A method for optimizing the arrangement of sensors on offshore wind turbine jackets

Abstract

This invention relates to the field of offshore wind turbine structure inspection, specifically to a method for optimizing the arrangement of sensors on offshore wind turbine jacket structures. The method includes the following steps: extracting the target order modal shape information of the offshore wind turbine jacket structure, constructing a candidate node set for sensor placement, and forming a working mode matrix; establishing a sensor placement evaluation model based on the working mode matrix, and reducing numerical instability caused by ill-conditioned information matrix under high-correlation mode conditions by introducing a regularization term; determining the scanning interval, evaluating the number of sensors for different numbers, and selecting the minimum number that meets the requirements as the recommended number of sensors; performing independent optimization by gradually replacing the measuring points, and selecting the placement scheme with higher information content, better stability, and stronger repeatability from multiple optimization results as the final result. This invention is more applicable to engineering applications of offshore wind turbine jacket structures, and can reduce engineering costs while ensuring monitoring effectiveness.

Description

Technical Field

[0001] This invention relates to the field of offshore wind turbine structure inspection, and more specifically to a method for optimizing the arrangement of sensors on offshore wind turbine jackets. Background Technology

[0002] With the continued advancement of the "dual carbon" target and the ongoing development of offshore wind power towards deeper waters and larger capacities, my country is increasingly focused on the safe operation and long-term service performance of offshore wind power infrastructure. Jacket foundations, due to their high overall rigidity, load-bearing capacity, and good adaptability to medium to deep waters, have become one of the important forms of support structures for offshore wind turbines. To better ensure the safe operation of offshore wind turbine jacket structures in complex marine environments and to monitor their service status in a timely manner, structural health monitoring technology has gradually become an important research direction in this field.

[0003] In recent years, with the continuous development of structural health monitoring and optimal sensor placement technologies, methods based on modal kinetic energy, modal guarantee criteria, effective independence method (EFI), optimal information matrix criteria, multi-objective optimization, and heuristic search have been gradually applied. These methods can improve the scientific nature of sensor placement to a certain extent and reduce the number of measurement points. However, existing methods are mostly designed for conventional civil structures such as buildings and bridges, and their applicability to offshore wind turbine jacket structures remains insufficient. In particular, for spatial three-dimensional truss structures like jackets, their low-order modes typically exhibit stronger three-dimensional coupling characteristics, and under the combined effects of wind, waves, currents, and turbine operating loads, the modes are dense and the response correlation is high, making it difficult for traditional methods to determine the number of measurement points and optimize their locations.

[0004] Compared to conventional structures, offshore wind turbine jacket structures typically feature a large number of nodes, complex spatial topology, significant modal coupling, and a large number of candidate measurement points. Relying solely on empirical methods for sensor placement often leads to an excessive number of sensors, severe information redundancy, and low placement efficiency, making it difficult to balance monitoring effectiveness with engineering costs. Jacket structures also face practical constraints in engineering applications, such as installation accessibility, power and communication reliability, long-term maintenance accessibility, and limited offshore construction windows. Under these conditions, sensor placement must not only ensure sufficient data volume but also consider numerical stability, noise resistance, and engineering feasibility. Existing methods often focus on optimizing single information indicators or only on the selection of measurement point locations, lacking systematic and effective solutions for how to objectively determine the number of sensors and how to ensure the robustness and reproducibility of placement results under complex constraints.

[0005] To address the shortcomings of existing sensor placement methods in offshore wind turbine jacket structures and improve the rationality, stability, and engineering applicability of measurement point placement, it is necessary to propose a sensor placement method that balances the determination of the number of sensors with their optimal locations. Based on this, this application proposes a sensor placement optimization method for offshore wind turbine jacket structures. By comprehensively considering information coverage, marginal benefit decay, numerical stability, and robust feasibility, the method optimizes the determination of the number and placement of sensors on the jacket structure. Summary of the Invention

[0006] To address the problems of difficulty in reasonably determining the number of sensors, large redundancy of measurement point information, insufficient stability of arrangement results, and weak engineering applicability in existing sensor arrangement methods for offshore wind turbine jacket structures, the purpose of this invention is to propose an optimized sensor arrangement method for offshore wind turbine jacket structures, so as to optimize the determination of the number and arrangement position of sensors, improve the monitoring information acquisition capability, the stability of the arrangement scheme, and the applicability of engineering applications.

