A method for evaluating the adjustable capacity of offshore wind farms considering the calculation of submarine cable losses

Through spectral clustering and BP neural network, the submarine cable loss mapping model of offshore wind farms was established, which solved the problem of large calculation error of offshore wind farms in the calculation of submarine cables, realized real-time adjustable capacity evaluation, improved AGC regulation performance, and promoted offshore Coordinated development of wind power and power grid.

CN114723299BActive Publication Date: 2024-11-01JIANGSU FRONTIER ELECTRIC TECH
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202210382206.X
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2022-04-12
Publication Date
2024-11-01
Estimated Expiration
2042-04-12

AI Technical Summary

Technical Problem

The existing offshore wind farm submarine cable loss calculation model has a large error, which affects the accuracy and response speed of automatic power generation control (AGC) adjustment, and cannot accurately and quickly estimate the adjustable capacity of the network connection point.

Method used

By inputting the SCADA historical data of the fan of offshore wind farm and the historical data of the operation and maintenance system, based on spectral clustering and BP neural network, a mapping of wind turbine clustering power to offshore wind farm collecting line power and submarine cable power loss is established, and a adjustable Capacity evaluation model to achieve real-time adjustable capacity evaluation.

Benefits of technology

It effectively improves the accuracy of adjustable capacity estimates, improves AGC regulation performance, and promotes the coordinated development of offshore wind power and power grid.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN114723299B_ABST
    Figure CN114723299B_ABST
Patent Text Reader

Abstract

The present invention discloses an offshore wind farm adjustable capacity evaluation method considering the calculation of submarine cable line loss: 1) inputting offshore wind farm wind turbine SCADA historical data and offshore wind farm operation and maintenance system historical data; 2) performing spectral clustering on wind turbines based on offshore wind farm wind turbine SCADA historical data; 3) establishing a mapping of wind turbine clustering power to offshore wind farm collector line power; 4) establishing a mapping of offshore wind farm collector line power to submarine cable power loss; 5) establishing an adjustable capacity evaluation model based on offshore wind farm collector line power and submarine cable power loss; 6) evaluating the adjustable capacity of the wind farm based on offshore wind farm wind turbine SCADA real-time information.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention belongs to the field of offshore wind power generation, and in particular, relates to an offshore wind farm adjustable capacity assessment method taking into account the calculation of submarine cable line losses. Background Art

[0002] Offshore wind power generation is an important way to utilize new energy. With the increasing penetration of offshore wind power in the power grid, its randomness and volatility characteristics have brought certain impacts to the power grid. After the wind turbines in the offshore wind farm generate power, they are collected by the power collection system, and then transmitted to the onshore grid through the high-voltage submarine cable after being boosted at the booster station, generating power loss in the process. The existing cable loss calculation model does not have a specific calculation standard specifically for submarine cables, resulting in large errors in the calculation of submarine cable losses in offshore wind farms. Automatic Generation Control (AGC) cannot accurately and quickly estimate the adjustable capacity of the grid connection point, which in turn affects the accuracy and response speed of AGC regulation. Considering the use of collector power to calculate the cable loss, a mapping of wind turbine power to grid connection point power is established based on a neural network, and finally the use of the real-time maximum power generation capacity of wind turbines to evaluate the real-time maximum adjustable capacity of wind farms is realized, providing a basis for grid dispatching and promoting offshore wind power consumption. Summary of the invention

[0003] Technical problem: A method for evaluating the adjustable capacity of offshore wind farms taking into account the calculation of submarine cable line losses is proposed. This method can provide a basis for grid dispatching and promote the coordinated development of offshore wind power and power grids.

[0004] Technical solution: The method for evaluating the adjustable capacity of an offshore wind farm taking into account the calculation of submarine cable line loss of the present invention comprises the following steps:

[0005] 1) Input the offshore wind farm wind turbine SCADA historical data and the offshore wind farm operation and maintenance system historical data;

[0006] 2) Spectral clustering of wind turbines based on offshore wind farm wind turbine SCADA historical data;

[0007] 3) Establish a mapping from the cluster power of wind turbines to the power of offshore wind farm collector lines;

[0008] 4) Establish a mapping from offshore wind farm collector power to submarine cable power loss;

[0009] 5) Establish an adjustable capacity assessment model based on offshore wind farm collector power and submarine cable power loss;

[0010] 6) Evaluate the adjustable capacity of the wind farm based on the real-time SCADA information of offshore wind farm wind turbines.

