Digital twin system of an engineering vessel integrating lifting and piling functions

Abstract

The application discloses an engineering ship digital twin system integrating hoisting and piling functions, comprising a multi-source data acquisition and fusion module, an operation digital twin model construction module, a dynamic working condition perception and analysis module, an elastic rule knowledge management and updating module, an intelligent decision reasoning module, a multi-target dynamic collaborative optimization module, a decision execution monitoring and feedback module, and a digital twin model verification and iteration module; the application realizes mutual influence simulation of hoisting and piling functions by constructing a high-fidelity coupled digital twin model; the piling model can dynamically adapt to actual construction conditions by using a geological real-time feedback mechanism; and a multi-target optimization algorithm is adopted to establish a collaborative optimization framework integrating operation efficiency, structural health and safety risk. The application can intelligently perceive operation intention, dynamically adapt to construction conditions and collaboratively optimize multi-targets, and significantly improves the intelligent operation level and construction safety of the engineering ship under complex sea conditions.

Description

Technical Field

[0001] This invention relates to the intersection of marine engineering and digital twin technology, and more specifically, to a digital twin system for engineering vessels that integrates lifting and piling functions. Background Technology

[0002] Currently, in the field of digital twin technology for engineering vessels, there are mainly several technical approaches, including digital twin systems based on a single operation mode, static piling models based on fixed parameters, and isolated optimization operation decision-making systems. Looking at existing technologies, the following core bottlenecks remain unresolved in achieving intelligent operation of integrated engineering vessels: First, there is a lack of multi-functional coupling modeling: the existing system cannot accurately describe the mutual influence of lifting and piling functions when they work simultaneously or switch modes. Second, there is a lack of real-time adaptability: the piling process lacks the ability to dynamically adjust according to actual geological conditions, which poses construction risks. Third, there is a lack of collaborative optimization capabilities: there is a lack of a unified optimization framework for multiple objectives such as operational efficiency, structural health, and safety risks; Therefore, there is an urgent need in this field for a new type of digital twin system that can intelligently perceive operational intentions, dynamically adapt to construction conditions, and collaboratively optimize multiple objectives. Summary of the Invention

[0003] The purpose of this invention is to provide a digital twin system for engineering vessels that integrates lifting and piling functions, in order to solve the above-mentioned problems existing in the prior art.

[0004] The application is as follows: A digital twin system for engineering vessels integrating lifting and piling functions includes: The multi-source data acquisition and fusion module is used to integrate multi-source scene data of the target engineering vessel's operating environment. The multi-source scene data includes three-dimensional structural data of the vessel and equipment, equipment operating parameters, and hydrological data of the operating sea area. After preprocessing, the vessel equipment operating status is output. The operational digital twin model construction module is used to construct an operational digital twin model of the target engineering vessel based on the multi-source scene data, and to perform coupling effect simulation verification on the operational digital twin model. Based on the deviation analysis between the simulation results and the physical monitoring data, the module performs dynamic parameter calibration and mechanism model supplementation to obtain an optimized operational digital twin model. The dynamic working condition perception and analysis module, based on the working condition data output by the optimized digital twin model of the operation, uses a recurrent neural network to identify the key working condition characteristics of the target engineering vessel and its lifting and piling equipment, and obtains a first working condition feature set including equipment health status, real-time operation progress and the working environment. The flexible rule knowledge management and update module is used to perform online transfer learning based on the first working condition feature set and update the flexible rule knowledge base. The flexible rule knowledge base stores multi-layered operation management and safety control rules oriented towards integrated lifting and piling functions. The intelligent decision-making reasoning module is used to construct a decision-making reasoning model, and to perform multi-objective game operation on the elastic rule knowledge base through the decision-making reasoning model to generate the first operation management decision for the coordinated operation of lifting and piling. The multi-objective dynamic collaborative optimization module, based on the first operation management decision, simultaneously processes multiple sub-objectives such as operation safety risk, operation efficiency and equipment energy consumption through a multi-objective optimization algorithm to generate a second operation management decision; The decision execution monitoring and feedback module is used to convert the second operation management decision into specific collaborative operation instructions for the lifting and piling equipment of the engineering vessel, send them to the ship-shore control center, and monitor the execution process and effect of the collaborative operation instructions in real time. The digital twin model verification and iteration module is used to verify and iteratively update the collaborative operation instructions through the operation digital twin model.

[0005] Furthermore: the operational digital twin model construction module constructs an operational digital twin model of the target engineering vessel, including: Based on the three-dimensional structural data of the vessel and equipment, a three-dimensional geometric model of the target engineering vessel and lifting and piling equipment is constructed, and the hydrological data of the operating sea area is mapped into the three-dimensional geometric model to generate an initial operating scene model that integrates environmental features. A real-time dynamic model of the equipment is constructed based on the equipment operating parameters. The spatial coordinate system of the real-time dynamic model of the equipment is calibrated and matched with that of the initial operation scenario model. The calibrated real-time dynamic model of the equipment is then coupled with the initial operation scenario model to construct an operational digital twin model of the target engineering vessel. The coupling effect of the digital twin model of the operation was simulated and verified, and the verification results are as follows: The design includes a set of verification scenarios for the typical coupling effects of lifting and piling. These scenarios include data on the impact of ship heeling caused by lifting during ship lifting operations on the verticality of piling, and the reaction of ship vibration caused by piling impact loads on the stability of the crane. The input parameters of the verification working condition set are input into the digital twin model of the operation for simulation calculation to obtain the twin simulation results; Simultaneously collect physical monitoring data of the target engineering vessel when it performs the corresponding operations of the verification condition set during actual operations; The twin simulation results are spatiotemporally aligned and compared with the physical monitoring data to calculate the simulation accuracy deviation of each coupling link. The simulation accuracy deviation includes the prediction deviation of the dynamic response of the crane hoisting, the prediction deviation of the ship attitude disturbance during the pile driving process, and the prediction deviation of the energy transfer between equipment. The distribution characteristics of simulation accuracy deviations are analyzed to identify the weak links in the model that cause the deviations. These weak links include inaccurate parameter settings, missing sub-models, and insufficient modeling of coupling effects. Based on the analysis results of the weak links, targeted model calibration and enhancement are initiated. Through the process of dynamic parameter calibration and mechanism model supplementation, an optimized digital twin model of the operation is generated.

