A method and system for reducing the order of FPSO models by integrating time-recurrent networks and fuzzy logic

Abstract

This invention proposes a method and system for reducing the order of FPSO models by integrating time-recurrent neural networks and fuzzy logic, belonging to the field of marine engineering structural analysis and intelligent modeling technology. Addressing the problem of large degrees of freedom and low computational efficiency in FPSO finite element models, this invention extracts key degrees of freedom from the high-dimensional finite element model to construct low-dimensional feature vectors. It then uses a fuzzy logic system to perform regularized modeling of the dominant mechanical properties of the structure, achieving initial order reduction. Furthermore, a time-recurrent neural network is introduced to compensate for high-order residual effects that are difficult to accurately characterize using the fuzzy logic model, ultimately resulting in a low-dimensional, high-fidelity reduced-order model. The proposed method significantly reduces the degree of freedom and parameter dimension of the finite element model while maintaining the main mechanical properties of the structure, effectively reducing computational complexity and improving the model's applicability and practical engineering value in applications such as rapid finite element analysis, online structural response prediction, and engineering decision support.

Description

Technical Field

[0001] This invention belongs to the field of marine engineering structure analysis and intelligent modeling technology, and particularly relates to a method for reducing the order of FPSO models and system finite element models that integrates time-cyclic networks and fuzzy logic. Background Technology

[0002] The statements in this section are merely background information related to the present invention and do not necessarily constitute prior art.

[0003] With the continuous development of deep-water and ultra-deep-water marine oil and gas resources, Floating Production Storage and Offloading (FPSO) units have become typical large and complex structural systems in the field of marine engineering due to their large structural dimensions, long service life, and complex operating conditions. High-dimensional finite element models refer to numerical models developed during finite element analysis to precisely describe the FPSO hull structure, mooring system, and their interactions. This is because the number of nodes, elements, and degrees of freedom increases exponentially, leading to a rapid expansion of the dimensions of the mass matrix, stiffness matrix, and related parameter matrices to N×N or even higher. Such models can comprehensively reflect key mechanical properties such as structural stiffness distribution, load transfer paths, and local stress concentrations.

[0004] In engineering practice, while the aforementioned high-dimensional finite element models possess high offline analysis accuracy, they also face a series of practical problems. On the one hand, the storage and computation of large-scale finite element parameter matrices place extremely high demands on computing resources, leading to a significant decrease in simulation efficiency. On the other hand, in application scenarios requiring efficient responses, such as multi-condition parameter analysis, structural state assessment, and rapid prediction, complete high-dimensional finite element models often fail to meet the engineering requirements for computational speed, limiting their further promotion and application.

[0005] To alleviate the computational burden caused by the excessive size of finite element models, researchers have attempted various model simplification or order reduction methods. However, existing technologies still have significant limitations in applications involving complex marine structures. Finite element order reduction methods based on linear assumptions are usually built upon the linear response of the structure. However, there is a strong coupling relationship between FPSO structures and mooring systems, and the parameter distribution exhibits highly nonlinear characteristics. When model parameters or structural forms change, the order reduction effect and applicability of such methods are severely limited. Order reduction methods based on modal truncation achieve model order compression by retaining a small number of dominant modes. However, when facing marine structures with complex parameter distributions and obvious local structural features, it is often difficult to completely preserve the parameter characteristics and overall mechanical features of the original model. Parameter simplification methods that rely on experience or rules are usually designed for specific structural forms or operating conditions and lack a systematic characterization of the overall distribution characteristics of finite element parameters. When the model size or parameter range changes, the accuracy and stability of the reduced model are difficult to guarantee.

[0006] The aforementioned difficulties have led to the need to effectively compress high-dimensional parameter systems into low-dimensional expressions while maintaining the structural characteristics and parameter coupling relationships of finite element models. This has become a key problem that urgently needs to be solved in the field of finite element analysis of large marine structures such as FPSOs. Summary of the Invention

[0007] To address the issues of massive parameter numbers, high matrix dimensionality, and low computational efficiency in existing FPSO finite element models, this invention proposes a method and system for reducing the order of FPSO models by integrating a time-recurrent neural network and fuzzy logic. By introducing a fuzzy logic system to regularize the distribution of dominant structural parameters in the FPSO finite element model, high-dimensional parameter matrices are compressed. Simultaneously, a time-recurrent neural network is used to learn and compensate for the complex nonlinear mapping relationships between high-dimensional parameters in the finite element model. While maintaining the overall structural characteristics and parameter correlations of the finite element model, this method effectively reduces the order of the model from high-dimensional to low-dimensional parameters, significantly decreasing model size and computational complexity, while also ensuring stability, interpretability, and accuracy in the reduction process.

[0008] On the one hand, a method for reducing the order of FPSO models by integrating time-recurrent networks and fuzzy logic is provided, including:

[0009] A high-dimensional finite element parametric model is constructed based on the structural response information and external load information of the floating production storage and unloading device.