[0011] In step 1), the input wind turbine SCADA historical data includes the historical active power P of all wind turbines in the offshore wind farm. wl ; The input offshore wind farm operation and maintenance system historical data includes offshore collector power P gl , power P at the head end of the submarine cable sl , power at the end of the submarine cable P ml 、Offshore booster station power P hl , Power of onshore centralized control center P ll ; The input time of historical data is 30 days, the time interval is 15 minutes, and the time dimension of various time series data is T = 30 × 24 × 4 = 2880; historical active power P wl The dimension is a×T, where a is the number of wind turbines; the offshore collector power P gl The dimension is b×T, where b is the number of collector wires; the power at the head end of the submarine cable is P sl , power at the end of the submarine cable P ml , offshore booster station power P hl and the power of the onshore centralized control center P ll The dimensions are all 1×T.

[0012] In the step 2), the method for spectral clustering of wind turbines based on the offshore wind farm wind turbine SCADA historical data is as follows:

[0013] 201) For any power sequence phasor combination of wind turbine i and wind turbine j in an offshore wind farm, calculate its Pearson correlation coefficient and denote it as p ij , thus constructing the similarity matrix P:

[0014]

[0015] 202) Based on the similarity matrix P, calculate the degree matrix H:

[0016]

[0017] The matrix is ​​a diagonal matrix, and the element h i is the sum of the elements in the i-th row of the matrix P;

[0018] 204) Construct the Laplace matrix L and normalize it:

[0019] L=HP (3)

[0020] L′=H -0.5 LH -0.5 (4)

[0021] 205) Determine the number of spectral clusters m, find the first m minimum eigenvalues ​​and corresponding eigenvectors of L′, normalize the eigenvectors and construct a new matrix U a×m ;

[0022] 206) For the matrix U a×m The row vectors of are clustered using K-means to obtain the m clusters of wind turbines C1, C2, ..., C m , realizing spectral clustering of wind turbines.

[0023] In step 3), the mapping method of establishing the wind turbine cluster power to the offshore wind farm collector power is as follows:

[0024] 301) Based on the cluster division results in the spectral clustering of wind turbines, the historical sequence Z of the total power of wind turbines in m clusters is calculated:

[0025]

[0026] Among them, the i-th row vector of Z represents the historical time series of the total power of wind turbines in the i-th cluster;

[0027] 302) Based on BP neural network, a mapping from wind turbine cluster power to collector power is established:

[0028] The 2880 column vectors of the total power history sequence Z of the wind turbines in the m clusters are used as the input of the BP neural network training set. gl The 2880 column vectors are used as the output of the BP neural network training set. The number of input and output samples of the neural network training is 2880.

[0029] In the BP neural network, the input vector is X=(x1,x2,...,x n ) T ; The hidden layer output is Y = (y1, y2, ..., y m ) T , the output layer output is O = (o1, o2, ..., o l ) T , the expected value is D=(d1,d2,...,d l ) T ,ω ij ,ω jk are the weights from the input layer to the hidden layer and the weights from the hidden layer to the output layer respectively;

[0030] Calculate the hidden layer output

[0031]

[0032] Calculate the output layer output

[0033]

[0034] In the above two formulas, f(g) is the activation function of the BP neural network, and the output error E of the network is as follows:

[0035]

[0036] Substituting equations (6) and (7) into error expression (8), we get equation (9):

[0037]

[0038] From formula (9), we can see that the output error E of the network is a weight value ω ij ,ω jk The function adjusts the weights to reduce the output error of the network. When the output error E of the network is less than the expected value or the number of training times reaches the standard, the network training ends.

[0039] The weights are updated along the negative gradient direction of the network's output error to reduce the error.

