Modeling method of FPSO riser support structure

Abstract

The present invention discloses a modeling method for a FPSO riser support structure, which belongs to the field of simulation modeling. The modeling method has the following specific steps: I. basic parameters are obtained through data acquisition equipment and field experiments, and the application scenarios and working conditions of the FPSO riser support are analyzed to determine the modeling requirements; II. a preliminary riser support structure model of the riser support structure is constructed based on the demand analysis results; the present invention can dynamically adjust design and operation decisions, optimize long-term benefits, ensure that the structure can maintain high reliability and benefits during long-term operation, and help to improve the flexibility and adaptability of structural design and maintenance plans; enhance the security and privacy of the method, realize real-time adaptation to multi-field coupling effects in long-term operation, help to improve the performance of the riser support structure from different angles, reduce computing costs and improve computing efficiency, and avoid delays that may be caused by centralized computing.

Description

Technical Field

[0001] The invention relates to the field of simulation modeling, and in particular to a modeling method for a FPSO riser support structure. Background Art

[0002] With the continuous development of marine oil and gas resources, especially in deep and ultra-deep water areas, FPSO, as the core facility of offshore oil and gas production, plays an increasingly important role. Its main function is to transfer seabed oil and gas resources to floating platforms through risers, and store and unload them. Due to its unique floating structure, the FPSO platform can be flexibly deployed in a dynamic marine environment and has high adaptability. However, as a vital component of the FPSO platform, the riser support structure faces complex engineering challenges in its design and operation. The main function of the riser support structure is to support the riser from the seabed to the floating platform and ensure its stability and safety under dynamic environmental loads such as waves, wind, and currents. Due to the high uncertainty of environmental factors and the complexity of the riser structure itself, traditional design methods often rely on experience and static models, which cannot fully reflect the actual performance of the system in different operating environments; therefore, it is particularly important to invent a modeling method for the FPSO riser support structure.

[0003] The existing FPSO riser support structure modeling method cannot dynamically adjust the design and operation decisions, cannot guarantee that the structure can maintain high reliability and efficiency during long-term operation, and reduces the flexibility and adaptability of the structure design and maintenance plan; in addition, the existing FPSO riser support structure modeling method has poor security and privacy, and cannot adapt to the multi-field coupling effects in real time during long-term operation, which is not conducive to improving the performance of the riser support structure from different angles; therefore, we propose a FPSO riser support structure modeling method. Summary of the invention

[0004] The purpose of the present invention is to solve the defects in the prior art and to propose a modeling method for a FPSO riser support structure.

[0005] In order to achieve the above object, the present invention adopts the following technical solutions:

[0006] FPSO riser support structure modeling method, the specific steps of the modeling method are as follows:

[0007] Ⅰ. Obtain basic parameters through data acquisition equipment and field experiments, analyze the application scenarios and working conditions of FPSO riser support to determine modeling requirements;

[0008] II. Based on the demand analysis results, a preliminary riser support structure model is constructed;

[0009] Ⅲ. Combine real-time environmental data to conduct multi-field coupling simulation and optimize the geometric parameters and material configuration of the support structure;

[0010] IV. Analyze simulation data, extract implicit rules and key features, and optimize the riser support structure model;

[0011] V. Optimize the global riser support structure model through federated learning and analyze the benefits and risks of support structure design decisions in long-term operation;

[0012] VI. Based on the analysis results, the model design parameters are globally searched and optimized, and multiple rounds of simulation verification are performed to verify the performance under extreme conditions and diverse working conditions;

[0013] VII. Integrate the supporting structure model into the FPSO design system and conduct full system testing and verification of the model to complete the deployment and generate an operations manual.

[0014] As a further solution of the present invention, the specific steps of analyzing the application scenarios and working conditions of the FPSO riser support in step I to determine the modeling requirements are as follows:

[0015] S1.1: Use sensors and satellite data collection technology to obtain environmental parameters of the sea area, including wave characteristics, current velocity, and wind field distribution data, and describe the data collection results in time series;

[0016] S1.2: Use the Morison equation to calculate the impact of environmental loads on FPSO risers, select candidate materials based on the FPSO riser support structure requirements, collect the physical and chemical properties of each candidate material, including tensile strength, compressive strength and corrosion resistance, and calculate the allowable stress of the material based on the material mechanics formula;

[0017] S1.3: Calculate the power spectrum density of wave loads using the Pierson-Moskowitz spectrum analysis method, and evaluate the impact and fatigue effects of the dynamic characteristics of environmental loads on the riser based on the calculated power spectrum density of wave loads;

[0018] S1.4: According to the FPSO riser support structure goals set by the staff, the first-level evaluation criteria and second-level evaluation indicators are determined, and a hierarchical model corresponding to the target layer, criterion layer and scheme layer is generated. For each pair of criteria in the criterion layer, a comparison matrix is ​​constructed to compare them pairwise, determine their relative importance, and assign weights through the 9-level scaling method;

[0019] S1.5: After the weights are assigned, each column of the comparison matrix is ​​normalized to obtain the corresponding standardized matrix. Then, the average value of each row is calculated to obtain the weight vector of each criterion. The consistency ratio of the standardized matrix is ​​calculated. If the ratio is less than 0.1, it means that the consistency of the matrix is ​​acceptable. Otherwise, the comparison matrix is ​​readjusted.

[0020] S1.6: Based on the standardized matrix that meets the requirements, the weight vector of each criterion is used as the importance weight of the criterion, and the corresponding FPSO riser support structure model construction index is generated.

