Method for predicting overall structural response of floating structure on basis of small number of sensors

Abstract

The present application relates to the technical field of ship and marine engineering. Provided is a method for predicting the overall structural response of a floating structure on the basis of a small number of sensors. Main steps thereof comprise: establishing a finite element simulation model of the full structural field of a floating structure; on the basis of the finite element simulation model, determining a first simulation response data set of the full structural field of the floating structure under a plurality of wave conditions; on the basis of the correlation between a response data set of candidate monitoring points and a response data set of a prediction point in the first simulation response data set, determining a monitoring point; on the basis of the response data set of the prediction point in the first simulation response data set under the plurality of wave conditions, and by means of a preset optimization algorithm, selecting a plurality of wave conditions as a characteristic wave load condition combination; then by means of using the response data set of the monitoring point under the plurality of wave conditions and characteristic wave load weighting coefficients of the corresponding wave conditions as training data, and, on the basis of a preset algorithm, training a surrogate model; and finally, outputting the surrogate model, such that by means of the trained surrogate model, any actual wave condition can be simulated as a combination of a plurality of characteristic wave load conditions and the weighting coefficients thereof, thereby achieving the prediction of the overall structural response of the floating structure. The present method for predicting structural responses can achieve real-time prediction of the full structural field and has relatively high prediction accuracy, and can further achieve calculation of fatigue damages of structural fatigue hot spots and prediction of remaining service lifetime of structures.

Description

A method for predicting the overall structural response of floating structures based on a small number of sensors Technical Field

[0001] This invention relates to the field of marine engineering technology, and in particular to a method for predicting the overall structural response of floating structures based on a small number of sensors. Background Technology

[0002] Floating structures encompass various types of transport or engineering vessels, floating oil, gas, and mineral extraction and storage facilities, as well as floating wind power, photovoltaic, and hydrogen / ammonia production platforms. Because floating structures are continuously subjected to varying wind, wave, and current loads at sea, their structures may experience yielding failure of structural materials or fatigue damage at structural hotspots under the constant action of alternating loads, thus affecting the safety and functional operation of the floating structures. Therefore, conducting full life-cycle, real-time structural response monitoring of floating structures is of great significance for their operation, maintenance, and asset integrity management.

[0003] Currently, traditional structural stress monitoring schemes based on sensor measurement data are limited by the number and space constraints of sensor deployment, making it difficult to reflect the entire structural field, especially the structural response at key locations. Schemes that utilize real-time calculations using measured wave data from the sea and finite element simulation models suffer from significant simulation errors due to the inaccuracy of wave data, resulting in a large computational burden for structural response data and hindering real-time monitoring. Additionally, some monitoring technologies that collect data from sensors, perform load inversion, and then obtain structural response data through simulation calculations have poor anti-interference capabilities and, similarly, are difficult to implement in real-time due to excessive computational demands.

[0004] Therefore, there is a need for a high-precision monitoring technology that can collect structural response data from a small number of sensors to monitor the structural response data of structural points of floating structures in real time. Summary of the Invention

[0005] One objective of this application is to provide a method for predicting the overall structural response of floating structures based on a small number of sensors, in order to solve the problem that it is difficult to perform high-precision real-time monitoring of the entire structural field of floating structures under the existing technology.

[0006] To achieve the above objectives, some embodiments of this application provide a method for predicting the overall structural response of floating structures based on a small number of sensors. The method includes:

[0007] Establish a finite element simulation model of the entire structural field of the floating structure;

[0008] Based on the finite element simulation model, the first simulation response dataset of the floating structure under various wave conditions is determined. The parameters of the wave conditions include at least the wave direction angle, period, and phase angle.

[0009] The monitoring points are determined based on the correlation between the response datasets of candidate monitoring points and the response datasets of predicted points in the first simulation response dataset;

[0010] Based on the response dataset of the predicted points in the first simulation response dataset under various wave conditions, several wave conditions are selected as characteristic wave load condition combinations through a preset optimization algorithm. The characteristic wave load condition combinations include multiple characteristic wave load conditions.

[0011] The response datasets of monitoring points under various wave conditions and the corresponding characteristic wave load weighting coefficients of wave conditions are used as training data. The proxy model is trained based on a preset algorithm. The characteristic wave load weighting coefficients are determined according to the response datasets of monitoring points under wave conditions and the response datasets of monitoring points under a combination of characteristic wave load conditions.

[0012] The output proxy model simulates any actual wave condition as a combination of several characteristic wave load conditions and their weighting coefficients, thereby enabling the prediction of the overall structural response of the floating structure.

[0013] Furthermore, based on the finite element simulation model, the first simulation response dataset of the floating structure under various wave conditions was determined, including:

[0014] Based on the step size and range of each parameter in the initial wave condition, a first wave condition set for simulation is generated.

[0015] Based on the first wave condition set and the finite element simulation model, the first simulation response dataset of the floating structure under the first wave condition set is determined.

[0016] Furthermore, the response dataset includes multiple response data points, which may include structural response data or motion response data.

[0017] Structural response data includes one or more combinations of the following: stress data, strain data, and deformation data.

[0018] Motion response data includes one or more combinations of the following: roll data, pitch data, sway data, and acceleration data.

[0019] Furthermore, based on the correlation between the response datasets of candidate monitoring points and the response datasets of predicted points in the first simulation response dataset, monitoring points are determined, including:

[0020] Candidate monitoring points in the first simulation response dataset that have high correlation with the response datasets of the predicted points and low correlation between any two response datasets are determined as the final monitoring points.

