Scouring detection method based on proxy model

Abstract

The invention relates to a scouring detection method based on an agent model. The method comprises the steps of obtaining to-be-corrected target response data of an offshore wind power pile foundation; on the basis of the obtained to-be-corrected target response data, constructing a finite element numerical model consistent with the physical model test size, and correcting the finite element numerical model by adopting an improved PSO algorithm to obtain a corrected finite element numerical model; wherein a plurality of particles with good fitness are selected as initial particles by adopting a Latin hypercube sampling method, and optimization of a PSO (Particle Swarm Optimization) algorithm is carried out after a search space is reduced; basic response data under different scour depths are calculated through the corrected finite element numerical model; by taking the basic response data as an input training set and the scouring depth as an output training set, constructing a BP neural network and performing training to obtain an agent model capable of being used for detecting the scouring depth, and taking the test data of each working condition as a test set to invert the scouring depth; the agent model constructed after model correction can detect the scouring of the pile foundation in real time.

Description

Technical Field

[0001] The present invention relates to a scour detection method based on a surrogate model, belonging to the technical field of engineering detection. Background Art

[0002] In recent years, the offshore wind power market has developed rapidly, and fixed pile foundations mainly composed of single piles and multi-pile foundations are the main foundation forms. Scour is an important factor causing the reduction of the durability of wind turbines and the instability and failure of foundations. The existing scour detection methods mainly rely on multi-beam and sonar surveys.

[0003] For example, the invention with the application number 202211000642.2 discloses a method and system for detecting scour of offshore wind power pile foundations based on an underwater robot. After determining the area to be detected, multiple scanning areas are determined, including an outer ring area and an inner ring area. First, the outer ring area is established. When multiple outer ring areas cannot cover all the detection areas, the inner ring area is set to ensure that the entire area to be detected can be covered. Then, the robot is deployed. After the robot scans each scanning area through the mechanical scanning sonar module, a scour image is finally obtained. The invention with the application number 202410665927.0 proposes an intelligent perception monitoring system for offshore wind turbines, including: an application layer, a real-time operating system, a communication connection layer, a logic control layer, and an environmental hardware layer. Among them, the communication connection layer communicates with the sensors and PLC controllers in the logic control layer; the sensors are used to obtain the measurement data of the components of the wind turbine in the environmental hardware layer, and the PLC controller is used to receive control signals and transmit them to the components of the wind turbine; the real-time operating system realizes the interface between the operating system kernel and the hardware based on the hardware abstraction layer, and the following one or more are deployed in the application layer: C++ / Rust / Python compatible libraries, real-time databases, graphics processing frameworks, SCADA systems, structural safety monitoring systems, fault diagnosis and analysis systems, and real-time control systems for wind turbines. Here, the sensors are used to detect the operation data, meteorological data, seabed environment data, and marine environment data of offshore wind turbines, and the structural safety equipment jointly uses a multi-beam sounding system and a side-scan sonar for scour detection and continuous detection of key parameters.

[0004] The multi-beam and sonar surveys provided by the above-mentioned existing technologies cannot achieve real-time detection. Therefore, a means is needed in engineering to detect the progress of pile foundation scour in real time. Summary of the Invention

[0005] The present invention provides a scour detection method based on a surrogate model. The surrogate model constructed after model correction can detect the scour of pile foundations in real time.

[0006] The technical solution adopted by the present invention to solve its technical problems is:

[0007] A scour detection method based on a surrogate model specifically includes the following steps:

[0008] Step S1, obtaining the target response data to be corrected of the offshore wind turbine pile foundation;

[0009] Step S2, based on the obtained target response data to be corrected, constructing a finite element numerical model with the same size as the physical model test, and using an improved PSO algorithm to correct the finite element numerical model to obtain a corrected finite element numerical model;

[0010] Among them, the Latin hypercube sampling method is used to select several particles with good fitness as the initial particles, and at the same time, after shrinking the search space, the optimization of the PSO algorithm is carried out;

[0011] Step S3, calculating the foundation response data at different scour depths through the corrected finite element numerical model;

