Offshore wind turbine blade corrosion rate prediction method based on IBSO-ESN

Abstract

The offshore wind turbine blade corrosion rate prediction method based on IBSO-ESN comprises the following steps: analyzing an offshore wind turbine blade corrosion principle, and confirming main corrosion factors; preprocessing the model input data, and sorting the preprocessed data into training set data and test set data; combining a differential evolution algorithm fused with a self-adaptive mechanism with a BSO algorithm, and proposing an IBSO algorithm; a variable inertia weight search strategy is adopted, so that the optimization capacity of the IBSO algorithm is further improved; performing parameter optimization on the echo state network by using an IBSO algorithm to obtain an IBSO-ESN model with optimal parameters; inputting the training set data into the IBSO-ESN model to obtain a trained prediction model; and inputting the data of the test set into the trained prediction model, and performing a corrosion rate prediction task of the offshore wind turbine blade. According to the method, the blade corrosion rate can be predicted under the operation state of the offshore wind turbine, and the prediction result and various indexes are superior to those of a traditional ESN model.

Description

Technical Field

[0001] The present invention relates to the technical field of offshore wind turbine maintenance, and in particular to an offshore wind turbine blade corrosion rate prediction method based on IBSO-ESN. Background Art

[0002] Composite materials are widely used in offshore wind turbines, particularly in the manufacture of wind turbine blades, due to their lightweight, high-strength, and corrosion-resistant properties. However, long-term exposure to the complex marine environment can lead to various corrosion and aging issues. After a period of exposure, a certain degree of fiber exposure occurs on the composite surface. Over time, this exposure increases rapidly, and the composite resin becomes increasingly pitted and roughened. Therefore, accurately predicting the corrosion rate of offshore wind turbine blades using factors influencing blade corrosion and promptly understanding the corrosion status of blades are crucial for developing blade maintenance strategies and ensuring their proper operation.

[0003] Although offshore wind turbine blades currently utilize corrosion-resistant composite materials, they are still subject to a range of corrosion in marine environments, leading to structural abnormalities and reduced power generation efficiency. To mitigate the losses caused by corrosion, domestic and international researchers have conducted extensive research on predicting the lifespan and corrosion rate of composite materials. Key prediction methods include machine learning, mechanical properties, empirical models, and physical and chemical models. While these methods offer promising prediction results, they still suffer from several drawbacks. For example, neural networks suffer from high computational complexity and time consumption when used with large sample sizes. Most of these studies focus on static material corrosion predictions in the laboratory, failing to consider the complex dynamic corrosion environment faced by wind turbine blades under actual operating conditions. Furthermore, when considering multiple factors, the crucial offshore salt spray environment is omitted. Summary of the Invention

[0004] To solve the above technical problems, the present invention proposes a method for predicting the corrosion rate of offshore wind turbine blades based on IBSO-ESN. The method designs a corrosion rate prediction model for offshore wind turbine blades; the main factors affecting the corrosion rate are input into the trained offshore wind turbine blade corrosion rate prediction model, and the corrosion rate prediction result is output. The offshore wind turbine blade corrosion rate prediction model of the present invention improves the population diversity of the beetle swarm optimization (BSO) algorithm by using an adaptive differential evolution algorithm to improve the BSO algorithm to obtain an improved beetle swarm optimization (IBSO) algorithm. Subsequently, the improved beetle swarm optimization IBSO algorithm is used to optimize the parameters of the echo state network (ESN) to improve the prediction accuracy of the model. The results show that the present invention has a good effect on predicting the corrosion rate of offshore wind turbine blades.

[0005] The technical solution adopted by the present invention is:

[0006] The offshore wind turbine blade corrosion rate prediction method based on IBSO-ESN includes the following steps:

[0007] Step 1: Analyze the corrosion principle of offshore wind turbine blades, identify the main corrosion factors, and use the main corrosion factors as model input data;

[0008] Step 2: Preprocess the model input data and organize the preprocessed data into training set data and test set data;

[0009] Step 3: Combining the differential evolution algorithm with the adaptive mechanism with the beetle swarm optimization (BSO) algorithm, an improved beetle swarm optimization (IBSO) algorithm is proposed to improve the population diversity of the beetle swarm during the iteration process.