[0007] To achieve the above objectives, the technical solution adopted by the present invention is: a method for optimizing the arrangement of sensors on offshore wind turbine jackets, comprising the following steps: Step 1: Extract the target order mode shape information of the offshore wind turbine jacket structure and obtain the modal displacement components of each node in the x, y, and z directions. Combined with the sensor installation constraints of the wind turbine jacket, screen all nodes to construct a candidate node set where sensors can be placed. Then, extract and normalize the modal responses corresponding to the candidate nodes to form the working mode matrix. Step 2: Establish a sensor deployment evaluation model based on the working mode matrix. For any combination of sensor nodes, extract the corresponding rows from the working mode matrix to form a sub-matrix and establish a regularized information matrix. Use its logarithmic determinant as the information content evaluation index. By introducing a regularization term, reduce the numerical instability caused by the ill-conditioned information matrix under high correlation mode conditions. Step 3: Determine the scanning range by considering three indicators: comprehensive information coverage, marginal benefit decay, and numerical stability. Use a combination of coarse and fine scanning to evaluate the number of sensors for different numbers of sensors, and select the minimum number that meets the requirements as the recommended number of sensors. Step 4: Construct multiple initial measurement point layouts under a fixed number of recommendations. Use single-point exchange and local search to gradually replace the measurement points and optimize them independently. Select the layout scheme with high information content, good stability and strong repeatability from the multiple optimization results as the final result.

[0008] The above-mentioned method for optimizing the arrangement of sensors on offshore wind turbine jackets, step 1 includes: Step 1-1: Establish a finite element analysis model of the offshore wind turbine jacket structure, obtain the mode shape data of the first few target modes of the structure, extract the displacement response components of each node in the three directions of the global coordinate system for each target mode, and arrange them in order of mode order to form the original mode matrix. Unify the mode order and the optimization space dimension, denoted as: , d The dimension of the modal feature vector corresponding to each candidate node. The modal order; Steps 1-2: Based on installation accessibility, power and communication conditions, maintenance convenience, security, and local facility space constraints, all nodes are screened, and nodes that do not meet the installation conditions are eliminated, resulting in a set of candidate nodes that can be used to deploy sensors. ,in The set of arrangeable candidate nodes to satisfy engineering constraints. The number of candidate nodes. Size of the set; Steps 1-3: Normalize the modal responses corresponding to candidate nodes to reduce the impact of differences in dimensions of different directions and modes, and establish a working modal matrix. Each row of the working modal matrix corresponds to the modal feature vector of a candidate node, thus transforming the engineering measurement point selection problem into a mathematical matrix row selection optimization problem.

[0009] In the above-mentioned method for optimizing the arrangement of sensors on offshore wind turbine jackets, step 2 includes: Step 2-1: Using the D-opt information content based on the regularized Fisher information matrix as the core evaluation index for the deployment quality, for any given combination of sensor nodes, the corresponding rows are extracted from the working mode matrix to form a sub-matrix, and the information matrix is ​​established by combining it with the observation noise covariance matrix. ,in, This is the noise-aware regularized information matrix. The regularization coefficient is . To obtain from the full candidate set Extract The submatrix formed by the corresponding rows, To observe the noise covariance matrix, I d The identity matrix of d Step 2-2: Introduce a regularization term into the information matrix. By taking the logarithmic determinant of the regularized information matrix, obtain the information content index of the sensor combination. ,in, For sensor set The D-opt information content is used to uniformly evaluate different numbers of sensors and different combinations of measurement points.

[0010] The aforementioned method for optimizing the sensor placement of offshore wind turbine jacket structures includes, in step 3: constructing a joint quantitative decision-making mechanism for high-correlation modal scenarios of jacket structures, comprising three indicators: information coverage index, marginal benefit decay index, and numerical stability index. The information coverage index... This is used to determine whether the acquisition of target modal information at the current quantity has reached a sufficient level, where... For information percentage, For the first Step information content, Maximum number of scans, Baseline information content; The marginal revenue decay index This is used to determine whether there is still a significant benefit in continuing to add sensors. This is the normalized ratio of marginal gain. For the first The incremental information brought by each sensor For information content; The numerical stability index is used to determine whether there is an obvious ill-conditioning problem in the current information matrix, thereby avoiding the selection of a number of sensors that are theoretically high in information content but numerically unstable.

