Offshore wind turbine foundation selection method based on multi-model fusion

Through multi-model fusion and random forest model optimization algorithm, the basic selection process of offshore fan is simplified, the complex and complex selection process is solved, and efficient and accurate basic selection of offshore fan is achieved, reducing costs and relying on designer experience is achieved. demand.

CN118690664BActive Publication Date: 2024-11-05CHINA POWER CONSRTUCTION GRP GUIYANG SURVEY & DESIGN INST CO LTD
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202411170682.0
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2024-08-26
Publication Date
2024-11-05
Estimated Expiration
2044-08-26

AI Technical Summary

Technical Problem

The basic selection process of offshore fan is complicated and relies heavily on the work experience and ability of the selection designer, which leads to difficulty in selecting and design, affecting the progress of project feasibility report preparation and cost control.

Method used

The basic selection method of offshore fan based on multi-model fusion is adopted, and the basic selection database is constructed by standardizing and labeling the model input feature data, and the random forest model, grid search algorithm, and Bayesian algorithm are used to optimize the parameters. The optimal random forest model realizes basic selection prediction of offshore fan.

Benefits of technology

The basic selection process of offshore fan is simplified, the dependence on designer experience is reduced, the selection efficiency is accelerated, the accuracy and efficiency of basic selection is improved, and the cost is reduced.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118690664B_ABST
    Figure CN118690664B_ABST
Patent Text Reader

Abstract

The present invention discloses an offshore wind turbine foundation selection method based on multi-model fusion, which belongs to the technical field of offshore wind power engineering. By considering the influencing factors of the offshore wind turbine foundation type under extreme operating conditions, the model input feature data is standardized respectively, and the corresponding foundation types are labeled so that the classification parameters correspond to the label parameters, thereby constructing a foundation selection database and dividing the training data set and the verification data set, and then establishing a foundation selection random forest model. Finally, a grid search algorithm and a Bayesian algorithm are used for secondary optimization respectively to determine the optimization parameters to obtain the optimal random forest model, and the classification parameters of the offshore wind power engineering project that needs to be selected are used as input to obtain the offshore wind turbine foundation selection prediction result, which solves the problem of selection and design difficulties caused by the cumbersome and complicated existing offshore wind turbine foundation selection process and the serious reliance on the work experience and ability of selection and design personnel, and accelerates the foundation selection efficiency.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention belongs to the technical field of offshore wind power engineering, and specifically relates to a method for selecting offshore wind turbine foundations based on multi-model fusion. Background Art

[0002] Offshore wind energy resources are abundant, and offshore wind power has developed rapidly due to its proximity to the power load center and its small footprint. The foundation of an offshore wind turbine is the supporting structure of a wind turbine, and plays a vital role in the safe operation of an offshore wind farm. The foundation provides key support for wind turbines for at least 25 years. While being subjected to wind loads, wind turbine operating loads, waves, currents and other loads, it also withstands the harsh test of the harsh offshore environment. At the same time, in the total investment of an offshore wind farm, the foundation cost accounts for 20%-30%, which is much higher than the same proportion of an onshore wind farm. Therefore, on the premise of in-depth analysis of the structural characteristics of different offshore wind turbine foundations, the geological, wind energy resources, marine hydrology and other environmental conditions of the sea area where the wind farm is located, reasonable foundation selection is one of the main ways to promote the reduction of offshore wind power costs and ensure the long-term safe operation of wind turbines.

[0003] Traditionally, the selection of offshore wind turbine foundations requires the use of professional marine engineering calculation software SACS to conduct design modeling and calculation verification of various foundation forms that may be suitable for the project site. The selection process is cumbersome and complicated, which seriously affects the progress of the preparation of the project feasibility report. It also relies heavily on the work experience and ability of the selection designers, making the selection design threshold relatively high. Summary of the invention