[0010] Finite element feature information for model order reduction is extracted from the high-dimensional finite element parameter model, and a low-dimensional finite element feature vector is constructed based on the extracted finite element feature information.

[0011] A fuzzy logic system is constructed based on the low-dimensional finite element feature vectors, and the main mechanical behaviors of the structure represented by the low-dimensional finite element feature vectors are modeled to obtain a preliminary reduced-order finite element model.

[0012] Based on the parameter error of the preliminary reduced-order finite element model, the estimated value of the dynamic residual term is calculated using a time-recurrent neural network, and the estimation result of the reduced-order finite element residual force is obtained.

[0013] The preliminary reduced-order finite element model is fused with the estimation results of the reduced-order residual force of the finite element model to obtain the final reduced-order finite element model.

[0014] Furthermore, the structural response information includes displacement, velocity, or acceleration information at key finite element nodes or equivalent degrees of freedom of the floating production storage and offloading device; the external load information includes wind load, wave load, ocean current load, and mooring system force information.

[0015] Furthermore, the high-dimensional finite element parameter model is a multi-degree-of-freedom structural finite element model that includes a mass matrix, a damping matrix, and a stiffness matrix.

[0016] Furthermore, key degrees of freedom that significantly affect the overall mechanical response of the FPSO are selected from the high-dimensional finite element parameter model to construct a low-dimensional finite element feature vector. The selection of the key degrees of freedom is based on at least one of the following factors: the overall stress path of the FPSO structure and the distribution of the main load-bearing components; the structural response characteristics of the midship and bow and stern regions of the hull; the high stress concentration points at mooring connection points and deck equipment support areas; and the sensor layout and available measurement information during engineering operation.

[0017] Furthermore, the fuzzy logic system adopts IF-THEN form fuzzy rules to segmentally model the structural mechanical behavior within different finite element feature intervals; the activation degree of each fuzzy rule is calculated through the membership function, and the finite element parameter outputs corresponding to each rule are weighted and fused to obtain the finite element order reduction result of the fuzzy logic system.

[0018] Furthermore, the membership function adopts one or more of Gaussian, triangular, or trapezoidal functions to describe the membership relationship of low-dimensional finite element features in different fuzzy sets.

[0019] Furthermore, the time-recurrent neural network model takes low-dimensional finite element feature vectors, structural response information, and external load information as inputs, and uses the finite element parameter errors generated by the fuzzy logic system as the learning object.

[0020] On the other hand, a system for reducing the order of FPSO models by integrating time-recurrent networks and fuzzy logic is provided, including:

[0021] The model building module is configured to: construct a high-dimensional finite element parametric model based on the structural response information and external load information of the floating production storage and unloading device;

[0022] The feature processing module is configured to: extract finite element feature information for model order reduction from the high-dimensional finite element parameter model, and construct a low-dimensional finite element feature vector based on the extracted finite element feature information;

[0023] The preliminary order reduction module is configured to: construct a fuzzy logic system based on the low-dimensional finite element feature vectors, model the main mechanical behaviors of the structure represented by the low-dimensional finite element feature vectors, and obtain a preliminary order-reduced finite element model;

[0024] The model compensation module is configured to: based on the parameter error of the preliminary reduced-order finite element model, use a time-recurrent neural network to model the dynamic residual term and obtain the estimation result of the reduced-order finite element residual force;

[0025] The fusion module is configured to fuse the preliminary reduced-order finite element model with the estimation result of the reduced-order finite element residual force to obtain the final reduced-order finite element model.

[0026] In another aspect, a computer device is also provided, including a computer-readable storage medium, a processor, and a computer program stored on the computer-readable storage medium and executable on the processor, wherein when the processor executes the program, it performs the method described in the first aspect.

[0027] In another aspect, a computer-readable storage medium is also provided, on which a computer program is stored, which, when executed by a processor, performs the method described in the first aspect.

[0028] The above technical solution has the following advantages or beneficial effects:

[0029] (1) By extracting key parameters from the high-dimensional finite element model to construct a low-dimensional feature vector, and then integrating the fuzzy logic system and the time loop neural network on this basis, the parameters of the FPSO finite element model were effectively reduced from N×N to r×r, which significantly reduced the model size and computational complexity.

[0030] (2) Compared with traditional reduction methods based on linear assumptions or modal truncation, this invention introduces a fuzzy logic system to model the a priori laws of finite element parameter distribution in a regularized manner, dividing the complex high-dimensional parameter relationships into parameter intervals with physical meaning, which effectively improves the stability and interpretability of the reduction process and overcomes the shortcomings of existing methods in preserving the mechanical characteristics of the original model in marine structures with complex parameter distributions and obvious local features.