[0011] In the above-mentioned method for optimizing the arrangement of sensors on offshore wind turbine jackets, step 3 includes: Step 3-1: Construct a quantity scanning interval so that subsequent scans will not fall into areas with obviously insufficient information or waste computing resources in obviously unstable areas. At the same time, appropriate redundancy is given to the quantity interval near the diminishing returns to avoid missing situations where a slight increase in quantity can significantly improve robustness and feasibility. Step 3-2: After obtaining the scanning range, a two-level quantity determination mechanism combining coarse scanning and fine scanning is adopted. Based on the coarse scanning results, a set of candidate quantities with potential feasibility is selected. Fine scanning significantly increases the initial layout quantity, the number of random seeds, and the search intensity in the position optimization process, so that each candidate quantity can be evaluated under more sufficient search conditions.

[0012] In the above-mentioned method for optimizing the arrangement of sensors on offshore wind turbine jackets, in step 3-2, the satisfaction rate and tail risk are still statistically analyzed during the fine scanning stage. A strict robustness and feasibility definition is set. If there is more than one number that meets the strict robustness and feasibility condition, the smallest number is selected as the recommended number of sensors. Under the premise of meeting the information and stability requirements, the number of sensors and engineering costs are reduced, and alternative numbers with robustness priority or information quality priority are output for selection according to different engineering needs.

[0013] The aforementioned method for optimizing the arrangement of sensors on offshore wind turbine jackets, wherein the strictly robust and feasible behavior is as follows: under multiple initial conditions and multiple disturbance scenarios, the satisfaction rate is not lower than a preset threshold, and the tail risk is controlled; and the stability constraint can still be satisfied under more unfavorable scenarios.

[0014] In the aforementioned method for optimizing the sensor layout of offshore wind turbine jackets, step 4 involves introducing a disturbance mechanism to adjust the current layout when the search stalls, and then continuing the search. This is further enhanced by performing strong neighbor re-examination and double-point replacement to reduce the risk of local optima. Step 4 includes: Step 4-1: Divide the location optimization process into two levels: outer search and inner improvement. The outer search adopts a multi-starting point initialization strategy to construct several sets of initial sensor layouts with large differences. By running multiple random seeds independently, the search can be improved to cross different local optimum basins. Step 4-2: In each outer search trajectory, the inner improvement performs a local exchange search on the current layout. The local exchange search gradually replaces individual measurement points in the current layout. The goal is to find a new layout that can increase information content while satisfying stability constraints, where the current layout is... , The layout is defined by swapping a point, where i represents a point in the original layout and j represents the swapped point. To exchange the incremental information brought by the sensor; Step 4-3: When the local exchange search stalls, an adaptive perturbation mechanism is introduced to perform a removal-backfill perturbation on the current layout, actively changing the layout topology and jumping out of the current local optimum. When the search achieves significant improvement again, the perturbation intensity is reduced. Step 4-4: After the perturbation is completed, perform the local exchange search again to form a perturbation-convergence cyclical improvement mechanism in the search process.

[0015] In the above-mentioned method for optimizing the arrangement of sensors on offshore wind turbine jackets, step 4 further includes: Steps 4-5: After single-point swapping fails to improve the layout, a stronger neighborhood re-examination is performed on the current layout, and paired test points are replaced for verification. If a better layout is found during the re-examination, the current layout is updated and the perturbation-convergence loop is re-entered, where the current layout is... , The layout after swapping two points is shown, where i1 and i2 are points in the original layout, and j1 and j2 are the swapped points. The information increment brought by the exchange sensor; if no improvement is found in the re-examination, the layout is considered to have passed the strong neighborhood test and can be used as the candidate optimal solution corresponding to the current random seed; Steps 4-6: By comparing the improvement of the search results of the new batch with the current best result, examine whether the fluctuation of the best value obtained under different random seeds tends to converge, determine the marginal benefit of continuing the search, and stop the position optimization process when several consecutive batches of search have not achieved substantial improvement and the fluctuation of the optimal value distribution tends to stabilize, and output the final optimal layout.

[0016] The beneficial effects of the proposed method for optimizing the arrangement of sensors on offshore wind turbine jackets are as follows: This invention does not first determine the number of sensors independently and then optimize their placement separately. Instead, it conducts a feasibility assessment for each candidate number in conjunction with the corresponding position optimization results, and then further optimizes the sensor placement under the constraint of the recommended number. This couples the determination of the number of sensors with the optimization of the placement, achieving a collaborative optimization determination of the number of sensors and their placement.

[0017] This invention is based on working mode matrix modeling, which organizes the modal response information of candidate nodes, so that the selection of measurement points is directly oriented towards the target modal information. It adopts the D-opt information content criterion of regularized information matrix, and maximizes the logarithmic determinant to make the selected measurement points provide more sufficient and balanced information coverage of the target mode. When selecting candidate nodes, engineering constraints are considered and modal data is normalized to reduce the influence of invalid and redundant points, which can improve the effectiveness of jacket structure monitoring information and modal identification capability.