[0012] Step S4, using the foundation response data obtained in Step S3 as the input training set and the scour depth as the output training set, constructing a BP neural network and training it to obtain a surrogate model that can be used to detect the scour depth, and using the test data of each working condition as the test set to invert the scour depth;

[0013] Further, the target response data to be corrected includes pile body moment, pile top displacement, jacket node stress, tower top displacement, tower top acceleration, and tower top rotation angle;

[0014] Further, the specific steps for obtaining the corrected finite element numerical model in Step S2 are as follows

[0015] Step S21, selecting and calculating half of the symmetric numerical model in the constructed finite element numerical model, and selecting the time history response data at the same position as the physical model under cyclic loading for the parameters to be corrected;

[0016] Step S22, calculating the normalized root mean square error between the finite element simulation result and the actual measurement of the physical model;

[0017] Step S23, continuing to normalize the root mean square error;

[0018] Step S24, evaluating the similarity of the spatial distribution of the target response to be corrected through the Pearson correlation coefficient;

[0019] Step S25, obtaining the objective function of the improved PSO algorithm based on the Pearson correlation coefficient;

[0020] Further, in Step S22, the calculation formula for the normalized root mean square error between the finite element simulation result and the actual measurement is:

[0021]

[0022] In formula (1), RMSE is the root mean square error, n is the number of responses of the measurement points where the target to be corrected is located, and y i is the response data of the measurement points obtained by numerical simulation, and x i is the response data of the measurement points measured in the physical model;

[0023] In step S23, the calculation formula for continuous normalization processing is:

[0024]

[0025] In formula (2), RMSE′ is the normalized root mean square error, is the average value of the response data of the measurement points measured in the physical model, is the average value of the response data of the measurement points obtained by numerical simulation;

[0026] In step S24, the calculation formula for Pearson correlation coefficient evaluation is:

[0027]

[0028] In formula (3), r is the Pearson correlation coefficient;

[0029] In step S25, the objective function of the improved PSO algorithm obtained is:

[0030]

[0031] In formula (4), f min (x) is the fitness function, α is the weight of the normalized root mean square error, β is the weight of the Pearson coefficient, and α + β = 1;

[0032] Furthermore, in step S2, before conducting the optimization of the PSO algorithm, several initial particle positions are randomly selected through random sampling, the fitness functions of several initial particles are calculated, and based on the fitness functions, several initial particle positions with good fitness are selected from several particles to narrow the search space, and PSO optimization is performed within the narrowed search space;

[0033] Furthermore, when correcting the finite element numerical model, the number of particles in each iteration is at least 20;

[0034] Furthermore, in step S4, the basic response data used as the training set is obtained based on sensitivity analysis, and the specific steps are as follows:

[0035] Step S41: Construct a surrogate model with the scour depth as the input sample set and the pile body moment, pile top displacement, jacket node stress, tower top displacement, tower top acceleration, and tower top rotation angle as the output sample set.

[0036] Step S42: Adopt the Latin hypercube sampling method to randomly select 1000 groups of scour depths and substitute them into the surrogate model to obtain 1000 groups of input-output pairs.

[0037] Step S43: Calculate the sensitivity of the input-output pairs and select an appropriate input training set.

[0038] Through the above technical solutions, compared with the prior art, the present invention has the following beneficial effects:

[0039] 1. The scour detection method based on the surrogate model provided by the present invention can detect the scour situation of the pile foundation in real time after pre-data collection, model correction, and surrogate model construction.

[0040] 2. The scour detection method based on the surrogate model provided by the present invention provides a reference basis for the health detection of the offshore wind turbine foundation, reduces the operation and maintenance costs, and has high detection accuracy at the same time. BRIEF DESCRIPTION OF THE DRAWINGS

[0041] The present invention will be further described below with reference to the drawings and embodiments.