[0031] (3) By using a time-recurrent neural network to learn the high-dimensional nonlinear mapping relationship between finite element parameters, the high-order residual effects that are difficult to accurately characterize in fuzzy logic systems can be compensated and modeled. Compared with the simplification method that relies on empirical rules, the accuracy and generalization ability of the reduced-order model under complex working conditions are significantly improved.

[0032] (4) The constructed reduced-order finite element model is suitable for multi-condition parameter analysis, rapid simulation and engineering optimization design. It can significantly improve the calculation efficiency while maintaining the main mechanical properties of the structure, and has good engineering application value. Attached Figure Description

[0033] The accompanying drawings, which form part of this invention, are used to provide a further understanding of the invention. The illustrative embodiments of the invention and their descriptions are used to explain the invention and do not constitute an improper limitation of the invention.

[0034] Figure 1 This is an overall flowchart of the method in Embodiment 1 of the present invention;

[0035] Figure 2 This is a flowchart of the fuzzy logic system order reduction in Embodiment 1 of the present invention;

[0036] Figure 3 This is a schematic diagram illustrating the function of the time-recurrent neural network in Embodiment 1 of the present invention;

[0037] Figure 4 This is a schematic diagram of the combination of a time-recurrent neural network and a fuzzy logic system in Embodiment 1 of the present invention. Detailed Implementation

[0038] To make the objectives, technical solutions, and advantages of the present invention clearer, the embodiments of the present invention will be described in further detail below with reference to the accompanying drawings. Those skilled in the art should understand that the specific embodiments described herein are for illustrative purposes only and are not intended to limit the scope of the invention.

[0039] It should be noted that the following detailed description is illustrative and intended to provide further explanation of the invention. Unless otherwise specified, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains.

[0040] As described in the background section, FPSOs operate in complex marine environments for extended periods, presenting numerous challenges to their structural finite element modeling and analysis. First, the large scale and complex configuration of FPSO structures necessitate the introduction of numerous nodes and elements during finite element discretization, resulting in a vast number of degrees of freedom. This leads to extremely high dimensions in the mass matrix, stiffness matrix, and related parameter matrices, resulting in a massive computational scale for the finite element model. Second, a significant coupling relationship exists between the FPSO structure and the mooring system. Mooring cable constraints, structural deformation, and environmental loads interact at the finite element level, causing the finite element parameter distribution to exhibit highly nonlinear and strongly coupled characteristics, further increasing the complexity of model analysis and solution. Furthermore, marine environmental loads exhibit significant randomness and time-varying characteristics. The structural stress and deformation characteristics continuously change under different operating conditions. Traditional finite element models based on fixed parameters or linear assumptions struggle to balance accuracy with computational efficiency, particularly limiting their application in rapid simulations or online evaluation scenarios.

[0041] To address the aforementioned issues, this invention proposes a method and system for reducing the order of FPSO models by integrating time-recurrent neural networks and fuzzy logic. This method mainly comprises two parts: the first part involves constructing a fuzzy logic modeling framework to characterize the main structural features and parameter distribution patterns in the FPSO finite element model. By regularizing the modeling of key parameters and characteristic quantities, structural compression of the high-dimensional finite element parameter matrix is ​​achieved. The second part is a parameter error compensation modeling method based on time-recurrent neural networks. This method uses a data-driven approach to learn and correct the approximation errors generated by the fuzzy logic model under complex parameter relationships and nonlinear distributions, thereby improving the reduced-order model's ability to approximate the original finite element model.

[0042] In its specific implementation, the high-dimensional finite element modeling problem of FPSO structures is first transformed into a parameter mapping and reconstruction problem, extracting and characterizing the key parameter features in the finite element model. Then, these parameter features are input into a reduced-order modeling framework composed of a fuzzy logic system and a time-recurrent neural network, obtaining a finite element parameter representation with significantly reduced dimensionality. Based on this, a low-order finite element model is reconstructed. Through this method, the present invention achieves a significant reduction in the dimensionality and computational scale of the finite element model parameters while ensuring that the main finite element characteristics of the FPSO structure are not significantly lost, thus significantly improving the computational efficiency and practicality of finite element analysis in engineering applications.

[0043] Example 1

[0044] This embodiment provides a method for reducing the order of an FPSO model by integrating time-cyclic networks and fuzzy logic, such as... Figure 1 As shown, Figure 1 This is an overall flowchart of the method according to Embodiment 1 of the present invention. The method includes the following steps:

[0045] S1: Based on the structural response information and external load information of the floating production storage and unloading device, a high-dimensional finite element parameter model is constructed;

[0046] S2: Extract finite element feature information from the high-dimensional finite element parametric model for model order reduction, and construct a low-dimensional finite element feature vector based on the extracted finite element feature information;

[0047] S3: Construct a fuzzy logic system based on low-dimensional finite element eigenvectors, model the main mechanical behaviors of the structure represented by the low-dimensional finite element eigenvectors, and obtain a preliminary reduced-order finite element model;

[0048] S4: Based on the parameter error of the preliminary reduced-order finite element model, the estimated value of the dynamic residual term is calculated using a time-recurrent neural network to obtain the estimation result of the reduced-order finite element residual force;

[0049] S5: The preliminary reduced-order finite element model is fused with the estimation results of the reduced-order residual force of the finite element model to obtain the final reduced-order finite element model.