[0018] This invention introduces a regularization term when constructing the information matrix to avoid the information matrix approaching singularity under highly correlated modes, thereby reducing ill-conditionedness and improving numerical stability. In the quantity determination stage, stability indicators are incorporated into the quantity criteria, not only pursuing a large amount of information, but also ensuring that the information matrix has sufficient stability in the weakest direction. By using coarse-scan and fine-scan statistics to measure the satisfaction rate and tail risk under different quantities, the feasibility of the scheme under disturbances and adverse conditions is examined. This is actually enhancing noise resistance and robustness, which can improve the numerical stability and noise resistance performance of the layout scheme.

[0019] This invention employs multi-starting-point independent search to avoid locking the result to a local optimum after a single initialization; it uses single-point exchange local search to gradually improve the current layout; when the search stalls, it introduces an adaptive perturbation mechanism to jump out of the current local optimum region through "removal-backfilling"; and it uses strong neighborhood re-checking to check whether there are better alternatives, avoiding situations where the solution appears to converge but a better one has not yet been found, thus reducing the risk of local optima and improving the robustness and reproducibility of the layout results.

[0020] It is more suitable for engineering applications of offshore wind turbine jacket structures, and can reduce engineering costs while ensuring monitoring effectiveness. Attached Figure Description

[0021] Figure 1 This is a flowchart illustrating the optimization of the number and arrangement of sensors in an embodiment of the present invention; Figure 2 This is a flowchart illustrating the inner layer position optimization in an embodiment of the present invention. Detailed Implementation

[0022] To enable those skilled in the art to better understand the technical solution of the present invention, the technical solution of the present invention will be described below in conjunction with specific embodiments and accompanying drawings.

[0023] Example 1 A method for optimizing the arrangement of sensors on offshore wind turbine jackets is provided, applicable to the joint optimization and determination of the number and placement of sensors in a jacket structure health monitoring system. Addressing the characteristics of jacket structures, such as significant modal coupling, a large number of candidate nodes, high similarity of measurement point responses, and strong disturbances in the marine environment, this invention divides the sensor arrangement process into two stages: number determination and placement optimization, to improve the information acquisition capability, numerical stability, and engineering applicability of the arrangement scheme.

[0024] First, the target order modal shape information of the offshore wind turbine jacket structure is extracted, and the modal displacement components of each node in the x, y, and z directions are obtained. Considering installation accessibility, maintenance conditions, and engineering constraints, all nodes are screened to construct a candidate node set where sensors can be deployed. Then, the modal responses corresponding to the candidate nodes are extracted and normalized to form the working mode matrix.

[0025] Secondly, a sensor deployment evaluation model is established based on the operating mode matrix. For any combination of sensor nodes, corresponding rows are extracted from the operating mode matrix to form a submatrix, and a regularized information matrix is ​​established, with its logarithmic determinant used as the information content evaluation index. By introducing a regularization term, the numerical instability caused by the ill-conditioned information matrix under high-correlation mode conditions can be reduced.

[0026] In the sensor number determination stage, the scanning interval is determined by three indicators: comprehensive information coverage, marginal benefit decay, and numerical stability. A combination of coarse and fine scanning is used to evaluate the number of sensors for different numbers. The coarse scanning is used to quickly screen potentially feasible numbers, while the fine scanning is used to determine whether each candidate number meets the robustness and feasibility requirements under more comprehensive search conditions. Finally, the minimum number that meets the requirements is selected as the recommended number of sensors.

[0027] In the location optimization phase, multiple initial measurement point layouts are constructed under a fixed number of recommendations, and each is optimized independently. During optimization, a single-point exchange local search method is used to gradually replace measurement points. When the search stalls, a perturbation mechanism is introduced to adjust the current layout before continuing the search. If necessary, strong neighborhood re-examination is performed to conduct double-point replacement and search again, reducing the risk of local optima. Finally, the layout scheme with higher information content, better stability, and stronger repeatability is selected from the multiple optimization results as the final result.

[0028] This technical solution ultimately outputs a recommended set of sensor quantities and corresponding placement locations, thereby achieving coordinated optimization of the number and location of sensors on the offshore wind turbine jacket structure. Compared with existing methods, the technical solution in this embodiment can improve the stability, robustness, and engineering application value of the placement scheme while ensuring effective acquisition of monitoring information.