[0021] Furthermore, based on the response datasets of the predicted points in the first simulation response dataset under various wave conditions, several wave conditions are selected as characteristic wave load condition combinations using a preset optimization algorithm, including:

[0022] Based on the response datasets of the predicted points under various wave conditions in the first simulation response dataset, multiple wave conditions are determined as the initial combination of characteristic wave load conditions through preset optimization objectives and corresponding optimization algorithms. The optimization objectives include at least one of the following: minimizing the sum of the dot products of the vectors composed of the response data of the predicted points under each characteristic wave load condition; or minimizing the average error of the predicted point response dataset obtained by solving the combination of characteristic wave load conditions in a large number of random verification samples based on the characteristic wave load weighting coefficients.

[0023] Furthermore, the response datasets of monitoring points under various wave conditions and the corresponding characteristic wave load weighting coefficients for each wave condition are used as training data to train a surrogate model based on a pre-defined method, including:

[0024] The response datasets of monitoring points under multiple wave conditions and the corresponding characteristic wave load weighting coefficients of wave conditions are used as training data. Multiple surrogate model training methods are used to train multiple surrogate models. The surrogate models are used to output the characteristic wave load weighting coefficients corresponding to the wave conditions based on the response datasets of monitoring points under wave conditions.

[0025] Based on the response dataset of the monitoring points under the verification sample working conditions, the characteristic wave load weighting coefficients of the corresponding verification sample working conditions output by multiple surrogate models are obtained. Based on the characteristic wave load weighting coefficients of the corresponding verification sample working conditions and the simulation response dataset of the prediction points under the characteristic wave load working conditions, the prediction response dataset of the prediction points under the verification sample working conditions is determined.

[0026] The optimal surrogate model is determined based on the average error of the predicted response dataset of the predicted point under the verification sample condition and the simulated or measured response dataset of the predicted point under the verification sample condition.

[0027] Furthermore, surrogate model training methods include radial basis functions, kriging, and deep neural networks.

[0028] Furthermore, following the output proxy model, it also includes:

[0029] Acquire measured response data from multiple sensors installed at monitoring points under actual wave conditions;

[0030] Input the measured response data into the proxy model to obtain the characteristic wave load weighting coefficients corresponding to the actual wave conditions;

[0031] Based on the characteristic wave load weighting coefficients corresponding to the actual wave conditions and the simulation response data of the full structural field of the floating structure under multiple characteristic wave load conditions, the predicted response data of the full structural field of the floating structure under the actual wave conditions are determined.

[0032] Output the predicted response data of the entire structural field of the floating structure under actual wave conditions.

[0033] Some embodiments of this application also provide a computer-readable medium having computer-readable instructions stored thereon, which can be executed by a processor to implement the aforementioned method for predicting the overall structural response of floating structures based on a small number of sensors.

[0034] Some embodiments of this application also provide an electronic device, which includes a memory for storing computer program instructions and a processor for executing the computer program instructions, wherein when the computer program instructions are executed by the processor, the electronic device performs the aforementioned method for predicting the overall structural response of floating structures based on a small number of sensors.

[0035] Compared with existing technologies, the solution provided in this application establishes a finite element simulation model of the entire structural field of the floating structure, and determines the first simulation response dataset of the entire structural field of the floating structure under various wave conditions based on the finite element simulation model. Based on the correlation between the response datasets of candidate monitoring points and the response datasets of predicted points in the first simulation response dataset, monitoring points are determined. Then, based on the response datasets of predicted points under various wave conditions in the first simulation response dataset, several wave conditions are selected as characteristic wave load combinations using a preset optimization algorithm. The response datasets of monitoring points under various wave conditions and the corresponding characteristic wave load weighting coefficients are used as training data to train a surrogate model based on a preset algorithm. Finally, the surrogate model is output. This surrogate model can simulate any actual wave condition as a combination of several characteristic wave load conditions and their weighting coefficients, thereby achieving prediction of the overall structural response of the floating structure. The structural response prediction method of this application can achieve real-time prediction of the entire structural field with high prediction accuracy. Furthermore, it can also calculate fatigue damage at structural fatigue hotspots and predict the remaining life of the structure. Attached Figure Description

[0036] Other features, objects, and advantages of this application will become more apparent from the following detailed description of non-limiting embodiments with reference to the accompanying drawings:

[0037] Figure 1 is a flowchart of a method for predicting the overall structural response of a floating structure based on a small number of sensors, provided by some embodiments of this application.

[0038] Figure 2 is a flowchart of a proxy model trained based on the response dataset of monitoring points and the weighted coefficients of characteristic wave loads provided in some embodiments of this application.

[0039] Figure 3 is a flowchart illustrating the use of a surrogate model to predict the response data of a prediction point under actual wave conditions, as provided in some embodiments of this application. Detailed Implementation

[0040] The present invention will now be described in detail with reference to the accompanying drawings and specific embodiments. These embodiments are based on the technical solution of the present invention and provide detailed implementation methods and specific operating procedures. However, the scope of protection of the present invention is not limited to the following embodiments.

[0041] Here, the method for predicting the overall structural response of floating structures based on a small number of sensors in this application embodiment is suitable for scenarios where the structural response of the entire structural field of a floating structure is predicted in real time.