[0050] Furthermore, in step S1: based on the structural form and engineering modeling requirements of FPSO in complex marine environments, a high-dimensional finite element model of FPSO is established.

[0051] Specifically, the hull structure, deck components, key connections, and mooring areas are finely discretized to construct a high-dimensional finite element parametric model that includes mass parameters, stiffness parameters, and related coupling terms. The dimension of each finite element parameter matrix is ​​[value missing]. This embodiment fully considers the engineering background of FPSO system operating for a long time in actual marine environment. As a typical large marine floating structure, FPSO's hull structure, deck system, cabin structure, as well as key connections and mooring areas, are subjected to complex spatial forces and structural responses under the combined action of random marine environmental loads such as wind, waves, and currents, as well as the mooring system.

[0052] To accurately characterize the overall mechanical properties of FPSO structures under complex working conditions, based on marine engineering structural analysis and finite element theory, the overall FPSO structure is spatially discretized, dividing the continuous structure into a finite number of elements and nodes, and a high-dimensional finite element mechanical model of the FPSO is established on this basis.

[0053] Let the total number of degrees of freedom of the FPSO finite element model be... Its overall finite element mechanical behavior can be uniformly represented in the following matrix form:

[0054] (1);

[0055] in, This represents the displacement vector of the FPSO finite element node; These represent the nodal velocity vector and acceleration vector, respectively. This is the finite element mass matrix, used to characterize the mass distribution characteristics of the FPSO structure; This is the damping correlation matrix, used to describe the energy dissipation characteristics of the structure; This is the stiffness matrix, used to characterize the overall and local stiffness properties of the FPSO structure; It represents the equivalent external force vector acting on the structure by environmental loads such as wind, waves, and currents, as well as the mooring system.

[0056] In the finite element modeling process, the geometric dimensions, material parameters, component connection methods, and mooring constraints of the FPSO structure are all reflected through the aforementioned finite element parameter matrix, enabling the established model to comprehensively reflect the mechanical response characteristics of the FPSO structure under complex environmental conditions.

[0057] Furthermore, rewriting equation (1) in an explicit expression of acceleration, we get:

[0058] (2);

[0059] As can be seen from equation (2), the nodal acceleration in the FPSO finite element model is determined by multiple factors such as displacement, velocity, and external loads, and its mapping relationship exhibits obvious high dimensionality and strong coupling characteristics. However, due to the limited number of degrees of freedom in the finite element model... N Typically quite large, the dimensions of the mass matrix, stiffness matrix, and damping matrix are all [missing information]. N×N This results in a large number of model parameters and complex matrix operations.

[0060] While the aforementioned high-dimensional finite element model has advantages in terms of structural description accuracy, in engineering applications such as rapid structural response assessment, parameter analysis, online simulation, and digital twins, direct calculation based on equations (1) or (2) often fails to meet the requirements for computational efficiency and real-time performance. Therefore, under the premise of preserving as much as possible the main mechanical properties of the FPSO structure and the distribution law of finite element parameters, the high-dimensional finite element model is effectively reduced in order, transforming the original... Converting the 3D finite element model into a lower-dimensional form becomes a key problem that needs to be solved in subsequent steps.

[0061] Studies have shown that under complex marine environmental loads and mooring constraints, the contributions of different regions, nodes, and degrees of freedom of an FPSO structure to the overall mechanical response vary significantly. Some finite element degrees of freedom and structural parameters play a dominant role in the overall structural response, while others mainly reflect local deformation or high-frequency detail characteristics, and their impact on the overall mechanical behavior is relatively small. Therefore, without significantly weakening the overall stiffness distribution, mass characteristics, and main force paths of the FPSO structure, preliminary dimensionality reduction of the high-dimensional finite element model can be achieved by reasonably selecting and reorganizing the finite element degrees of freedom and parameters.

[0062] Therefore, in step S2: based on the finite element modeling mechanism analysis of FPSO structures and long-term engineering operation experience, the parameter matrix in the high-dimensional finite element model established in step S1 is further analyzed. Key parameters that are highly sensitive to the overall structural response and parameter changes are selected to construct a finite element reduction parameter space. Finite element feature information for model reduction is extracted from the high-dimensional finite element parameter model, including but not limited to structural stiffness distribution characteristics, mass distribution characteristics, parameters of locally reinforced areas, and parameter coupling relationship characteristics, forming the original parameter feature set for finite element model reduction.