[0004] The purpose of the present invention is to address the problem that the existing offshore wind turbine foundation selection process is cumbersome and complicated, and is heavily dependent on the work experience and ability of the selection and design personnel, resulting in selection and design difficulties. An offshore wind turbine foundation selection method based on multi-model fusion is proposed. By considering the influencing factors of the offshore wind turbine foundation type under extreme operating conditions, the model input feature data is standardized, and the corresponding foundation types are labeled so that the classification parameters correspond to the label parameters, thereby constructing a basic selection database and dividing the training data set and the verification data set, and then establishing a basic selection random forest model. Finally, the grid search algorithm and the Bayesian algorithm are used for secondary optimization, respectively, to determine the optimization parameters to obtain the optimal random forest model, and the classification parameters of the offshore wind power project that needs to be selected are used as input to obtain the offshore wind turbine foundation selection prediction results.

[0005] In order to solve the above technical problems, the present invention provides an offshore wind turbine foundation selection method based on multi-model fusion, including: the offshore wind turbine foundation selection method based on multi-model fusion, characterized in that it includes the following steps:

[0006] S1. Obtain offshore wind turbine foundation type data and corresponding construction condition data, consider the influencing factors of selecting offshore wind turbine foundation type under extreme operating conditions, and select construction condition data as model input feature data;

[0007] S2, respectively perform standardization processing and labeling processing to obtain classification parameters and labeling parameters, and then build a basic selection database, and divide the basic selection database into a training data set and a verification data set;

[0008] S21, inputting the model into feature data for standardization to obtain classification parameters;

[0009] S22, labeling each selected basic type to obtain label parameters;

[0010] S23, making the classification parameters correspond to the label parameters, building a basic selection database, and randomly selecting a training data set and a verification data set;

[0011] The labeling process is to convert the foundation type classification from text description to digital label description so that the model can perform calculation and recognition. Specifically, the gravity foundation, single pile foundation, jacket foundation, suction cylinder foundation, and high pile cap foundation are calibrated with 1, 2, 3, 4, and 5 respectively;

[0012] S3, using the training data set to train, and through the validation data set to evaluate to obtain several decision tree models, and then integrate to establish a basic selection random forest model;

[0013] S4, respectively using the grid search algorithm and the Bayesian algorithm to perform secondary optimization on the basic selection random forest model, so as to determine the optimization parameters of the basic selection random forest model;

[0014] S5. Based on the optimization parameters and the basic selection random forest model, the optimal random forest model is obtained, and the classification parameters of the offshore wind power project that needs to be selected are used as input to obtain the prediction results of the offshore wind turbine basic selection.

[0015] Preferably, the influencing factors of the offshore wind turbine foundation type include wind turbine conditions, ocean hydrological conditions and geographical geological conditions; among which, six load components under extreme operating conditions are selected from the wind turbine conditions: Fx , F , F , Mx , My , Mz ;in, Fx , F , F They are x , y , z Directional force; Mx , My , Mz They are x , y , z The bending moment in the direction. And the upper limit and lower limit of the first-order frequency of the whole machine; in the marine hydrological conditions, the average water depth, the maximum wave height, the wave period corresponding to the maximum wave height, the maximum current velocity, the extreme high tide level, the extreme low tide level, the average growth thickness and density of marine organisms are selected; in the geographical geological conditions, the strength of the topsoil, the shallowest depth of the bearing layer, and the distance from the shore are selected.

[0016] Preferably, S3 comprises the following steps:

[0017] S31, setting the number of decision tree models and the maximum number of training times;

[0018] S32, determining whether the current training times are less than the maximum training times, if so, executing step S33; otherwise, executing step S35;

[0019] S33, take random sampling with replacement from the training data set, and sample w A new sub-training dataset D is formed and randomly selected m features, among which m < M ;in, M Indicates the number of classification parameters, that is, the number of influencing factors;

[0020] S34. Use new sub-training dataset D and m The decision tree model is trained with features, and the effect of the decision tree model is evaluated using the validation data set. A c , proceed to step S32;

[0021] Among them, the prediction accuracy ,in, F i For the i The number of correct prediction samples corresponding to the basic form, N is the total number of samples in the test set;

[0022] S35, select the corresponding Ac The largest decision tree model is used as a submodel of the random forest model;

[0023] S36, repeat steps S32 to S35 until the number of decision tree models reaches N tree ;

[0024] S37, will N treeThe decision tree models are integrated to build a random forest model.