[0010] Step 4: A search strategy with variable inertia weights is used to further enhance the optimization capability of the improved beetle swarm optimization IBSO algorithm. Step 5: The improved beetle swarm optimization IBSO algorithm is used to optimize the parameters of the echo state network (ESN) and obtain the optimal IBSO-ESN model.

[0011] Step 6: Input the training set data into the IBSO-ESN model to obtain the trained prediction model;

[0012] Step 7: Input the test set data into the trained prediction model to perform the offshore wind turbine blade corrosion rate prediction task.

[0013] Step 8: Select evaluation indicators to judge the prediction performance of the IBSO-ESN model, and verify the advancedness of the IBSO-ESN model through comparative experiments.

[0014] In step 2, the model input data is preprocessed, and the specific steps include:

[0015] First, data is obtained from offshore wind farms;

[0016] Secondly, principal components analysis (PCA) is used to reduce the data dimension to obtain more streamlined data, as follows:

[0017] a1: Standardize the sample data and calculate the standardized matrix. Assuming the total number of samples is q and the number of indicators is p, the calculation process is as follows:

[0018]

[0019] In formula (1): is the value of the corresponding item in the standardized matrix; e i,j is the value of the jth indicator in the i-th group of sample data, i = 1, 2, ..., q, j = 1, 2, ..., p; is the average value of the j-th indicator; S j is the sample variance.

[0020] a2: Calculate the correlation coefficient matrix. The calculation process is as follows:

[0021]

[0022] In formula (2): i,j is the value of the corresponding item in the correlation coefficient matrix; is the average value of the i-th indicator; e z,i is the value of the i-th indicator of the z-th group sample; e z,j is the value of the jth indicator of the zth group of samples;

[0023] a3: Calculate the contribution rate A of each indicator j , the calculation process is as follows:

[0024]

[0025] In step 3, an improved beetle swarm optimization (IBSO) algorithm is proposed, which is as follows:

[0026] First, the population is initialized, and during the iteration process, individual longicorn beetles are randomly selected from the population for mutation. The specific calculation formula is as follows:

[0027] X i (k) = [X i,1 (k),X i,2 (k),…,X i,j (k)] (4);

[0028] In formula (1): X i (k) is the individual vector of the i-th longicorn beetle in the k-th iteration; X i,j (k) is the jth gene of the i-th longicorn individual; after the population is initialized, a longicorn individual is randomly selected from the k-th iteration population for mutation, and the mutation vector B is obtained. i (k). The specific calculation formula is as follows:

[0029] B i (k) = X p1 (k)+F i (X p2 (k)-X p3 (k)) (5);

[0030]

[0031] In the above formula: X p1 (k), X p2 (k), X p3 (k) are three randomly selected longicorn individual vectors from the population in the kth iteration, and X p1 (k)≠X p2 (k)≠X p3 (k); F i is the mutation operator after the adaptive mechanism is added. The larger its value is, the less likely it is to fall into a local extreme point. r1 is an independent random number, 0≤r1≤1; ε1 is the adjustment of F i probability.

[0032] Then, in the kth iteration, the newly generated mutant individuals are allowed to cross-recombined with individuals in the original population with a certain probability to enhance the diversity of the population and select the most suitable next generation. The specific calculation formula is as follows:

[0033]

[0034] Where: C i,j (k) is the gene of the crossover offspring; B i,j (k) is the gene of the mutant offspring; is the crossover probability after adding the adaptive mechanism; r2 and r3 are independent random numbers, 0≤r2, r3≤1; ε2 is the adjustment probability.

[0035] After obtaining the crossover offspring, the next generation X with high fitness is selected from the population through the fitness function. i (k+1), the specific calculation formula is as follows:

[0036]

[0037] In formula (9): f(·) is the fitness function; C i (k) is the individual vector after crossover recombination.