[0063] Specifically, suppose the FPSO finite element model established in step S1 includes For each structural degree of freedom, the corresponding nodal displacement vector can be expressed as:

[0064] (3);

[0065] in, Let represent the initial displacement variables of the FPSO. Through analysis of finite element modal characteristics, structural stress distribution, and the response of key components, key degrees of freedom that significantly affect the overall mechanical response of the FPSO are selected from the above high-dimensional displacement vectors to construct the finite element eigenvectors of the low-dimensional FPSO.

[0066] (4);

[0067] in, To select the key node displacement components from the original finite element degrees of freedom, the selection criteria for key degrees of freedom include at least one of the following factors: the overall stress path of the FPSO structure and the distribution of the main load-bearing components; the structural response characteristics of key areas such as the midship and bow and stern of the hull; high stress concentration areas such as mooring connection points and deck equipment support areas; and the sensor layout and available measurement information during the operation of the project.

[0068] Through the above methods, the original The finite element space is compressed into 12-dimensional space. The low-dimensional feature space significantly reduces the number of degrees of freedom in the finite element model while preserving the main mechanical properties and key response information of the FPSO structure. This low-dimensional vector not only characterizes the main motion features of the FPSO system, but also provides a compact and efficient input space for subsequent reduced-order modeling based on fuzzy logic systems and time-recurrent neural networks, which is beneficial to improving model training efficiency and computational stability.

[0069] Low-dimensional finite element eigenvectors From the original finite element degree of freedom nodal displacement vector The key nodes or structural parameters in the model have a significantly lower dimension than the original model. . Not equivalent to Instead of discarding the degrees of freedom, the model retains the degrees of freedom that contribute most to the main mechanical properties of the FPSO structure. For the discarded degrees of freedom, fuzzy logic and neural networks are used for error compensation, thereby ensuring that the low-dimensional finite element model approximates the original high-dimensional model while maintaining the main mechanical properties.

[0070] Therefore, in step S3, as Figure 2 As shown, Figure 2 A flowchart for reducing the order of a fuzzy logic system. This involves constructing low-dimensional finite element eigenvectors in step S2. Including environmental load information, a fuzzy logic system is constructed to model the distribution patterns and structural characteristics of the main parameters in the FPSO finite element model according to rules. Through fuzzy inference and weighted fusion mechanisms, the high-dimensional finite element parameter model is structurally simplified to obtain a preliminary reduced-order model characterizing the main mechanical properties of the FPSO structure. Specifically, this includes the following sub-steps:

[0071] S3.1, Fuzzy rule construction. In low-dimensional finite element eigenvectors... Building upon this foundation, a fuzzy logic system is constructed to characterize the dominant mechanical features of FPSO structures under complex operating conditions. By introducing fuzzy sets and fuzzy rules, the fuzzy logic system decomposes the high-dimensional, nonlinearly coupled finite element parameter relationships into multiple locally linear or quasi-linear sub-models, thereby achieving a structural approximation of the high-dimensional finite element model as a whole. The fuzzy logic rules are expressed in "IF–THEN" form as follows:

[0072] IF is THEN (5);

[0073] in, For the first The fuzzy set corresponding to each rule is used to describe the feature interval of the low-dimensional finite element eigenvector; This is the low-dimensional finite element parameter matrix under this rule; The number of fuzzy rules is defined as follows. By designing a reasonable fuzzy set and number of rules, the fuzzy logic system can characterize the dominant mechanical change trend of the FPSO system in a low-dimensional parameter space, achieving an initial reduction in the order of the high-dimensional finite element model.

[0074] S3.2 Membership Function Design and Rule Activation. To achieve a smooth mapping across different parameter ranges, it is necessary to design a corresponding membership function for each low-dimensional feature component. Its form can be triangular, Gaussian, or trapezoidal to ensure the continuity and robustness of the system output during parameter changes. Membership functions are used to calculate the activation degree of each fuzzy rule under the current low-dimensional parameter conditions.

[0075] (6);

[0076] The activation values ​​were then normalized to obtain the weight coefficients for each rule:

[0077] (7);

[0078] Among them, normalized weights It can reflect the contribution of each rule to the overall structural parameter output under the current low-dimensional feature conditions.

[0079] S3.3, Weighted Fusion Output. After calculating the rule activation weights, the outputs of each local finite element model are weighted and fused to obtain the overall output of the fuzzy logic system:

[0080] (8);

[0081] Through this weighted fusion process, the fuzzy logic system can approximate the dominant mechanical structure of the FPSO finite element model in a low-dimensional feature space, thereby achieving high-dimensional... Parameter matrix to low dimension Structural order reduction of a matrix.