[0021] As a further solution of the present invention, the specific calculation formula of the Morison equation is as follows:

[0022] ;

[0023] In the formula, Represents the total load acting on the riser, in N; represents the drag coefficient; Represents the density of seawater, in kg / m³; Represents the cross-sectional area of ​​the structure, in m²; Represents the speed of ocean current or wave, in m / s; represents the coefficient of inertia; Represents the relative flow velocity in the riser, in m / s; Represents the acceleration of the relative flow velocity, in m / s²;

[0024] The specific calculation formula of Pierson-Moskowitz spectrum is as follows:

[0025] ;

[0026] In the formula, Represents the wave energy spectral density, in m² / Hz; Represents the wave frequency, unit Hz; and Represent the empirical coefficients respectively; Represents the acceleration due to gravity, in m / s²; Represents the peak frequency, in Hz; Represents significant wave height.

[0027] It is characterized in that the specific steps of constructing the preliminary riser support structure model of the riser support structure in step II are as follows:

[0028] S2.1: According to the demand analysis results, determine the geometric shape, main dimensions and boundary conditions of the FPSO riser support structure, then set the material properties of the riser support results, and use computer-aided design tools to create a three-dimensional geometric model of the riser support structure, including the riser body, support nodes and connectors;

[0029] S2.2: Discretize the created riser support structure model, and divide the riser support structure model into a finite number of units according to the structural characteristics of the beam structure, panel and shell structure, and the solid area, and generate a preliminary mesh division. Then, with the optimization goal of optimizing the unit quality and satisfying the mesh division density requirement, initialize the number of division volumes, pheromone concentration, volatility factor, pheromone influence factor, and heuristic function influence factor;

[0030] S2.3: According to the local characteristics of the grid division, a heuristic function is constructed based on the inverse of the unit quality index. Then, the initial position of each division body is initialized, and the heuristic value of each selectable node is calculated through the heuristic function. The corresponding selection probability is calculated based on the heuristic value and pheromone concentration of each node. According to the selection probability of each node, the next node is selected by random selection.

[0031] S2.4: Each partition body gradually completes the meshing of the entire model according to the selection probability. After each round, the meshing schemes of all partition bodies and their unit quality and meshing efficiency evaluation indicators are saved. The fitness of each scheme is evaluated based on the aspect ratio of the mesh of each partition scheme, the uniformity of size change, and the smoothness of the unit size transition area.

[0032] S2.5: Arrange the fitness values ​​of each partition from high to low, and based on the arrangement results, update the pheromone concentration of the path between each node based on the volatility factor, repeat the path selection and pheromone concentration update until the fitness value of the partition scheme changes below the preset threshold or reaches the maximum number of iterations, and output the best grid partition scheme with the highest fitness value as the final partition result of the riser support structure model.

[0033] It is characterized in that the specific steps of optimizing the geometric parameters and material configuration of the support structure in step III are as follows:

[0034] S3.1: Obtain real-time environmental data from sensors or external databases, normalize the acquired data, and smooth the data through interpolation or filtering algorithms to remove excess data noise. Then, based on the real-time environmental data, obtain the solid mechanics field, fluid dynamics field, and thermal field physical fields involved in the support structure in the actual application scenario, and clarify the interaction relationship between the physical fields. Generate a corresponding multi-field coupling simulation model based on the constructed riser support structure model;

[0035] S3.2: Apply fluid pressure and shear force to the solid surface, and dynamically adjust the fluid boundary according to the displacement of the structure surface to impose coupled boundary conditions on the interface between the solid and the fluid. If a thermal field exists, the heat transferred by the fluid is used as the heat load of the solid and the temperature change of the solid affects the thermal distribution of the fluid as well as the coupled boundary conditions;

[0036] S3.3: For the fluid field part of the riser support structure model after meshing, the computational fluid dynamics meshing is used, and a shared mesh is established at the fluid-solid interface or the two fields are coupled by interpolation method, and the solid and fluid of the riser support structure model are solved separately. If the interaction between the fluid and the structure is higher than the preset threshold, the solid and fluid are solved separately, and the information on the coupling interface is exchanged in each time iteration. Otherwise, the control equations of the solid and fluid are solved simultaneously;

[0037] S3.4: Solve each physical field in the preset time step, so that the solution of each physical field gradually approaches the stable solution until the calculated value reaches the preset convergence criterion or time step. After the calculation stops, extract the stress and deformation distribution of the solid structure, the pressure and velocity distribution of the fluid field, and the temperature change of the thermal field;

[0038] S3.5: Initialize the immune population set, each antibody represents a geometric parameter and material configuration scheme, and set the geometric parameter and material configuration range of the antibody, substitute the geometric parameter and material configuration of each antibody into the multi-field coupling simulation model, calculate its fitness value, and obtain the stress distribution, thermal deformation and fluid load distribution of the structure of different antibodies under the multi-field coupling action;

[0039] S3.6: For antibodies with fitness values ​​higher than a preset threshold, multiple copies with identical geometric parameters and material configurations as the original antibody are generated by cloning, and the cloned antibodies are mutated using a random perturbation method to generate new candidate solutions. The fitness values ​​of all mutated copies of the antibody are recalculated and compared with the fitness values ​​of the original antibody;

[0040] S3.7: If the mutated antibody has a higher fitness, the mutation scheme is retained; otherwise, the scheme is discarded and the original antibody is retained. The new antibody is merged with the current population, the optimal antibody is screened out according to the fitness value, the immune memory set is updated, and the iterative update is repeated until the fitness change is lower than the preset threshold or the number of iterations reaches the preset upper limit. The geometric parameters and material configuration scheme with the highest fitness value in the immune population set are output, and the geometric parameters and material configuration of the current multi-field coupling simulation model are updated.

[0041] It is characterized in that the specific steps of optimizing the riser support structure model in step IV are as follows:

[0042] S4.1: Collect the multi-field data generated by the simulation of the riser support structure model, normalize the data, format the processed multi-field data, and generate a time series input format that meets the conditions ,in Representative The input data vector at each moment, input multiple fields of data into the pre-trained RNN model;

[0043] S4.2: The RNN model forward propagates each group of input multi-field data, processes the input time series step by step through recursive calculation, updates the hidden state, and extracts the characteristic pattern in the time series. The output layer calculates the output corresponding to the hidden state sequence of each time step, and obtains the predicted value of the dynamic behavior and the corresponding feature extraction result;

[0044] S4.3: The principal component analysis method is used to reduce the dimension of the hidden state sequence generated by the RNN model, extract the dominant features, analyze the relationship between the features extracted by the RNN and the physical model parameters, and optimize the riser support structure model by adjusting the physical parameters corresponding to the key features.