[0042] In this scenario, floating structures continuously endure varying wind, wave, and current loads at sea. Under the constant action of alternating loads, the structure itself may experience yielding failure of structural materials or fatigue damage at structural hotspots, thus affecting the functionality and safe operation of the floating structure. Therefore, it is necessary to monitor the structural response of floating structures throughout their entire life cycle and in real time to improve their safety and ensure functionality. However, many critical structural points in floating structures, especially fatigue hotspots, cannot be monitored in real time by installing sensors. Furthermore, existing methods for monitoring structural stress through simulation technology are greatly affected by the accuracy of wave data, resulting in large errors. At the same time, the acquisition of structural response is also difficult to achieve in real time due to limitations in computing power.

[0043] In actual sea conditions, ocean waves are irregular. In engineering applications, to simplify the problem, irregular wave conditions can be considered as a superposition of many regular wave conditions with different wave directions, periods, amplitudes, and random phases. The formula for the rise of an irregular wave face can be expressed as follows:

[0044] Where, ξ Ai Let ε represent the amplitude of the i-th component wave, x and t represent the position and time of the recording point, respectively. i The initial phase of the i-th long-peaked irregular wavelet can take any value between 0 and 2π, k i ω is a parameter representing the wave number of the i-th component wave, whose value is equal to 2π times the number of waves per unit distance. i This indicates the angular frequency of the wave.

[0045] Therefore, under certain assumptions, the structural or kinematic response of a floating structure caused by actual sea conditions can also be considered as the superposition of structural or kinematic responses caused by many regular wave conditions. At any point in time within actual sea conditions, the formula for the structural or kinematic response of a floating structure can be expressed as follows: R = R 1 +R 2 +…+R i

[0046] Where R is the structural or kinematic response of the floating structure at that point in time caused by the actual sea conditions, R 1 To R i The structural or kinematic response is caused by the i regular wave conditions derived from the irregular wave condition at that time point.

[0047] Furthermore, assuming that the structural or kinematic response of a floating structure caused by regular wave conditions with arbitrary wave direction, period, phase, and amplitude can be characterized by a linear weighted sum of the structural or kinematic responses caused by characteristic wave load conditions of several regular wave conditions, where characteristic wave load conditions refer to wave conditions with unit wave amplitude under several specific wave directions, periods, and phases, the linear weighted formula for the structural or kinematic response can be expressed as follows:

[0048] Among them, R1 to R n Let n be the structural or kinematic responses caused by each of the n characteristic wave load conditions. to For n characteristic wave load conditions, the characteristic wave load weighting coefficients are applied to the regular wave load condition.

[0049] Therefore, at any point in time in actual sea conditions, the formula for the structural or kinematic response of a floating structure caused by irregular wave conditions can be expressed as follows:

[0050] That is, R = a1R1 + a2R2 + ... + a n R n

[0051] Among them, R1 to R n Let a1 to a2 be the structural or kinematic responses caused by n characteristic wave load conditions. n The weighting coefficients for the characteristic wave loads at this time point in the actual sea conditions for n characteristic wave load conditions.

[0052] By installing sensors at multiple designated locations on the floating structure, such as strain sensors, attitude sensors, and acceleration sensors, m response data points can be obtained at any given time. The response data includes structural response data and motion response data, where m is the number of sensors installed. m is greater than or equal to n, meaning that the number of sensors installed must be greater than or equal to the number of characteristic wave load conditions.

[0053] Under n characteristic wave load conditions, the response data collected by these m sensors can be represented by the following matrix M:

[0054] Where, r 11 to r mn These are the response data collected by m sensors under n characteristic wave load conditions.

[0055] Therefore, for any point in time in actual sea conditions, the structural or kinematic response of a floating structure caused by irregular wave conditions can be expressed in the following matrix form:

[0056] The above matrix form can be simplified as follows: M·A=R′

[0057] in, This is a vector of weighted coefficients for n characteristic wave loads. Let m be the response data vectors collected by m sensors.

[0058] Therefore, the characteristic wave load weighting coefficient vector A can be expressed as follows: A = M -1 ·R′

[0059] For the pseudo-inverse matrix M -1 Solve using R ′ By using the response data collected by the sensor, the characteristic wave load weighting coefficient vector A can be calculated, thereby obtaining the characteristic wave load weighting coefficients corresponding to n characteristic wave loads at any time point.

[0060] After obtaining the characteristic wave load weighting coefficients corresponding to n characteristic wave loads at any time point, we can further apply the formula R = a1R1 + a2R2 + ... + a n R n The structural and kinematic responses of the floating structure under actual sea conditions at any time point were calculated.

[0061] This application provides a method for predicting the overall structural response of floating structures based on a limited number of sensors. The method establishes a finite element simulation model of the entire structural field of the floating structure and determines a first simulated response dataset of the entire structural field under various wave conditions based on the finite element simulation model. Monitoring points are determined based on the correlation between the response datasets of candidate monitoring points and the response datasets of predicted points in the first simulated response dataset. Several wave conditions are selected as combinations of characteristic wave load conditions using a preset optimization algorithm based on the response datasets of predicted points under various wave conditions in the first simulated response dataset. The response datasets of monitoring points under various wave conditions and the corresponding characteristic wave load weighting coefficients are used as training data to train a surrogate model based on a preset algorithm. Finally, the surrogate model is output, which can simulate any actual wave condition as a combination of several characteristic wave load conditions and their weighting coefficients, thereby predicting the overall structural response of the floating structure. This structural response prediction method can achieve real-time prediction of the entire structural field with high accuracy, and can also calculate fatigue damage at structural fatigue hotspots and predict the remaining life of the structure.

[0062] In some embodiments of this application, a process for predicting the overall structural response of a floating structure based on a small number of sensors is executed via electronic devices, as shown in Figure 1. The method may include the following steps:

[0063] Step S101: Establish a finite element simulation model of the entire structural field of the floating structure.