[0082] It is worth noting that although fuzzy logic systems can reflect the dominant finite element parameter characteristics of FPSO systems well, some mechanical phenomena that are difficult to capture through regularization still exist in the complex marine environment: nonlinear coupling of local structures, such as the mutual influence between the mooring system and the local stiffness of the hull; the randomness and strong time-varying nature of external loads, which means that some low-dimensional features cannot fully represent the changes in high-dimensional parameters; and the possibility of discontinuous or abrupt responses of high-order finite element parameters near the boundaries of fuzzy rules. Therefore, relying solely on fuzzy logic systems for order reduction may result in residual errors, requiring the subsequent introduction of neural networks to learn and compensate for these errors, thereby improving the accuracy of the low-dimensional finite element model in approximating the original high-dimensional structural characteristics.

[0083] Therefore, to further improve the accuracy of the reduced-order model, step S4 involves constructing a time-recurrent neural network to learn and compress the complex high-dimensional nonlinear mapping relationships between the finite element parameters. For example... Figure 3 As shown, Figure 3 This is a schematic diagram illustrating the function of a time-recurrent neural network. By further introducing a time-recurrent neural network, equivalent compensation modeling is performed on the high-order mechanical residual effects that the fuzzy logic system failed to accurately characterize during the finite element model reduction process of FPSO, thereby improving the accuracy and adaptability of the overall finite element reduction model under complex marine conditions.

[0084] Specifically, it includes the following sub-steps:

[0085] S4.1 Definition of the reduced-order residual force term in finite element model. Based on the high-dimensional FPSO finite element model established in step S1 and the preliminary reduced-order finite element model obtained through the fuzzy logic system in step S3, the dynamic residual term between the two is defined in the equivalent low-dimensional degree-of-freedom space.

[0086] Under low-dimensional degrees of freedom, the projection form of the high-dimensional finite element model can be expressed as:

[0087] (9);

[0088] in, It is the projected finite element mass matrix. It is the damping matrix of the projection. It is the stiffness matrix of the projection.

[0089] The preliminary order reduction model obtained based on the fuzzy logic system is as follows:

[0090] (10);

[0091] in, It is the finite element mass matrix estimated by the fuzzy logic system. It is the damping matrix estimated by the fuzzy logic system. It is the stiffness matrix estimated by the fuzzy logic system.

[0092] Therefore, the reduced-order residual force of the finite element method can be defined at the level of the equivalent low-dimensional dynamic equations: (11);

[0093] in, It is the estimation error of the finite element mass matrix. It is the estimation error of the damping matrix. It is the estimation error of the stiffness matrix.

[0094] The residual force term, from a mechanical perspective, uniformly characterizes the unmodeled higher-order dynamic effects introduced by the truncation of finite element degrees of freedom, the finiteness of fuzzy rules, and local nonlinear effects.

[0095] S4.2 Construction of the Residual Force Compensation Model Using a Time-Recurrent Neural Network. Addressing the characteristics of the reduced-order residual force in the finite element method (FEB) method—strong nonlinearity, significant time-varying nature, and condition-dependent behavior—this invention employs a time-recurrent neural network (TRNN) for data-driven modeling and online estimation of the residual force term. By introducing time recursion and internal state memory mechanisms into the network structure, the TRNN can continuously update the historical evolution information of the structural dynamic response, thereby effectively characterizing the dynamic characteristics of the residual force over time.

[0096] The input vector of the time-recurrent neural network is selected as low-dimensional feature information that can reflect the current motion state of the structure, external loads, and environmental influences, and its form is defined as:

[0097] (12);

[0098] in, These are low-dimensional finite element eigenvectors. For the corresponding velocity vector, For the equivalent external load vector, This refers to the sensor response information acquired by the FPSO structure monitoring system.

[0099] In a time-recurrent neural network, hidden state variables are introduced to represent historical information. h ( t Its recursive update relationship over time can be expressed as:

[0100] (13);

[0101] in, h ( t- 1) Represents the time-recurrent neural network at time step [time]. t- The hidden state of 1 Z (.) represents a state update function with time-recursive properties. Let the input vector be the input vector at the current time. This is a set of network parameters.

[0102] Under the aforementioned recursive mechanism, the output of the time-recurrent neural network is defined as the estimation result of the reduced residual force of the finite element method:

[0103] (14);

[0104] in, This represents the predicted value of the residual force by the time-recurrent neural network. Q(.) is a network mapping function that includes time-recurrence properties.

[0105] S4.3, Introduction of Mean Squared Error Loss Function and Construction of Optimization Objective. To quantitatively evaluate the sequence approximation accuracy of the time-series recurrent neural network for the reduced-order residual force of the finite element method and to guide the training process of the network parameters, the mean squared error (MSE) for time series data is introduced as a loss function, defined as:

[0106] (15);

[0107] in, The number of training samples. This represents the reference residual force calculated by the high-fidelity finite element model and the fuzzy logic reduced-order model. This represents the predicted value of the recurrent time-series neural network for the corresponding time series samples. By minimizing the aforementioned loss function, the recurrent time-series neural network achieves optimal approximation of the time series characteristics of the reduced-order residual force in the finite element method under the mean square principle, thereby ensuring the continuity, rationality, and numerical stability of the introduced residual force compensation term at the overall dynamics and energy levels.

