dual-dimensional monitoring. Specifically, firstly, a unified data channel ID coding rule is established according to common international standards for ship data (such as ISO 19848). This coding rule includes a rule version number, model version number, functional device identifier, attribute label, and remarks. Then, the multi-source heterogeneous raw datasets are processed according to this coding rule to unify naming and format standardization, generating standardized raw data. That is, the following eight unified naming and format standardization checks are performed sequentially:

[0053] 1) Check the original signal list to verify that all data is within the preset signal list;

[0054] 2) Data model construction: Define the data structure (field name, data type, unit, value range) for each type of data.

[0055] 3) Functional device matching: Associate each piece of data with the corresponding functional device (such as host, generator, exhaust gas treatment device).

[0056] 4) Attribute label selection: Select matching labels (such as RPM speed, TEMP temperature, PRESS pressure) from the standard attribute label library.

[0057] 5) Secondary function matching: Perform secondary verification on the association results to correct matching errors;

[0058] 6) Signal uniqueness check to ensure that the same device has only one code for the same attribute to avoid duplication;

[0059] 7) Cross-ship naming consistency check: verify whether the codes for the same equipment and the same attributes are consistent across different ships;

[0060] 8) Encoding generation and binding: Encoding generation and binding are performed according to the established unified data channel ID encoding rules. A unique data channel ID is generated for each type of data and the ID is bound to the corresponding data, thereby completing the unified naming and format standardization of data, generating standardized raw data, and completely solving the problems of data barriers and misunderstandings.

[0061] Then, based on the multi-source heterogeneous original dataset, the actual number of ship measurement points received by each ship is counted, and the ship data integrity is calculated by combining this with the preset number of ship measurement points that the ship should receive (ship data integrity = actual number of ship measurement points received / preset number of ship measurement points received × 100%). Similarly, the actual number of device measurement points received by each device on the ship is counted, and the device data integrity is calculated by combining this with the preset number of device measurement points that the device should receive (device data integrity = actual number of device measurement points / preset number of device measurement points received × 100%). A two-dimensional data quality monitoring system is constructed based on ship data integrity and device data integrity. The specific construction method of the two-dimensional data quality monitoring system is as follows:

[0062] Pre-set ship-level integrity thresholds (e.g., 95%) and equipment-level integrity thresholds (e.g., 98%). When the ship data integrity of a vessel is lower than the ship-level integrity threshold, the vessel is marked as a high-risk object for data loss. When the equipment data integrity of a device is lower than the equipment-level integrity threshold, the device is marked as a priority object for data anomaly investigation. Based on the marking results, dynamically adjust the execution priority of subsequent data cleaning, data trimming, data synchronization, and data processing to ensure that the data of high-risk objects is processed first.

[0063] Finally, a two-dimensional data quality monitoring system is used to monitor the quality standards of the standardized raw data. The monitored indicators include data correctness, timeliness, accuracy, completeness, relevance, and accessibility. For data that does not meet the preset quality standards, the following processes are performed sequentially: data cleaning (removing outliers and filling in missing values, using linear interpolation or previous value imputation), data pruning (removing duplicate or meaningless redundant fields), data synchronization (unifying timestamps to UTC millisecond-level time base), and data processing (converting the format to JSON, standardizing units to MPa, ℃, etc.). The processed standardized data is then output. This standardized data is stored in the GaussDB distributed database and simultaneously written to a Redis cache.

[0064] III. Collaborative Simulation and Knowledge-Driven Steps: Relying on the elastic computing resources of the Group's cloud PaaS service, a file management system is constructed to store multi-format model algorithm files. These multi-format model algorithm files include multiple ship system simulation models, such as hull fluid model, main engine performance model, exhaust gas treatment model, and navigation control model. A unique file identifier is assigned to each model algorithm file. Fault phenomena, causes, and handling measures extracted from pre-acquired historical fault data are structured using fault tree analysis to construct a ship knowledge graph containing fault nodes, cause nodes, symptom nodes, and their causal relationships. The target model algorithm file is then read from the model algorithm file management system based on the file identifier and loaded. According to the physical measurement point location and signal type represented by the unified data channel ID encoding rule, each data field in the standardized data is associated and bound with the input variables of the corresponding physical measurement points in the target model algorithm file to form a simulation task instance. Based on the simulation task instance, collaborative simulation of multiple ship system simulation models is executed, and during the simulation process, the ship knowledge graph is called to perform attribution reasoning on abnormal simulation results, outputting simulation result data.