[0038] Finally, the speed v of the traditional BSO algorithm is calculated by the above method. i (k) and position x i (k) is updated and modified. The specific calculation formula is as follows:

[0039] T i (k)=ω(k)v i (k)+c1R and +(P i best -x i (k))+c2R and (G best -x i (k)) (10);

[0040]

[0041] H i (k) = x i (k)+v i (k+1) (12);

[0042]

[0043] Where: T i (k) is the transition velocity vector of the kth iteration of the i-th individual longhorn beetle; ω(k) is the inertia weight of the kth iteration; P i best is the optimal position vector of the i-th individual longicorn; G best is the optimal position vector of the longicorn group; c1 and c2 are learning factors; R and is a random vector of dimension j×1, each element of which R and (j,1) are all random numbers in [0,1]; v i (k) is the velocity vector of the k-th iteration of the i-th longicorn individual; H i(k) is the transition position vector of the kth iteration of the i-th individual longicorn; x i (k) is the position vector of the k-th iteration of the i-th individual longicorn; P(·) is the probability function; v i (k+1) is the velocity vector of the k+1th iteration of the i-th longicorn individual; x i (k+1) is the position vector of the i-th longicorn individual at the k+1th iteration.

[0044] In step 4, a search strategy with variable inertia weight is adopted, specifically as follows:

[0045] Maintaining a high inertia weight in the early stages of the search ensures a strong global search capability, thereby effectively exploring the solution space. As the search progresses, the inertia weight gradually decreases, causing the algorithm to gradually shift to local search, thereby improving the accuracy of the solution and the level of detail of the local search. The specific calculation formula is as follows:

[0046]

[0047] In formula (11): k max is the maximum number of iterations; ω max With ω min are the maximum and minimum inertia weights respectively;

[0048] In step 5, obtaining the IBSO-ESN model with optimal parameters includes the following steps:

[0049] First, the preprocessed data is normalized to eliminate the impact of the input data dimension and ensure the generalization ability of the model. At the same time, the variable parameter set θ of the echo state network (ESN) model is initialized and the number of iterations is set to zero to facilitate subsequent operations. The normalization calculation formula is as follows:

[0050]

[0051] In formula (15): a is the original data; is the normalized data; a min and a max are the minimum and maximum values ​​of the original data, respectively.

[0052] Secondly, use the IBSO algorithm to optimize ESN parameters:

[0053] During this process, the position update of the longhorn beetle is achieved through random initialization. The longhorn beetle uses its antennae to sense the odor concentration of the surrounding environment and uses this as a basis for decision-making to determine the movement direction of the population. The initialization process is as follows:

[0054] X i (k) =]Xi,1 (k),X i,2 (k),…,X i,j (k)] (16);

[0055] During the iteration process, the beetle population continuously adjusts its search location and search range to optimize the parameters. The specific adjustment process is as follows:

[0056] T i (k)=ω(k)v i (k)+c1R and +(P i best -x i (k))+c2R and (G best -x i (k)) (17);

[0057]

[0058] H i (k) = x i (k)+v i (k+1) (19);

[0059]

[0060] In the above formula: T i (k) is the transition velocity vector of the kth iteration of the i-th individual longhorn beetle; ω(k) is the inertia weight of the kth iteration; P i best is the optimal position vector of the i-th individual longicorn; G best is the optimal position vector of the longicorn group; c1 and c2 are learning factors; R and is a random vector of dimension j×1, each element of which R and (j,1) are all random numbers in [0,1]; v i (k) is the velocity vector of the k-th iteration of the i-th longicorn individual; H i (k) is the transition position vector of the kth iteration of the i-th individual longicorn; x i (k) is the position vector of the k-th iteration of the i-th individual longicorn; P(·) is the probability function; v i (k+1) is the velocity vector of the k+1th iteration of the i-th longicorn individual; x i (k+1) is the position vector of the i-th longicorn individual at the k+1th iteration.

[0061] When the preset maximum number of iterations is reached, the algorithm stops and the movement of the population also stops. The specific judgment process is as follows:

[0062] k <k max (twenty one);

[0063] Where: k is the current iteration number; k max is the maximum number of iterations.

[0064] Finally, by analyzing the various indicators recorded during the iteration process, the optimization effect of the IBSO algorithm is evaluated, and the optimal parameter model is obtained. The specific indicator calculation process is as follows:

[0065]

[0066]

[0067]

[0068] Where: σ RMSE is the RMSE value; σ MAPE is the MAPE value; y(s) is the sth data in the true value; y'(s) is the sth data in the predicted value; is the average of the true values; n is the number of data.