[0045] It is characterized in that the specific steps of optimizing the global riser support structure model in step V are as follows:

[0046] S5.1: The central server sends the riser support structure model to each group of different simulation sites or experimental equipment, sets the global model parameters and the local model parameters of each node through random initialization, and adds a unique number label to each local model parameter, and uses the initial global model parameters as the main river position and the local model parameters of each node as the tributary position;

[0047] S5.2: Each node trains the riser support structure model based on its own local data with the purpose of minimizing the local objective function. After the training is completed, each node sends the update of the local model parameters to the central server. Then the master node aggregates the local model parameters using the weighted average method and updates the global parameters.

[0048] S5.3: According to the flow direction update rule, the position of each tributary is updated by simulating the aggregation effect of water flow, that is, for each tributary, the local model parameters are corresponding to each tributary. After the tributary is updated, the position of the main river is optimized according to the gradient information of the global loss function, and then the position of the tributary is randomly adjusted through the evaporation mechanism;

[0049] S5.4: Calculate the loss value of each tributary and evaluate the impact of the new local model parameters on the model performance. Then update the main river position to the tributary parameters with the best fitness. Repeat the parameter update. When the rate of change of the global loss function is less than the set threshold or the maximum number of iterations is reached, stop the parameter update.

[0050] S5.5: Weightedly combine the optimized global parameters with the global parameter aggregation results of federated learning, use verification data to evaluate the fitness of the parameters after collaborative modeling, and record performance indicators. If the local loss function value of a node is higher than the preset threshold, further adjust the weight ratio, and repeatedly update the local parameters and global parameters. When the rate of change of the global loss function is less than the set threshold, select the weight ratio with the highest fitness as the optimal weight ratio, and dynamically update the model in real time.

[0051] As a further solution of the present invention, the specific calculation formula for random initialization is as follows:

[0052] ;

[0053] In the formula, represents the initial position of the main river; Represents the total number of child nodes; Represents the index of each child node; Representative Local model parameters of each node;

[0054] The specific calculation formula of the local objective function is as follows:

[0055] ;

[0056] In the formula, Representative The local loss function of each node; Representative Number of node data samples; represents the loss function; Representative The node Input data and labels;

[0057] The specific calculation formula of weighted average is as follows:

[0058] ;

[0059] In the formula, Representative The local model parameters after the node weights are adjusted.

[0060] It is characterized in that the specific steps of analyzing the benefits and risks of the design decision of the support structure in long-term operation in step V are as follows:

[0061] S6.1: Collect all the participants in the design and operation of the riser support structure, define the strategy space of each participant based on all the strategies that each participant may choose in the decision-making process, and construct the corresponding utility function based on the goal of each participant to maximize the long-term benefits or minimize the risks by choosing a specific strategy to represent the relationship between the benefits of the participant and the strategy he chooses;

[0062] S6.2: Collect all state information of the riser support structure and construct the corresponding state space. According to all actions that each participant can take in each state, construct the action space corresponding to different participants, and set the corresponding reward function based on design cost, operation income and maintenance cost;

[0063] S6.3: Calculate the probability of a state in the state space transferring to the next state by randomly taking any action in the action space, set a discount factor based on the operation cycle of the riser support structure and the information of various long-term planning objectives, and calculate the expected value of the maximum long-term utility that the participant can obtain in the subsequent time under the current state through the value function based on the benefits of each state and the expected future benefits;

[0064] S6.4: Use the Bellman equation to iteratively solve the problem to obtain the optimal action of each participant at each time. In the iterative solution process, recursively calculate the optimal strategy of each participant at each time point to find the equilibrium solution, that is, the equilibrium point is the point where the participant cannot increase the utility by unilaterally changing its strategy;

[0065] S6.5: Analyze the benefits and risks of each participant under Nash equilibrium, comprehensively evaluate the long-term effects of the design plan, calculate the utility, cost and risk factors of different design plans and operation strategy combinations, and evaluate the most feasible and sustainable design plan in long-term operation.

[0066] As a further solution of the present invention, the specific calculation formula of the utility function is as follows:

[0067] ;

[0068] In the formula, Representative The utility of each participant, ; Represents the total number of child nodes The payoff function of each participant; Representative The cost function of each participant represents the cost of their decision. ;

[0069] The specific calculation formula of the value function is as follows:

[0070] ;

[0071] In the formula, Representative Participants in the state The value of the following, that is, its long-term utility; Represents the sub-action space Middle Actions; Represents in state Next, take action Instant rewards when represents the discount factor; Represents in state , take action Transfer to state The transition probability of Representation status Next, the expected utility in the future.

[0072] Compared with the prior art, the present invention has the following beneficial effects:

[0073] 1. The FPSO riser support structure modeling method collects all participants in the design and operation of the riser support structure, defines the strategy space of each participant based on all the strategies that each participant will choose in the decision-making process, and constructs a corresponding utility function based on the goal of each participant to obtain the maximum long-term benefit or minimize the risk by selecting a specific strategy to express the relationship between the benefits of the participant and the strategy selected by it. The corresponding state space, the action space corresponding to different participants and the reward function are constructed, and the transition probability of randomly taking any action in the action space under any state in the state space so that the state transfers to the next state is calculated. The discount factor is set based on the operation cycle of the riser support structure and the information of various long-term planning objectives. According to the benefits of each state and the expected benefits in the future, the value function is used to calculate the participants' subsequent benefits in the current state. The expected value of the maximum long-term utility that can be obtained within a certain period of time is solved iteratively using the Bellman equation to obtain the optimal action of each participant at each moment. In the process of iterative solution, the equilibrium solution is found by recursively calculating the optimal strategy of each participant at each time point, that is, the inability of the participant to increase the utility by unilaterally changing its strategy is taken as the equilibrium point. The benefits and risks of each participant under the Nash equilibrium are analyzed, the long-term effect of the design scheme is comprehensively evaluated, the utility, cost and risk factors of different combinations of design schemes and operation strategies are calculated, and the most feasible and sustainable design scheme in long-term operation is evaluated. The design and operation decisions can be adjusted dynamically, the long-term benefits are optimized, and the high reliability and benefits of the structure can be ensured in the long-term operation process, which is helpful to improve the flexibility and adaptability of the structural design and maintenance scheme.