[0065] This step aims to achieve unified management of model assets through a model algorithm file management system, enable fault diagnosis and attribution reasoning through knowledge graphs, and achieve automatic integration of standardized data and simulation models through data binding. Specifically, it first constructs a model algorithm file management system based on a three-layer architecture: "metadata layer - file entity layer - description and association layer." The metadata layer describes the author, version, creation time, and dependencies of the files; the file entity layer stores multi-format model algorithm files, including Simulink model files, Python algorithm script files, and FMU co-simulation module files, through a distributed file system; and the description and association layer records the relationships between files and between files and data. Based on this, a metadata model is constructed using the ISO 10303 (STEP) standard to ensure that model information can be exchanged between different systems. Each model algorithm file is assigned a unique file identifier, and the version number, author, creation time, dependencies and access permissions of each file are recorded. It also supports version control (generating a new version number with each modification and retaining historical versions), permission management (assigning permissions according to three levels: read-only, edit, and administrator, and authorizing according to user roles), collaborative editing (multiple users can edit different modules at the same time, and automatically or manually merge when there are conflicts), and full lifecycle traceability (recording the entire process of creation, modification, use and disposal of each file).

[0066] Then, fault tree analysis (FTA) is used to perform structured modeling of the fault phenomena, causes, and handling measures extracted from pre-acquired historical fault data, constructing a ship knowledge graph containing fault nodes, cause nodes, symptom nodes, and their causal relationships. Specifically, based on Fault Tree Analysis (FTA), typical ship fault patterns (i.e., fault phenomena, causes, and handling measures) are extracted from pre-acquired historical fault data, such as excessively high main engine cylinder liner temperature, decreased exhaust gas treatment efficiency, and abnormally high fuel consumption. Based on this, fault nodes, cause nodes, symptom nodes, and their causal relationships are identified, and these nodes (as vertices of the graph) and causal relationships (as edges of the graph) are stored in the Neo4j graph database, thus constructing the ship knowledge graph. Subsequently, a web-based question-and-answer system is developed based on the Flask framework, supporting users to query for fault diagnosis and maintenance decisions using natural language queries (e.g., "What might be the cause of excessively high main engine temperature?") or structured queries. After receiving user input, the question-and-answer system parses the user's intent, queries the Neo4j graph database, and finally returns fault diagnosis results and maintenance suggestions.

[0067] Next, the target model algorithm file is read from the model algorithm file management system based on the file identifier and a loading operation is performed (i.e., the target model algorithm file is loaded into the K8S container cloud environment based on the file identifier). Based on the physical measurement point location and signal type represented by the unified data channel ID encoding rules (physical measurement point location includes equipment identifier and measurement point identifier; signal type includes speed signal, temperature signal, pressure signal, and flow signal), each data field in the processed standardized data is associated and bound with the corresponding input variables of the physical measurement points in the target model algorithm file to form a simulation task instance.

[0068] The specific implementation of the association binding is as follows: The target model algorithm file is read from the model algorithm file management system according to the file identifier and a loading operation is performed. After the loading operation, a mapping configuration table is established between the unified data channel ID code and the input variables in the target model algorithm file. The mapping configuration table uses the unified data channel ID code as the key and the memory address of the input variable in the target model algorithm file as the value. Based on the physical measurement point location and signal type represented by the unified data channel ID code rules, each data field in the processed standardized data is written to the memory address of the corresponding input variable in the target model algorithm file according to the mapping configuration table, thereby completing the association binding, forming a simulation task instance, and thus achieving seamless data and model integration.

[0069] Finally, based on simulation task instances, multiple ship system simulation models are used for collaborative simulation. Under the premise of protecting core intellectual property rights (by encapsulating the core algorithm as a black box and employing model encryption technology), data exchange between multi-source heterogeneous models is achieved through standardized interfaces, enabling system-level performance virtual simulation or semi-physical simulation. During the simulation process, a ship knowledge graph is invoked to perform attribution reasoning on abnormal simulation results. Specifically, the attribution reasoning method is as follows: during the joint simulation of multiple ship system simulation models, the intermediate output results of each simulation model are monitored in real time. When any intermediate output result (such as exhaust gas emission concentration or main engine temperature) exceeds a preset safety threshold, the intermediate output result exceeding the threshold is marked as an abnormal simulation result, and the corresponding fault phenomenon is extracted as a query condition. Based on the query condition, fault nodes with the same or similar fault phenomena as the query condition are matched in the ship knowledge graph. The causal relationship of the matched fault nodes is traced back to the cause node to obtain a list of candidate fault causes leading to the abnormal simulation result and a confidence score for each candidate fault cause. The list of candidate fault causes and the confidence scores are output as simulation result data to assist researchers in root cause analysis. The simulation results include the performance curves of each system (the actual ship systems corresponding to multiple ship system simulation models, namely the hull system, main engine system, exhaust gas treatment system, navigation control system, etc.), the overall ship energy efficiency index, and the simulation results of failure modes.

[0070] IV. Data Sharing Steps: The standardized data and simulation results data are shared externally using a dual-mode approach of Restful API and MQTT.