[0069] This method not only improves the prediction accuracy, but also provides an effective new way to optimize ESN parameters.

[0070] In step 6, the training set data is input into the IBSO-ESN model to obtain a trained prediction model, as follows:

[0071] The training set data in the dataset was used and input into the IBSO-ESN model according to the number of data required for each iteration. The number of model runs was set to 50, the population size was set to 30, and the maximum number of iterations was set to 100. After the training, the best parameter set was saved, and the offshore wind turbine blade corrosion rate prediction model training was completed.

[0072] The present invention provides a method for predicting the corrosion rate of offshore wind turbine blades based on IBSO-ESN, and the technical effects are as follows:

[0073] 1) This invention proposes a method for predicting offshore wind turbine blade corrosion rates based on IBSO-ESN. Compared to the traditional BSO optimization algorithm, the IBSO algorithm of this invention combines the original algorithm with an adaptive differential evolution algorithm. Compared to the traditional ESN model, the IBSO-ESN model of this invention utilizes the IBSO algorithm for parameter optimization. Comparative tests have shown that this invention can predict blade corrosion rates during offshore wind turbine operation, and the prediction results and various indicators are superior to those of the traditional ESN model.

[0074] 2) The method of the present invention is used in the intelligent operation and maintenance of wind turbine blades. This method not only improves the BSO algorithm by combining it with the adaptive differential evolution algorithm to solve the problem of population diversity attenuation in its iterative process and improve the optimization accuracy, but also uses the IBSO algorithm to optimize the parameters of the ESN model to solve the problem that the model parameters are difficult to optimize and improve the prediction accuracy of the model. BRIEF DESCRIPTION OF THE DRAWINGS

[0075] The present invention will be further described below with reference to the accompanying drawings and examples:

[0076] Figure 1 1 is an overall flow chart of the method of the present invention.

[0077] Figure 2 This is a flow chart of the IBSO-ESN prediction model in the method of the present invention. DETAILED DESCRIPTION

[0078] The present invention will be described in further detail below with reference to the accompanying drawings and specific embodiments.

[0079] Example 1:

[0080] This embodiment provides a method for predicting the corrosion rate of offshore wind turbine blades based on IBSO-ESN. First, the IBSO algorithm is proposed by combining the adaptive differential evolution algorithm with the BSO algorithm, which effectively solves the problem that the BSO algorithm is prone to low population diversity and easy to fall into local optimality during the iteration process. Secondly, based on the traditional ESN model, the IBSO algorithm is used to optimize the parameters of the ESN model to obtain the ESN model with the optimal parameters to improve the prediction accuracy. The specific process is as follows: Figure 1 shown.

[0081] Step 1: Collect measured data of different wind turbine blades in offshore wind farms and perform dimensionality reduction on the data using the PCA method to obtain more streamlined data to construct a prediction dataset for offshore wind turbine blade corrosion rate. The constructed dataset is divided into a training set and a test set.

[0082] The specific operation method is as follows: First, obtain data such as corrosion rate, salt spray concentration, light intensity, humidity, temperature, wind speed, and air pressure at the offshore wind farm. Then, use the PCA method to analyze each data and calculate the contribution rate, avoiding data with a contribution rate close to 0. The contribution rate calculation process is as follows: First, standardize the sample data and calculate the normalization matrix. Assuming the total number of samples is q and the number of indicators is p, the calculation process is as follows:

[0083]

[0084] Where: is the value of the corresponding item in the standardized matrix; e i,jis the value of the jth indicator in the i-th group of sample data, i = 1, 2, ..., q, j = 1, 2, ..., p; is the average value of the j-th indicator; S j is the sample variance. Next, find the correlation coefficient matrix. The calculation process is as follows:

[0085]

[0086] Where: i,j is the value of the corresponding item in the correlation coefficient matrix; is the average value of the i-th indicator; e z,i is the value of the i-th indicator of the z-th group sample; e z,j is the value of the jth indicator of the zth group sample. Then calculate the contribution rate A of each indicator j , the calculation process is as follows:

[0087]