[0029] The technical problems solved by the technical solution provided in this embodiment also include: how to reduce the number of unnecessary sensors and reduce information redundancy and engineering costs while ensuring sufficient acquisition of monitoring information; how to improve the ability of the sensor layout scheme to identify the target mode when the low-order mode of the jacket structure is obviously coupled in three directions and the candidate node response similarity is high; and how to improve the stability, feasibility and reproducibility of the sensor layout results under the conditions of considering noise disturbance, numerical ill-conditioning and engineering constraints.

[0030] Example 2 Specifically, this embodiment will provide a detailed description of the technical solution based on the technology of embodiment 1.

[0031] First, a finite element analysis model of the offshore wind turbine jacket structure is established to obtain the mode shape data of the first few target modes. For each target mode, the displacement response components of each node in the three directions of the global coordinate system are extracted and arranged sequentially according to the mode order to form the original mode matrix. The mode order and the optimization space dimension are unified and denoted as: , d The dimension of the modal feature vector corresponding to each candidate node. Let be the modal order, then each node can be mapped to a d-dimensional modal feature vector. Considering that not all nodes are suitable for sensor placement in practical engineering, further screening is performed on all nodes based on engineering constraints such as installation accessibility, power and communication conditions, maintenance convenience, security, and local facility space occupation. Nodes that do not meet the installation conditions are eliminated, resulting in a candidate node set that can be used for sensor placement. S is the set of deployable candidate nodes that satisfy the engineering constraints, and NC is the number of candidate nodes.

[0032] Based on this, the modal responses corresponding to candidate nodes are normalized to reduce the impact of differences in dimensions of different directions and modes, ultimately establishing a working modal matrix. Each row of this working modal matrix corresponds to the modal eigenvector of a candidate node, thus transforming the engineering problem of measuring point selection into a mathematical problem of matrix row selection optimization.

[0033] Regarding the evaluation criteria, this embodiment uses the D-opt information content based on the regularized Fisher information matrix as the core evaluation index for the deployment quality. For any given combination of sensor nodes, the corresponding rows are extracted from the working mode matrix to form a sub-matrix, and an information matrix is ​​established by combining it with the observation noise covariance matrix. ,in, For noise-sensing regularized information matrix; The regularization coefficient is . To obtain from the full candidate set Extract The submatrix formed by the corresponding rows, To observe the noise covariance matrix, I d Let be the identity matrix of size d. To avoid problems such as near-singularity of the matrix, difficulty in calculating the logarithmic determinant, and noise amplification when there is high modal correlation or a small number of sensors, a regularization term is introduced into the information matrix to improve numerical stability. By calculating the logarithmic determinant of the regularized information matrix, the information content index of the sensor combination is obtained. ,in, For sensor set The D-opt information content. The larger this index, the more comprehensive the coverage of the target modal space and the more balanced the information distribution of the selected sensor combination. Based on this index, different numbers of sensors and different combinations of measurement points can be uniformly evaluated.

[0034] In the sensor quantity determination phase, this embodiment does not directly determine the optimal quantity based on the inflection point of a single information curve. Instead, it constructs a three-indicator joint quantity decision-making mechanism for the high-correlation modal scenario of the jacket structure. The three indicators include information coverage indicators. Marginal revenue decay index And numerical stability index, Information percentage; For the first Step information content; Maximum number of scans, For baseline information content, This is the normalized ratio of marginal gain. For the first The incremental information brought by each sensor The information content is defined as follows: Information Coverage Index is used to determine whether the acquisition of target modal information at the current number of sensors has reached a sufficient level; Marginal Return Decay Index is used to determine whether there is still a significant benefit in continuing to add sensors; Numerical Stability Index is used to determine whether there are obvious ill-conditioned problems in the current information matrix, thereby avoiding the selection of a number of sensors that are theoretically high in information content but numerically unstable.

[0035] Based on the above three types of indicators, we first construct a quantity scanning interval so that subsequent scans will not fall into areas with obviously insufficient information or waste computing resources in obviously unstable areas. At the same time, we also give appropriate redundancy to the quantity interval near the diminishing returns to avoid missing situations where a slight increase in quantity could significantly improve robustness and feasibility.