[0071] Specifically, the RESTful API mode is suitable for scenarios with high data integrity requirements and low interaction frequency (such as report generation and offline analysis). It adopts JSON data format and HTTPS interaction protocol, combined with token authentication, RBAC role-based access control and call monitoring, and supplemented by rate limiting and circuit breaking mechanisms to ensure that only authorized users (such as specific designers or maintenance experts) can access standardized data and simulation reports of specific ships or models.

[0072] MQTT mode: Suitable for scenarios with high real-time requirements (such as remote real-time monitoring and dynamic decision-making). It establishes strict TOPIC naming rules (e.g., vessel / {IMO} / realtime), access authorization policies, QoS usage principles (QoS=0 for non-critical data, QoS=1 for general data, QoS=2 for critical security data), and packet size limits (single packet not exceeding 64KB; exceeding this limit results in split packets). It pushes standardized data streams and key warning information from simulation results in real time.

[0073] At the same time, a multi-layered security protection system has been constructed:

[0074] At the interface level: Configuration includes an authentication interface (for user login and token issuance and refresh), a customized data sharing interface (for customizing data content according to user needs), a standardized data channel interface (for acquiring data via ISO19848 encoded channels), a standardized time-series data interface (for acquiring time-series data within a time range), and a customized data push interface (for proactively pushing data to subscribers). Furthermore, each standardized interface supports cross-system compatibility, enabling "one-time data collection, multi-party sharing and reuse," significantly improving the utilization rate of R&D resources.

[0075] At the transmission level: The transmitted content is encrypted using the AES encryption algorithm (using the AES-256 algorithm) to prevent data from being stolen during transmission; a firewall is used for network isolation to ensure data security, and audit logs are recorded for security traceability.

[0076] At the ownership level: Clearly define the data ownership and sharing permissions for five roles to avoid stakeholders bearing risks and losses. Specifically, platform users (ship owners) own the raw data generated by their vessels; platform providers (platform operators) own the metadata and governed data generated by the platform; solution providers own the models and algorithms they develop; solution users only access data within the scope of their licenses; and data purchasers obtain limited usage rights according to the purchase agreement. For example, a ship owner owns the raw data, but through platform authorization, can share standardized data and simulation results for a specific time period with equipment suppliers for fault diagnosis; the permissions are automatically revoked after the diagnosis is completed.

[0077] V. Shore-based Intelligent Application Development and Iterative Closed-Loop Steps: Based on the standardized data and simulation results, develop intelligent applications including ship energy efficiency analysis, performance prediction, voyage management, and remote operation and maintenance of shipboard equipment; and feed back the feedback data generated during the operation of the intelligent applications to the resource pool of the group's cloud PaaS service, which includes storage resources and elastic computing resources. The feedback data includes model simulation errors, energy efficiency prediction deviations, and data quality monitoring results; use the model simulation errors to correct the model parameters of the model algorithm file; use the energy efficiency prediction deviations to trigger incremental learning or model retraining of the intelligent applications; and use the data quality monitoring results to update the monitoring rules of the dual-dimensional data quality monitoring system or optimize the unified data channel ID encoding rules, forming an iterative closed loop of data processing and application optimization.

[0078] This step aims to develop various intelligent applications based on data and simulation results, and to input operational feedback data in reverse to form an iterative closed loop, enabling the system to continuously evolve. Specifically, based on the standardized data and simulation results, at least one intelligent application is developed. These applications include: single-ship energy efficiency and consumption analysis (based on function fitting and LSTM algorithms to predict fuel consumption and power), multi-ship energy efficiency and consumption comparison analysis (using NARX network to deeply fuse time-series operational data and interference factors), ship performance prediction (including dynamic prediction of main engine performance and pollution assessment), voyage data management, voyage segment analysis, route calculation, CII level assessment (calculating CII rating according to IMO regulations and providing improvement suggestions), and remote operation and maintenance of marine exhaust gas aftertreatment equipment (using a three-layer architecture of perception layer, network layer, and application layer to achieve real-time monitoring of equipment operating status, fault diagnosis, and operation and maintenance decisions).

[0079] Then, the feedback data generated during the operation of the intelligent application is fed back to the resource pool of the group's cloud PaaS service, which includes storage resources and elastic computing resources. The feedback data includes model simulation errors, energy efficiency prediction deviations, and data quality monitoring results. The model simulation error is calculated as follows: the simulation result data and the processed standardized data (such as actual fuel consumption and simulated fuel consumption) within the same time window are selected, and the root mean square error (RMSE) of the two is calculated; when the root mean square error exceeds the preset error threshold for N consecutive time windows (N is a positive integer greater than or equal to 3, for example, N=5), it is determined that the model parameters need to be corrected, and the model parameter calibration procedure is automatically triggered.