[0025] Preferably, the random forest model H The specific expression is:

[0026] ;

[0027] In the formula, h i For the i A decision tree; Y Output variables for the model, i.e. label parameters; I For evaluation criteria.

[0028] As a preferred method, under the condition that all decision tree parameters are set to default values, a grid search algorithm is used to optimize the architecture parameters of the random forest model to obtain the optimal architecture optimization parameters; then the architecture optimization parameters are used to optimize each decision tree parameter through the Bayesian algorithm to obtain the optimal parameters of each decision tree; thereby determining the optimization parameters of the basic selection random forest model.

[0029] Preferably, the basic steps of the grid search algorithm are:

[0030] Sa, defines the search space of optimized parameters;

[0031] Sb, initialize the grid, traverse every combination in the parameter search space, and train a model for each parameter combination;

[0032] Sc, evaluate the performance of the random forest model using k-fold cross validation;

[0033] Sc1, divide the data set equally into k share;

[0034] Sc2, use it k-1 One set is used as the training set, and the remaining one set is used as the validation set to train the model and evaluate its performance;

[0035] Sc3. Repeat this process k times, each time selecting a different set of data as the validation set;

[0036] Sc4. Calculate the average value of the prediction accuracy as the model performance of the group of parameter combinations;

[0037] Sd, compare the model performance of all parameter combinations, and select the parameter combination with the best performance as the optimal parameter combination.

[0038] Beneficial effects of the present invention:

[0039] 1. This scheme considers the influencing factors of offshore wind turbine foundation types under extreme operating conditions, standardizes the model input feature data, labels the corresponding foundation types, and makes the classification parameters correspond to the label parameters, thereby constructing a foundation selection database and dividing the training data set and the verification data set, and then establishing a foundation selection random forest model. Finally, the grid search algorithm and the Bayesian algorithm are used for secondary optimization to determine the optimization parameters to obtain the optimal random forest model, and the classification parameters of the offshore wind power project to be selected are used as input to obtain the prediction results of the offshore wind turbine foundation selection, which solves the problem of the difficulty of selection and design caused by the cumbersome and complicated selection process of the existing offshore wind turbine foundation and the heavy reliance on the work experience and ability of the selection designers, and speeds up the efficiency of foundation selection.

[0040] 2. This scheme introduces the random forest algorithm to construct an offshore wind turbine foundation selection method and system. Compared with the traditional method, it effectively avoids the design modeling, calculation verification and other steps of large-scale software, and speeds up the foundation selection efficiency.

[0041] 3. This scheme optimizes the parameters of the random forest model by using the grid search algorithm and the Bayesian algorithm for different model parameter items to obtain the optimal random forest model. The classification parameters of the offshore wind turbine foundation selection are input into the optimal random forest model to accurately obtain the prediction results of the foundation selection.

[0042] 4. Since the number of architectural parameters of the random forest model is small, in order to ensure the optimization efficiency, this scheme first uses a grid search algorithm to optimize the architectural parameters of the random forest model to obtain the optimal architectural optimization parameters; then, using the optimal architectural parameters, the Bayesian algorithm is used to optimize the parameters of each decision tree to obtain the optimal parameters of each decision tree; thereby determining the various optimization parameters of the random forest model, greatly improving the efficiency of model parameter optimization and model training.

[0043] 5. This solution uses a grid search algorithm to determine the best match by combining all possible parameters that need to be optimized. The optimization effect is good. In view of the fact that the architectural parameters of the random forest model do not affect each other, only one-to-one comparison and evaluation are required, and the parameter space is small. The grid search algorithm is used to obtain the best parameter combination. BRIEF DESCRIPTION OF THE DRAWINGS

[0044] Other features, objects and advantages of the present invention will become more apparent by reading the detailed description of non-limiting embodiments made with reference to the following drawings. The drawings are only for the purpose of illustrating preferred embodiments and are not to be considered as limiting the present invention. Also, the same reference symbols are used throughout the drawings to represent the same parts.