[0064] It is understood that electronic devices may include, but are not limited to, laptops, desktop computers, tablets, mobile phones, wearable devices, head-mounted displays, servers, mobile email devices, portable game consoles, portable music players, e-reader devices, televisions in which one or more processors are embedded or coupled, or other electronic devices capable of accessing networks.

[0065] Here, floating structures may include, but are not limited to: various types of marine transport vessels, engineering vessels, floating devices for oil, gas and mineral extraction and storage, floating wind power, photovoltaic, hydrogen and ammonia production platforms, etc.

[0066] The overall structure field refers to the entire structure of a floating structure. For example, a ship is the entire ship structure, and an offshore engineering platform is the entire platform structure. Specifically, it can include materials such as outer plates, bulkheads, and skeletons, steel grades, plate thickness specifications, etc., as well as the position of each structural component in space.

[0067] Finite Element Analysis (FEA) is a method for simulating real physical systems (geometry and load conditions) using mathematical approximations. FEA uses simple, interacting elements (elements) to approximate a real system with an infinite number of unknowns using a finite number of unknowns. FEA solves a complex problem by substituting a simpler one. It considers the solution domain as composed of many small, interconnected subdomains called finite elements. For each element, a simpler approximate solution is assumed, and then the overall conditions satisfying the domain, such as the equilibrium conditions of the structure, are derived to obtain the solution. Because the real problem is replaced by a simpler one, this solution is not an exact solution but an approximate one. Since most real-world problems are difficult to solve accurately, and FEA offers high computational accuracy and can adapt to various complex shapes, it has become an effective engineering analysis tool.

[0068] A finite element simulation model is a model established using the finite element analysis method. It is a combination of elements that are connected only at the nodes, transmit forces only through the nodes, and are constrained only at the nodes. It is the result of discretization of the mechanical model and is a digital model for numerical calculation.

[0069] Here, a finite element simulation model of the entire structural field for floating structures can be established using third-party finite element simulation software, such as Det Norske Veritas (DNV)'s SESAM software package, FEmap, patran / nastran, Ansys, etc.

[0070] Step S102: Based on the finite element simulation model, determine the first simulation response dataset of the floating structure under various wave conditions.

[0071] After obtaining the finite element simulation model, the structural response dataset of the floating structure under a typical loading condition and different wave conditions is obtained through simulation of the entire structural field. This dataset is the first simulation response dataset, which contains the simulation response data of all structural points in the floating structure.

[0072] Here, the response dataset includes multiple response data points. In some embodiments of this application, the response data may include structural response data and motion response data. Structural response data includes, but is not limited to, the following: stress data, strain data, deformation data, etc. For example, the yield stress data of a certain structural point is 150 MPa, and the deformation data is a vertical deformation of 5 mm. Motion response data may include, but is not limited to, the following: roll data, pitch data, heave data, acceleration data, etc. For example, the roll data of the floating structure is 15°, the pitch data is 5°, the heave data is -2 m, and the acceleration of a certain structural point is 1 m / s² in the x-direction. 2 y-direction 2m / s 2 0.5m / s in the z direction 2 .

[0073] It is understood that the parameters used to describe wave conditions include at least the wave direction angle, period, and phase angle, and may include many other parameters. No restrictions are placed on the parameters used to describe wave conditions here.

[0074] Here, various wave conditions can be generated within the sea state range that the floating structure may actually encounter by setting a certain step size for the variation of wave parameters. In some embodiments of this application, a first wave condition set for simulation can be generated based on the step size and range of each parameter of the initial wave condition. The first wave condition set includes multiple wave conditions. Then, based on the first wave condition set and the finite element simulation model, a first simulation response dataset of the entire structural field of the floating structure under the first wave condition set is determined. For example, the initial wave parameters are a period of 3s, a wave direction angle of 0°, and a phase angle of 0°. The step size for each parameter is set to a period step size of 1s, a wave direction angle step size of 15°, and a phase angle step size of 30°. Based on the initial wave condition and the step size of each parameter, wave condition A is obtained. Based on wave condition A and the finite element simulation model, the simulation response dataset of the entire structural field under wave condition A is obtained. The simulation response dataset of the entire structural field under wave condition A is added to the first simulation response dataset to obtain a new first simulation response dataset. Then, based on wave condition A and the step size of each parameter, wave condition B is obtained. The simulation response dataset of the entire structural field under wave condition B is obtained and added to the first simulation response dataset. This process is repeated until the parameters of the wave condition are exceeded.

[0075] In some embodiments of this application, a full-ship finite element simulation model of the Floating Production Storage and Offloading System (FPSO) is established using the SESAM software package from Det Norske Veritas (DNV). The simulation model has approximately 700,000 elements, i.e., structural points. The first simulation response dataset is then established through the following process: defining wave parameter variables, performing hydrodynamic analysis using SESAM-HydroD, performing structural response analysis using SESAM-Sestra, and extracting simulation data using SESAM-Xtract. The wave parameter variables include: period 3s–30s, wave direction angle 0°–360°, and phase angle 0°–360°.

[0076] In some embodiments of this application, sensors can be installed at multiple key structural locations of the already constructed and operational floating structure to collect measured response datasets of the floating structure under actual sea conditions, and the union of the measured response datasets and the response datasets obtained through simulation can be used as the first simulation response dataset.

[0077] Step S103: Determine the monitoring points based on the correlation between the response datasets of candidate monitoring points and the response datasets of predicted points in the first simulation response dataset.