[0074] 2. The present invention sends the riser support structure model to each group of different simulation sites or experimental equipment through the central server, sets the global model parameters and the local model parameters of each node through random initialization, and adds a unique number label to each local model parameter, takes the initial global model parameters as the main river position, and the local model parameters of each node as the tributary position. Each node trains the riser support structure model based on its own local data with the purpose of minimizing the local objective function. After the training is completed, each node sends the update of the local model parameters to the central server, and then the master node aggregates the local model parameters using the weighted average method, updates the global parameters, and updates the position of each tributary according to the flow direction update rule to simulate the aggregation effect of the water flow. After the tributary is updated, the main river position is optimized according to the gradient information of the global loss function, and then the position of the tributary is randomly adjusted through the evaporation mechanism, the loss value of each tributary is calculated, and the new local model parameters are evaluated. The impact of the number on the model performance is then calculated, and the main river position is updated to the tributary parameters with the best fitness. The parameter update is repeated. When the rate of change of the global loss function is less than the set threshold or the maximum number of iterations is reached, the parameter update is stopped, and the optimized global parameters are weightedly combined with the global parameter aggregation results of federated learning. The fitness of the parameters after collaborative modeling is evaluated using the verification data, and the performance indicators are recorded. If the local loss function value of a node is higher than the preset threshold, the weight ratio is further adjusted, and the local parameters and global parameters are repeatedly updated. When the rate of change of the global loss function is less than the set threshold, the weight ratio with the highest fitness is selected as the optimal weight ratio, and the model is dynamically updated in real time to enhance the security and privacy of the method, realize real-time adaptation to the multi-field coupling effects in long-term operation, and help to improve the performance of the riser support structure from different angles, reduce computing costs and improve computing efficiency, and avoid delays that may be caused by centralized computing. BRIEF DESCRIPTION OF THE DRAWINGS

[0075] The accompanying drawings are used to provide further understanding of the present invention and constitute a part of the specification. They are used to explain the present invention together with the embodiments of the present invention and do not constitute a limitation of the present invention.

[0076] Figure 1 This is a flowchart of the FPSO riser support structure modeling method proposed in the present invention. DETAILED DESCRIPTION Example 1

[0077] Reference Figure 1 , FPSO riser support structure modeling method, the specific steps of the modeling method are as follows:

[0078] Basic parameters are obtained through data acquisition equipment and field experiments, and the application scenarios and working conditions of FPSO riser supports are analyzed to determine modeling requirements.

[0079] Specifically, the environmental parameters of the sea area, including wave characteristics, current velocity and wind field distribution data, are obtained through sensors and satellite data collection technology, and the data collection results are described in time series. The Morison equation is used to calculate the impact of environmental loads on FPSO risers. According to the requirements of the FPSO riser support structure, candidate materials are selected, and the tensile strength, compressive strength and corrosion resistance of each candidate material are collected. The allowable stress of the material is calculated according to the material mechanics formula, and the power spectral density of the wave load is calculated using the Pierson-Moskowitz spectrum analysis method. Based on the calculated power spectral density of the wave load, the impact and fatigue effects of the dynamic characteristics of the environmental load on the riser are evaluated. According to the FPSO set by the staff, The riser support structure target is determined, the first-level evaluation criteria and the second-level evaluation indicators are determined, and a hierarchical model corresponding to the target layer, criterion layer and scheme layer is generated. For each pair of criteria in the criterion layer, a comparison matrix is ​​constructed for pairwise comparison to determine their relative importance, and weights are assigned through the 9-level scaling method. After the weight assignment is completed, each column of the comparison matrix is ​​normalized to obtain the corresponding standardized matrix, and then the average value of each row is calculated to obtain the weight vector of each criterion, and the standardized matrix consistency ratio is calculated. If the ratio is less than 0.1, it means that the matrix consistency is acceptable. Otherwise, the comparison matrix is ​​readjusted. Based on the standardized matrix that meets the requirements, the weight vector of each criterion is used as the importance weight of the criterion, and the corresponding FPSO riser support structure model construction indicator is generated.

[0080] In this embodiment, the specific calculation formula of the Morison equation is as follows:

[0081] ;

[0082] In the formula, Represents the total load acting on the riser, in N; represents the drag coefficient; Represents the density of seawater, in kg / m³; Represents the cross-sectional area of ​​the structure, in m²; Represents the speed of ocean current or wave, in m / s; represents the coefficient of inertia; Represents the relative flow velocity in the riser, in m / s; Represents the acceleration of the relative flow velocity, in m / s²;

[0083] The specific calculation formula of Pierson-Moskowitz spectrum is as follows:

[0084] ;

[0085] In the formula, Represents the wave energy spectral density, in m² / Hz; Represents the wave frequency, unit Hz; and Represent the empirical coefficients respectively; Represents the acceleration due to gravity, in m / s²; Represents the peak frequency, in Hz; Represents significant wave height.

[0086] Based on the demand analysis results, a preliminary riser support structure model of the riser support structure is constructed.