[0042] Figure 1 is a schematic diagram of the reduced search space provided by the present invention;

[0043] Figure 2 is a schematic diagram of the sample space obtained by sampling in the three-dimensional sampling space through the Latin hypercube sampling method provided by the present invention;

[0044] Figure 3 is a schematic diagram of evaluating the quality of the initial particle position by calculating the fitness provided by the present invention;

[0045] Figure 4 is a graph of the fitness change of the PSO algorithm in two groups of model correction tests provided by the present invention;

[0046] Figure 5 is a schematic diagram of the pile body moment sensitivity provided by the present invention;

[0047] Figure 6 is a schematic diagram of the scour detection result with the moment as the input provided by the present invention;

[0048] Figure 7 is a schematic diagram of the scour detection result with the displacement as the input provided by the present invention;

[0049] Figure 8 is a physical model diagram provided by the present invention;

[0050] Figure 9 It is a schematic diagram of the finite element numerical model provided by the present invention;

[0051] Figure 10 It is a scour simulation diagram of the test carried out through the physical model provided by the present invention. Specific embodiments

[0052] Now, the present invention will be further described in detail with reference to the accompanying drawings. In the description of the present application, it should be understood that the orientation or positional relationship indicated by terms such as "left side", "right side", "upper part", "lower part", etc. is based on the orientation or positional relationship shown in the accompanying drawings, and is only for the convenience of describing the present invention and simplifying the description, rather than indicating or implying that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation. "First", "second", etc. do not represent the importance of components, so they cannot be understood as limiting the present invention. The specific dimensions adopted in this embodiment are only for illustrating the technical solution by way of example, and do not limit the protection scope of the present invention.

[0053] As described in the background art, currently, the methods for detecting the progress of pile foundation scour mainly rely on multi-beam and sonar surveys. These detection methods have two main defects. One is the relatively high detection cost, and the other is the inability to achieve real-time detection.

[0054] To solve the above problems, the present application provides a scour detection method based on a surrogate model. Its main technical means is to collect the response data during the operation of offshore wind turbines. After model correction and surrogate model construction, real-time pile foundation scour detection can be achieved. The specific steps are as follows:

[0055] Step S1, obtaining the target response data to be corrected of the offshore wind power pile foundation; the target response data to be corrected includes pile body moment, pile top displacement, jacket node stress, tower top displacement, tower top acceleration, and tower top rotation angle;

[0056] Step S2, based on the obtained target response data to be corrected, constructing a finite element numerical model with the same size as the physical model test, and using an improved PSO algorithm to correct the finite element numerical model to obtain a corrected finite element numerical model;

[0057] Among them, the Latin hypercube sampling method is used to select several particles with good fitness as the initial particles, and at the same time, after shrinking the search space, the optimization of the PSO algorithm (particle swarm optimization algorithm) is carried out;

[0058] Step S3, calculating the foundation response data at different scour depths through the corrected finite element numerical model;

[0059] Step S4: Using the basic response data obtained in Step S3 as the input training set and the scouring depth as the output training set, construct a BP neural network and train it to obtain a surrogate model that can be used to detect the scouring depth, and use the test data of each working condition as the test set to invert the scouring depth.

[0060] In the above steps, as the biggest innovation point of this application, an improved PSO algorithm is provided to correct the finite element numerical model, and the calculation amount is reduced by pre - narrowing the search range to improve the calculation accuracy.

[0061] The specific steps for obtaining the corrected finite element numerical model are as follows:

[0062] Step S21: Select half of the symmetric numerical model in the constructed finite element numerical model for calculation, and select the time - history response data at the same position as the physical model under cyclic loading for the parameters to be corrected.

[0063] Step S22: Calculate the root - mean - square error (RMSE) of the normalization between the finite - element simulation results and the actual measurement of the physical model. The calculation formula for the root - mean - square error of the normalization between the finite - element simulation results and the actual measurement is:

[0064]

[0065] In formula (1), RMSE is the root - mean - square error, n is the number of responses at the measurement points where the target to be corrected is located, y i is the response data of the measurement points obtained from numerical simulation, and x i is the response data of the measurement points measured in the physical model.

[0066] Step S23: Continue to normalize the root - mean - square error. The calculation formula for the continued normalization process is:

[0067]

[0068] In formula (2), RMSE′ is the normalized root - mean - square error, is the average value of the response data of the measurement points measured in the physical model, is the average value of the response data of the measurement points obtained from numerical simulation.