[0080] Furthermore, the incremental learning method involves using historical data from periods when energy efficiency prediction deviations exceed a preset deviation threshold as incremental samples. The online gradient descent (OGD) algorithm is then used to update the weight parameters of the energy efficiency analysis model in the intelligent application, without reloading the entire historical training data. This achieves lightweight, real-time model iteration. The incremental learning approach using online gradient descent eliminates the need to load the entire historical training dataset; it only relies on incremental samples from periods with deviations exceeding the threshold to update the model weight parameters. This significantly reduces computational overhead and data reading time during model iteration, making it suitable for parallel multi-ship applications on cloud platforms. It enables lightweight, high-frequency, real-time fine-tuning of the energy efficiency analysis model, allowing it to quickly correct prediction deviations following changes in ship navigation conditions and continuously improve the accuracy of short-term energy efficiency forecasts.

[0081] Furthermore, the specific method of model retraining is as follows: when the cumulative duration of the energy efficiency prediction deviation exceeding the preset deviation threshold reaches the preset retraining trigger condition, all historical training data and newly added data are reloaded to perform complete training on the energy efficiency analysis model in the intelligent application, updating all model parameters. As a fallback optimization strategy, the model retraining mechanism performs complete training based on all historical data combined with newly added business data when the cumulative duration of deviation exceeding the limit reaches the trigger condition. This fully learns the long-term operating condition distribution characteristics, corrects the cumulative model offset problem, and compensates for the insufficient global fitting ability caused by incremental learning's local parameter fine-tuning. Through the dual-mode iterative mechanism of incremental lightweight iteration + full retraining, both real-time model update efficiency and long-term prediction accuracy are balanced. While saving cloud platform elastic computing resources, it continuously ensures the output reliability of intelligent applications such as ship energy efficiency analysis and performance prediction, achieving efficient and stable closed-loop optimization iteration of the application model.

[0082] Finally, the model parameters in the model algorithm file are corrected using the simulation error. Incremental learning or model retraining of intelligent applications is triggered by energy efficiency prediction deviations. The monitoring rules of the dual-dimensional data quality monitoring system are updated or the unified data channel ID encoding rules are optimized using data quality monitoring results, forming an iterative closed loop of data processing and application optimization. This creates a complete iterative closed loop encompassing data access, standardized governance, collaborative simulation, intelligent application development, feedback collection, iterative optimization, and re-accessing data. This continuously improves the efficiency of shore-based collaborative R&D, helps address IMO low-carbon regulations, and promotes the high-end, intelligent, and green transformation of the shipbuilding industry.

[0083] This invention also relates to an intelligent ship shore-based collaborative R&D system based on a group cloud PaaS service. This system corresponds to the aforementioned intelligent ship shore-based collaborative R&D method based on a group cloud PaaS service and can be understood as a system that implements the aforementioned method. Figure 2As shown, this invention relies on the group's cloud PaaS service to build shore-based services, adopting a cloud PaaS deployment model. It configures cloud service resources such as GaussDB distributed database, Redis cache, RabbitMQ message queue, K8S container cloud, and ELK operation and maintenance log analysis, constructing a four-layer architecture from an engineering deployment perspective: "Data Source Access Layer - Basic Tools and Resources Layer - Core Function Application Layer - Collaborative R&D Application Layer". The data source access layer aggregates various types of business data sources, specifically including ship-side data, AIS and other meteorological shore-side data, simulation models, algorithm programs, and knowledge base data. Data is uploaded via ship-to-shore transmission and shore-side access, serving as the data input source. The basic tools and resources layer, relying on the group's cloud PaaS service, provides infrastructure services such as network, K8s environment, GaussDB cloud storage, and Redis cloud cache, providing hardware and cloud environment support for all upper-layer business functions, including elastic computing power, distributed storage, and cache scheduling. The core function application layer includes system management (including operation management, data interface application service management, monitoring reports, data list management, log management, data access control, and user management), data aggregation, and other functions. The system comprises six major backend support modules: storage design, data storage, multi-source access, and heterogeneous data; data monitoring (including ship-side system equipment status, ship-side signal acquisition status, ship-to-shore transmission status, and ship-side data quality monitoring); data governance (including standardized naming, data cleaning, data processing, and data management requirements); data interaction and services (including secure interaction, data services, interface services, and application services); and visualization and access (including portal integration, identity authentication, and single sign-on). These modules provide underlying capabilities to support front-end business process processing. The collaborative R&D application layer provides basic applications such as decision support, data collaboration, model collaboration, innovative applications, algorithm collaboration, and knowledge collaboration, supporting the R&D, deployment, and iterative operation of various innovative intelligent applications on ships and shore.