[0006] Furthermore, the operational digital twin model construction module also includes the steps of model calibration and mechanism supplementation based on the verification results, and generating an optimized model, specifically including: N31. Analyze the simulation verification results of coupling effects, and based on the analysis of simulation accuracy deviation, identify and extract the list of model parameters that need to be calibrated and the list of coupling mechanism model items that need to be supplemented. N32. For model parameters that need to be calibrated, by comparing the differences between simulation predictions and physical monitoring values, a parameter calibration model based on inversion analysis or system identification is established, and cross-validation is performed using a historical optimization case library to complete the dynamic update of key dynamic parameters, material parameters, and fluid dynamic coefficients. N33. For the coupling mechanism model items that need to be supplemented, analyze whether the current model accurately describes the interactive physical process. If it does not accurately describe it, generate the corresponding supplementary mechanism sub-model by calling the mechanism model library or starting the first principle derivation, and integrate it into the original model framework in a modular way. N34. For the operational digital twin model after parameter dynamic calibration and mechanism model supplementation, the verification case set is used to perform secondary coupling effect simulation verification, and the updated simulation accuracy deviation is calculated. N35. Determine whether the updated simulation accuracy deviation meets the preset model fidelity threshold. If it does, mark the job digital twin model as the optimized job digital twin model and output it. If it does not meet the threshold, return to step N31, generate a new calibration and supplementary requirement list based on the secondary verification results, and start a new round of model iteration optimization process.

[0007] Furthermore: In the dynamic working condition perception and analysis module, the recurrent neural network used is a hybrid network structure of LSTM and GRU. This hybrid network structure includes an input layer, a dual-branch processing layer, a fusion layer, a feature extraction layer, and a fully connected layer. The processing flow of this hybrid network structure includes: The input layer of the hybrid network structure receives a sequence of time-series work condition data output from the optimized digital twin model of the operation, and the input dimension is consistent with the feature dimension of the time-series work condition data. The time-series operating condition data is simultaneously input into the LSTM branch and the GRU branch for parallel processing; The LSTM branch embeds a spatiotemporal attention mechanism, and the formula for calculating the attention weights at time step t is as follows: α t =Softmax(W h h t-1 +W x x t +b), Where, α t Represented as an attention weight vector, W h Represented as the hidden state weight matrix, h t-1 Represented as the hidden state at time step t-1, W x Represented as the input weight matrix, x t denoted as the ship equipment operating condition at time step t, and b represents the bias term; The output of the LSTM branch at each time step t Output of the GRU branch The fusion is performed using dynamic weighting to obtain the fused output sequence; The fused output sequence is further mapped through a fully connected layer to generate a high-dimensional vector containing time-dependent features, which serves as the final output of the dynamic working condition perception and analysis module.

[0008] Furthermore: In the flexible rule knowledge management and update module, the update triggering conditions for the flexible rule knowledge base include: Signals of sudden changes in equipment health status and signals of changes in the working environment from the dynamic working condition perception and analysis module; The decision simulation deviation detected by the digital twin simulation verification and iteration module exceeds a preset threshold; And manual rule revision or priority adjustment instructions input by operators through human-computer interaction terminals.

[0009] Furthermore: the construction and operation of the decision reasoning model in the intelligent decision reasoning module includes: Each operation management and safety control rule in the flexible rule knowledge base is transformed into a basic processing unit in a fuzzy Petri net. The preconditions of each rule are mapped to the input location, the conclusion or action of the rule is mapped to the output location, and the activation logic of the rule itself is mapped to the transition connecting the two. The lifting safety rule, piling accuracy rule, and operation efficiency rule are defined as input layer nodes of the fuzzy Petri net. By setting the activation threshold of the transition and the fuzzy inference function, the competition and cooperation relationship between rules is simulated. According to the preset operation scenario and the real-time input working condition feature set, the trigger strength of each path in the fuzzy Petri net is calculated. Based on the trigger strength and the preset priority of the rule, a dynamic priority decision tree for coordinating multiple rule conflicts is generated. Input the current first working condition feature set into the dynamic priority decision tree, traverse the tree structure to obtain all feasible initial candidate decision sets, call the operation digital twin model for simulation for each initial candidate decision, and calculate the estimated performance score after simulation. A deep Q-network is constructed as a policy optimizer. The state space of the policy optimizer is defined by the first condition feature set and the initial candidate decision. The action space of the policy optimizer is a set of operations that adjust the initial candidate decision. The policy optimizer uses the estimated performance score as the initial reward signal and activates the optimization process when the policy optimization condition is triggered.

[0010] Furthermore: In the intelligent decision-making reasoning module, the flexible rule knowledge base is used to perform multi-objective game operations through the decision-making reasoning model, specifically including: Define job rule containers corresponding to different job types. Each job rule container stores a set of constraints and rules for a specific job type. Receive activation level signals from each rule in the elastic rule knowledge base, and parse the logical relationships between the rules; The activation level signals of each rule are input into the fuzzy inference unit of the decision reasoning model to calculate the comprehensive activation value of each task rule container, specifically expressed as follows: , where A c R is represented as the overall activation value of the job rule container c. c w represents the number of rules associated with job container c. c,r Let b represent the weight coefficient of the r-th rule in the job container c. c,r This indicates the actual activation level of the r-th rule under the current operating conditions; Based on the comprehensive activation value of each operation rule container, a multi-level dynamic priority decision tree is constructed. The decision tree uses the operation stage, equipment status and risk assessment as nodes, and the activation value of the rule container as the node connection strength. Based on the dynamic target weights set for the current task, a set of candidate decision paths is generated through a dynamic priority decision tree; The candidate decision paths are evaluated using a deep Q-network with a reinforcement learning mechanism, and the optimal decision path that maximizes long-term cumulative rewards is selected as the first task management decision output.

[0011] Furthermore, the multi-objective dynamic collaborative optimization module employs multi-objective optimization algorithms including the improved NSGA-Ⅲ algorithm and the improved ant colony algorithm. The conflict objectives handled include at least minimizing operational safety risks, minimizing the cycle time of single-pile operations, minimizing overall energy consumption, and balancing the wear of key equipment.

[0012] Furthermore: In the multi-objective dynamic collaborative optimization module, the optimization process for multiple conflicting objectives using the improved NSGA-Ⅲ algorithm includes: Cluster analysis of conflict targets is performed based on historical operation management decision data to generate representative feature vectors under different decision-making models; The first operation management decision and the current real-time operating condition data are encoded into an initial population, and the reference point is initialized using the representative feature vector. The offspring population is generated by simulated binary crossover and polynomial mutation. During the fitness assessment phase, a first penalty term is applied to individuals that violate the lifting load safety threshold, and a second penalty term is applied to individuals that violate the pile driving verticality safety threshold. Excellent individuals are selected based on non-dominated ranking and environmental selection mechanisms. At the same time, a dynamic weight matrix is ​​constructed, which assigns time-varying weights to each conflict target according to the current operational situation. By incorporating the dynamic weight matrix into the environment selection process, the optimization priority of each conflicting objective is dynamically adjusted according to the operational situation, thereby generating a second operational management decision.