[0108] While maintaining the interpretability of the fuzzy logic system structure, the ability of the time recurrent neural network to remember and model historical dynamic response information is utilized to effectively compensate for the high-order time-varying information missed in the finite element reduction process, thereby constructing an FPSO finite element reduction model that combines time consistency, modeling accuracy and computational efficiency.

[0109] like Figure 4 As shown, Figure 4 This is a schematic diagram illustrating the combination of a time-recurrent neural network and a fuzzy logic system. Specifically, in step S5, the preliminary finite element reduced-order model output by the fuzzy logic system in step S3 is superimposed and fused with the finite element reduced-order residual force predicted by the time-recurrent neural network in step S4 to construct the final finite element reduced-order model of the FPSO system, as shown below:

[0110] (16);

[0111] The fusion-reduced-order model shown in Equation (16) enables an effective approximation of the dynamic response of the FPSO system without explicitly calculating the complete coupling form of the mass matrix, damping matrix, and stiffness matrix in the original high-dimensional finite element model, thereby significantly reducing the computational complexity of the model. This fusion-reduced-order finite element model significantly reduces the degree of freedom of the FPSO finite element model through low-dimensional degree-of-freedom modeling, avoids the repeated assembly of high-dimensional finite element matrices and large-scale matrix operations, greatly reduces the computational cost, and improves the real-time performance and scalability of the model while retaining the main mechanical properties and global response laws of the FPSO structure. Furthermore, it has good robustness and adaptability to complex environmental loads, structural parameter uncertainties, and local high-order effects through the residual force compensation mechanism of the time-recurrent neural network.

[0112] In engineering applications, the fused finite element reduced-order model in this embodiment can be widely used in scenarios such as motion response prediction, online rapid simulation, control algorithm design, and structural safety assessment of FPSO systems. By calling this reduced-order model in real time for dynamic calculations, the efficiency of engineering analysis and control decisions related to FPSO systems can be significantly improved while ensuring computational accuracy.

[0113] It should be noted that although the reference low-dimensional finite element parameter matrix can be obtained by projecting the high-dimensional finite element model, this process is computationally expensive and highly dependent on the complete high-dimensional model, making it difficult to meet the engineering requirements of online analysis and rapid simulation. Therefore, in this embodiment, the reference matrix is ​​not directly used as the final reduced-order model. Instead, it is used as supervisory information, and a time-recurrent neural network is introduced to learn the mapping relationship between low-dimensional structural features and high-order mechanical residual effects. This allows for the efficient construction and online updating of the FPSO finite element reduced-order model without the need for a high-dimensional finite element model.

[0114] Example 2

[0115] This embodiment provides a FPSO model order reduction system that integrates time-cyclic networks and fuzzy logic, including:

[0116] The model building module is configured to: construct a high-dimensional finite element parametric model based on the structural response information and external load information of the floating production storage and unloading device;

[0117] The feature processing module is configured to: extract finite element feature information from the high-dimensional finite element parametric model for model order reduction, and construct a low-dimensional finite element feature vector based on the extracted finite element feature information;

[0118] The preliminary order reduction module is configured to: construct a fuzzy logic system based on low-dimensional finite element eigenvectors, model the main mechanical behaviors of the structure represented by the low-dimensional finite element eigenvectors, and obtain a preliminary order-reduced finite element model;

[0119] The model compensation module is configured to: based on the parameter error of the initially reduced finite element model, use a time recurrent neural network to model the dynamic residual term and obtain the estimation result of the reduced finite element residual force;

[0120] The fusion module is configured to fuse the preliminary reduced-order finite element model with the estimation results of the reduced-order residual force of the finite element model to obtain the final reduced-order finite element model.

[0121] It should be noted that each module in this embodiment corresponds one-to-one with each step in Embodiment 1, and their specific implementation process is the same, so it will not be repeated here.

[0122] Example 3

[0123] This embodiment also provides a computer device, including a computer-readable storage medium, a processor, and a computer program stored on the computer-readable storage medium and executable on the processor. When the processor executes the program, it completes the method described in Embodiment 1.

[0124] It should be understood that in this embodiment, the processor can be a central processing unit (CPU), or it can be other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor can be a microprocessor or any conventional processor, etc.

[0125] Memory may include read-only memory and random access memory, and provides instructions and data to the processor. A portion of memory may also include non-volatile random access memory. For example, memory may also store information about the device type.

[0126] In the implementation process, each step of the above method can be completed by the integrated logic circuits in the processor hardware or by software instructions.