[0045] Figure 1is a flow chart of the overall method of an embodiment of the present invention;

[0046] Figure 2 This is a schematic diagram of the process of step 4 of the present invention;

[0047] Figure 3 It is a flow chart of optimizing parameters of the grid search algorithm of the present invention;

[0048] Figure 4 The parameter optimization flow chart of the Bayesian optimization algorithm of the present invention;

[0049] Figure 5 It is a system structure diagram of the present invention. DETAILED DESCRIPTION

[0050] In order to make the objectives, technical solutions and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific implementation method described herein is only an optimal embodiment of the present invention, which is only used to explain the present invention and does not limit the scope of protection of the present invention. All other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention.

[0051] Example 1: Figure 1 As shown in the figure, the offshore wind turbine foundation selection method based on multi-model fusion includes:

[0052] Step 1: Collect construction condition data of existing offshore wind power projects and corresponding selected foundation type data.

[0053] Step 2: Model input features and output category selection. When selecting the foundation, it is necessary to consider the foundation's bearing capacity, stiffness, transportation and installation, material usage and other requirements, and select appropriate construction conditions as the model input features. The corresponding selected foundation form will be used as the model's output category.

[0054] According to the factors affecting the selection of offshore wind farm foundation, appropriate construction condition related parameters are selected, mainly including: wind turbine conditions, marine hydrological conditions and geographical geological conditions. Among them, six load components under extreme operating conditions are selected in wind turbine conditions: Fx , F , F , Mx , My , Mz , and the upper limit of the first-order frequency of the whole machine f max , lower limit f min ; Select the average water depth in the ocean hydrological conditions h , maximum wave height H , Maximum wave height corresponds to wave period T, Maximum current velocity V max , extreme high tide E H , Extreme Low Tide E L , Average growth thickness of marine organisms t ,density ρ ; Select the topsoil strength in geographical and geological conditions c , shallowest depth of bearing layer d , Distance from shore L .

[0055] Among them, the extreme operating conditions of wind turbines, average water depth, maximum wave height, wave period corresponding to maximum wave height, maximum current velocity, extreme high tide level, extreme low tide level, average growth thickness of marine organisms, and density are the main factors affecting the foundation load, which in turn can affect the amount of foundation materials; the strength of the topsoil, the shallowest burial depth of the bearing layer, and the strength of the bearing layer affect the bearing capacity of the foundation; the distance from the shore and the average water depth affect the transportation and installation of the foundation; the average water depth, extreme high tide level, extreme low tide level, average growth thickness of marine organisms, density, strength of the topsoil, the shallowest burial depth of the bearing layer, and the strength of the bearing layer are the main factors affecting the foundation stiffness.

[0056] The selected foundation type can adapt to different requirements of bearing capacity, rigidity, transportation and installation, and material consumption, mainly including: gravity foundation, single pile foundation, jacket foundation, suction tube foundation, and high pile cap foundation.

[0057] Step 3: Data processing. The model input feature data is standardized to obtain classification parameters, and the corresponding basic types are labeled to obtain label parameters. The classification parameters correspond to the label parameters one by one, and they are combined to form a basic selection database. Data , randomly select 80% of them as training data set Train The remaining 20% ​​is used as a validation dataset. Validation .

[0058] Since different data sources usually lead to differences in the dimensions and magnitude of the data, in order to make these data comparable, it is necessary to use standardization methods to eliminate these differences. Standardization is to scale the data of each influencing factor proportionally, remove the unit restrictions of the data, and convert it into a dimensionless pure value. min-max Method:

[0059] , i , j =1, 2, 3, …, n , n is the total data volume.

[0060] in, Yi is the processed data set, X i This is the dataset before processing.

[0061] Labeling processing is to convert the foundation type classification from "text description" to "digital label classification" so that the model can perform calculation and recognition. Specifically, the gravity foundation, single pile foundation, jacket foundation, suction tube foundation, and high pile cap foundation are calibrated with 1, 2, 3, 4, and 5 respectively.