[0069] Step S24: Evaluate the similarity of the spatial distribution of the response of the target to be corrected through the Pearson correlation coefficient. The calculation formula for the evaluation using the Pearson correlation coefficient is:

[0070]

[0071] In formula (3), r is the Pearson correlation coefficient;

[0072] Step S25, obtain the objective function of the improved PSO algorithm based on the Pearson correlation coefficient, and the obtained objective function of the improved PSO algorithm is:

[0073]

[0074] In formula (4), f min (x) is the fitness function, α is the weight of the normalized root mean square error, β is the weight of the Pearson coefficient, and α + β = 1.

[0075] After obtaining the objective function of the improved PSO algorithm through the above means, optimization operations need to be performed through it. Before optimization in this application, the search space is first reduced by the method of random sampling pre-experiment. The specific method is to randomly select several initial particle positions through random sampling, calculate the fitness function of the particle, select several initial particle positions with better fitness from these particles, reduce the search space while including as many of these better initial particle positions as possible, or the search space can also be divided into several small search sub-spaces, and PSO optimization is performed on each search sub-space.

[0076] A schematic diagram of the reduction of the search space is given here ( Figure 1 as shown). It should be noted that even though the position with low fitness of the sample particles may find the global optimal solution nearby, the probability of this situation occurring in a search space with a relatively uniform fitness distribution is low. To improve efficiency, the search range can be reduced by this method first, and if the expected result cannot be achieved, the search range can be appropriately enlarged until the structure meets the expectation.

[0077] Regarding the method adopted for random sampling, the currently most common random sampling method is the Monte Carlo sampling method. This method starts from the perspective of probability distribution, making the samples randomly fall at any position in the sampling space, which may lead to too high sampling density in some areas and too low sampling density in other areas. In the engineering field, the probability of some extreme situations occurring is low, but these situations are often the key concerns, and the Monte Carlo sampling method may not be able to obtain samples in these areas. Preferably, this application adopts the Latin hypercube sampling method, and its process can be summarized as layering, sampling, and scrambling. First, the sampling space is layered, randomly sampled between each layer, and then the order of the obtained samples is scrambled. In this way, it can be ensured that there is at least one sample in all areas, so that the sampling is more uniform and has stronger coverage.

[0078] When this application conducts experimental verification, the elastic modulus E of the physical model pile is selected p , the elastic modulus E of the soil body s , and the internal friction angle φ of the soil body are used for model optimization. The initial search intervals of the three parameters are respectively: Ep ∈ (1 - 40) GPa, E s ∈ (5 - 60) MPa, φ ∈ (10 - 45)°. Through the Latin hypercube sampling method, 100 samplings are carried out in the three-dimensional sampling space, and the obtained sample space is as Figure 2 shown. The parameter values of the 100 groups of sample particles obtained by sampling are input into the finite element numerical model to calculate the fitness and evaluate the quality of the initial particle positions. The results are as Figure 3 shown. It can be seen from Figure 3 that the fitness of the initial particles at the sampling points is between 0.47 - 0.96. The fitness of a large number of initial particle positions is very low, indicating that the values of these particles are not reasonable, and it is difficult to search for the global optimal solution near them. Therefore, it can be inferred that it is reasonable to adopt the method of shrinking the search space.

[0079] In short, when shrinking the search space, it should contain as many relatively good initial particles as possible. At the same time, to ensure that there is sufficient search space near each particle, it is necessary to avoid the initial particle positions being located on the boundary of the search space.

[0080] After the evaluation of the initial particle positions and the shrinking of the search space, the PSO optimization algorithm is carried out. Generally, in order to ensure the accuracy of the search, the PSO optimization algorithm needs to set 20 - 100 particles in each generation and carry out more than 100 iterations. In this application, since the initial positions obtained after the LHS sampling and screening of the initial positions have a high probability of being near the optimal position, the number of particles and the number of iterations in the PSO optimization process can be appropriately reduced.