[0127] The method in Embodiment 1 can be directly implemented by a hardware processor, or implemented by a combination of hardware and software modules within the processor. The software modules can reside in readily available storage media in the art, such as random access memory, flash memory, read-only memory, programmable read-only memory, electrically erasable programmable memory, or registers. This storage medium is located in memory; the processor reads information from the memory and, in conjunction with its hardware, completes the steps of the above method. To avoid repetition, a detailed description is not provided here.

[0128] Those skilled in the art will recognize that the units and algorithm steps described in connection with the various examples of this embodiment can be implemented in electronic hardware or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this invention.

[0129] Example 4

[0130] This embodiment also provides a computer-readable storage medium for storing computer instructions, which, when executed by a processor, complete the method described in Embodiment 1.

[0131] The above description is merely a preferred embodiment of the present invention and is not intended to limit the invention. Various modifications and variations can be made to the present invention by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A method for reducing the order of FPSO models by integrating time-cyclic networks and fuzzy logic, characterized in that, include: A high-dimensional finite element parametric model is constructed based on the structural response information and external load information of the floating production storage and unloading device. Finite element feature information for model order reduction is extracted from the high-dimensional finite element parameter model, and a low-dimensional finite element feature vector is constructed based on the extracted finite element feature information. A fuzzy logic system is constructed based on the low-dimensional finite element feature vectors, and the main mechanical behaviors of the structure represented by the low-dimensional finite element feature vectors are modeled to obtain a preliminary reduced-order finite element model. Based on the parameter error of the preliminary reduced-order finite element model, the estimated value of the dynamic residual term is calculated using a time-recurrent neural network, and the estimation result of the reduced-order finite element residual force is obtained. The preliminary reduced-order finite element model is fused with the estimation results of the reduced-order residual force of the finite element model to obtain the final reduced-order finite element model.

2. The method according to claim 1, characterized in that, The structural response information includes displacement, velocity, or acceleration information at key finite element nodes or equivalent degrees of freedom of the floating production storage and offloading device; the external load information includes wind load, wave load, ocean current load, and mooring system force information.

3. The method according to claim 1, characterized in that, The high-dimensional finite element parameter model is a multi-degree-of-freedom structural finite element model that includes a mass matrix, a damping matrix, and a stiffness matrix.

4. The method according to claim 1, characterized in that, Key degrees of freedom that significantly affect the overall mechanical response of the FPSO are selected from the high-dimensional finite element parameter model, and low-dimensional finite element feature vectors are constructed. The selection of the key degrees of freedom is based on at least one of the following factors: the overall force path of the FPSO structure and the distribution of the main load-bearing components; the structural response characteristics of the midship and bow and stern regions of the hull; the high stress concentration points at mooring connection points and deck equipment support areas; and the sensor layout and available measurement information during engineering operation.

5. The method according to claim 1, characterized in that, The fuzzy logic system uses IF-THEN fuzzy rules to model the structural mechanical behavior within different finite element feature intervals; the activation degree of each fuzzy rule is calculated through a membership function, and the finite element parameter outputs corresponding to each rule are weighted and fused to obtain the finite element order reduction result of the fuzzy logic system.

6. The method according to claim 5, characterized in that, The membership function is one or more of Gaussian, triangular, or trapezoidal functions, used to describe the membership relationship of low-dimensional finite element features in different fuzzy sets.

7. The method according to claim 1, characterized in that, The time-recurrent neural network model takes low-dimensional finite element feature vectors, structural response information, and external load information as inputs, and uses the finite element parameter errors generated by the fuzzy logic system as the learning object.

8. A system for reducing the order of FPSO models by integrating time-cyclic networks and fuzzy logic, characterized in that, include: The model building module is configured to: construct a high-dimensional finite element parametric model based on the structural response information and external load information of the floating production storage and unloading device; The feature processing module is configured to: extract finite element feature information for model order reduction from the high-dimensional finite element parameter model, and construct a low-dimensional finite element feature vector based on the extracted finite element feature information; The preliminary order reduction module is configured to: construct a fuzzy logic system based on the low-dimensional finite element feature vectors, model the main mechanical behaviors of the structure represented by the low-dimensional finite element feature vectors, and obtain a preliminary order-reduced finite element model; The model compensation module is configured to: based on the parameter error of the preliminary reduced-order finite element model, use a time-recurrent neural network to model the dynamic residual term and obtain the estimation result of the reduced-order finite element residual force; The fusion module is configured to fuse the preliminary reduced-order finite element model with the estimation result of the reduced-order finite element residual force to obtain the final reduced-order finite element model.

9. A computer device comprising a computer-readable storage medium, a processor, and a computer program stored on the computer-readable storage medium and executable on the processor, characterized in that, When the processor executes the program, it implements the steps in the FPSO model reduction method that integrates time-cyclic networks and fuzzy logic as described in any one of claims 1-7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the program is executed by the processor, it implements the steps in the FPSO model reduction method that integrates time-cyclic networks and fuzzy logic as described in any one of claims 1-7.

Images