[0062] Step 4: Build a random forest model. Use the basic selection training data set Train Training and validation dataset Validation Multiple optimal decision tree models were evaluated and then integrated to establish a basic selection random forest model.

[0063] like Figure 2 As shown, the specific steps are as follows:

[0064] 4.1 Setting the number of decision tree models N tree and maximum number of training sessions;

[0065] 4.2 Determine whether the current number of training times is less than the maximum number of training times. If so, proceed to step 4.3. Otherwise, proceed to step 4.6.

[0066] 4.3 Take random sampling with replacement from the training data set Train and sample w A new sub-training data set is formed D ;

[0067] 4.4 Random Selection m features, among which m < M ;

[0068] 4.5 Using a new sub-training dataset D and m Features, train the decision tree model, and use the validation data set to evaluate the effect of the decision tree model A c , go to step 4.2;

[0069] Among them, the prediction accuracy ,in, F i For the i The number of correct prediction samples corresponding to the basic form, N is the total number of samples in the test set.

[0070] 4.6 Select the corresponding A cThe largest decision tree model is used as a submodel of the random forest model;

[0071] 4.7 Repeat steps 4.2 to 4.6 until the number of decision tree models reaches N tree ;

[0072] 4.8 N tree Integrate decision tree models to build a random forest model:

[0073] ;

[0074] In the formula, H is a random forest model; h i For the i A decision tree; Y Output variables for the model, i.e. label parameters; I For evaluation criteria.

[0075] Step 5: Determine the optimal parameters of the random forest model, that is, find a set of optimal parameter combinations to make the model prediction effect the best. The parameter combination with the highest prediction accuracy is the optimal parameter of the model. The parameters to be optimized include the random forest architecture parameters and the random forest decision tree parameters.

[0076] Among them, random forest architecture parameters: number of decision trees N tree , Evaluation Criteria I ; Random Forest Decision Tree Parameters: Maximum Number of Features n max , Maximum Depth d max , the minimum number of samples that can be divided into nodes l min , leaves contain the smallest sample ls min , Maximum number of leaf nodes l max .

[0077] The grid search algorithm is to traverse all possible combinations of the parameters to be optimized to determine the best match. The optimization effect is good. When the parameter space is large, the calculation amount of the grid search will become very large, resulting in low search efficiency. The architecture parameters of the random forest model do not affect each other, so only one-to-one comparison evaluation is required. The parameter space is small, so the grid search algorithm is used to obtain the best parameter combination.

[0078] The Bayesian optimization algorithm is a hyperparameter search algorithm based on Bayesian theory. It finds the optimal hyperparameter combination by establishing a probability model representing the objective function. Bayesian optimization uses a sequential approach to search for hyperparameters, trying only a portion of hyperparameter combinations each time, and continuously updating the probability model based on the existing data, so that the global optimal solution can be found in a relatively small number of experiments. The decision tree parameters are not independent parameters, and different decision tree parameters will produce a large number of parameter combinations. The Bayesian algorithm can improve efficiency while ensuring the optimization results.

[0079] Since the number of architectural parameters of the random forest model is relatively small, in order to ensure the optimization efficiency, the grid search algorithm is first used to optimize the architectural parameters of the random forest model (all decision tree parameters are set to default values) to obtain the optimal architectural optimization parameters; then, using the optimal architectural parameters, the Bayesian algorithm is used to optimize the parameters of each decision tree to obtain the optimal parameters of each decision tree; thereby determining the various optimization parameters of the random forest model.

[0080] like Figure 3 As shown, the basic steps of the grid search algorithm are:

[0081] (1) Define the search space of the parameters to be optimized: number of decision trees N tree If this value is too small, the model is prone to underfitting, and if it is too large, the model is prone to overfitting. Therefore, a moderate value is needed. For the offshore wind turbine foundation selection model, the probability of construction conditions influencing factors is analyzed, and the classification parameter types are between 10 and 100. According to the empirical search space, it is determined to be [20, 200]; Evaluation criteria I This value is a measure of the impurity of the classification results of the random forest model. The search space has two options: Gini coefficient or information gain. Other parameters use the default values.