[0036] After obtaining the scanning interval, this embodiment employs a two-stage quantity determination mechanism combining coarse and fine scanning. In the coarse scanning stage, each candidate quantity within the scanning interval is solved independently multiple times. Each solution uses a different initial layout and a weaker inner-layer search intensity to quickly estimate the robustness and feasibility of the layout scheme under that quantity with low computational cost. The coarse scanning mainly collects two types of information: first, the proportion that satisfies stability constraints, which characterizes the probability that the quantity will meet the stability requirements in multiple independent solutions; and second, the tail risk level under adverse perturbation scenarios, which identifies quantities that, while performing reasonably well on average, may become significantly unstable in a few scenarios. Based on the coarse scanning results, a set of candidate quantities with potential feasibility is selected.

[0037] Subsequently, this embodiment performs a fine scan of the aforementioned candidate quantity set. Compared to a coarse scan, a fine scan significantly increases the number of initial layouts, the number of random seeds, and the search intensity during the position optimization process, allowing each candidate quantity to be evaluated under more comprehensive search conditions. The fine scan stage still calculates the satisfaction rate and tail risk, defining "strictly robust feasibility" as follows: under multiple initial values ​​and multiple perturbation scenarios, the satisfaction rate is not lower than a preset threshold, and the tail risk is controlled, meaning that stability constraints are still met even under unfavorable scenarios. If more than one quantity meets the strict robust feasibility condition, the smallest quantity is selected as the recommended number of sensors to reduce the number of sensors and engineering costs while meeting information and stability requirements; alternative quantities prioritizing robustness or information quality can also be output for selection based on different engineering needs.

[0038] After determining the recommended number of sensors, this embodiment enters the location optimization stage. Since candidate nodes in the jacket structure often exhibit strong modal similarity, different combinations of measuring points may show little difference in information indicators, but significant differences in spatial distribution rationality, directional coverage, and sensitivity to disturbances. Therefore, location optimization with a fixed number of sensors needs to balance the quality of the solution and the reproducibility of the results. To this end, this embodiment divides the location optimization process into two levels: outer search and inner improvement. The outer search employs a multi-starting-point initialization strategy, constructing several sets of initially different sensor layouts. The initial layout can come from high-quality starting points quickly generated by heuristic rules, or from diverse starting points after randomization and feasible repair. By independently running multiple random seeds, the ability to search across different local optimum basins is improved.

[0039] In each outer search trajectory, the inner improvement first performs a local swap search on the current layout. The local swap search initially employs a single-point neighborhood swap (1-swap) approach, progressively replacing individual measurement points in the current layout. Specifically, this local swap search progressively replaces individual measurement points in the current layout. The aim is to find new layouts that can increase information content and satisfy stability constraints, thereby converging quickly to locally optimal solutions with lower computational cost. The current layout is... , The layout is defined by swapping a point, where i represents a point in the original layout and j represents the swapped point. To exchange the incremental information brought by the sensor.

[0040] When the local exchange search stalls, this embodiment further introduces an adaptive perturbation mechanism, performing a "removal-backfilling" perturbation on the current layout to proactively change the layout topology and escape the current local optimum. The perturbation intensity can be adjusted according to the degree of search stalling; the more obvious the stall, the larger the perturbation range. When the search achieves significant improvement again, the perturbation intensity is reduced to avoid invalid walks. After the perturbation is completed, the local exchange search is performed again, forming a "perturbation-convergence" cyclical improvement mechanism in the search process.

[0041] Considering that in scenarios with high relevance candidate points, stagnation in single-point replacement does not necessarily mean that the layout has reached a truly stable local optimum, this embodiment also includes a strong neighborhood review step. This step performs a stronger neighborhood review of the current layout after single-point swapping fails to improve it further, using a two-point neighborhood swapping method (2-swap) to perform replacement checks on paired test points. For example, performing replacement checks on paired test points... This excludes situations where "a single point cannot be improved, but two points can still be significantly improved." The current layout is... , The layout after swapping two points is shown, where i1 and i2 are points in the original layout, and j1 and j2 are the swapped points. To exchange the incremental information from the sensors. If a better layout is found during the re-examination, the current layout is updated and the perturbation-convergence loop is re-entered; if no improvement is found during the re-examination, the layout is considered to have passed the strong neighborhood test and can be used as a candidate optimal solution corresponding to the current random seed.

[0042] Regarding the stopping criteria, this embodiment does not employ a simple fixed number of iterations, but rather uses evidence-based stopping based on the statistical characteristics of results from multiple starting points. Specifically, by comparing the improvement of the search results of newly added batches relative to the current optimal result, and examining whether the fluctuations of the optimal values ​​obtained under different random seeds tend to converge, it is determined whether the marginal benefit of continuing the search is sufficiently small. When several consecutive batches of searches have not yielded substantial improvement, and the fluctuations in the distribution of optimal values ​​tend to stabilize, the position optimization process is stopped, and the final optimal layout is output. The final output not only includes the information-optimal layout with a fixed number of iterations, but also evidence of convergence consistency of the layout under conditions of multi-starting point search and strong neighborhood verification, thereby providing support for the robustness and reproducibility of the layout results.