[0013] Furthermore: In the multi-objective dynamic collaborative optimization module, an improved ant colony algorithm is used to handle multiple conflicting objectives, including: L11. In the multi-objective dynamic collaborative optimization module, an improved ant colony algorithm is used to handle multiple conflicting objectives, specifically including: L12. Decompose the first task management decision into multiple task sub-task nodes and construct a network graph model based on task timing and resource dependencies. An independent pheromone matrix is ​​established for each conflict target, and a multi-target pheromone update rule is designed, specifically as follows: For the pheromone update of target k on the path from node i to node j: , , in, Let ρ be the pheromone concentration for target k on path (i,j) at iteration number t, ρ represent the pheromone evaporation coefficient, and M represent the number of ants. Let Q be the pheromone increment left by the m-th ant in this iteration, and let Q be the pheromone intensity constant. The overall performance score of the path constructed by the m-th ant, where δ is the adjustment coefficient, and Rm is the overall performance score of the path constructed by the m-th ant. k The real-time risk assessment coefficient for target k; L13. Combining the physical constraints and efficiency requirements of lifting and piling operations, design a heuristic factor that integrates the characteristics of the operations, specifically expressed as follows: Heuristic factor for node i to node j of target k: , Where, d ij τ represents the estimated time cost required to execute task j on node j. ij D represents the angle change required to switch from the job mode of node i to the job mode of node j. k This represents the deviation of the current solution from the target k, where φ and ν represent the weighting coefficients, respectively. L14. Ants select paths based on the probability of the product of pheromone concentration and heuristic factor, thus constructing a complete sequence of operation plans. L15. Iterate through steps L12 to L14. When the maximum number of iterations is reached or the solution quality converges, select the operation plan with the highest overall performance from the Pareto optimal solution set as the second operation management decision output.

[0014] Compared with the prior art, the present invention achieves the following beneficial effects: This invention establishes modules for multi-source data acquisition and fusion, digital twin model construction, dynamic working condition perception and analysis, flexible rule knowledge management and updating, intelligent decision-making and reasoning, multi-objective dynamic collaborative optimization, decision execution monitoring and feedback, and digital twin model verification and iteration. By constructing a high-fidelity coupled digital twin model, it simulates the mutual influence between lifting and piling functions. Utilizing a real-time geological feedback mechanism, the piling model dynamically adapts to actual construction conditions. Furthermore, it employs a multi-objective optimization algorithm to establish a collaborative optimization framework that unifies operational efficiency, structural health, and safety risks. This invention intelligently perceives operational intentions, dynamically adapts to construction conditions, and collaboratively optimizes multiple objectives, significantly improving the intelligent operation level and construction safety of engineering vessels in complex sea conditions. Attached Figure Description

[0015] Figure 1 This is an architecture diagram of a digital twin system for engineering vessels that integrates lifting and piling functions, provided by an embodiment of the present invention. Detailed Implementation

[0016] The present invention will now be described in detail with reference to the accompanying drawings.

[0017] Example 1

[0018] This invention provides a digital twin system for engineering vessels that integrates lifting and piling functions, such as... Figure 1 As shown, it includes: The multi-source data acquisition and fusion module is used to integrate multi-source scene data of the target engineering vessel's operating environment. The multi-source scene data includes three-dimensional structural data of the vessel and equipment, equipment operating parameters, and hydrological data of the operating sea area. After preprocessing, the vessel equipment operating status is output. The operational digital twin model construction module is used to construct an operational digital twin model of the target engineering vessel based on the multi-source scene data, and to perform coupling effect simulation verification on the operational digital twin model. Based on the deviation analysis between the simulation results and the physical monitoring data, the module performs dynamic parameter calibration and mechanism model supplementation to obtain an optimized operational digital twin model. The dynamic working condition perception and analysis module, based on the working condition data output by the optimized digital twin model of the operation, uses a recurrent neural network to identify the key working condition characteristics of the target engineering vessel and its lifting and piling equipment, and obtains a first working condition feature set including equipment health status, real-time operation progress and the working environment. The flexible rule knowledge management and update module is used to perform online transfer learning based on the first working condition feature set and update the flexible rule knowledge base. The flexible rule knowledge base stores multi-layered operation management and safety control rules oriented towards integrated lifting and piling functions. The intelligent decision-making reasoning module is used to construct a decision-making reasoning model, and to perform multi-objective game operation on the elastic rule knowledge base through the decision-making reasoning model to generate the first operation management decision for the coordinated operation of lifting and piling. The multi-objective dynamic collaborative optimization module, based on the first operation management decision, simultaneously processes multiple sub-objectives such as operation safety risk, operation efficiency and equipment energy consumption through a multi-objective optimization algorithm to generate a second operation management decision; The decision execution monitoring and feedback module is used to convert the second operation management decision into specific collaborative operation instructions for the lifting and piling equipment of the engineering vessel, send them to the ship-shore control center, and monitor the execution process and effect of the collaborative operation instructions in real time. The digital twin model verification and iteration module is used to verify and iteratively update the collaborative operation instructions through the operation digital twin model.

[0019] Specifically, this invention establishes modules for multi-source data acquisition and fusion, digital twin model construction, dynamic working condition perception and analysis, flexible rule knowledge management and updating, intelligent decision-making and reasoning, multi-objective dynamic collaborative optimization, decision execution monitoring and feedback, and digital twin model verification and iteration. By constructing a high-fidelity coupled digital twin model, it simulates the mutual influence of lifting and piling functions. Utilizing a real-time geological feedback mechanism, the piling model dynamically adapts to actual construction conditions. Furthermore, a multi-objective optimization algorithm is employed to establish a collaborative optimization framework that unifies operational efficiency, structural health, and safety risks. This embodiment intelligently perceives operational intent, dynamically adapts to construction conditions, and collaboratively optimizes multiple objectives, significantly improving the intelligent operation level and construction safety of engineering vessels in complex sea conditions. It is particularly suitable for composite operational vessels with both lifting and piling functions, such as offshore wind power installation vessels and bridge construction vessels. This system, by perceiving operational intent, fusing multi-physics models, and achieving online optimization of the construction process, solves the core challenges faced by integrated functional vessels in complex operating environments.

[0020] In the above embodiments, specifically: the operational digital twin model construction module constructs an operational digital twin model of the target engineering vessel, including: Based on the three-dimensional structural data of the vessel and equipment, a three-dimensional geometric model of the target engineering vessel and lifting and piling equipment is constructed, and the hydrological data of the operating sea area is mapped into the three-dimensional geometric model to generate an initial operating scene model that integrates environmental features. A real-time dynamic model of the equipment is constructed based on the equipment operating parameters. The spatial coordinate system of the real-time dynamic model of the equipment is calibrated and matched with that of the initial operation scenario model. The calibrated real-time dynamic model of the equipment is then coupled with the initial operation scenario model to construct an operational digital twin model of the target engineering vessel. The coupling effect of the digital twin model of the operation was simulated and verified, and the verification results are as follows: The design includes a set of verification scenarios for the typical coupling effects of lifting and piling. These scenarios include data on the impact of ship heeling caused by lifting during ship lifting operations on the verticality of piling, and the reaction of ship vibration caused by piling impact loads on the stability of the crane. The input parameters of the verification working condition set are input into the digital twin model of the operation for simulation calculation to obtain the twin simulation results; Simultaneously collect physical monitoring data of the target engineering vessel when it performs the corresponding operations of the verification condition set during actual operations; The twin simulation results are spatiotemporally aligned and compared with the physical monitoring data to calculate the simulation accuracy deviation of each coupling link. The simulation accuracy deviation includes the prediction deviation of the dynamic response of the crane hoisting, the prediction deviation of the ship attitude disturbance during the pile driving process, and the prediction deviation of the energy transfer between equipment. The distribution characteristics of simulation accuracy deviations are analyzed to identify the weak links in the model that cause the deviations. These weak links include inaccurate parameter settings, missing sub-models, and insufficient modeling of coupling effects. Based on the analysis results of the weak links, targeted model calibration and enhancement are initiated. Through the process of dynamic parameter calibration and mechanism model supplementation, an optimized digital twin model of the operation is generated.