[0082] (2) Initialize the grid, traverse every combination in the parameter search space, and train a model for each parameter combination;

[0083] (3) Utilization k The performance of the model was evaluated by the cross-validation method (the offshore wind turbine foundation selection model relies on data collection from actual wind farm projects, and the amount of data is medium, so k =4 can ensure both cross-validation efficiency and accuracy).

[0084] k The basic steps of cross validation are:

[0085] (3.1) Divide the data set equally into k share;

[0086] (3.2) Use k-1 as the training set, and the remaining 1 as the validation set to train the model and evaluate its performance;

[0087] (3.3) Repeat this process k times, each time selecting a different set of data as the validation set;

[0088] (3.4) Calculation k Evaluation results (prediction accuracy A c ) is taken as the model performance of this set of parameter combinations.

[0089] (4) Compare the model performance of all parameter combinations and select the parameter combination with the best performance as the optimal parameter combination.

[0090] like Figure 4 As shown, the basic steps of the Bayesian optimization algorithm are:

[0091] (1) Setting the objective function of Bayesian optimization f ( x ).

[0092] Use any combination of parameters to pass k The negative root mean square error obtained by fold cross validation is used as the objective function of Bayesian optimization f ( x ): ;

[0093] in, f(x) It is the objective function of Bayesian optimization algorithm; RMSE To pass k The average of the root mean square differences between the predicted values ​​and the true values ​​obtained by cross-validation; x ij Indicates i Second-rate k The first fold in the validation data set during cross validation j individual data; y ij Indicates i Second-rate k The first part of the predicted value in cross validation j individual data; k is the number of cross validations; n for k The amount of data in the offshore wind turbine foundation selection validation dataset during the fold cross validation.

[0094] 2) Determine the parameter search space that needs to be optimized.

[0095] The number of classification parameters of the random forest model for offshore wind turbine foundation selection is between 10 and 100. According to the number of offshore wind farms under construction, the foreseeable amount of model data is between 10 and 100.4 Within, it belongs to the model of medium data, and the appropriate decision tree parameter search space is set as follows:

[0096] Maximum number of features n max , which indicates the maximum number of features considered during the division, and the optimization is performed between 10% and 99% of the classification feature parameters, that is, [0.1 m , 0.99 m ];in, m is The number of input features selected in step 2;

[0097] Maximum Depth d max ,To ensure that the model can be successfully fitted when there is a large amount of data and features, it is necessary to limit the maximum depth of the decision tree, and the search space is [10, 100];

[0098] Minimum number of samples that can be divided into nodes l min It is related to the total amount of data. The search space is determined as [2, 4] according to the amount of data of the basic selection model.

[0099] Leaves contain the smallest sample ls min The minimum number of samples of leaf nodes is limited. If the number of leaf nodes is less than the number of samples, it will be pruned together with its sibling nodes. The search space is determined to be [1, 4] based on the possible amount of data of the model.

[0100] Maximum number of leaf nodes l max The random forest model limits this value to prevent model overfitting. Considering the number of features of the basic selection model, the search space is determined to be [10, 50].

[0101] (3) Determine the number of iterations of Bayesian optimization I ter =[10, 30] and the number of steps of random search S t =[1, 4].

[0102] (4) Select several optimization parameter combinations: X 1 , X 2 , X 3 , … , X M (If the model is not initialized, it is randomly selected; otherwise, the set of maximum points of the previous round of acquisition function is used), train the random forest model, and obtain the corresponding objective function value Y 1 ,Y 2 , Y 3 , …, Y M , combined with the parameters to form a set of known points D ={( X 1 , Y 1 ), ( X 2 , Y 2 ), ( X 3 , Y 3 ), …, ( X M , Y M )};

[0103] (5) Fitting the surrogate function using a set of known points G ( X ), using a proxy function G(X) Calculate the posterior probability distribution of the objective function at any test point.