[0088] Step 2: Combining the differential evolution algorithm with the adaptive mechanism with the beetle swarm optimization (BSO) algorithm, an improved beetle swarm optimization (IBSO) algorithm is proposed to improve the population diversity of the beetle swarm during the iteration process;

[0089] The specific operation is: first initialize the population, and then randomly select longicorn individuals in the population for mutation during the iteration process. The specific calculation formula is as follows:

[0090] X i (k) =]X i,1 (k),X i,2 (k),…,X i,j (k)]

[0091] Where: X i (k) is the individual vector of the i-th longicorn beetle in the k-th iteration; X i,j (k) is the jth gene of the i-th longicorn individual. After the population is initialized, a longicorn individual is randomly selected from the k-th iteration population for mutation, and the mutation vector B is obtained. i (k). The specific calculation formula is as follows:

[0092] B i (k) = X p1 (k)+F i (X p2 (k)-X p3 (k))

[0093]

[0094] Where: X p1 (k), X p2 (k), X p3 (k) are three randomly selected longicorn individual vectors from the population in the kth iteration, and X p1 (k)≠X p2 (k)≠X p3 (k); F i is the mutation operator after the adaptive mechanism is added. The larger its value is, the less likely it is to fall into a local extreme point. r1 is an independent random number, 0≤r1≤1; ε1 is the adjustment of F i Then, in the kth iteration, the newly generated mutant individuals are allowed to cross-recombine with individuals in the original population with a certain probability to enhance the diversity of the population and select the most suitable next generation. The specific calculation formula is as follows:

[0095]

[0096] Where: C i,j (k) is the gene of the crossover offspring; B i,j (k) is the gene of the mutant offspring; is the crossover probability after adding the adaptive mechanism; r2 and r3 are independent random numbers, 0≤r2, r3≤1; ε2 is the adjustment After obtaining the crossover offspring, the next generation X with high fitness is selected from the population through the fitness function. i (k+1). The specific calculation formula is as follows:

[0097]

[0098] Where: f(·) is the fitness function; C i (k) is the individual vector after cross-recombination. Finally, the speed v of the traditional BSO algorithm is calculated by the above method. i (k) and position x i (k) is updated and modified. The specific calculation formula is as follows:

[0099] T i (k)=ω(k)v i (k)+c1R and +(P i best -x i (k))+c2R and (G best -x i (k))

[0100]

[0101] Hi (k) = x i (k)+v i (k+1)

[0102]

[0103] Where: T i (k) is the transition velocity vector of the kth iteration of the i-th individual longhorn beetle; ω(k) is the inertia weight of the kth iteration; P i best is the optimal position vector of the i-th individual longicorn; G best is the optimal position vector of the longicorn group; c1 and c2 are learning factors; R and is a random vector of dimension j×1, each element of which R and (j,1) are all random numbers in [0,1]; v i (k) is the velocity vector of the k-th iteration of the i-th longicorn individual; H i (k) is the transition position vector of the kth iteration of the i-th individual longicorn; x i (k) is the position vector of the kth iteration of the i-th individual longicorn; P(·) is the probability function. i (k+1) is the velocity vector of the k+1th iteration of the i-th longicorn individual; x i (k+1) is the position vector of the i-th longicorn individual at the k+1th iteration.

[0104] Step 3: Adopt a search strategy with variable inertia weight to further improve the optimization capability of IBSO;

[0105] The specific operation is: in the early stages of the search, a high inertia weight is maintained to ensure a strong global search capability, thereby effectively exploring the solution space. As the search progresses, the inertia weight gradually decreases, causing the algorithm to gradually shift to local search, thereby improving the accuracy of the solution and the level of detail of the local search. The specific calculation formula is as follows:

[0106]

[0107] Where: k max is the maximum number of iterations; ω max With ω min are the maximum and minimum inertia weights respectively.