[0021] In the above embodiments, specifically: the task digital twin model construction module further includes the step of performing model calibration and mechanism supplementation based on the verification results, and generating an optimized model, specifically including: N31. Analyze the simulation verification results of coupling effects, and based on the analysis of simulation accuracy deviation, identify and extract the list of model parameters that need to be calibrated and the list of coupling mechanism model items that need to be supplemented. N32. For model parameters that need to be calibrated, by comparing the differences between simulation predictions and physical monitoring values, a parameter calibration model based on inversion analysis or system identification is established, and cross-validation is performed using a historical optimization case library to complete the dynamic update of key dynamic parameters, material parameters, and fluid dynamic coefficients. N33. For the coupling mechanism model items that need to be supplemented, analyze whether the current model accurately describes the interactive physical process. If it does not accurately describe it, generate the corresponding supplementary mechanism sub-model by calling the mechanism model library or starting the first principle derivation, and integrate it into the original model framework in a modular way. The interactive physical process is represented as: the nonlinear coupling vibration transmission path between the hull, crane and pile driver, or the additional resistance effect of the ship caused by the hoisting operation. N34. For the operational digital twin model after parameter dynamic calibration and mechanism model supplementation, the verification case set is used to perform secondary coupling effect simulation verification, and the updated simulation accuracy deviation is calculated. N35. Determine whether the updated simulation accuracy deviation meets the preset model fidelity threshold. If it does, mark the job digital twin model as the optimized job digital twin model and output it. If it does not meet the threshold, return to step N31, generate a new calibration and supplementary requirement list based on the secondary verification results, and start a new round of model iteration optimization process.

[0022] It should be noted that the multi-source data acquisition and fusion module and the operational digital twin model construction module in this embodiment integrate the three-dimensional structural data of the ship, lifting equipment, and piling equipment, equipment operating parameters, and marine hydrological data to construct a unified operational digital twin model. This model achieves coupled modeling of the mutual influence between the lifting and piling equipment during operation, overcoming the problem that existing systems can only handle a single operational mode. Specifically, this is reflected in: Coupled effect simulation: By coupling the real-time dynamic model of the equipment with the operation scenario model, the influence of the ship's attitude change on the pile driving accuracy during the lifting operation, as well as the feedback effect of the pile driving impact on the stability of the crane, can be accurately simulated.

[0023] Dynamic evaluation and compensation mechanism: By constructing an integrity evaluation and defect compensation process, the model can be ensured to discover and correct defects caused by insufficient modeling of functional coupling relationships, thereby improving the accuracy of the model in describing multifunctional interactions.

[0024] Verification results: In practical applications, the system can provide early warnings of coupled risks such as "the ship's rolling caused by pile driving operations may lead to increased swaying of the suspended load," which is something that traditional independent modeling systems cannot achieve.

[0025] In the above embodiments, specifically: the dynamic working condition perception and analysis module uses a recurrent neural network structure that is a hybrid network structure of LSTM and GRU. The hybrid network structure includes an input layer, a dual-branch processing layer, a fusion layer, a feature extraction layer, and a fully connected layer. The processing flow of the hybrid network structure includes: The input layer of the hybrid network structure receives a sequence of time-series work condition data output from the optimized digital twin model of the operation, and the input dimension is consistent with the feature dimension of the time-series work condition data. The time-series operating condition data is simultaneously input into the LSTM branch and the GRU branch for parallel processing; The LSTM branch embeds a spatiotemporal attention mechanism, and the formula for calculating the attention weights at time step t is as follows: α t =Softmax(W h h t-1 +W x x t +b), Where, α t Represented as an attention weight vector, W h Represented as the hidden state weight matrix, h t-1 Represented as the hidden state at time step t-1, W x Represented as the input weight matrix, x t denoted as the ship equipment operating condition at time step t, and b represents the bias term; The output of the LSTM branch at each time step t Output of the GRU branch The fusion is performed using dynamic weighting to obtain the fused output sequence; The fused output sequence is further mapped through a fully connected layer to generate a high-dimensional vector containing time-dependent features, which serves as the final output of the dynamic working condition perception and analysis module.

[0026] In the above embodiments, specifically: in the flexible rule knowledge management and update module, the update triggering conditions for the flexible rule knowledge base include: Signals of sudden changes in equipment health status and signals of changes in the working environment from the dynamic working condition perception and analysis module; The decision simulation deviation detected by the digital twin simulation verification and iteration module exceeds a preset threshold; And manual rule revision or priority adjustment instructions input by operators through human-computer interaction terminals.

[0027] In the above embodiments, specifically: the construction and operation of the decision reasoning model in the intelligent decision reasoning module includes: Each operation management and safety control rule in the flexible rule knowledge base is transformed into a basic processing unit in a fuzzy Petri net. The preconditions of each rule are mapped to the input location, the conclusion or action of the rule is mapped to the output location, and the activation logic of the rule itself is mapped to the transition connecting the two. The lifting safety rule, piling accuracy rule, and operation efficiency rule are defined as input layer nodes of the fuzzy Petri net. By setting the activation threshold of the transition and the fuzzy inference function, the competition and cooperation relationship between rules is simulated. According to the preset operation scenario (such as "heavy lifting" and "precision piling") and the real-time input working condition feature set, the trigger strength of each path in the fuzzy Petri net is calculated. Based on the trigger strength and the preset priority of the rule, a dynamic priority decision tree for coordinating multiple rule conflicts is generated. Each branch of the decision tree represents a possible decision path, and its weight is determined by the comprehensive activation strength of the corresponding rule set. Input the current first working condition feature set into the dynamic priority decision tree, traverse the tree structure to obtain all feasible initial candidate decision sets, call the operation digital twin model for simulation for each initial candidate decision, and calculate the estimated performance score after simulation. A deep Q-network is constructed as a policy optimizer. The state space of the policy optimizer is defined by the first working condition feature set and the initial candidate decisions. The action space of the policy optimizer is the set of operations that adjust the initial candidate decisions (such as adjusting the job sequence or modifying equipment parameters). The policy optimizer uses the estimated performance score as the initial reward signal. When a policy optimization condition is triggered (such as the estimated safety score being lower than a threshold, or the existence of multiple candidate decisions with similar but conflicting scores), the optimization process is activated. The optimization process uses a combination of offline historical job data and online simulation data to iteratively train the deep Q-network. By maximizing the long-term cumulative reward, it learns to select and generate the optimal policy for the first job management decision from the candidate decision set under the current complex working condition.