[0040]

[0041]

[0042] η is the learning rate, ranging from 0 to 1; the network weights are iteratively updated according to the increments obtained in the above formula.

[0043] ω ij (n+1)=ω ij (n)+Δω ij (13)

[0044] ω jk (n+1)=ω jk (n)+Δω jk (14)

[0045] After the update is completed, the information is forwarded again until the number of training times is greater than the target value or the error is less than the target value. Based on the above BP neural network training process, a mapping from wind turbine cluster power to collector line power is established:

[0046] P G =F1(P Z ) (15)

[0047] Among them, F1 represents the trained neural network model, P Z represents the clustered power of a group of wind turbines, P G Indicates the corresponding collector power.

[0048] In step 4), a mapping method for establishing the power loss of the offshore wind farm collector line to the power loss of the submarine cable is as follows:

[0049] 401) Calculate the historical data of submarine cable power loss:

[0050] P xl =P sl -P ml (16)

[0051] Among them, P xl is the historical data of submarine cable power loss, with a dimension of 1×b, where b is the number of collector wires;

[0052] 402) Taking the power of the offshore wind farm collector line P gl As input, the submarine cable power loss P xl For output, based on the BP neural network described in step 302), a mapping of the offshore wind farm collector power to the submarine cable power loss is established:

[0053] P X =F2(P G ) (17)

[0054] Among them, P G is a group of collector power, P X Represents the corresponding submarine cable power loss.

[0055] In step 5), an adjustable capacity assessment model based on the offshore wind farm collector power and submarine cable power loss is established as follows:

[0056] P K =sum(P G )-F2(P G )-ave(P hl +P ll ) (18)

[0057] Among them, sum() represents vector summation, and ave() represents vector mean.

[0058] In step 6), the model for evaluating the adjustable capacity of the wind farm based on the real-time SCADA information of the wind turbines in the offshore wind farm is as follows:

[0059] 601) reads the real-time maximum power generation capacity of the wind turbine in the offshore wind farm wind turbine SCADA real-time information, and based on the spectral clustering results of the m clusters of wind turbines in step 206), calculates the sum of the maximum power generation capacity of the wind turbines in the m clusters to form an m-dimensional vector, denoted as P ZS ;

[0060] 602) The calculation formula for the real-time adjustable capacity of an offshore wind farm is:

[0061] P KS =sum(F1(PZS ))-F2(F1(P ZS ))-ave(P hl +P ll ) (19)

[0062] Among them, P KS Indicates the real-time maximum adjustable capacity of the wind farm.

[0063] Beneficial effects: Compared with the prior art, the present invention has the following advantages: using the power of the collector line to calculate the line loss of the submarine cable, establishing a mapping of the wind turbine power to the grid connection point power based on a neural network, and finally realizing the use of the real-time maximum power generation capacity of the wind turbine to evaluate the real-time maximum adjustable capacity of the wind farm, providing a basis for grid dispatching and promoting offshore wind power consumption; this method can effectively improve the accuracy of adjustable capacity estimation, thereby improving the AGC regulation performance of offshore wind farms and promoting the coordinated development of offshore wind power and power grids. BRIEF DESCRIPTION OF THE DRAWINGS

[0064] Figure 1 is a flow chart of the present invention;

[0065] Figure 2 This is the spectral clustering result diagram of wind turbines;

[0066] Figure 3 This is the result diagram of the collector power calculation;

[0067] Figure 4 This is the result diagram of submarine cable loss calculation;

[0068] Figure 5 This is the result diagram of adjustable capacity estimation. DETAILED DESCRIPTION

[0069] In order to make the purpose, technical solution and advantages of the present invention more clearly understood, the present invention is described in detail below in conjunction with the accompanying drawings and implementation examples. It should be understood that the specific implementation examples described herein are only used to explain the present invention and are not used to limit the invention.