[0108] Step 4: If Figure 2 As shown in the figure, IBSO is used to optimize the parameters of the echo state network (ESN) and the IBSO-ESN model with the optimal parameters is obtained;

[0109] The specific operation is as follows: First, the original data is normalized to eliminate the influence of the input data dimension and ensure the generalization ability of the model. At the same time, the variable parameter set θ of the ESN model is initialized and the number of iterations is set to zero to facilitate subsequent operations. The specific calculation formula for normalization is as follows:

[0110]

[0111] Where: a is the original data; is the normalized data; a min and a max are the minimum and maximum values ​​of the original data, respectively. Secondly, the IBSO algorithm is used to optimize the ESN parameters. In this process, the position update of the longhorn beetle is achieved through random initialization. The longhorn beetle perceives the odor concentration of the surrounding environment through its tentacles, and uses this as a basis for decision-making to determine the movement direction of the population. During the iteration process, the longhorn beetle population continuously adjusts its search position and search range to achieve parameter optimization. When the preset maximum number of iterations is reached, the algorithm stops, and the movement of the population is also terminated accordingly. Finally, by analyzing the optimal prediction error and the corresponding optimal position recorded during the iteration process, the optimization effect of the IBSO algorithm is evaluated, the optimal parameter set of the ESN is determined, and an ESN network with optimal parameters is obtained. This method not only improves the accuracy of the prediction, but also provides an effective new way to optimize ESN parameters.

[0112] Step 5: Train the model to obtain a trained offshore wind turbine blade crack corrosion rate prediction model. Select evaluation indicators to judge the prediction performance of the model, and verify the advancement of the model through comparative tests.

[0113] The specific operation is as follows: Use the training set data in the dataset and input it into the designed IBSO-ESN offshore wind turbine blade corrosion rate prediction model according to the number of data required for each iteration. The number of model runs is set to 50, the population size is set to 30, and the maximum number of iterations is set to 100. After the training is completed, the best parameter set is saved, and the offshore wind turbine blade corrosion rate prediction model training is completed. The network performance is expressed as root mean square error (RMSE), mean absolute percentage error (MAPE), and determination coefficient R 2 Three key indicators are quantitatively analyzed. The specific calculation formula is as follows:

[0114]

[0115]

[0116]

[0117] Where: σ RMSE is the RMSE value; σ MAPE is the MAPE value; y(s) is the sth data in the true value; y'(s) is the sth data in the predicted value; is the average of the true values; n is the number of data.

[0118] Step 6: Input the test set data into the trained IBSO-ESN offshore wind turbine blade corrosion rate prediction model and output the corrosion rate prediction results.

[0119] The present invention completes the ESN and IBSO-ESN model training on the actual offshore wind farm data set training set, and uses the test set to evaluate the model prediction effect. Table 1 shows the comparison results of the model prediction effect.

[0120] Table 1 Comparison of prediction performance of each model

[0121]

[0122] The RMSE value, MAPE value, R 2 They reached 0.188%, 1.526%, and 0.958 respectively. Compared with the traditional ESN, the RMSE value decreased by 0.306, the MAPE value decreased by 4.381 percentage points, and the R 2 The improvement is 9.93%, which proves that the method of the present invention has certain advancement.

[0123] This method for predicting offshore wind turbine blade corrosion rates addresses the low population diversity and local optima inherent in the beetle swarm optimization (BSO) algorithm. By integrating the IBSO algorithm with a differential evolution algorithm, this method optimizes the parameters of an ESN. Finally, the ESN with the optimal parameters is used to predict offshore wind turbine blade corrosion rates. Results demonstrate that this method outperforms traditional ESN in predicting performance.

Claims

1. The offshore wind turbine blade corrosion rate prediction method based on IBSO-ESN is characterized by The following steps are involved: Step 1: Analyze the corrosion principle of offshore wind turbine blades, identify the corrosion factors, and use the corrosion factors as model input data; Step 2: Preprocess the model input data and organize the preprocessed data into training set data and test set data; Step 3: Combine the differential evolution algorithm with the adaptive mechanism with the beetle swarm optimization BSO algorithm to propose an improved beetle swarm optimization IBSO algorithm to improve the population diversity of the beetle swarm during the iteration process; Step 4: Use a search strategy with variable inertia weight to further enhance the optimization capability of the improved beetle swarm optimization IBSO algorithm; Step 5: Use the improved beetle swarm optimization IBSO algorithm to optimize the parameters of the echo state network (ESN) and obtain the IBSO-ESN model with the optimal parameters. Step 6: Input the training set data into the IBSO-ESN model to obtain the trained prediction model; Step 7: Input the test set data into the trained prediction model to perform the offshore wind turbine blade corrosion rate prediction task.