[0104] Among them, the proxy function G(X) Gaussian fitting is used, and the fitting process is as follows:

[0105] (5.1) Create kernel function for Gaussian fitting model k : ;

[0106] in,( X i , X j ) represents i and j parameter combinations, k is the kernel function;

[0107] (5.2) The covariance matrix of a known data set can be calculated using a known data set and a kernel function: ;

[0108] (5.3) The Gaussian distribution of the objective function is established through the covariance matrix.

[0109] Known Datasets D The objective function value in Y 1:M It obeys normal distribution, that is Y 1:M ~ N (0, K ), so as toD Gaussian fitting can obtain the Gaussian distribution function G ( X )~ N (0, K ).

[0110] Since the known data set D The objective function value in Y 1:M And the objective function value of any subsequent test point Y t ( t > M ) satisfies the joint Gaussian distribution, that is, ;

[0111] ;

[0112] (5.4) According to the joint Gaussian distribution, the posterior probability distribution of any test point is calculated as: ;

[0113] in, Any test point The objective function value of for expectations, for The variance of .

[0114] (6) Establish acquisition function based on posterior probability distribution EI ( X );

[0115] (6.1) For a known data set D in front M Calculated times is the optimal value of the current function. M +1 calculation, the optimal objective function value is obtained Y M+1 < , the optimal function value remains unchanged; otherwise, the optimal function value is corrected to Y M+1 . Based on this, the improved function is defined as follows:

[0116] ;

[0117] (6.2) Calculation of known data sets D The expected value of any point in the M +1 sampling point:

[0118] ;

[0119] (6.3) According to the posterior expectation and variance obtained from the posterior probability distribution of the test point, the improved probability density function is: ;

[0120] in, is the maximum function value of the current test point, is the mean square error, For expectation.

[0121] (6.4) According to the improved probability density function, the acquisition function is obtained by integrating over (0, +∞) EI ( X ): ;

[0122] in, is the probability distribution function of the standard normal distribution; is the distribution function of the standard normal distribution.

[0123] (7) Taking the maximum value of the acquisition function as the parameter combination, repeat steps (4) to (7) until the number of cycles is greater than the maximum number of cycles, thereby obtaining the optimal parameter combination of the random forest model;

[0124] (8) Input the optimal parameter combination into the random forest model established in step 4 to obtain the optimal random forest model.

[0125] Step 6: Train and optimize the random forest model according to the determined optimization parameters and basic selection training library to obtain the optimal random forest model.

[0126] Step 7: Input the classification parameters of the offshore wind power project that needs to be selected into the optimal random forest model established in step 6, and output the prediction results of the offshore wind turbine foundation selection.

[0127] Example 2: Figure 5 As shown, the present scheme also discloses an offshore wind turbine foundation selection system based on multi-model fusion applicable to the offshore wind turbine foundation selection method based on multi-model fusion, including: a data acquisition module, which can collect construction conditions and foundation selection data corresponding to the offshore wind power project;

[0128] The data processing and classification module can standardize and label the collected data to obtain the classification parameters and label parameters of the offshore wind turbine foundation selection, and then combine them to form a basic selection database Data , and then the model training set data Train and validation set data Validation Automatic division of .

[0129] The random forest model for offshore wind turbine foundation selection can select the corresponding offshore wind turbine foundation based on the input offshore wind power project construction conditions.

[0130] The selection result output module can output the selection results of the corresponding offshore wind turbine foundation.

[0131] The above specific embodiments are preferred embodiments of the present invention, and are not intended to limit the specific implementation scope of the present invention. The scope of the present invention includes but is not limited to the present specific embodiments. All equivalent changes made in accordance with the shape, structure, and method of the present invention are within the protection scope of the present invention.