[0084] In terms of engineering deployment, the ship-side system adopts industrial-grade hardware and redundant design to adapt to the harsh marine environment, demonstrating promising prospects for engineering applications. The shore-side system relies on the group's cloud PaaS service for elastic scaling, supporting large-scale data storage and high-concurrency access. Meanwhile, the R&D system employs a modular design, with each functional module capable of independent upgrades, adapting to new R&D needs such as ammonia-fueled ships and intelligent energy efficiency systems, exhibiting excellent engineering adaptability and scalability.

[0085] From the perspective of protecting business processes, the R&D system is abstracted into five sequentially connected business function modules. These modules operate based on the backend support components within the aforementioned four-layer architecture. Specifically, they include a data access and aggregation module, a multi-source data standardization and quality governance module, a collaborative simulation and knowledge-driven module, a data sharing module, and a shore-based intelligent application development and iteration closed-loop module, all connected sequentially. The correspondence between each business function module and the backend support components, as well as their specific functional limitations, are as follows:

[0086] The data access and aggregation module relies on the network and dynamically scalable storage resources provided by the group's cloud PaaS service. It uses the MQTT protocol to receive ship-side measurement data uploaded by at least one green intelligent vessel in real time, and simultaneously acquires supplementary shore-side data. The ship-side measurement data and shore-side supplementary data are then uniformly aggregated to the storage resources of the group's cloud PaaS service to form a multi-source heterogeneous raw dataset. This business module is mainly implemented based on the data source access layer in the four-layer architecture and the data aggregation module and data monitoring module under the core function application layer.

[0087] The multi-source data standardization and quality governance module formulates unified data channel ID encoding rules according to the general international standards for ship data. It then uses these rules to standardize the naming and formatting of the multi-source heterogeneous raw datasets, generating standardized raw data. Based on these datasets, it calculates the total number of measurement points that should be received and the actual number of measurement points received for each ship and each piece of equipment on board. It calculates the ship data integrity based on the actual number of received measurement points and the number that should be received, and calculates the equipment data integrity based on the actual number of received measurement points and the number that should be received. A two-dimensional data quality monitoring system is then constructed based on these two dimensions. This system monitors the standardized raw data for quality standards, and for data that does not meet the preset quality standards, it sequentially performs data cleaning, data trimming, data synchronization, and data processing, outputting the processed standardized data. This business module relies on the data governance module of the core functional application layer.

[0088] The collaborative simulation and knowledge-driven module, relying on the elastic computing resources of the group's cloud PaaS service, constructs a file management system that stores multi-format model algorithm files. These multi-format model algorithm files include multiple ship system simulation models, such as a hull fluid model, main engine performance model, exhaust gas treatment model, and navigation control model. Each model algorithm file is assigned a unique file identifier. Fault phenomena, causes, and handling measures extracted from pre-acquired historical fault data are structured using fault tree analysis to construct a ship knowledge graph containing fault nodes, cause nodes, symptom nodes, and their causal relationships. Then, based on the file identifiers, the model... The algorithm file management system reads the target model algorithm file; based on the physical measurement point location and signal type represented by the unified data channel ID encoding rule, it associates and binds each data field in the standardized data with the input variables of the corresponding physical measurement points in the target model algorithm file to form a simulation task instance; based on the simulation task instance, it performs collaborative simulation of multiple ship system simulation models, and calls the ship knowledge graph to perform attribution reasoning on abnormal simulation results during the simulation process, and outputs simulation result data; this module runs on the elastic computing power environment of the basic tools and resource layer, and calls the model collaboration and data collaboration capabilities of the collaborative R&D application layer to complete multi-system joint simulation and fault knowledge reasoning analysis.

[0089] The data sharing module employs RESTful API and MQTT modes to share standardized data and simulation results externally. The RESTful API mode combines token authentication, role-based access control, and call monitoring to ensure that only authorized users can access standardized data and simulation reports for specific ships or models. The MQTT mode establishes TOPIC naming rules, access authorization policies, quality of service principles, and data packet size limits, pushing standardized data streams and key warning information from simulation results in real time. It also configures authentication interfaces, customized data sharing interfaces, standardized data channel interfaces, standardized time-series data interfaces, and customized data push interfaces, and combines AES encrypted transmission and firewall isolation to ensure data security. This module relies on the core functional application layer's data interaction and service modules, visualization and access modules to achieve bidirectional data sharing and interaction between multiple R&D entities through secure interaction and various standardized business interfaces, and combines identity authentication and access control to achieve secure data distribution and access management.

[0090] The shore-based intelligent application development and iteration closed-loop module, based on the standardized data and simulation results, develops intelligent applications including ship energy efficiency analysis, performance prediction, voyage management, and remote operation and maintenance of shipboard equipment. It then feeds back the feedback data generated during the operation of these intelligent applications to a resource pool within the group's cloud PaaS service, including storage and elastic computing resources. This feedback data includes model simulation errors, energy efficiency prediction deviations, and data quality monitoring results. The module uses the model simulation errors to correct the model parameters in the model algorithm file, the energy efficiency prediction deviations to trigger incremental learning or model retraining in the intelligent applications, and the data quality monitoring results to update the monitoring rules of the dual-dimensional data quality monitoring system or optimize the unified data channel ID encoding rules, forming an iterative closed loop of data processing and application optimization. This module leverages the innovative application capabilities of the collaborative R&D application layer to complete business system development and feeds back the data to optimize the entire data and model governance rules, achieving system iterative upgrades.