[0070] like Figure 1 As shown, a method for evaluating the adjustable capacity of an offshore wind farm considering the calculation of submarine cable line loss in an embodiment of the present invention includes the following steps:

[0071] 1) Input the offshore wind farm wind turbine SCADA historical data and the offshore wind farm operation and maintenance system historical data;

[0072] 2) Spectral clustering of wind turbines based on offshore wind farm wind turbine SCADA historical data;

[0073] 3) Establish a mapping from the cluster power of wind turbines to the power of offshore wind farm collector lines;

[0074] 4) Establish a mapping from offshore wind farm collector power to submarine cable power loss;

[0075] 5) Establish an adjustable capacity assessment model based on offshore wind farm collector power and submarine cable power loss;

[0076] 6) Evaluate the adjustable capacity of the wind farm based on the real-time SCADA information of offshore wind farm wind turbines.

[0077] In step 1), the input wind turbine SCADA historical data includes the historical active power P of all wind turbines in the offshore wind farm. wl ; The input offshore wind farm operation and maintenance system historical data includes offshore collector power P gl , power P at the head end of the submarine cable sl , power at the end of the submarine cable P ml 、Offshore booster station power P hl , Power of onshore centralized control center P ll ; The input time of historical data is 30 days, the time interval is 15 minutes, and the time dimension of various time series data is T = 30 × 24 × 4 = 2880; historical active power P wl The dimension is a×T, where a is the number of wind turbines; the offshore collector power P gl The dimension is b×T, where b is the number of collector wires; the power at the head end of the submarine cable is P sl , power at the end of the submarine cable P ml , offshore booster station power P hl and the power of the onshore centralized control center P ll The dimensions are all 1×T.

[0078] In the step 2), the method for spectral clustering of wind turbines based on the offshore wind farm wind turbine SCADA historical data is as follows:

[0079] 201) For any power sequence phasor combination of wind turbine i and wind turbine j in an offshore wind farm, calculate its Pearson correlation coefficient and denote it as p ij , thus constructing the similarity matrix P:

[0080]

[0081] 202) Based on the similarity matrix P, calculate the degree matrix H:

[0082]

[0083] The matrix is ​​a diagonal matrix, and the element h i is the sum of the elements in the i-th row of the matrix P;

[0084] 204) Construct the Laplace matrix L and normalize it:

[0085] L=HP (3)

[0086] L′=H -0.5 LH -0.5 (4)

[0087] 205) Determine the number of spectral clusters m, find the first m minimum eigenvalues ​​and corresponding eigenvectors of L′, normalize the eigenvectors and construct a new matrix U a×m ;

[0088] 206) For the matrix U a×m The row vectors of are clustered using K-means to obtain the m clusters of wind turbines C1, C2, LC m , realizing spectral clustering of wind turbines.

[0089] In step 3), the mapping method of establishing the wind turbine cluster power to the offshore wind farm collector power is as follows:

[0090] 301) Based on the cluster division results in the spectral clustering of wind turbines, the historical sequence Z of the total power of wind turbines in m clusters is calculated:

[0091]

[0092] Among them, the i-th row vector of Z represents the historical time series of the total power of wind turbines in the i-th cluster;

[0093] 302) Based on BP neural network, a mapping from wind turbine cluster power to collector power is established:

[0094] The 2880 column vectors of the total power history sequence Z of the wind turbines in the m clusters are used as the input of the BP neural network training set. gl The 2880 column vectors are used as the output of the BP neural network training set. The number of input and output samples of the neural network training is 2880.

[0095] In the BP neural network, the input vector is X=(x1,x2,...,x n ) T ; The hidden layer output is Y = (y1, y2, ..., y m ) T , the output layer output is O = (o1, o2, ..., o l ) T , the expected value is D=(d1,d2,...,d l ) T ,ω ij ,ω jk are the weights from the input layer to the hidden layer and the weights from the hidden layer to the output layer respectively;

[0096] Calculate the hidden layer output

[0097]

[0098] Calculate the output layer output

[0099]

[0100] In the above two formulas, f(g) is the activation function of the BP neural network, and the output error E of the network is as follows:

[0101]

[0102] Substituting equations (6) and (7) into error expression (8), we get equation (9):

[0103]

[0104] From formula (9), we can see that the output error E of the network is a weight value ω ij ,ω jk The function adjusts the weights to reduce the output error of the network. When the output error E of the network is less than the expected value or the number of training times reaches the standard, the network training ends.