[0028] It should be noted that the first working condition feature set refers to a comprehensive working condition feature vector set output by the dynamic working condition perception and analysis module, which has undergone intelligent recognition and feature extraction. The comprehensive working condition feature vector set specifically includes the following quantitative features: Equipment health status characteristics include the time-series statistical characteristics of the tension and deformation of the crane's main hook wire rope, the abnormal index of the pressure and flow of the main hydraulic system, the efficiency attenuation coefficient of the impact energy of the pile hammer, and the real-time stress concentration factor of each major structural component (such as the pile frame and boom hinge). Real-time progress characteristics of the operation include: stage indicators of the current hoisting operation cycle (such as "lifting", "slewing", "positioning"), average cumulative deviation of verticality of completed piles, percentage of the current instantaneous penetration rate of piles to the design value, and real-time positioning error of the vessel relative to the target operation position. Characteristics of the operating environment include local wind field characteristics (such as wind speed, wind direction and turbulence intensity) estimated in real time by shipborne sensors, surface flow velocity and direction, coupling index of significant wave height and ship encounter period, and seabed sediment type (such as soft soil and hard soil classification probability based on acoustic inversion). Each feature dimension of the feature set has been encoded and dimensionality reduced by the recurrent neural network, transforming the high-dimensional original working condition data into a low-dimensional feature representation with clear physical or engineering significance that is suitable for subsequent rule matching and decision reasoning.

[0029] It should be noted that, through the dynamic working condition perception and analysis module and the flexible rule knowledge management and update module, this system can dynamically adjust the model and behavior rules according to real-time working conditions, thus overcoming the limitations of the traditional static piling model.

[0030] Real-time geological adaptation: The dynamic working condition perception module adopts a hybrid network of LSTM and GRU, which can identify the characteristics of geological parameter changes during the operation process in real time (such as the abrupt change pattern of penetration resistance). When these characteristics are identified, the elastic rule knowledge base is triggered to be updated, and the rules of relevant piling parameters (such as hammer energy and frequency) will be dynamically corrected.

[0031] Intelligent decision adaptation: The intelligent decision reasoning module uses deep Q-networks for strategy optimization, which can quickly adjust the piling strategy based on real-time working conditions (such as current geological hardness and sea conditions), realizing the transformation from "preset parameter construction" to "perception-decision-adaptive construction".

[0032] Application example: During the construction of pile foundations for an offshore wind farm, the system automatically deduced that a hard interlayer had been encountered by sensing an abnormal decrease in the penetration rate in real time, and made a decision to suggest "switching to low-frequency high-energy mode", which avoided the risk of pile damage and improved construction efficiency by about 15%.

[0033] In the above embodiments, specifically: the intelligent decision-making reasoning module performs multi-objective game operations on the elastic rule knowledge base through the decision-making reasoning model, specifically including: Define job rule containers corresponding to different job types. Each job rule container stores a set of constraints and rules for a specific job type. Receive activation level signals from each rule in the elastic rule knowledge base, and parse the logical relationships between the rules; The activation level signals of each rule are input into the fuzzy inference unit of the decision reasoning model to calculate the comprehensive activation value of each task rule container, specifically expressed as follows: , where A c R is represented as the overall activation value of the job rule container c. c w represents the number of rules associated with job container c. c,r Let b represent the weight coefficient of the r-th rule in the job container c. c,r This indicates the actual activation level of the r-th rule under the current operating conditions; Based on the comprehensive activation value of each operation rule container, a multi-level dynamic priority decision tree is constructed. The decision tree uses the operation stage, equipment status and risk assessment as nodes, and the activation value of the rule container as the node connection strength. Based on the dynamic target weights set for the current task, a set of candidate decision paths is generated through a dynamic priority decision tree; The candidate decision paths are evaluated using a deep Q-network with a reinforcement learning mechanism, and the optimal decision path that maximizes long-term cumulative rewards is selected as the first task management decision output.

[0034] In the above embodiments, specifically: the multi-objective dynamic collaborative optimization module uses multi-objective optimization algorithms including the improved NSGA-Ⅲ algorithm and the improved ant colony algorithm, and the conflict objectives handled include at least minimizing operational safety risks, minimizing the cycle time of single pile operation, minimizing comprehensive energy consumption, and balancing the wear of key equipment.

[0035] In the above embodiments, specifically: the multi-objective dynamic collaborative optimization module employs an improved NSGA-III algorithm to handle the optimization process of multiple conflicting objectives, including: Cluster analysis of conflict targets is performed based on historical operation management decision data to generate representative feature vectors under different decision-making models; The first operation management decision and the current real-time operating condition data are encoded into an initial population, and the reference point is initialized using the representative feature vector. The offspring population is generated by simulated binary crossover and polynomial mutation. During the fitness assessment phase, a first penalty term is applied to individuals that violate the lifting load safety threshold, and a second penalty term is applied to individuals that violate the pile driving verticality safety threshold. Excellent individuals are selected based on non-dominated ranking and environmental selection mechanisms. At the same time, a dynamic weight matrix is ​​constructed. The dynamic weight matrix assigns time-varying weights to each conflict target according to the current operational situation (such as sea state level and mission urgency). By incorporating the dynamic weight matrix into the environment selection process, the optimization priority of each conflicting objective is dynamically adjusted according to the operational situation, thereby generating a second operational management decision.

[0036] In the above embodiments, specifically: the multi-objective dynamic collaborative optimization module employs an improved ant colony algorithm to handle multiple conflicting objectives, including: L11. In the multi-objective dynamic collaborative optimization module, an improved ant colony algorithm is used to handle multiple conflicting objectives, specifically including: L12. Decompose the first task management decision into multiple task sub-task nodes and construct a network graph model based on task timing and resource dependencies. An independent pheromone matrix is ​​established for each conflict target, and a multi-target pheromone update rule is designed, specifically as follows: For the pheromone update of target k on the path from node i to node j: , , in, Let ρ be the pheromone concentration for target k on path (i,j) at iteration number t, ρ represent the pheromone evaporation coefficient, and M represent the number of ants. Let Q be the pheromone increment left by the m-th ant in this iteration, and let Q be the pheromone intensity constant. The overall performance score of the path constructed by the m-th ant, where δ is the adjustment coefficient, and Rm is the overall performance score of the path constructed by the m-th ant. k The real-time risk assessment coefficient for target k; L13. Combining the physical constraints and efficiency requirements of lifting and piling operations, design a heuristic factor that integrates the characteristics of the operations, specifically expressed as follows: Heuristic factor for node i to node j of target k: , Where, d ij τ represents the estimated time cost required to execute task j on node j. ij D represents the angle change required to switch from the job mode of node i to the job mode of node j. k This represents the deviation of the current solution from the target k, where φ and ν represent the weighting coefficients, respectively. L14. Ants select paths based on the probability of the product of pheromone concentration and heuristic factor, thus constructing a complete sequence of operation plans. L15. Iterate through steps L12 to L14. When the maximum number of iterations is reached or the solution quality converges, select the operation plan with the highest overall performance from the Pareto optimal solution set as the second operation management decision output.