[0078] Here, multiple candidate monitoring points and multiple prediction points for the floating structure are pre-determined structural critical points. Specifically, based on the monitoring needs of the floating structure and practical engineering experience, some structural critical points can be pre-selected as candidate monitoring points and prediction points in the entire structural field of the floating structure. Candidate monitoring points can be structural points that easily characterize the overall stress state of the floating structure. These structural points can have their corresponding response data collected by setting sensors at these points. Prediction points can be key structural strength points or fatigue hotspots in the floating structure, which often cannot have their response data collected by setting sensors. For example, 81 points can be selected as candidate monitoring points from approximately 700,000 structural points in the floating structure, and 136 key strength points and fatigue hotspots can be selected as prediction points.

[0079] It is understandable that the number of candidate monitoring points is greater than the number of monitoring points. The final monitoring points are only a small portion of the candidate monitoring points. The corresponding measured response data is obtained by setting up sensors at the final monitoring points. For example, 19 candidate monitoring points can be selected from 81 candidate monitoring points as the final monitoring points, and strain sensors can be set up at these 19 monitoring points to obtain measured strain response data.

[0080] Here, the final monitoring point is selected from multiple candidate monitoring points based on the correlation between the response datasets of the candidate monitoring points and the prediction points, and the correlation between the response datasets of each candidate monitoring point.

[0081] In some embodiments of this application, candidate monitoring points in the first simulation response dataset that have a high correlation with the response dataset of the prediction point and a low correlation between any two response datasets are determined as the final monitoring points.

[0082] Correlation analysis between response datasets can be performed using various methods, such as Pearson, Spearman, and Kendall. Specifically, correlation analysis can be conducted using third-party software that implements correlation analysis algorithms or by developing custom software. For example, the built-in correlation analysis function in the third-party software Optimus can be used to analyze the correlation between the response datasets of candidate monitoring points and prediction points, as well as the correlation between the response datasets of each pair of candidate monitoring points.

[0083] After obtaining the correlation analysis results, the candidate monitoring points that simultaneously satisfy the conditions of high correlation with the response dataset of the prediction point and low correlation with the response datasets of other candidate monitoring points are determined as the final monitoring points.

[0084] Step S104: Based on the response dataset of the predicted points in the first simulation response dataset under various wave conditions, select several wave conditions as characteristic wave load condition combinations using a preset optimization algorithm. The characteristic wave load condition combinations include multiple characteristic wave load conditions.

[0085] After determining all prediction points, the corresponding response datasets of all prediction points under various wave conditions are obtained from the first simulation response dataset, and the characteristic wave load condition combination is determined through the response dataset.

[0086] It is understood that the pre-set optimization algorithm can be different optimization analysis algorithms, such as differential evolution algorithm, sequential quadratic programming algorithm, generalized simplified gradient algorithm, adaptive evolution algorithm and single-objective particle swarm algorithm, etc. The embodiments of this application do not impose specific restrictions on the optimization algorithm used.

[0087] In some embodiments of this application, the pre-set optimization algorithm is the Differential Evolution (DE) algorithm. The Differential Evolution algorithm is a type of population-based adaptive global optimization algorithm, belonging to the evolutionary algorithm family. It features simple structure, ease of implementation, fast convergence, and strong robustness, and is widely used in data mining, pattern recognition, digital filter design, artificial neural networks, electromagnetism, and many other fields. Similar to genetic algorithms, the Differential Evolution algorithm is also an optimization algorithm based on modern intelligence theory. It guides the optimization search direction through the collective intelligence generated by the cooperation and competition among individuals within the population. The basic idea of ​​this algorithm is to start with a randomly generated initial population, generate new individuals by summing the vector difference between any two individuals in the population with a third individual, and then compare the new individual with the corresponding individual in the current population. If the fitness of the new individual is better than that of the current individual, then the new individual replaces the old individual in the next generation; otherwise, the old individual is retained. Through continuous evolution, superior individuals are retained, inferior individuals are eliminated, and the search is guided towards the optimal solution.

[0088] In some embodiments of this application, multiple wave conditions can be determined as a combination of characteristic wave load conditions based on the response dataset of the predicted points in the first simulation response dataset under multiple wave conditions, through a preset optimization objective and a corresponding optimization algorithm. The combination of characteristic wave load conditions includes multiple characteristic wave load conditions.

[0089] Here, the optimization objective includes at least one of the following: minimizing the sum of the dot products of the response dataset of the predicted points under the combination of characteristic wave load conditions, or obtaining the weighting coefficient of the characteristic wave load by solving the combination of characteristic wave load conditions, and then minimizing the average error of the predicted point response dataset obtained based on the weighting coefficient of the characteristic wave load in a large number of random validation samples.

[0090] Under each wave condition, the response data of all predicted points are treated as a vector. The smaller the dot product of the response data vectors of all predicted points under two different wave conditions, the better the orthogonality of the response datasets under those two wave conditions, i.e., the lower the correlation. Therefore, by selecting multiple wave conditions with low correlation among them as a combination of characteristic wave load conditions, irregular wave conditions can be described using a small number of characteristic wave load conditions.

[0091] In some embodiments of this application, the sum of the dot products of the response datasets of the predicted points under characteristic wave load conditions is less than the sum of the dot products of the response datasets of the predicted points under non-characteristic wave load conditions. Here, the sum of the dot products of the response data of all predicted points under different combinations of wave load conditions is calculated, and the corresponding combination of wave characteristic loads is determined based on the sum of the dot products. For example, the sums of the dot products are sorted in ascending order, and the top-ranked combinations of wave load conditions are determined as the combinations of characteristic wave load conditions.