[0087] Specifically, according to the results of the demand analysis, the geometric shape, main dimensions and boundary conditions of the FPSO riser support structure are determined, and then the material properties after the riser support results are set. A three-dimensional geometric model of the riser support structure is created using a computer-aided design tool, including the riser body, support nodes and connectors. The created riser support structure model is discretized, and the riser support structure model is divided into a finite number of units according to the structural characteristics of the beam structure, panel and shell structure and the solid area, and a preliminary grid division is generated. Then, in order to optimize the unit quality and meet the grid division density requirements as the optimization goal, the number of division bodies, pheromone concentration, volatility factor, pheromone influence factor and heuristic function influence factor are initialized. According to the local characteristics of the grid division and based on the inverse of the unit quality index, a heuristic function is constructed. Then, the initial position of each division body is initialized, and the heuristic function is used to calculate each The heuristic value of the node is selected, and the corresponding selection probability is calculated based on the heuristic value and pheromone concentration at each node. According to the selection probability of each node, the next node is selected by random selection. Each partition body gradually completes the grid division of the entire model according to the selection probability. After each round, the grid division schemes of all partition bodies and their unit quality and division efficiency evaluation indicators are saved. The fitness of each scheme is evaluated according to the aspect ratio of the grid of each division scheme, the uniformity of size change and the smoothness of the unit size transition area. The fitness value of each partition body is arranged from high to low, and according to the arrangement result, the pheromone concentration of the path between each node is updated based on the volatility factor. The path selection and pheromone concentration update are repeated until the fitness value of the division scheme changes below the preset threshold or reaches the maximum number of iterations. The best grid division scheme with the highest fitness value is output as the final division result of the riser support structure model.

[0088] Combine real-time environmental data to perform multi-field coupling simulation and optimize the geometric parameters and material configuration of the support structure.

[0089] Specifically, real-time environmental data is obtained from sensors or external databases, the obtained data is normalized, and the data is smoothed through interpolation or filtering algorithms to remove excess data noise. Then, the solid mechanics field, fluid dynamics field, and thermal field physical fields involved in the support structure in the actual application scenario are obtained based on the real-time environmental data, and the interaction relationship between the physical fields is clarified. Based on the constructed riser support structure model, a corresponding multi-field coupling simulation model is generated, and the fluid pressure and shear force are applied to the solid surface. According to the displacement of the structure surface, the fluid boundary is dynamically adjusted to impose coupling boundary conditions on the interface between the solid and the fluid. If the thermal field exists, the fluid transmission Heat is used as the heat load of the solid and the temperature change of the solid affects the thermal distribution of the fluid. At the same time, it is used as a coupling boundary condition. In the fluid field part of the riser support structure model after meshing, computational fluid dynamics meshing is adopted, and a shared grid is established at the fluid-solid interface or the two fields are coupled through interpolation method. The solid and fluid of the riser support structure model are solved separately. If the interaction between the fluid and the structure is higher than the preset threshold, the solid and the fluid are solved separately, and the information on the coupling interface is exchanged in each time iteration. Otherwise, the control equations of the solid and the fluid are solved jointly. In the preset time step, each physical field is solved so that the solution of each physical field gradually approaches the stable solution until the solid and the fluid are connected. When the calculated value reaches the preset convergence criterion or time step, the calculation stops, and the stress and deformation distribution of the solid structure, the pressure and velocity distribution of the fluid field, and the temperature change of the thermal field are extracted. The immune population set is initialized, and each antibody represents a geometric parameter and material configuration scheme. The geometric parameter and material configuration range of the antibody are set, and the geometric parameter and material configuration of each antibody are substituted into the multi-field coupling simulation model to calculate its fitness value, and the stress distribution, thermal deformation and fluid load distribution of the structure of different antibodies under the multi-field coupling are obtained. For antibodies with fitness values ​​higher than the preset threshold, multiple copies of the geometric parameters and material configuration that are exactly the same as the original antibody are generated by cloning, and then the geometric parameters and material configurations of the original antibody are used. The random perturbation method mutates the cloned antibodies to generate new candidate solutions, recalculates the fitness values ​​of all mutated replica antibodies, and compares them with the fitness of the original antibodies. If the mutated antibody has a higher fitness, the mutation scheme is retained. Otherwise, the scheme is discarded and the original antibody is retained. The new antibody is merged with the current population, the optimal antibody is screened out according to the fitness value, and the immune memory set is updated. The iterative update is repeated until the fitness change is lower than the preset threshold or the number of iterations reaches the preset upper limit, and the geometric parameters and material configuration scheme with the highest fitness value in the immune population set are output, and the geometric parameters and material configuration of the current multi-field coupling simulation model are updated.

[0090] Analyze simulation data, extract implicit rules and key features, and optimize the riser support structure model.

[0091] Specifically, collect the multi-field data generated by the simulation of the riser support structure model, normalize the data, format the multi-field data, and generate a time series input format that meets the conditions. ,in Representative The input data vector at each moment is obtained, and the multi-field data is input into the pre-trained RNN model. The RNN model performs forward propagation on each group of multi-field data input, and gradually processes the input time series through recursive calculation, updates the hidden state, and extracts the characteristic patterns in the time series. The output layer calculates the output corresponding to the hidden state sequence of each time step, and obtains the predicted value of the dynamic behavior and the corresponding feature extraction result. The principal component analysis method is used to reduce the dimension of the hidden state sequence generated by the RNN model, extract the dominant features, analyze the relationship between the features extracted by the RNN and the physical model parameters, and optimize the riser support structure model by adjusting the physical parameters corresponding to the key features. Example 2

[0092] Reference Figure 1 , FPSO riser support structure modeling method, the specific steps of the modeling method are as follows:

[0093] The global riser support structure model is optimized through federated learning, and the benefits and risks of the support structure design decisions in long-term operation are analyzed.