[0091] This invention provides an objective and scientific method and system for collaborative R&D of intelligent ships on shore, based on the group's cloud PaaS service. By combining multi-source data access, standardized governance, knowledge-driven collaborative simulation, intelligent application development, and iterative optimization in a closed loop, a complete collaborative R&D technology system is constructed. While ensuring reliable data quality, high collaborative efficiency, and strong secure sharing capabilities, it systematically improves the technological iteration efficiency of green and intelligent ship R&D, providing a feasible, scalable, and reliable new solution for collaborative R&D across all stages of ship design, construction, equipment, and operation.

[0092] It should be noted that the specific embodiments described above enable those skilled in the art to more fully understand the present invention, but do not limit the present invention in any way. Therefore, although the present invention has been described in detail with reference to the accompanying drawings and embodiments, those skilled in the art should understand that modifications or equivalent substitutions can still be made to the present invention. In short, all technical solutions and improvements that do not depart from the spirit and scope of the present invention should be covered within the protection scope of the present invention patent.

Claims

1. A method for collaborative R&D of intelligent ships on shore based on group cloud PaaS services, characterized in that, Includes the following steps: Data access and aggregation steps: Relying on the network and storage resources that are dynamically scalable according to business load provided by the Group's cloud PaaS service, the MQTT protocol is used to receive ship-side measurement point data uploaded by at least one green intelligent ship in real time, and at the same time, supplementary shore-side data is obtained; the ship-side measurement point data and shore-side supplementary data are uniformly aggregated to the storage resources of the Group's cloud PaaS service to form a multi-source heterogeneous original dataset. Multi-source data standardization and quality governance steps: A unified data channel ID encoding rule is established according to the general international standard for ship data. This rule is then used to standardize the naming and formatting of the multi-source heterogeneous raw datasets, generating standardized raw data. Based on this multi-source heterogeneous raw dataset, for each ship and each piece of equipment on board, the total number of measurement points that should have been received and the number of measurement points actually received are calculated. Ship data integrity is calculated based on the actual number of received ship measurement points and the number of ship measurement points that should have been received. Similarly, equipment data integrity is calculated based on the actual number of received equipment measurement points and the number of equipment measurement points that should have been received. A two-dimensional data quality monitoring system is then constructed based on both ship data integrity and equipment data integrity. The standardized raw data is monitored by a two-dimensional data quality monitoring system. For data that does not meet the preset quality standards, data cleaning, data trimming, data synchronization and data processing are performed in sequence to output the processed standardized data. Collaborative simulation and knowledge-driven steps: Relying on the elastic computing resources of the group's cloud PaaS service, a file management system is constructed to store multi-format model algorithm files. These multi-format model algorithm files include multiple ship system simulation models, such as hull fluid model, main engine performance model, exhaust gas treatment model, and navigation control model. A unique file identifier is assigned to each model algorithm file. Fault phenomena, causes, and handling measures extracted from pre-acquired historical fault data are structured using fault tree analysis to construct a ship knowledge graph containing fault nodes, cause nodes, symptom nodes, and their causal relationships. The target model algorithm file is then read from the model algorithm file management system based on the file identifier. According to the physical measurement point location and signal type represented by the unified data channel ID encoding rules, each data field in the standardized data is associated and bound with the input variables of the corresponding physical measurement points in the target model algorithm file to form a simulation task instance. Based on the simulation task instance, multiple ship system simulation models are executed in a collaborative simulation, and the ship knowledge graph is called during the simulation process to perform attribution reasoning on abnormal simulation results and output simulation result data. Closed-loop steps for shore-based intelligent application development and iteration: Based on the standardized data and simulation results, develop intelligent applications including ship energy efficiency analysis, performance prediction, voyage management, and remote operation and maintenance of shipboard equipment; The feedback data generated during the operation of the intelligent application is then fed back to the resource pool of the group's cloud PaaS service, which includes storage resources and elastic computing resources. The feedback data includes model simulation errors, energy efficiency prediction deviations, and data quality monitoring results. The model parameters of the model algorithm file are corrected using the model simulation errors, the incremental learning or model retraining of the intelligent application is triggered using the energy efficiency prediction deviations, and the monitoring rules of the dual-dimensional data quality monitoring system or the unified data channel ID encoding rules are updated using the data quality monitoring results, forming an iterative closed loop of data processing and application optimization.