[0092] Here, a variety of possible wave load combinations can be calculated using a preset optimization algorithm. For example, by setting the optimization algorithm to calculate 20,000 times, the optimal wave load combinations can be found from the results, and one of the optimal wave load combinations can be selected as the characteristic wave load combination.

[0093] For example, a differential evolution algorithm can be used to find k wave conditions within a parameter range that have a relatively small sum of dot products and can effectively cover the wave conditions that floating structures may encounter in practice. k may be between 5 and 20. These k wave conditions can be used as the characteristic load condition combination for the floating structure. Furthermore, the number of characteristic load conditions in the characteristic load condition combination can be determined by comprehensively considering the number of input signals from the monitoring point sensors.

[0094] In some embodiments of this application, the average error of the response dataset of all predicted points obtained by solving the weighting coefficient of the characteristic wave load through a combination of characteristic wave load conditions in a large number of random verification samples is less than the average error of the response dataset of all predicted points obtained by solving the weighting coefficient of the characteristic wave load through a combination of non-characteristic wave load conditions in a large number of random verification samples.

[0095] Specifically, multiple wave conditions can be selected from various wave conditions, with several wave conditions selected each time, resulting in multiple wave condition combinations. For each wave condition combination, a corresponding characteristic wave load weighting coefficient is determined through a surrogate model. Based on the obtained characteristic wave load weighting coefficients, the response dataset of all predicted points is calculated. Then, the average error of the response dataset of all predicted points in a large number of random validation samples is calculated. The wave condition combination with the smallest average error is taken as the characteristic wave load combination. This ensures that the average error of the response dataset of all predicted points obtained by solving for the characteristic wave load weighting coefficients through the characteristic wave load combination in a large number of random validation samples is less than the average error of the response dataset of all predicted points obtained by solving for the characteristic wave load weighting coefficients through the non-characteristic wave load combination in a large number of random validation samples.

[0096] Step S105: Use the response dataset of monitoring points under various wave conditions and the weighted coefficients of characteristic wave loads of the corresponding wave conditions as training data to train a proxy model based on a preset algorithm.

[0097] It is understood that the response dataset of all monitoring points under various wave conditions can be a simulated response dataset obtained by simulating various wave conditions through finite element simulation models or artificially generating them, causing the corresponding response of the floating structure at all monitoring points. Alternatively, it can be a measured response dataset collected by sensors installed at monitoring points in a built and operating floating structure under various actual wave conditions. It can also be a combined response dataset of simulated response dataset and measured response dataset. This application embodiment does not impose specific restrictions on the source of the response dataset of monitoring points under wave conditions.

[0098] Here, the characteristic wave load weighting coefficient corresponding to a certain wave condition is determined based on the response data vector of all monitoring points under that wave condition and the response dataset of all monitoring points under the combination of characteristic wave load conditions. Specifically, it can be determined according to the aforementioned formula A = M. -1 ·R ′ The calculation yields M here. -1 R is the pseudo-inverse matrix of the response data matrix of all monitoring points under multiple characteristic wave load conditions. ′ This is the response data vector for all monitoring points under this wave condition.

[0099] The surrogate model training method is a pre-selected method, which may include, but is not limited to, radial basis functions (RBF), kriging, and deep neural network algorithms (DNN). It can be understood that the surrogate model training method only needs to output the corresponding feature wave load weighting coefficient vector based on the input response dataset; this application does not impose specific restrictions on the surrogate model training method used.

[0100] Figure 2 illustrates the process of training a surrogate model based on the response dataset of monitoring points and the weighted coefficients of characteristic wave loads in some embodiments of this application. As shown in Figure 2, the response datasets of all monitoring points under various wave conditions and the weighted coefficients of characteristic wave loads for the corresponding wave conditions are used as training data. The surrogate model is trained according to a preset surrogate model training method to obtain the surrogate model, which may include the following steps:

[0101] Step S201: The response data of the monitoring point under multiple wave conditions and the weighted coefficients of the characteristic wave loads of the corresponding wave conditions are used as training data and trained by multiple surrogate model training methods to obtain multiple surrogate models.

[0102] Response data and corresponding characteristic wave load weighting coefficients under multiple wave conditions are used as training samples. An appropriate proportion of sample points are reserved as a validation set and not used in the surrogate model training. The remaining sample points are used as the training set for surrogate model training. Specifically, the response data of the monitoring point under each sample condition in the training set is used as the input data of the surrogate model, and the characteristic wave load weighting coefficients corresponding to that sample condition are used as the output data of the surrogate model. The surrogate model is then trained to obtain the final surrogate model.

[0103] Similarly, the response data of all monitoring points under various wave conditions in the training samples and the corresponding weighted coefficients of characteristic wave loads are used as training data to train other surrogate model training methods, and finally multiple trained surrogate models are obtained.

[0104] Each trained surrogate model can be used to output the characteristic wave load weighting coefficient corresponding to a certain wave condition based on the response data of all monitoring points under a certain wave condition.

[0105] Step S202: Based on the response data of the monitoring points under the verification sample working conditions, obtain the characteristic wave load weighting coefficients of the corresponding verification sample working conditions output by multiple surrogate models, and determine the predicted response data of the prediction points under the verification sample working conditions based on the characteristic wave load weighting coefficients of the corresponding verification sample working conditions and the simulation response dataset of the prediction points under the characteristic wave load working conditions.