2. The offshore wind turbine blade corrosion rate prediction method based on IBSO-ESN according to claim 1 is characterized by: The method also includes step 8: selecting evaluation indicators to judge the prediction performance of the IBSO-ESN model, and verifying the advanced nature of the IBSO-ESN model through comparative experiments.

3. The offshore wind turbine blade corrosion rate prediction method based on IBSO-ESN according to claim 1 is characterized by: In step 2, the model input data is preprocessed, and the specific steps include: First, data is obtained from offshore wind farms; Secondly, principal component analysis (PCA) is used to reduce the data dimension to obtain more streamlined data.

4. The offshore wind turbine blade corrosion rate prediction method based on IBSO-ESN according to claim 3 is characterized by: The principal component analysis method PCA is used to reduce the dimension of the data, which includes the following steps: a1: Standardize the sample data and calculate the standardized matrix. Assuming the total number of samples is q and the number of indicators is p, the calculation process is as follows: In formula (1): is the value of the corresponding item in the standardized matrix; e i,j is the value of the jth indicator in the i-th group of sample data, i = 1, 2, ..., q, j = 1, 2, ..., p; is the average value of the j-th indicator; S j is the sample variance; a2: Calculate the correlation coefficient matrix. The calculation process is as follows: In formula (2): i,j is the value of the corresponding item in the correlation coefficient matrix; is the average value of the i-th indicator; e z,i is the value of the i-th indicator of the z-th group sample; e z,j is the value of the jth indicator of the zth group of samples; a3: Calculate the contribution rate A of each indicator j , the calculation process is as follows:

5. The offshore wind turbine blade corrosion rate prediction method based on IBSO-ESN according to claim 1 is characterized by: In step 3, an improved beetle swarm optimization IBSO algorithm is proposed, as follows: First, the population is initialized, and during the iteration process, individual longicorn beetles in the population are randomly selected for mutation. The specific calculation formula is as follows: X i (k)=[X i,1 (k),X i,2 (k),…,X i,j (k)] (4); In formula (1): X i (k) is the individual vector of the i-th longicorn beetle in the k-th iteration; X i,j (k) is the jth gene of the i-th longicorn individual; after the population is initialized, a longicorn individual is randomly selected from the k-th iteration population for mutation, and the mutation vector B is obtained. i (k); the specific calculation formula is as follows: B i (k)=X p1 (k)+F i (X p2 (k)-X p3 (k)) (5); In the above formula: X p1 (k), X p2 (k), X p3 (k) are three randomly selected longicorn individual vectors from the population in the kth iteration, and X p1 (k)≠X p2 (k)≠X p3 (k); F i is the mutation operator after the adaptive mechanism is added. The larger its value is, the less likely it is to fall into a local extreme point. r1 is an independent random number, 0≤r1≤1; ε1 is the adjustment of F i probability; Then, in the kth iteration, the newly generated mutant individuals are allowed to cross-recombined with individuals in the original population with a certain probability to enhance the diversity of the population and select the most suitable next generation; the specific calculation formula is as follows: Where: C i,j (k) is the gene of the crossover offspring; B i,j (k) is the gene of the mutant offspring; is the crossover probability after adding the adaptive mechanism; r2 and r3 are independent random numbers, 0≤r2, r3≤1; ε2 is the adjustment probability; After obtaining the crossover offspring, the next generation X with high fitness is selected from the population through the fitness function. i (k+1), the specific calculation formula is as follows: In formula (9): f(·) is the fitness function; C i (k) is the individual vector after crossover recombination; Finally, the speed v of the traditional BSO algorithm is calculated by the above method. i (k) and position x i (k) Update and modify; the specific calculation formula is as follows: T i (k)=ω(k)v i (k)+c1R and +(P i best -x i (k))+c2R and (G best -x i (k)) (10); H i (k)=x i (k)+v i (k+1) (12); Where: T i (k) is the transition velocity vector of the kth iteration of the i-th individual longhorn beetle; ω(k) is the inertia weight of the kth iteration; P i best is the optimal position vector of the i-th individual longicorn; G best is the optimal position vector of the longicorn group; c1 and c2 are learning factors; R and is a random vector of dimension j×1, each element of which R and (j,1) are all random numbers in [0,1]; v i (k) is the velocity vector of the k-th iteration of the i-th longicorn individual; H i (k) is the transition position vector of the kth iteration of the i-th individual longicorn; x i (k) is the position vector of the k-th iteration of the i-th individual longicorn; P(·) is the probability function; v i (k+1) is the velocity vector of the k+1th iteration of the i-th longicorn individual; x i (k+1) is the position vector of the i-th longicorn individual at the k+1th iteration.