Claims

1. The offshore wind turbine foundation selection method based on multi-model fusion is characterized by: The steps include: S1. Obtain offshore wind turbine foundation type data and corresponding construction condition data, consider the influencing factors of selecting offshore wind turbine foundation type under extreme operating conditions, and select construction condition data as model input feature data; The influencing factors of the offshore wind turbine foundation type include wind turbine conditions, marine hydrological conditions and geographical geological conditions; among them, the load components under extreme operating conditions, as well as the upper limit and lower limit of the first-order frequency of the whole machine are selected in the wind turbine conditions; the average water depth, maximum wave height, wave period corresponding to the maximum wave height, maximum current velocity, extreme high tide level, extreme low tide level, average growth thickness and density of marine organisms are selected in the marine hydrological conditions; the topsoil strength, the shallowest burial depth of the bearing layer and the distance from the shore are selected in the geographical geological conditions; S2, respectively perform standardization processing and labeling processing to obtain classification parameters and labeling parameters, and then build a basic selection database, and divide the basic selection database into a training data set and a verification data set; S21, inputting the model into feature data for standardization to obtain classification parameters; S22, labeling each selected basic type to obtain label parameters; S23, making the classification parameters correspond to the label parameters, building a basic selection database, and randomly selecting a training data set and a verification data set; The labeling process is to convert the foundation type classification from text description to digital label description so that the model can perform calculation and recognition. Specifically, the gravity foundation, single pile foundation, jacket foundation, suction cylinder foundation, and high pile cap foundation are calibrated with 1, 2, 3, 4, and 5 respectively; S3, using the training data set to train, and through the validation data set to evaluate to obtain several decision tree models, and then integrate to establish a basic selection random forest model; S4, respectively using the grid search algorithm and the Bayesian algorithm to perform secondary optimization on the basic selection random forest model, so as to determine the optimization parameters of the basic selection random forest model; S5. Based on the optimization parameters and the basic selection random forest model, the optimal random forest model is obtained, and the classification parameters of the offshore wind power project that needs to be selected are used as input to obtain the prediction results of the offshore wind turbine basic selection.

2. The offshore wind turbine foundation selection method based on multi-model fusion according to claim 1 is characterized in that: The S3 comprises the following steps: S31, setting the number of decision tree models and the maximum number of training times; S32, determining whether the current training times are less than the maximum training times, if so, executing step S33; Otherwise, execute step S35; S33, take random sampling with replacement from the training data set, and sample w A new sub-training dataset D is formed and randomly selected m features, among which m < M ; in, M Indicates the number of classification parameters, that is, the number of influencing factors; S34. Use new sub-training dataset D and m The decision tree model is trained with features, and the effect of the decision tree model is evaluated using the validation data set. A c , proceed to step S32; Among them, the prediction accuracy ,in, F i For the i The number of correct prediction samples corresponding to the basic form, N is the total number of samples in the test set; S35, select the corresponding Ac The largest decision tree model is used as a submodel of the random forest model; S36, repeat steps S32 to S35 until the number of decision tree models reaches N tree ; S37, will N tree The decision tree models are integrated to build a random forest model.

3. The offshore wind turbine foundation selection method based on multi-model fusion according to claim 2 is characterized in that: The random forest model H The specific expression is: ; In the formula, h i For the i A decision tree; Y Output variables for the model, i.e. label parameters; I For evaluation criteria.

4. The offshore wind turbine foundation selection method based on multi-model fusion according to claim 1 is characterized in that: Under the condition that all decision tree parameters are set to default values, the grid search algorithm is used to optimize the architecture parameters of the random forest model to obtain the optimal architecture optimization parameters; then the architecture optimization parameters are used to optimize the parameters of each decision tree through the Bayesian algorithm to obtain the optimal parameters of each decision tree; thus determining the optimization parameters of the basic selection random forest model.

5. The offshore wind turbine foundation selection method based on multi-model fusion according to claim 1 is characterized in that: The basic steps of the grid search algorithm are: Sa, defines the search space of optimized parameters; Sb, initialize the grid and train a model for each parameter combination; Sc, evaluate the performance of the random forest model using k-fold cross validation; Sc1, divide the data set equally into k share; Sc2, use it k-1 One set is used as the training set, and the remaining one set is used as the validation set to train the model and evaluate its performance; Sc3. Repeat this process k times, each time selecting a different set of data as the validation set; Sc4. Calculate the average value of the prediction accuracy as the model performance of the group of parameter combinations; Sd, compare the model performance of all parameter combinations, and select the parameter combination with the best performance as the optimal parameter combination.