[0094] Specifically, the central server sends the riser support structure model to different groups of simulation sites or experimental equipment, sets the global model parameters and the local model parameters of each node through random initialization, adds a unique number label to each local model parameter, takes the initial global model parameters as the main river position, and the local model parameters of each node as the tributary position. Each node trains the riser support structure model based on its own local data with the purpose of minimizing the local objective function. After the training is completed, each node sends the update of the local model parameters to the central server, and then the master node aggregates the local model parameters using the weighted average method, updates the global parameters, and updates the position of each tributary according to the flow direction update rule to simulate the aggregation effect of the water flow, that is, for the local model parameters corresponding to each tributary, after the tributary is updated, the main river position is updated according to the gradient information of the global loss function. The position is optimized, and then the position of the tributaries is randomly adjusted through the evaporation mechanism, the loss value of each tributary is calculated, and the impact of the new local model parameters on the model performance is evaluated. Then the main river position is updated to the tributary parameters with the best fitness, and the parameter update is repeated. When the rate of change of the global loss function is less than the set threshold or the maximum number of iterations is reached, the parameter update is stopped, and the optimized global parameters are weightedly combined with the global parameter aggregation results of federated learning. The validation data is used to evaluate the fitness of the parameters after collaborative modeling, and the performance indicators are recorded. If there is a node whose local loss function value is higher than the preset threshold, the weight ratio is further adjusted, and the local parameters and global parameters are repeatedly updated. When the rate of change of the global loss function is less than the set threshold, the weight ratio with the highest fitness is selected as the optimal weight ratio, and the model is dynamically updated in real time.

[0095] It should be further explained that the specific calculation formula for random initialization is as follows:

[0096] ;

[0097] In the formula, represents the initial position of the main river; Represents the total number of child nodes; Represents the index of each child node; Representative Local model parameters of each node;

[0098] The specific calculation formula of the local objective function is as follows:

[0099] ;

[0100] In the formula, Representative The local loss function of each node; Representative Number of node data samples; represents the loss function; Representative The node Input data and labels;

[0101] The specific calculation formula of weighted average is as follows:

[0102] ;

[0103] In the formula, Representative The local model parameters after the node weights are adjusted.

[0104] Specifically, the participants in the design and operation of the riser support structure are collected, and the strategy space of each participant is defined based on all the strategies that each participant will choose in the decision-making process. Based on the goal of each participant to obtain the maximum long-term benefit or minimize the risk by choosing a specific strategy, a corresponding utility function is constructed to represent the relationship between the benefit of the participant and the strategy selected by the participant. All state information of the riser support structure is collected and the corresponding state space is constructed. According to all the actions that each participant can take in each state, the action space corresponding to different participants is constructed, and the corresponding reward function is set based on the design cost, operation income and maintenance cost. The transition probability of randomly taking any action in the action space in any state in the state space so that the state transfers to the next state is calculated. Based on the riser support structure The discount factor is set based on the information of the operation cycle and various objectives of the long-term plan. According to the benefits of each state and the expected benefits in the future, the expected value of the maximum long-term utility that the participants can obtain in the subsequent time under the current state is calculated through the value function. The Bellman equation is used for iterative solution to obtain the optimal action of each participant at each moment. In the process of iterative solution, the equilibrium solution is found by recursively calculating the optimal strategy of each participant at each point in time, that is, the participant cannot increase the utility by unilaterally changing its strategy as the equilibrium point, analyze the benefits and risks of each participant under the Nash equilibrium, comprehensively evaluate the long-term effect of the design scheme, calculate the utility, cost and risk factors of different design schemes and operation strategy combinations, and evaluate the most feasible and sustainable design scheme in long-term operation.

[0105] In addition, it should be noted that the specific calculation formula of the utility function is as follows:

[0106] ;

[0107] In the formula, Representative The utility of each participant, ; Represents the total number of child nodes The payoff function of each participant; Representative The cost function of each participant represents the cost of their decision. ;

[0108] The specific calculation formula of the value function is as follows:

[0109] ;

[0110] In the formula, Representative Participants in the state The value of the following, that is, its long-term utility; Represents the sub-action space Middle Actions; Represents in state Next, take action Instant rewards when represents the discount factor; Represents in state , take action Transfer to state The transition probability of Representation status Next, the expected utility in the future.

[0111] Based on the analysis results, the model design parameters are globally searched and optimized, and multiple rounds of simulation verification are carried out to verify the performance under extreme conditions and diverse working conditions.

[0112] The supporting structure model is integrated into the FPSO design system and the model is fully tested and verified to complete the deployment and generate the operation manual.