[0106] It is understood that the response data of all monitoring points under the verification sample working conditions can be the simulation response data obtained from the finite element simulation model, the response data under the measured wave working conditions, or a combination of simulation response data and measured response data. This application embodiment does not impose specific limitations on this.

[0107] In addition, the simulation response data of the predicted points under the final characteristic wave load combination is generated based on the finite element simulation model of the floating structure. The specific method can be referred to the generation process of the first simulation response dataset mentioned above, and will not be repeated here.

[0108] Here, the proxy model outputs the characteristic wave load weighting coefficient corresponding to the verification sample condition based on the response data of all the input monitoring points under the verification sample condition. Then, the obtained characteristic wave load weighting coefficient is matrix-multiplied with the simulation response data of all the prediction points under the characteristic wave load condition to obtain the response data of all the prediction points under the verification sample condition. This response data is the predicted response data calculated based on the characteristic wave load weighting coefficient.

[0109] Multiple surrogate models output their respective characteristic wave load weighting coefficients based on the response data of all input monitoring points under the verification sample conditions. The obtained characteristic wave load weighting coefficients are then matrix-multiplied with the simulation response data of all prediction points under the characteristic wave load conditions to obtain the prediction point response data corresponding to different surrogate models.

[0110] Step S203: Determine the optimal surrogate model based on the predicted response dataset of the predicted point under the verification sample working condition and the average error of the response dataset of the predicted point under the verification sample working condition.

[0111] Here, the simulation or measured response dataset of all predicted points under the verification sample conditions is based on the response data obtained from the finite element simulation model or the response data obtained from the actual ship sensor measurements, and is used as the evaluation standard for judging the performance of the surrogate model.

[0112] All predicted response data obtained from different surrogate models are compared with the simulated or measured response data of all predicted points under the verification sample conditions. The corresponding average error is calculated, and the surrogate model corresponding to the predicted response data with the smallest average error is determined as the optimal surrogate model.

[0113] In some embodiments of this application, the optimal surrogate model can be further optimized by adding additional wave conditions and corresponding response datasets of monitoring points under the additional wave conditions.

[0114] Step S106: Output the proxy model. The proxy model simulates any actual wave condition as a combination of several characteristic wave load conditions and their weighting coefficients, thereby enabling the prediction of the overall structural response of the floating structure.

[0115] After selecting the best-performing proxy model from multiple proxy models, the best proxy model is output for users to use to predict the response data of all prediction points under actual wave conditions.

[0116] Figure 3 illustrates the process of using a surrogate model to predict the response data of a prediction point under actual wave conditions in some embodiments of this application. As shown in Figure 3, the process may specifically include the following steps:

[0117] Step S301: Obtain the measured response dataset collected by multiple sensors set at the monitoring point under actual wave conditions.

[0118] Here, the actual wave condition refers to the irregular wave condition experienced by the floating structure in the actual marine environment at a certain point in time. Multiple monitoring points in the floating structure are equipped with sensors to collect response data, and multiple sensors collect the measured response data set at that point in time.

[0119] Step S302: Input the measured response data into the proxy model to obtain the characteristic wave load weighting coefficients corresponding to the actual wave conditions.

[0120] The measured response data under actual wave conditions collected at all monitoring points are used as input data to the optimal surrogate model, and the optimal surrogate model outputs the characteristic wave load weighting coefficients corresponding to the actual wave conditions.

[0121] Step S303: Based on the weighting coefficient of the characteristic wave load corresponding to the actual wave condition and the simulation response data of the prediction point under multiple characteristic wave load conditions, determine the predicted response data of the prediction point under the actual wave condition.

[0122] After obtaining the weighted coefficients of the characteristic wave loads corresponding to the actual wave conditions, the coefficients are multiplied by matrix with the simulation response datasets of all predicted points under multiple characteristic wave load conditions to obtain the predicted response data of all predicted points under the actual wave conditions.

[0123] Step S304: Output the predicted response data of the predicted point under actual wave conditions.

[0124] Similarly, the measured response data of the monitoring points are input into the optimal surrogate model, which outputs the corresponding time-history characteristic wave load weighting coefficients, thereby outputting the time-history predicted response data for all predicted points. This allows for the calculation of fatigue damage to structural hotspots and the prediction of the remaining life of the structure. Furthermore, the output time-history predicted response data, fatigue damage calculation results, and the predicted remaining life of the structure can be displayed graphically and visually, providing users with an intuitive understanding of the structural fatigue status of all predicted points in the floating structure.

[0125] In summary, the solution provided in this application establishes a finite element simulation model of the entire structural field of a floating structure, determines the first simulation response dataset of the floating structure under various wave conditions based on the finite element simulation model, determines the monitoring points based on the correlation between the response datasets of candidate monitoring points and the response datasets of predicted points in the first simulation response dataset, selects several wave conditions as characteristic wave load combinations based on the response datasets of predicted points under various wave conditions through a preset optimization algorithm, and then uses the response datasets of monitoring points under various wave conditions and the corresponding characteristic wave load weighting coefficients as training data to train a surrogate model based on the preset algorithm. Finally, the surrogate model is output, which can simulate any actual wave condition as a combination of several characteristic wave load conditions and their weighting coefficients through the trained surrogate model, thereby realizing real-time prediction of the overall structural response of the floating structure. The structural response prediction method of this application can realize real-time prediction of the entire structural field with high prediction accuracy, and can also calculate fatigue damage of structural fatigue hotspots and predict the remaining life of the structure.