2. The intelligent ship shore-based collaborative R&D method based on group cloud PaaS service according to claim 1, characterized in that, Following the collaborative simulation and knowledge-driven steps, a data sharing step is also included: The standardized data and simulation results are shared externally using RESTful API and MQTT modes. The RESTful API mode combines token authentication, role-based access control, and call monitoring to ensure that only authorized users can access standardized data and simulation reports for specific ships or models. The MQTT mode establishes TOPIC naming rules, access authorization policies, quality of service usage principles, and data packet size limits, and pushes standardized data streams and key early warning information from simulation results in real time. Simultaneously, authentication interfaces, customized data sharing interfaces, standardized data channel interfaces, standardized time-series data interfaces, and customized data push interfaces are configured, and AES encrypted transmission and firewall isolation are used to ensure data security.

3. The intelligent ship shore-based collaborative R&D method based on group cloud PaaS service according to claim 1 or 2, characterized in that, In the multi-source data standardization and quality governance steps, the specific construction method of the dual-dimensional data quality monitoring system is as follows: pre-set ship-level integrity threshold and equipment-level integrity threshold; when the ship data integrity of a certain ship is lower than the ship-level integrity threshold, the ship is marked as a high-risk object for data missing. When the data integrity of a device is lower than the device-level integrity threshold, the device is marked as a priority target for data anomaly investigation; and the execution priority of subsequent data cleaning, data trimming, data synchronization and data processing is dynamically adjusted based on the marking result.

4. The intelligent ship shore-based collaborative R&D method based on group cloud PaaS service according to claim 1 or 2, characterized in that, In the collaborative simulation and knowledge-driven steps, the attribution reasoning of the ship knowledge graph specifically includes: during the joint simulation of multiple ship system simulation models, real-time monitoring of the intermediate output results of each simulation model; when any intermediate output result exceeds a preset safety threshold, marking the intermediate output result exceeding the preset safety threshold as an abnormal simulation result, and extracting the fault phenomenon corresponding to the abnormal simulation result as a query condition; based on the query condition, matching fault nodes with the same or similar fault phenomena as the query condition in the ship knowledge graph; tracing back the causal relationship of the matched fault nodes to obtain a list of candidate fault causes leading to the abnormal simulation result and a confidence score for each candidate fault cause; and using the list of candidate fault causes and the confidence score as simulation result data.

5. The intelligent ship shore-based collaborative R&D method based on group cloud PaaS service according to claim 1 or 2, characterized in that, In the collaborative simulation and knowledge-driven steps, the specific implementation method for associating and binding each data field in the standardized data with the input variables of the corresponding physical measurement points in the target model algorithm file is as follows: The target model algorithm file is read from the model algorithm file management system according to the file identifier and a loading operation is performed; after the loading operation, a mapping configuration table is established between the unified data channel ID code and the input variables in the target model algorithm file, with the unified data channel ID code as the key and the memory address of the input variable in the target model algorithm file as the value; based on the physical measurement point location and signal type represented by the unified data channel ID code rule, each data field in the standardized data is written to the memory address of the corresponding input variable in the target model algorithm file according to the mapping configuration table, thereby completing the association and binding, and forming a simulation task instance.

6. The intelligent ship shore-based collaborative R&D method based on group cloud PaaS service according to claim 1 or 2, characterized in that, In the closed-loop steps of the development and iteration of the shore-based intelligent application, the specific method of incremental learning is as follows: using historical data of the period when the energy efficiency prediction deviation exceeds the preset deviation threshold as incremental samples, the weight parameters of the ship energy efficiency analysis model in the intelligent application are updated using an online gradient descent algorithm, without reloading the full amount of historical training data. The specific method of model retraining is as follows: when the cumulative duration of the energy efficiency prediction deviation exceeding the preset deviation threshold reaches the preset retraining trigger condition, the full amount of historical training data and new data are reloaded to perform complete training on the energy efficiency analysis model in the intelligent application and update all model parameters.

7. The intelligent ship shore-based collaborative R&D method based on group cloud PaaS service according to claim 1 or 2, characterized in that, In the closed-loop steps of the development and iteration of the shore-based intelligent application, the calculation method of the model simulation error is as follows: select the simulation result data and the standardized data within the same time window, and calculate the root mean square error of the two; when the root mean square error exceeds the preset error threshold for N consecutive time windows, it is determined that the model parameters need to be corrected, where N is a positive integer greater than or equal to 3.