6. The offshore wind turbine blade corrosion rate prediction method based on IBSO-ESN according to claim 5 is characterized by: In step 4, a search strategy with variable inertia weight is adopted, and the specific calculation formula is as follows: In formula (11): k max is the maximum number of iterations; ω max With ω min are the maximum and minimum inertia weights respectively.

7. The offshore wind turbine blade corrosion rate prediction method based on IBSO-ESN according to claim 5 is characterized by: The step 5 comprises the following steps: First, the preprocessed data is normalized to eliminate the influence of the input data dimension. At the same time, the variable parameter set θ of the echo state network (ESN) model is initialized and the number of iterations is set to zero. The normalization calculation formula is as follows: In formula (15): a is the original data; is the normalized data; a min and a max are the minimum and maximum values ​​of the original data respectively; Secondly, use the IBSO algorithm to optimize ESN parameters: During this process, the position update of the longhorn beetle is achieved through random initialization. The longhorn beetle senses the odor concentration of the surrounding environment through its antennae and uses this as a basis for decision-making to determine the movement direction of the population. The initialization process is as follows: X i (k)=[X i,1 (k),X i,2 (k),…,X i,j (k)] (16); During the iteration process, the beetle population continuously adjusts its search location and search range to optimize the parameters. The specific adjustment process is as follows: T i (k)=ω(k)v i (k)+c1R and +(P i best -x i (k))+c2R and (G best -x i (k)) (17); H i (k)=x i (k)+v i (k+1) (19); In the above formula: T i (k) is the transition velocity vector of the kth iteration of the i-th individual longhorn beetle; ω(k) is the inertia weight of the kth iteration; P i best is the optimal position vector of the i-th individual longicorn; G best is the optimal position vector of the longicorn group; c1 and c2 are learning factors; R and is a random vector of dimension j×1, each element of which R and (j,1) are all random numbers in [0,1]; v i (k) is the velocity vector of the k-th iteration of the i-th longicorn individual; H i (k) is the transition position vector of the kth iteration of the i-th individual longicorn; x i (k) is the position vector of the k-th iteration of the i-th individual longicorn; P(·) is the probability function; v i (k+1) is the velocity vector of the k+1th iteration of the i-th longicorn individual; x i (k+1) is the position vector of the i-th longicorn individual at the k+1th iteration; When the preset maximum number of iterations is reached, the algorithm stops, and the movement of the population also stops accordingly. The specific judgment process is as follows: k<k max (21); Where: k is the current iteration number; k max is the maximum number of iterations; Finally, by analyzing the various indicators recorded during the iteration process, the optimization effect of the IBSO algorithm is evaluated, and the optimal parameter model is obtained.

8. The offshore wind turbine blade corrosion rate prediction method based on IBSO-ESN according to claim 7 is characterized by: In step 6, the training set data is input into the IBSO-ESN model to obtain a trained prediction model, specifically as follows: using the training set data in the dataset, inputting it into the IBSO-ESN model according to the number of data required for each iteration; setting the number of model runs to 50, the population size to 30, and the maximum number of iterations to 100; after the training is completed, saving the best parameter set, thus completing the training of the offshore wind turbine blade corrosion rate prediction model.

9. The offshore wind turbine blade corrosion rate prediction method based on IBSO-ESN according to claim 2 is characterized by: Evaluation indicators include: Where: σ RMSE is the RMSE value; σ MAPE is the MAPE value; y(s) is the sth data in the true value; y'(s) is the sth data in the predicted value; is the average of the true values; n is the number of data.

Images