[0043] The above embodiments are merely illustrative of the structural concept and features of the present invention, intended to enable those skilled in the art to understand the content of the present invention and implement it accordingly, and should not be construed as limiting the scope of protection of the present invention. All equivalent changes or modifications made based on the essence of the present invention should be covered within the scope of protection of the present invention.

Claims

1. A method for optimizing the arrangement of sensors on offshore wind turbine jackets, characterized in that, Includes the following steps: Step 1: Extract the target order mode shape information of the offshore wind turbine jacket structure, and obtain the modal displacement components of each node in the x, y, and z directions. Combined with the sensor installation constraints of the wind turbine jacket, screen all nodes to construct a candidate node set where sensors can be placed. Then, extract and normalize the modal responses corresponding to the candidate nodes to form the working mode matrix. Step 2: Establish a sensor deployment evaluation model based on the working mode matrix. For any combination of sensor nodes, extract the corresponding rows from the working mode matrix to form a sub-matrix and establish a regularized information matrix. Use its logarithmic determinant as the information content evaluation index. By introducing a regularization term, reduce the numerical instability caused by the ill-conditioned information matrix under high correlation mode conditions. Step 3: Determine the scanning range by considering three indicators: comprehensive information coverage, marginal benefit decay, and numerical stability. Use a combination of coarse and fine scanning to evaluate the number of sensors for different numbers of sensors, and select the minimum number that meets the requirements as the recommended number of sensors. Step 4: Construct multiple initial measurement point layouts under a fixed number of recommendations. Use single-point exchange and local search to gradually replace the measurement points and optimize them independently. Select the layout scheme with high information content, good stability and strong repeatability from the multiple optimization results as the final result.

2. The method for optimizing the arrangement of sensors on offshore wind turbine jackets according to claim 1, characterized in that, Step 1 includes: Step 1-1: Establish a finite element analysis model of the offshore wind turbine jacket structure, obtain the mode shape data of the first few target modes of the structure, extract the displacement response components of each node in the three directions of the global coordinate system for each target mode, and arrange them in order of mode order to form the original mode matrix. Unify the mode order and the optimization space dimension, denoted as: , d The dimension of the modal feature vector corresponding to each candidate node. The modal order; Steps 1-2: Based on installation accessibility, power and communication conditions, maintenance convenience, security, and local facility space constraints, all nodes are screened, and nodes that do not meet the installation conditions are eliminated, resulting in a set of candidate nodes that can be used to deploy sensors. ,in The set of arrangeable candidate nodes to satisfy engineering constraints. The number of candidate nodes. Size of the set; Steps 1-3: Normalize the modal responses corresponding to candidate nodes to reduce the impact of differences in dimensions of different directions and modes, and establish a working modal matrix. Each row of the working modal matrix corresponds to the modal feature vector of a candidate node, thus transforming the engineering measurement point selection problem into a mathematical matrix row selection optimization problem.

3. The method for optimizing the arrangement of sensors on offshore wind turbine jackets according to claim 2, characterized in that, Step 2 includes: Step 2-1: Using the D-opt information content based on the regularized Fisher information matrix as the core evaluation index for the deployment quality, for any given combination of sensor nodes, the corresponding rows are extracted from the working mode matrix to form a sub-matrix, and the information matrix is ​​established by combining it with the observation noise covariance matrix. ,in, This is the noise-aware regularized information matrix. The regularization coefficient is . To obtain from the full candidate set Extract The submatrix formed by the corresponding rows, To observe the noise covariance matrix, I d The identity matrix of d Step 2-2: Introduce a regularization term into the information matrix. By taking the logarithmic determinant of the regularized information matrix, obtain the information content index of the sensor combination. ,in, For sensor set The D-opt information content is used to uniformly evaluate different numbers of sensors and different combinations of measurement points.