[0126] It will be apparent to those skilled in the art that this application is not limited to the details of the exemplary embodiments described above, and that it can be implemented in other specific forms without departing from the spirit or essential characteristics of this application. Therefore, the embodiments should be considered exemplary and non-limiting in all respects, and the scope of this application is defined by the appended claims rather than the foregoing description. Thus, all variations falling within the meaning and scope of equivalents of the claims are intended to be embraced within this application. No reference numerals in the claims should be construed as limiting the scope of the claims. Furthermore, it is clear that the word "comprising" does not exclude other elements or steps, and the singular does not exclude the plural. Terms such as "first," "second," etc., are used to denote names and do not indicate any particular order.

Claims

1. A method for predicting the overall structural response of floating structures based on a small number of sensors, characterized in that, The method includes: Establish a finite element simulation model of the entire structural field of the floating structure; Based on the finite element simulation model, the first simulation response dataset of the floating structure under various wave conditions is determined, wherein the parameters of the wave conditions include at least the wave direction angle, period, and phase angle. The monitoring points are determined based on the correlation between the response datasets of candidate monitoring points and the response datasets of predicted points in the first simulation response dataset; Based on the response dataset of the predicted points in the first simulation response dataset under various wave conditions, several wave conditions are selected as characteristic wave load condition combinations through a preset optimization algorithm. The characteristic wave load condition combinations include multiple characteristic wave load conditions. The response datasets of the monitoring points under various wave conditions and the corresponding characteristic wave load weighting coefficients of the wave conditions are used as training data to train a proxy model based on a preset algorithm. The characteristic wave load weighting coefficients are determined based on the response datasets of the monitoring points under the wave conditions and the response datasets of the monitoring points under the combination of characteristic wave load conditions. The proxy model is output, and any actual wave condition is simulated as a combination of several characteristic wave load conditions and their weighting coefficients through the proxy model, so as to predict the overall structural response of the floating structure.

2. The prediction method according to claim 1, characterized in that, Based on the finite element simulation model, the first simulation response dataset of the floating structure under various wave conditions is determined, including: Based on the step size and range of each parameter in the initial wave condition, a first wave condition set for simulation is generated. Based on the first wave condition set and the finite element simulation model, the first simulation response dataset of the floating structure under the first wave condition set is determined.

3. The prediction method according to claim 1, characterized in that, The response dataset includes multiple response data points, which may include structural response data or motion response data. The structural response data includes one or more combinations of the following: stress data, strain data, and deformation data. The motion response data includes one or more combinations of the following: roll data, pitch data, sway data, and acceleration data.

4. The prediction method according to claim 1, characterized in that, Based on the correlation between the response datasets of candidate monitoring points and the response datasets of predicted points in the first simulation response dataset, monitoring points are determined, including: Candidate monitoring points in the first simulation response dataset that have high correlation with the response datasets of the prediction points and low correlation between any two response datasets are determined as the final monitoring points.

5. The prediction method according to claim 1, characterized in that, Based on the response datasets of predicted points in the first simulation response dataset under various wave conditions, several wave conditions are selected as characteristic wave load condition combinations using a preset optimization algorithm, including: Based on the response datasets of the predicted points in the first simulation response dataset under various wave conditions, multiple wave conditions are determined as a combination of characteristic wave load conditions through a preset optimization objective and a corresponding optimization algorithm. The optimization objective includes at least one of the following: the sum of the dot products of the vectors composed of the response data of the predicted points under each characteristic wave load condition is minimized; or the characteristic wave load weighting coefficient is obtained by solving the combination of characteristic wave load conditions, and the average error of the predicted point response dataset obtained based on the characteristic wave load weighting coefficient is minimized in a large number of random verification samples.

6. The prediction method according to claim 1, characterized in that, Using the response datasets of the monitoring points under various wave conditions and the corresponding characteristic wave load weighting coefficients for each wave condition as training data, a proxy model is trained based on a preset method, including: The response datasets of the monitoring points under multiple wave conditions and the corresponding characteristic wave load weighting coefficients of the wave conditions are used as training data. Multiple proxy models are trained using multiple proxy model training methods to obtain multiple proxy models. The proxy models are used to output the characteristic wave load weighting coefficients corresponding to the wave conditions based on the response datasets of the monitoring points under the wave conditions. Based on the response dataset of the monitoring point under the verification sample working condition set, the characteristic wave load weighting coefficients corresponding to the verification sample working condition output by multiple proxy models are obtained, and based on the characteristic wave load weighting coefficients corresponding to the verification sample working condition and the simulation response dataset of the prediction point under the characteristic wave load working condition, the prediction response dataset of the prediction point under the verification sample working condition is determined. The optimal surrogate model is determined based on the average error of the predicted response dataset of the predicted point under the verification sample condition and the simulated or measured response dataset of the predicted point under the verification sample condition.

7. The prediction method according to claim 1, characterized in that, The surrogate model training method includes radial basis functions, kriging, and deep neural networks.

8. The prediction method according to claim 1, characterized in that, After outputting the proxy model, the following is also included: Acquire measured response data from multiple sensors installed at the monitoring point under actual wave conditions; Input the measured response data into the proxy model to obtain the characteristic wave load weighting coefficients corresponding to the actual wave condition; Based on the characteristic wave load weighting coefficients corresponding to the actual wave conditions and the simulation response data of the full structural field of the floating structure under multiple characteristic wave load conditions, the predicted response data of the full structural field of the floating structure under the actual wave conditions are determined. Output the predicted response data of the entire structural field of the floating structure under the actual wave conditions.

Images