[0037] It should be noted that, through the multi-objective dynamic collaborative optimization module, this system establishes a unified framework and simultaneously performs collaborative optimization on multiple conflicting objectives such as operational safety risks, operational efficiency, equipment energy consumption, and equipment wear.

[0038] A unified optimization framework is established: the conflicting objectives addressed by the system include safety risks, work cycles, energy consumption, and equipment wear, and are solved using improved NSGA-III and ant colony optimization algorithms. This changes the situation in traditional systems where objectives are optimized in isolation and often contradictory.

[0039] Dynamic priority adjustment: The system constructs a dynamic weight matrix, which can dynamically adjust the optimization priority of each objective according to the real-time operational situation (such as the safety weight being automatically increased when the sea state deteriorates), thus achieving global optimization rather than local optimization decision-making.

[0040] Advantages of multi-algorithm fusion: The improved NSGA-Ⅲ algorithm excels at handling Pareto front searches under continuous variables and complex constraints; the improved ant colony algorithm is adept at finding efficient paths in discrete task sequencing and resource allocation problems. The combination of the two enables the system to optimize both piling process parameters (continuous variables) and job task sequences (discrete variables).

[0041] Overall benefits: In practical engineering applications, the system improves overall work efficiency by about 18% and reduces equipment energy consumption by about 12% while ensuring construction safety (such as reducing the peak hoisting stress by more than 20%), achieving a win-win situation in terms of safety, efficiency and economy.

[0042] It should be further explained that the intelligent sensing and recognition process of the above-mentioned operating mode is as follows: S11, Multi-source heterogeneous data acquisition and preprocessing: A data acquisition network is established through the multi-source data acquisition and fusion module, covering three types of data sources: operation commands, equipment status, and environmental parameters. Operation command data includes: hook lifting commands, pile hammer start and stop commands, ship propulsion commands, anchoring control commands, etc. Equipment status data includes: crane angle sensor readings, pile driving frame pressure data, hydraulic system oil pressure and temperature. Environmental parameter data includes: wave height period, wind speed and direction, and water flow speed and direction. S12, multi-dimensional feature engineering and fusion, extracts statistical features from time series data: mean, variance, peak value, trend slope, etc., and extracts logical features from operation instruction sequences: instruction combination pattern, execution order, time interval, etc., and uses principal component analysis (PCA) method for feature dimensionality reduction; S1.3, Job pattern recognition based on deep learning: Construct a deep neural network classifier and use the softmax function to output the probability distribution of each job pattern. P(m|X) = softmax(W·X + b), Where X is the feature vector, W is the weight matrix, b is the bias term, m represents the job mode, and the output is the probability distribution of each job mode.

[0043] The multi-objective dynamic collaborative optimization module employs multi-objective optimization algorithms, including the improved NSGA-Ⅲ algorithm and the improved ant colony algorithm. The implementation methods are illustrated using virtual test piling and process parameter optimization: S41, Parallel Simulation and Rapid Evaluation, runs hundreds of virtual test pile driving simulation tasks in parallel on a cloud computing platform, with simulation parameter space covering key parameters such as hammer energy, frequency, and hammer drop height. S42, Multi-objective optimization decision-making: Constructing a mathematical model for the pile driving operation optimization problem: min f(x) = [f1(x), f2(x), f3(x)], st g j (x) ≤ 0, j = 1,2,...,m, Where f1 is the negative penetration rate, f2 is the maximum stress, f3 is the energy cost, and g j (x) represents the j-th constraint function, and the constraint function g j(x) defines the feasible region of the decision variable x, and uses the improved NSGA-Ⅲ algorithm and the improved ant colony algorithm to solve for the optimal solution set; Multi-objective collaborative optimization and decision support are illustrated through a construction plan. The above-mentioned work plan includes at least a construction plan, and an example of a construction plan is provided below: S51, Construction of a comprehensive performance indicator system and establishment of a unified evaluation framework: Performance = ∑(w i ·I i ) Where I i For the normalized indicators, w i For the corresponding weights, Performance represents the evaluation framework; S52, Global optimization solution: A construction sequence optimization model is established through the multi-objective dynamic collaborative optimization module. max∑(Performance k ·x k ), st∑(Resource k · x k )≤Budget, Where x k is the decision variable, representing whether to select the k-th construction scheme. The multi-objective optimization algorithm in the multi-objective dynamic collaborative optimization module selects the construction scheme with the highest comprehensive performance from the k construction schemes as the decision output.

[0044] The engineering vessel digital twin system also includes an adaptive switching module, which is used to adaptively switch and fuse the operational digital twin model. Specifically, this module includes: S21, Construct digital twin sub-models specific to multiple operation modes, including: constructing a crane operation twin sub-model, which establishes the dynamic equation of the crane swing based on the Lagrange equation; constructing a pile driving operation twin sub-model, which uses the wave equation to describe the propagation of impact stress waves in the pile; and constructing a transient coupling twin sub-model of the crane and the pile driving frame, which describes the interaction between the two when they are working together through dynamic coupling terms. S22, Based on real-time pattern recognition and twin model fusion, a fuzzy logic-based model weight allocator is designed. It takes the probability distribution of the work pattern output by the integrated model of work pattern recognition and motion response prediction as input, calculates the real-time weight coefficients of each corresponding twin model, and generates the final digital twin state and response output through weighted fusion. y fused = Σ (w i * y i ) Among them, Σw i = 1, the weight coefficient is dynamically adjusted according to the pattern confidence to achieve a smooth transition between the outputs of different twin models; S23, Based on twin prediction, assess equipment interference risk by establishing a refined three-dimensional geometric model of equipment such as cranes and piling frames in a digital twin environment, and calculating the minimum real-time distance between equipment based on the motion response prediction sequence: d min (t) = min(dis(part i (t), part j (t))) for all i, j, Where d min (t) represents the minimum real-time distance, part i (t), part j (t) represents the position of the geometric center or key point of the i-th and j-th equipment components (such as crane boom, pile driver, pile body, etc.) in three-dimensional space at time t, respectively, dis represents the distance function, and for all i, j represents the traversal description; Prospective collision detection is performed based on predicted trajectories, and based on d min (t) is compared with the preset three-level safety threshold to generate a graded interference early warning signal.

[0045] This step enables precise switching and smooth transition of multiphysics models, ensuring simulation accuracy under different operating conditions.