8. The intelligent ship shore-based collaborative R&D method based on group cloud PaaS service according to claim 1 or 2, characterized in that, In the data access and aggregation step, the ship-side measurement point data includes main engine speed, fuel consumption, exhaust emission parameters, navigation attitude and equipment operating temperature, and the shore-side supplementary data includes AIS data, meteorological data, historical model data, algorithm parameter data and fault knowledge base data. And / or, in the multi-source data standardization and quality governance steps, the unified data channel ID encoding rule includes rule version number, model version number, functional device identifier, attribute label and remarks information; the unified naming and format standardization processing includes original signal list inspection, data model construction, functional device matching, attribute label selection, secondary function matching, signal uniqueness inspection, cross-ship naming consistency inspection and encoding generation and binding; the data cleaning includes removing outliers and filling missing values; the data pruning includes removing redundant fields; the data synchronization includes unifying timestamps; and the data processing includes format conversion and unit unification.

9. A smart ship shore-based collaborative R&D system based on a group cloud PaaS service, characterized in that, This includes, in sequence, a data access and aggregation module, a multi-source data standardization and quality governance module, a collaborative simulation and knowledge-driven module, and a shore-based intelligent application development and iteration closed-loop module. The data access and aggregation module relies on the network and storage resources that are dynamically scalable according to the business load provided by the Group's cloud PaaS service. It uses the MQTT protocol to receive ship-side measurement data uploaded by at least one green intelligent ship in real time, and at the same time obtains shore-side supplementary data. The ship-side measurement data and shore-side supplementary data are uniformly aggregated to the storage resources of the Group's cloud PaaS service to form a multi-source heterogeneous original dataset. The multi-source data standardization and quality governance module formulates unified data channel ID encoding rules according to the general international standards for ship data. It then uses these rules to uniformly name and format the multi-source heterogeneous raw datasets, generating standardized raw data. Based on these multi-source heterogeneous raw datasets, it calculates the total number of measurement points that should be received and the actual number of measurement points received for each ship and each piece of equipment on board. It calculates the ship data integrity based on the actual number of received measurement points and the number of ship measurement points that should be received, and calculates the equipment data integrity based on the actual number of received equipment measurement points and the number of equipment measurement points that should be received. A two-dimensional data quality monitoring system is then constructed based on both ship data integrity and equipment data integrity. The standardized raw data is monitored by a two-dimensional data quality monitoring system. For data that does not meet the preset quality standards, data cleaning, data trimming, data synchronization and data processing are performed in sequence to output the processed standardized data. The collaborative simulation and knowledge-driven module, relying on the elastic computing resources of the group's cloud PaaS service, constructs a file management system that stores multi-format model algorithm files. These multi-format model algorithm files include multiple ship system simulation models, such as a hull fluid model, main engine performance model, exhaust gas treatment model, and navigation control model. Each model algorithm file is assigned a unique file identifier. Fault phenomena, causes, and handling measures extracted from pre-acquired historical fault data are structured using fault tree analysis to construct a ship knowledge graph containing fault nodes, cause nodes, symptom nodes, and their causal relationships. The target model algorithm file is then read from the model algorithm file management system based on the file identifier. According to the physical measurement point location and signal type represented by the unified data channel ID encoding rules, each data field in the standardized data is associated and bound with the input variables of the corresponding physical measurement points in the target model algorithm file to form a simulation task instance. Based on the simulation task instance, multiple ship system simulation models are executed in a collaborative simulation, and the ship knowledge graph is called during the simulation process to perform attribution reasoning on abnormal simulation results and output simulation result data. The shore-based intelligent application development and iteration closed-loop module, based on the standardized data and the simulation results data, develops intelligent applications including ship energy efficiency analysis, performance prediction, voyage management, and remote operation and maintenance of shipboard equipment. The feedback data generated during the operation of the intelligent application is then fed back to the resource pool of the group's cloud PaaS service, which includes storage resources and elastic computing resources. The feedback data includes model simulation errors, energy efficiency prediction deviations, and data quality monitoring results. The model parameters of the model algorithm file are corrected using the model simulation errors, the incremental learning or model retraining of the intelligent application is triggered using the energy efficiency prediction deviations, and the monitoring rules of the dual-dimensional data quality monitoring system or the unified data channel ID encoding rules are updated using the data quality monitoring results, forming an iterative closed loop of data processing and application optimization.

10. The intelligent ship shore-based collaborative R&D system based on the group cloud PaaS service according to claim 9, characterized in that, Following the collaborative simulation and knowledge-driven module, a data sharing module is also included: The standardized data and simulation results are shared externally using RESTful API and MQTT modes. The RESTful API mode combines token authentication, role-based access control, and call monitoring to ensure that only authorized users can access standardized data and simulation reports for specific ships or models. The MQTT mode establishes TOPIC naming rules, access authorization policies, quality of service usage principles, and data packet size limits, and pushes standardized data streams and key early warning information from simulation results in real time. Simultaneously, authentication interfaces, customized data sharing interfaces, standardized data channel interfaces, standardized time-series data interfaces, and customized data push interfaces are configured, and AES encrypted transmission and firewall isolation are used to ensure data security.

Images