[0081] To verify the advantages of the initial particle position screening and search space scaling, this application provides two groups of model correction tests for PSO optimization:

[0082] The first group: Without the screening of the LHS initial particle positions and the scaling of the search space, 20 particles are set in each generation, and the iteration termination condition is that the number of iterations reaches 100 times;

[0083] The second group: Adopt the method of shrinking the search space. The initial search space is shrunk into two relatively good search sub-spaces, and PSO optimization is carried out separately. 20 particles are set in each generation, and the iteration termination condition is that the number of iterations reaches 20 times.

[0084] Respectively compare the computational amount, computational time, and model correction accuracy of the two groups of model correction tests. The fitness changes of the PSO algorithm in the two groups of model correction tests are as Figure 4 shown, Figure 4 in which 4a is the fitness change diagram of the first group, 4b is the fitness change diagram of the first search sub-space of the second group, and 4c is the fitness change diagram of the second search sub-space of the second group.

[0085] Next, regarding the acquisition of the basic response data as the training set in step S4, this application preferably uses sensitivity analysis for confirmation. Conducting local sensitivity analysis based on statistical data requires a large amount of data sets, and the surrogate model can greatly reduce the computational cost of finite element simulation. The specific steps are as follows: Construct a surrogate model with the scour depth as the input sample set and the pile body moment, pile top displacement, jacket node stress, tower top displacement, tower top acceleration, and tower top rotation angle as the output sample set; use the Latin hypercube sampling method to randomly select 1,000 groups of scour depths and substitute them into the surrogate model to obtain 1,000 groups of input-output pairs; finally, obtain the sensitivity of the input-output pairs.

[0086] Regarding the aforementioned target response data to be corrected, Table 1 shows the local sensitivities of each structure to scour.

[0087] Table 1 Local sensitivities of each structure response to scour

[0088]

[0089] In Table 1, the sensitivity of the pile top displacement is 0.93, and the sensitivity of the tower top displacement is 0.71. It can be seen that since the pile top displacement is close to the scour pit, relatively speaking, it is a local structure response, while the tower top displacement is a global structure response, making the sensitivity of the former greater than that of the latter and more sensitive to local scour pits. The sensitivity of the tower top acceleration is 1.06, indicating that the acceleration response is more sensitive to the scour pit than the displacement. Overall, however, the sensitivity coefficients of the pile top displacement, tower top displacement, tower top acceleration, and tower top rotation angle are around 1, indicating that these displacement, velocity, and acceleration-based structural responses are close in terms of the scour detection accuracy of the surrogate model when used as input training samples. The local sensitivity of the jacket node stress is between 0.07 and 0.16, which is much smaller than the previous structural responses. Therefore, the jacket node stress is not suitable as an input training sample for constructing the scour detection surrogate model.

[0090] Regarding the pile body moment, Figure 5 As shown, the moment sensitivity in the middle of the pile body is between 4 and 10, indicating that these moments are more suitable as the input training sample set compared to the former, and Figure 6 (Scour detection result with moment as input)- Figure 7 (Scour detection result with displacement as input) The shown scour detection results also verify the above conclusion.

[0091] Finally, this application also provides a physical model diagram capable of implementing the above detection method, as Figure 8 shown, Figure 9 is the constructed finite element numerical model, Figure 10 showing its scour simulation diagram.

[0092] Those skilled in the art can understand that, unless otherwise defined, all terms used herein (including technical and scientific terms) have the same meaning as the general understanding of those of ordinary skill in the art to which this application belongs. It should also be understood that terms such as those defined in a general dictionary should be understood as having a meaning consistent with the meaning in the context of the prior art, and will not be interpreted in an idealized or overly formal sense unless defined as such here.

[0093] As used in this application, the meaning of "and / or" includes both the case where each exists alone and the case where both exist simultaneously.

[0094] As used in this application, the meaning of "connection" can be a direct connection between components or an indirect connection between components through other components.