[0046] It should be understood that the above embodiments are one or more embodiments of the present invention, and there are many other embodiments and variations based on the present invention; any variations and modifications made by those skilled in the art through the present invention without making pioneering innovations are all within the protection scope of the present invention.

Claims

1. A digital twin system for engineering vessels integrating lifting and piling functions, characterized in that, include: The multi-source data acquisition and fusion module is used to integrate multi-source scene data of the target engineering vessel's operating environment. The multi-source scene data includes three-dimensional structural data of the vessel and equipment, equipment operating parameters, and hydrological data of the operating sea area. After preprocessing, the vessel equipment operating status is output. The operational digital twin model construction module is used to construct an operational digital twin model of the target engineering vessel based on the multi-source scene data, and to perform coupling effect simulation verification on the operational digital twin model. Based on the deviation analysis between the simulation results and the physical monitoring data, the module performs dynamic parameter calibration and mechanism model supplementation to obtain an optimized operational digital twin model. The dynamic working condition perception and analysis module, based on the working condition data output by the optimized digital twin model of the operation, uses a recurrent neural network to identify the key working condition characteristics of the target engineering vessel and its lifting and piling equipment, and obtains a first working condition feature set including equipment health status, real-time operation progress and the working environment. The flexible rule knowledge management and update module is used to perform online transfer learning based on the first working condition feature set and update the flexible rule knowledge base. The flexible rule knowledge base stores multi-layered operation management and safety control rules oriented towards integrated lifting and piling functions. The intelligent decision-making reasoning module is used to construct a decision-making reasoning model, and to perform multi-objective game operation on the elastic rule knowledge base through the decision-making reasoning model to generate the first operation management decision for the coordinated operation of lifting and piling. The multi-objective dynamic collaborative optimization module, based on the first operation management decision, simultaneously processes multiple sub-objectives such as operation safety risk, operation efficiency and equipment energy consumption through a multi-objective optimization algorithm to generate a second operation management decision; The decision execution monitoring and feedback module is used to convert the second operation management decision into specific collaborative operation instructions for the lifting and piling equipment of the engineering vessel, send them to the ship-shore control center, and monitor the execution process and effect of the collaborative operation instructions in real time. The digital twin model verification and iteration module is used to verify and iteratively update the collaborative operation instructions through the operation digital twin model.

2. The digital twin system for engineering vessels integrating lifting and piling functions according to claim 1, characterized in that: The operational digital twin model construction module constructs an operational digital twin model of the target engineering vessel, including: Based on the three-dimensional structural data of the vessel and equipment, a three-dimensional geometric model of the target engineering vessel and lifting and piling equipment is constructed, and the hydrological data of the operating sea area is mapped into the three-dimensional geometric model to generate an initial operating scene model that integrates environmental features. A real-time dynamic model of the equipment is constructed based on the equipment operating parameters. The spatial coordinate system of the real-time dynamic model of the equipment is calibrated and matched with that of the initial operation scenario model. The calibrated real-time dynamic model of the equipment is then coupled with the initial operation scenario model to construct an operational digital twin model of the target engineering vessel. The coupling effect of the digital twin model of the operation was simulated and verified, and the verification results are as follows: The design includes a set of verification scenarios for the typical coupling effects of lifting and piling. These scenarios include data on the impact of ship heeling caused by lifting during ship lifting operations on the verticality of piling, and the reaction of ship vibration caused by piling impact loads on the stability of the crane. The input parameters of the verification working condition set are input into the digital twin model of the operation for simulation calculation to obtain the twin simulation results; Simultaneously collect physical monitoring data of the target engineering vessel when it performs the corresponding operations of the verification condition set during actual operations; The twin simulation results are spatiotemporally aligned and compared with the physical monitoring data to calculate the simulation accuracy deviation of each coupling link. The simulation accuracy deviation includes the prediction deviation of the dynamic response of the crane hoisting, the prediction deviation of the ship attitude disturbance during the pile driving process, and the prediction deviation of the energy transfer between equipment. The distribution characteristics of simulation accuracy deviations are analyzed to identify the weak links in the model that cause the deviations. These weak links include inaccurate parameter settings, missing sub-models, and insufficient modeling of coupling effects. Based on the analysis results of the weak links, targeted model calibration and enhancement are initiated. Through the process of dynamic parameter calibration and mechanism model supplementation, an optimized digital twin model of the operation is generated.

3. The digital twin system for engineering vessels integrating lifting and piling functions according to claim 2, characterized in that: The operational digital twin model construction module also includes steps for model calibration and mechanism supplementation based on verification results, and for generating an optimized model, specifically including: N31. Analyze the simulation verification results of coupling effects, and based on the analysis of simulation accuracy deviation, identify and extract the list of model parameters that need to be calibrated and the list of coupling mechanism model items that need to be supplemented. N32. For model parameters that need to be calibrated, by comparing the differences between simulation predictions and physical monitoring values, a parameter calibration model based on inversion analysis or system identification is established, and cross-validation is performed using a historical optimization case library to complete the dynamic update of key dynamic parameters, material parameters, and fluid dynamic coefficients. N33. For the coupling mechanism model items that need to be supplemented, analyze whether the current model accurately describes the interactive physical process. If it does not accurately describe it, generate the corresponding supplementary mechanism sub-model by calling the mechanism model library or starting the first principle derivation, and integrate it into the original model framework in a modular way. N34. For the operational digital twin model after parameter dynamic calibration and mechanism model supplementation, the verification case set is used to perform secondary coupling effect simulation verification, and the updated simulation accuracy deviation is calculated. N35. Determine whether the updated simulation accuracy deviation meets the preset model fidelity threshold. If it does, mark the job digital twin model as the optimized job digital twin model and output it. If it does not meet the threshold, return to step N31, generate a new calibration and supplementary requirement list based on the secondary verification results, and start a new round of model iteration optimization process.

4. The digital twin system for engineering vessels integrating lifting and piling functions according to claim 1, characterized in that: The dynamic working condition perception and analysis module uses a hybrid recurrent neural network structure of LSTM and GRU. This hybrid network structure includes an input layer, a dual-branch processing layer, a fusion layer, a feature extraction layer, and a fully connected layer. The processing flow of this hybrid network structure includes: The input layer of the hybrid network structure receives a sequence of time-series work condition data output from the optimized digital twin model of the operation, and the input dimension is consistent with the feature dimension of the time-series work condition data. The time-series operating condition data is simultaneously input into the LSTM branch and the GRU branch for parallel processing; The LSTM branch embeds a spatiotemporal attention mechanism, and the formula for calculating the attention weights at time step t is as follows: α t =Softmax(W h h t-1 +W x x t +b), Where, α t Represented as an attention weight vector, W h Represented as the hidden state weight matrix, h t-1 Represented as the hidden state at time step t-1, W x Represented as the input weight matrix, x t denoted as the ship equipment operating condition at time step t, and b represents the bias term; The output of the LSTM branch at each time step t Output of the GRU branch The fusion is performed using dynamic weighting to obtain the fused output sequence; The fused output sequence is further mapped through a fully connected layer to generate a high-dimensional vector containing time-dependent features, which serves as the final output of the dynamic working condition perception and analysis module.