4. The method for optimizing the arrangement of sensors on offshore wind turbine jackets according to claim 3, characterized in that, In step 3: a joint quantitative decision-making mechanism is constructed for the high-relevance modal scenario of jacket stents, including an information coverage index, a marginal benefit decay index, and a numerical stability index. The information coverage index... This is used to determine whether the acquisition of target modal information at the current quantity has reached a sufficient level, where... For information percentage, For the first Step information content, Maximum number of scans, Baseline information content; The marginal revenue decay index This is used to determine whether there is still a significant benefit in continuing to add sensors. This is the normalized ratio of marginal gain. For the first The incremental information brought by each sensor For information content; The numerical stability index is used to determine whether there is an obvious ill-conditioning problem in the current information matrix, thereby avoiding the selection of a number of sensors that are theoretically high in information content but numerically unstable.

5. The method for optimizing the arrangement of sensors on offshore wind turbine jackets according to claim 4, characterized in that, Step 3 includes: Step 3-1: Construct a quantity scanning interval so that subsequent scans will not fall into areas with obviously insufficient information or waste computing resources in obviously unstable areas. At the same time, appropriate redundancy is given to the quantity interval near the diminishing returns to avoid missing situations where a slight increase in quantity can significantly improve robustness and feasibility. Step 3-2: After obtaining the scanning range, a two-level quantity determination mechanism combining coarse scanning and fine scanning is adopted. Based on the coarse scanning results, a set of candidate quantities with potential feasibility is selected. Fine scanning significantly increases the initial layout quantity, the number of random seeds, and the search intensity in the position optimization process, so that each candidate quantity can be evaluated under more sufficient search conditions.

6. The method for optimizing the arrangement of sensors on offshore wind turbine jackets according to claim 5, characterized in that, In step 3-2, the fine scanning stage still calculates the satisfaction rate and tail risk, sets a strict robustness and feasibility definition, and if there is more than one number that meets the strict robustness and feasibility condition, the smallest number is selected as the recommended number of sensors. Under the premise of meeting the information and stability requirements, the number of sensors and engineering costs are reduced, and alternative numbers with robustness priority or information quality priority are output for different engineering needs to choose from.

7. The method for optimizing the arrangement of sensors on offshore wind turbine jackets according to claim 6, characterized in that, The strictly robust and feasible behavior is as follows: under multiple initial values ​​and multiple perturbation scenarios, the satisfaction rate is not lower than the preset threshold, the tail risk is controlled, and the stability constraint can still be satisfied under more unfavorable scenarios.

8. The method for optimizing the arrangement of sensors on offshore wind turbine jackets according to claim 5, characterized in that, In step 4, when the search stalls, a perturbation mechanism is introduced to adjust the current layout before continuing the search. Further strong neighbor re-examination and double-point replacement are performed to reduce the risk of local optima. Step 4 includes: Step 4-1: Divide the location optimization process into two levels: outer search and inner improvement. The outer search adopts a multi-starting point initialization strategy to construct several sets of initial sensor layouts with large differences. By running multiple random seeds independently, the search can be improved to cross different local optimum basins. Step 4-2: In each outer search trajectory, the inner improvement performs a local exchange search on the current layout. The local exchange search gradually replaces individual measurement points in the current layout. The goal is to find a new layout that can increase information content while satisfying stability constraints, where the current layout is... , The layout is defined by swapping a point, where i represents a point in the original layout and j represents the swapped point. To exchange the incremental information brought by the sensor; Step 4-3: When the local exchange search stalls, an adaptive perturbation mechanism is introduced to perform a removal-backfill perturbation on the current layout, actively changing the layout topology and jumping out of the current local optimum. When the search achieves significant improvement again, the perturbation intensity is reduced. Step 4-4: After the perturbation is completed, perform the local exchange search again to form a perturbation-convergence cyclical improvement mechanism in the search process.

9. The method for optimizing the arrangement of sensors on offshore wind turbine jackets according to claim 8, characterized in that, Step 4 also includes: Steps 4-5: After single-point swapping fails to improve the layout, a stronger neighborhood re-examination is performed on the current layout, and paired test points are replaced for verification. If a better layout is found during the re-examination, the current layout is updated and the perturbation-convergence loop is re-entered, where the current layout is... , The layout after swapping two points is shown, where i1 and i2 are points in the original layout, and j1 and j2 are the swapped points. The information increment brought by the exchange sensor; if no improvement is found in the re-examination, the layout is considered to have passed the strong neighborhood test and can be used as the candidate optimal solution corresponding to the current random seed; Steps 4-6: By comparing the improvement of the search results of the new batch with the current best result, examine whether the fluctuation of the best value obtained under different random seeds tends to converge, determine the marginal benefit of continuing the search, and stop the position optimization process when several consecutive batches of search have not achieved substantial improvement and the fluctuation of the optimal value distribution tends to stabilize, and output the final optimal layout.

Images