Claims

1. A FPSO riser support structure modeling method, characterized in that: The specific steps of the modeling method are as follows: Ⅰ. Obtain basic parameters through data acquisition equipment and field experiments, analyze the application scenarios and working conditions of FPSO riser support to determine modeling requirements; II. Based on the demand analysis results, a preliminary riser support structure model is constructed; Ⅲ. Combine real-time environmental data to conduct multi-field coupling simulation and optimize the geometric parameters and material configuration of the support structure; IV. Analyze simulation data, extract implicit rules and key features, and optimize the riser support structure model; V. Optimize the global riser support structure model through federated learning and analyze the benefits and risks of support structure design decisions in long-term operation; VI. Based on the analysis results, the model design parameters are globally searched and optimized, and multiple rounds of simulation verification are performed to verify the performance under extreme conditions and diverse working conditions; VII. Integrate the support structure model into the FPSO design system and conduct full system testing and verification of the model to complete deployment and generate an operations manual; The specific steps of constructing the preliminary riser support structure model of the riser support structure described in step II are as follows: S2.1: According to the demand analysis results, determine the geometric shape, main dimensions and boundary conditions of the FPSO riser support structure, then set the material properties of the riser support results, and use computer-aided design tools to create a three-dimensional geometric model of the riser support structure, including the riser body, support nodes and connectors; S2.2: Discretize the created riser support structure model, and divide the riser support structure model into a finite number of units according to the structural characteristics of the beam structure, panel and shell structure, and the solid area, and generate a preliminary mesh division. Then, with the optimization goal of optimizing the unit quality and satisfying the mesh division density requirement, initialize the number of division volumes, pheromone concentration, volatility factor, pheromone influence factor, and heuristic function influence factor; S2.3: According to the local characteristics of the grid division, a heuristic function is constructed based on the inverse of the unit quality index. Then, the initial position of each division body is initialized, and the heuristic value of each selectable node is calculated through the heuristic function. The corresponding selection probability is calculated based on the heuristic value and pheromone concentration of each node. According to the selection probability of each node, the next node is selected by random selection. S2.4: Each partition body gradually completes the meshing of the entire model according to the selection probability. After each round, the meshing schemes of all partition bodies and their unit quality and meshing efficiency evaluation indicators are saved. The fitness of each scheme is evaluated based on the aspect ratio of the mesh of each partition scheme, the uniformity of size change, and the smoothness of the unit size transition area. S2.5: Arrange the fitness values ​​of each partition from high to low, and based on the arrangement result, update the pheromone concentration of the path between each node based on the volatility factor, repeat the path selection and pheromone concentration update until the fitness value of the partition scheme changes below the preset threshold or reaches the maximum number of iterations, and output the best grid partition scheme with the highest fitness value as the final partition result of the riser support structure model; The specific steps for optimizing the geometric parameters and material configuration of the support structure in step III are as follows: S3.1: Obtain real-time environmental data from sensors or external databases, normalize the acquired data, and smooth the data through interpolation or filtering algorithms to remove excess data noise. Then, based on the real-time environmental data, obtain the solid mechanics field, fluid dynamics field, and thermal field physical fields involved in the support structure in the actual application scenario, and clarify the interaction relationship between the physical fields. Generate a corresponding multi-field coupling simulation model based on the constructed riser support structure model; S3.2: Apply fluid pressure and shear force to the solid surface, and dynamically adjust the fluid boundary according to the displacement of the structure surface to impose coupled boundary conditions on the interface between the solid and the fluid. If a thermal field exists, the heat transferred by the fluid is used as the heat load of the solid and the temperature change of the solid affects the thermal distribution of the fluid as well as the coupled boundary conditions; S3.3: For the fluid field part of the riser support structure model after meshing, the computational fluid dynamics meshing is used, and a shared mesh is established at the fluid-solid interface or the two fields are coupled by interpolation method, and the solid and fluid of the riser support structure model are solved separately. If the interaction between the fluid and the structure is higher than the preset threshold, the solid and fluid are solved separately, and the information on the coupling interface is exchanged in each time iteration. Otherwise, the control equations of the solid and fluid are solved simultaneously; S3.4: Solve each physical field in the preset time step, so that the solution of each physical field gradually approaches the stable solution until the calculated value reaches the preset convergence criterion or time step. After the calculation stops, extract the stress and deformation distribution of the solid structure, the pressure and velocity distribution of the fluid field, and the temperature change of the thermal field; S3.5: Initialize the immune population set, each antibody represents a geometric parameter and material configuration scheme, and set the geometric parameter and material configuration range of the antibody, substitute the geometric parameter and material configuration of each antibody into the multi-field coupling simulation model, calculate its fitness value, and obtain the stress distribution, thermal deformation and fluid load distribution of the structure of different antibodies under the multi-field coupling action; S3.6: For antibodies with fitness values ​​higher than a preset threshold, multiple copies with identical geometric parameters and material configurations as the original antibody are generated by cloning, and the cloned antibodies are mutated using a random perturbation method to generate new candidate solutions. The fitness values ​​of all mutated copies of the antibody are recalculated and compared with the fitness values ​​of the original antibody; S3.7: If the mutated antibody has a higher fitness, the mutation scheme is retained. Otherwise, the scheme is discarded and the original antibody is retained. The new antibody is merged with the current population, and the optimal antibody is screened out according to the fitness value. The immune memory set is updated and iterative updates are repeated until the fitness change is lower than the preset threshold or the number of iterations reaches the preset upper limit. The geometric parameters and material configuration scheme with the highest fitness value in the immune population set are output, and the geometric parameters and material configuration of the current multi-field coupling simulation model are updated. The specific steps of optimizing the riser support structure model in step IV are as follows: S4.1: Collect the multi-field data generated by the simulation of the riser support structure model, normalize the data, format the processed multi-field data, and generate a time series input format that meets the conditions ,in Representative The input data vector at each moment, input multiple fields of data into the pre-trained RNN model; S4.2: The RNN model forward propagates each group of input multi-field data, processes the input time series step by step through recursive calculation, updates the hidden state, and extracts the characteristic pattern in the time series. The output layer calculates the output corresponding to the hidden state sequence of each time step, and obtains the predicted value of the dynamic behavior and the corresponding feature extraction result; S4.3: The hidden state sequence generated by the RNN model is reduced in dimension by principal component analysis, the dominant features are extracted, the relationship between the features extracted by the RNN and the physical model parameters is analyzed, and the riser support structure model is optimized by adjusting the physical parameters corresponding to the key features; The specific steps of optimizing the global riser support structure model in step V are as follows: S5.1: The central server sends the riser support structure model to different simulation sites or experimental equipment, sets the global model parameters and the local model parameters of each node through random initialization, and adds a unique number label to each local model parameter. The initial global model parameters are used as the main river position, and the local model parameters of each node are used as the tributary position. The specific calculation formula of random initialization is as follows: ； In the formula, represents the initial position of the main river; Represents the total number of child nodes; Represents the index of each child node; Representative Local model parameters of each node; S5.2: Each node trains the riser support structure model based on its own local data with the purpose of minimizing the local objective function. After the training is completed, each node sends the update of the local model parameters to the central server. Then the master node aggregates the local model parameters using the weighted average method and updates the global parameters. The specific calculation formula of the local objective function is as follows: ； In the formula, Representative The local loss function of each node; Representative Number of node data samples; represents the loss function; Representative The node Input data and labels; The specific calculation formula of weighted average is as follows: ； In the formula, Representative The local model parameters after the node weights are adjusted; S5.3: According to the flow direction update rule, the position of each tributary is updated by simulating the aggregation effect of water flow, that is, for each tributary, the local model parameters are corresponding to each tributary. After the tributary is updated, the position of the main river is optimized according to the gradient information of the global loss function, and then the position of the tributary is randomly adjusted through the evaporation mechanism; S5.4: Calculate the loss value of each tributary and evaluate the impact of the new local model parameters on the model performance. Then update the main river position to the tributary parameters with the best fitness. Repeat the parameter update. When the rate of change of the global loss function is less than the set threshold or the maximum number of iterations is reached, stop the parameter update. S5.5: Weightedly combine the optimized global parameters with the global parameter aggregation results of federated learning, use the verification data to evaluate the fitness of the parameters after collaborative modeling, and record the performance indicators. If the local loss function value of a node is higher than the preset threshold, further adjust the weight ratio, repeatedly update the local parameters and global parameters. When the rate of change of the global loss function is less than the set threshold, select the weight ratio with the highest fitness as the optimal weight ratio, and dynamically update the model in real time; The specific steps for analyzing the benefits and risks of the design decisions of the support structure in long-term operation as described in step V are as follows: S6.1: Collect all participants in the design and operation of the riser support structure, define the strategy space of each participant based on all the strategies that each participant will choose in the decision-making process, and construct a corresponding utility function based on the goal of each participant to obtain the maximum long-term benefit or minimize the risk by choosing a specific strategy to represent the relationship between the participant's benefits and the strategy he chooses. The specific calculation formula of the utility function is as follows: ； In the formula, Representative The utility of each participant, ; Represents the total number of child nodes The payoff function of each participant; Representative The cost function of each participant represents the cost of their decision. ; S6.2: Collect all state information of the riser support structure and construct the corresponding state space. According to all actions that each participant can take in each state, construct the action space corresponding to different participants, and set the corresponding reward function based on design cost, operation income and maintenance cost; S6.3: Calculate the probability of transferring from any state to the next state by randomly taking any action in the action space under any state in the state space, set a discount factor based on the operation cycle of the riser support structure and the information of various long-term planning objectives, and calculate the expected value of the maximum long-term utility that the participant can obtain in the subsequent time under the current state through the value function according to the benefits of each state and the expected future benefits. The specific calculation formula of the value function is as follows: ； In the formula, Representative Participants in the state The value of the following, that is, its long-term utility; Represents the sub-action space Middle Actions; Represents in state Next, take action Instant rewards when represents the discount factor; Represents in state , take action Transfer to state The transition probability of Representation status Next, the expected utility in the future; S6.4: Use the Bellman equation to iteratively solve the problem to obtain the optimal action of each participant at each time. In the iterative solution process, recursively calculate the optimal strategy of each participant at each time point to find the equilibrium solution, that is, the equilibrium point is the point where the participant cannot increase the utility by unilaterally changing its strategy; S6.5: Analyze the benefits and risks of each participant under Nash equilibrium, comprehensively evaluate the long-term effects of the design plan, calculate the utility, cost and risk factors of different design plans and operation strategy combinations, and evaluate the most feasible and sustainable design plan in long-term operation.