[0095] Taking the above-described ideal embodiments of the present invention as an inspiration, through the above description, relevant staff can fully make various changes and modifications without departing from the technical idea of this invention. The technical scope of this invention is not limited to the content in the specification, and its technical scope must be determined according to the scope of the claims.

Claims

1. A scour detection method based on a surrogate model, characterized in that: Specifically, it includes the following steps: Step S1, obtaining the target response data to be corrected of the offshore wind turbine pile foundation; Step S2, based on the obtained target response data to be corrected, constructing a finite element numerical model with the same size as the physical model test, and using the improved PSO algorithm to correct the finite element numerical model to obtain the corrected finite element numerical model; Among them, the Latin hypercube sampling method is used to select several particles with good fitness as the initial particles, and at the same time, after shrinking the search space, the optimization of the PSO algorithm is carried out; Step S3, calculating the foundation response data at different scour depths through the corrected finite element numerical model; Step S4, using the foundation response data obtained in Step S3 as the input training set and the scour depth as the output training set, constructing a BP neural network and training it to obtain a surrogate model that can be used to detect the scour depth, and using the test data of each working condition as the test set to invert the scour depth.

2. The scour detection method based on an agent model according to claim 1, characterized in that: The target response data to be corrected includes pile body moment, pile top displacement, jacket node stress, tower top displacement, tower top acceleration, and tower top rotation angle.

3. The scour detection method based on the surrogate model according to claim 1, characterized in that: The specific steps for obtaining the corrected finite element numerical model in Step S2 are as follows: Step S21, selecting and calculating half of the symmetric numerical model in the constructed finite element numerical model, and selecting the time history response data at the same position as the physical model under cyclic loading for the parameters to be corrected; Step S22, calculating the normalized root mean square error between the finite element simulation result and the actual measurement of the physical model; Step S23, continuing to normalize the root mean square error; Step S24, evaluating the similarity of the spatial distribution of the target response to be corrected through the Pearson correlation coefficient; Step S25, obtaining the objective function of the improved PSO algorithm based on the Pearson correlation coefficient.

4. The scour detection method based on the surrogate model according to claim 3, wherein: In Step S22, the calculation formula for the normalized root mean square error between the finite element simulation result and the actual measurement is: In formula (1), RMSE is the root mean square error, n is the number of responses of the measurement points where the target to be corrected is located, and y i is the response data of the measurement points obtained by numerical simulation, and x i is the response data of the measurement points measured in the physical model; In Step S23, the calculation formula for the continued normalization process is: In formula (2), RMSE′ is the normalized root mean square error, which is the average value of the measured response data of the measuring points in the physical model, and which is the average value of the response data of the measuring points obtained by numerical simulation; In Step S24, the calculation formula for the evaluation by the Pearson correlation coefficient is: In formula (3), r is the Pearson correlation coefficient; In Step S25, the obtained objective function of the improved PSO algorithm is: In formula (4), f min (x) is the fitness function, α is the weight of the normalized root mean square error, β is the weight of the Pearson coefficient, and α + β = 1.

5. The scour detection method based on the surrogate model according to claim 4, wherein: In Step S2, before carrying out the optimization of the PSO algorithm, several initial particle positions are randomly sampled, the fitness functions of several initial particles are calculated, and based on the fitness function, several initial particle positions with good fitness are selected from several particles, the search space is shrunk, and PSO optimization is carried out within the shrunk search space.

6. The scour detection method based on the surrogate model according to claim 5, characterized in that: When correcting the finite element numerical model, the number of particles in each iteration is at least 20.

7. The scour detection method based on the surrogate model according to claim 1, characterized in that: In Step S4, the foundation response data used as the training set is obtained based on sensitivity analysis, and the specific steps are as follows: Step S41, constructing a surrogate model with the scour depth as the input sample set and the pile body moment, pile top displacement, jacket node stress, tower top displacement, tower top acceleration, and tower top rotation angle as the output sample set; Step S42, using the Latin hypercube sampling method to randomly select 1000 groups of scour depths and substituting them into the surrogate model to obtain 1000 groups of input-output pairs; Step S43, calculating the sensitivity of the input-output pairs and selecting a suitable input training set.

Images