5. A digital twin system for engineering vessels integrating lifting and piling functions according to claim 1, characterized in that: In the flexible rule knowledge management and update module, the update triggering conditions for the flexible rule knowledge base include: Signals of sudden changes in equipment health status and signals of changes in the working environment from the dynamic working condition perception and analysis module; The decision simulation deviation detected by the digital twin simulation verification and iteration module exceeds a preset threshold; And manual rule revision or priority adjustment instructions input by operators through human-computer interaction terminals.

6. A digital twin system for engineering vessels integrating lifting and piling functions according to claim 1, characterized in that: The intelligent decision-making reasoning module includes the construction and operation of the decision-making reasoning model, comprising: Each operation management and safety control rule in the flexible rule knowledge base is transformed into a basic processing unit in a fuzzy Petri net. The preconditions of each rule are mapped to the input location, the conclusion or action of the rule is mapped to the output location, and the activation logic of the rule itself is mapped to the transition connecting the two. The lifting safety rule, piling accuracy rule, and operation efficiency rule are defined as input layer nodes of the fuzzy Petri net. By setting the activation threshold of the transition and the fuzzy inference function, the competition and cooperation relationship between rules is simulated. According to the preset operation scenario and the real-time input working condition feature set, the trigger strength of each path in the fuzzy Petri net is calculated. Based on the trigger strength and the preset priority of the rule, a dynamic priority decision tree for coordinating multiple rule conflicts is generated. Input the current first working condition feature set into the dynamic priority decision tree, traverse the tree structure to obtain all feasible initial candidate decision sets, call the operation digital twin model for simulation for each initial candidate decision, and calculate the estimated performance score after simulation. A deep Q-network is constructed as a policy optimizer. The state space of the policy optimizer is defined by the first condition feature set and the initial candidate decision. The action space of the policy optimizer is a set of operations that adjust the initial candidate decision. The policy optimizer uses the estimated performance score as the initial reward signal and activates the optimization process when the policy optimization condition is triggered.

7. A digital twin system for engineering vessels integrating lifting and piling functions according to claim 1, characterized in that: In the intelligent decision-making reasoning module, the flexible rule knowledge base is used to perform multi-objective game operations through the decision-making reasoning model, specifically including: Define job rule containers corresponding to different job types. Each job rule container stores a set of constraints and rules for a specific job type. Receive activation level signals from each rule in the elastic rule knowledge base, and parse the logical relationships between the rules; The activation level signals of each rule are input into the fuzzy inference unit of the decision reasoning model to calculate the comprehensive activation value of each task rule container, specifically expressed as follows: , where A c R is represented as the overall activation value of the job rule container c. c w represents the number of rules associated with job container c. c,r Let b represent the weight coefficient of the r-th rule in the job container c. c,r This indicates the actual activation level of the r-th rule under the current operating conditions; Based on the comprehensive activation value of each operation rule container, a multi-level dynamic priority decision tree is constructed. The decision tree uses the operation stage, equipment status and risk assessment as nodes, and the activation value of the rule container as the node connection strength. Based on the dynamic target weights set for the current task, a set of candidate decision paths is generated through a dynamic priority decision tree; The candidate decision paths are evaluated using a deep Q-network with a reinforcement learning mechanism, and the optimal decision path that maximizes long-term cumulative rewards is selected as the first task management decision output.

8. A digital twin system for engineering vessels integrating lifting and piling functions according to claim 1, characterized in that: The multi-objective dynamic collaborative optimization module employs multi-objective optimization algorithms including the improved NSGA-Ⅲ algorithm and the improved ant colony algorithm. The conflict objectives it addresses include at least minimizing operational safety risks, minimizing the cycle time of a single pile operation, minimizing overall energy consumption, and balancing the wear of key equipment.

9. A digital twin system for engineering vessels integrating lifting and piling functions according to claim 8, characterized in that: The multi-objective dynamic collaborative optimization module employs an improved NSGA-III algorithm to handle the optimization process of multiple conflicting objectives, including: Cluster analysis of conflict targets is performed based on historical operation management decision data to generate representative feature vectors under different decision-making models; The first operation management decision and the current real-time operating condition data are encoded into an initial population, and the reference point is initialized using the representative feature vector. The offspring population is generated by simulated binary crossover and polynomial mutation. During the fitness assessment phase, a first penalty term is applied to individuals that violate the lifting load safety threshold, and a second penalty term is applied to individuals that violate the pile driving verticality safety threshold. Excellent individuals are selected based on non-dominated ranking and environmental selection mechanisms. At the same time, a dynamic weight matrix is ​​constructed, which assigns time-varying weights to each conflict target according to the current operational situation. By incorporating the dynamic weight matrix into the environment selection process, the optimization priority of each conflicting objective is dynamically adjusted according to the operational situation, thereby generating a second operational management decision.

10. A digital twin system for engineering vessels integrating lifting and piling functions according to claim 8, characterized in that: The multi-objective dynamic collaborative optimization module employs an improved ant colony algorithm to handle multiple conflicting objectives, including: L11. In the multi-objective dynamic collaborative optimization module, an improved ant colony algorithm is used to handle multiple conflicting objectives, specifically including: L12. Decompose the first task management decision into multiple task sub-task nodes and construct a network graph model based on task timing and resource dependencies. An independent pheromone matrix is ​​established for each conflict target, and a multi-target pheromone update rule is designed, specifically as follows: For the pheromone update of target k on the path from node i to node j: ， ， in, Let ρ be the pheromone concentration for target k on path (i,j) at iteration number t, ρ represent the pheromone evaporation coefficient, and M represent the number of ants. Let Q be the pheromone increment left by the m-th ant in this iteration, and let Q be the pheromone intensity constant. The overall performance score of the path constructed by the m-th ant, where δ is the adjustment coefficient, and Rm is the overall performance score of the path constructed by the m-th ant. k The real-time risk assessment coefficient for target k; L13. Combining the physical constraints and efficiency requirements of lifting and piling operations, design a heuristic factor that integrates the characteristics of the operations, specifically expressed as follows: Heuristic factor for node i to node j of target k: ， Where, d ij τ represents the estimated time cost required to execute task j on node j. ij D represents the angle change required to switch from the job mode of node i to the job mode of node j. k This represents the deviation of the current solution from the target k, where φ and ν represent the weighting coefficients, respectively. L14. Ants select paths based on the probability of the product of pheromone concentration and heuristic factor, thus constructing a complete sequence of operation plans. L15. Iterate through steps L12 to L14. When the maximum number of iterations is reached or the solution quality converges, select the operation plan with the highest overall performance from the Pareto optimal solution set as the second operation management decision output.

Images