2. The FPSO riser support structure modeling method according to claim 1, characterized in that: The specific steps for analyzing the application scenarios and working conditions of FPSO riser supports and determining modeling requirements in step Ⅰ are as follows: S1.1: Use sensors and satellite data collection technology to obtain environmental parameters of the sea area, including wave characteristics, current velocity, and wind field distribution data, and describe the data collection results in time series; S1.2: Use the Morison equation to calculate the impact of environmental loads on FPSO risers. According to the requirements of the FPSO riser support structure, select candidate materials, collect the physical and chemical properties of tensile strength, compressive strength and corrosion resistance of each candidate material, and calculate the allowable stress of the material according to the material mechanics formula. The specific calculation formula of the Morison equation is as follows: ； In the formula, Represents the total load acting on the riser, in N; represents the drag coefficient; Represents the density of seawater, in kg / m³; Represents the cross-sectional area of ​​the structure, in m²; Represents the speed of ocean current or wave, in m / s; represents the coefficient of inertia; Represents the relative flow velocity in the riser, in m / s; Represents the acceleration of the relative flow velocity, in m / s²; S1.3: The power spectrum density of the wave load is calculated using the Pierson-Moskowitz spectrum analysis method. Based on the calculated power spectrum density of the wave load, the impact and fatigue effects of the dynamic characteristics of the environmental load on the riser are evaluated. The specific calculation formula of the Pierson-Moskowitz spectrum is as follows: ； In the formula, Represents the wave energy spectral density, in m² / Hz; Represents the wave frequency, unit Hz; and Represent the empirical coefficients respectively; Represents the acceleration due to gravity, in m / s²; Represents the peak frequency, in Hz; Represents significant wave height; S1.4: According to the FPSO riser support structure goals set by the staff, the first-level evaluation criteria and second-level evaluation indicators are determined, and a hierarchical model corresponding to the target layer, criterion layer and scheme layer is generated. For each pair of criteria in the criterion layer, a comparison matrix is ​​constructed to compare them pairwise, determine their relative importance, and assign weights through the 9-level scaling method; S1.5: After the weights are assigned, each column of the comparison matrix is ​​normalized to obtain the corresponding standardized matrix. Then, the average value of each row is calculated to obtain the weight vector of each criterion. The consistency ratio of the standardized matrix is ​​calculated. If the ratio is less than 0.1, it means that the consistency of the matrix is ​​acceptable. Otherwise, the comparison matrix is ​​readjusted. S1.6: Based on the standardized matrix that meets the requirements, the weight vector of each criterion is used as the importance weight of the criterion, and the corresponding FPSO riser support structure model construction index is generated.

Images