[0105] The weights are updated along the negative gradient direction of the network's output error to reduce the error.

[0106]

[0107]

[0108] η is the learning rate, ranging from 0 to 1; the network weights are iteratively updated according to the increments obtained in the above formula.

[0109] ω ij (n+1)=ω ij (n)+Δω ij (13)

[0110] ω jk (n+1)=ω jk (n)+Δω jk (14)

[0111] After the update is completed, the information is forwarded again until the number of training times is greater than the target value or the error is less than the target value. Based on the above BP neural network training process, a mapping from wind turbine cluster power to collector line power is established:

[0112] P G =F1(P Z ) (15)

[0113] Among them, F1 represents the trained neural network model, P Z represents the clustered power of a group of wind turbines, P G Indicates the corresponding collector power.

[0114] In step 4), a mapping method for establishing the power loss of the offshore wind farm collector line to the power loss of the submarine cable is as follows:

[0115] 401) Calculate the historical data of submarine cable power loss:

[0116] P xl =P sl -P ml (16)

[0117] Among them, P xl is the historical data of submarine cable power loss, with a dimension of 1×b, where b is the number of collector wires;

[0118] 402) Taking the power of the offshore wind farm collector line P gl As input, the submarine cable power loss P xl For output, based on the BP neural network described in step 302), a mapping of the offshore wind farm collector power to the submarine cable power loss is established:

[0119] P X =F2(P G ) (17)

[0120] Among them, P G is a group of collector power, P X Represents the corresponding submarine cable power loss.

[0121] In step 5), an adjustable capacity assessment model based on the offshore wind farm collector power and submarine cable power loss is established as follows:

[0122] P K =sum(P G )-F2(P G )-ave(P hl +P ll ) (18)

[0123] Among them, sum() represents vector summation, and ave() represents vector mean.

[0124] In step 6), the model for evaluating the adjustable capacity of the wind farm based on the real-time SCADA information of the wind turbines in the offshore wind farm is as follows:

[0125] 601) reads the real-time maximum power generation capacity of the wind turbine in the offshore wind farm wind turbine SCADA real-time information, and based on the spectral clustering results of the m clusters of wind turbines in step 206), calculates the sum of the maximum power generation capacity of the wind turbines in the m clusters to form an m-dimensional vector, denoted as P ZS ;

[0126] 602) The calculation formula for the real-time adjustable capacity of an offshore wind farm is:

[0127] P KS =sum(F1(P ZS ))-F2(F1(P ZS ))-ave(P hl +P ll ) (19)

[0128] Among them, P KS Indicates the real-time maximum adjustable capacity of the wind farm.

[0129] Example

[0130] In this embodiment, the collected data of 67 wind turbines in an offshore wind farm in Jiangsu are used, and the historical data of the offshore wind farm for 30 days is input, with an interval of 15 minutes. First, the wind turbines are spectrally clustered based on the historical power. The clustering results are as follows: Figure 2 As shown in the figure, the fan does not represent the actual spatial location. Figure 2 It can be seen that the 67 wind turbines are grouped into 12 categories.

[0131] According to the clustering results, the wind turbine power is processed, and then the processed wind turbine power is used to map the collector power. The calculation results are as follows: Figure 3 Compared with the case where the original power sequence is used directly for mapping without clustering, the mean absolute error (MAE) and root mean square error (RMSE) are used as evaluation indicators. The error comparison is shown in Table 1:

[0132] Table 1 Comparison of collector line power calculation errors with and without spectral clustering

[0133] MAE / MW RMSE / MW Spectral Clustering 1.63 2.14 No spectral clustering 1.96 2.83

[0134] It can be seen that the error of the power calculation of the collector line using spectral clustering is significantly reduced. The calculated collector line power is used to map the power loss of the submarine cable. The results are as follows: Figure 4 Compared with the calculation of submarine cable power loss using wind turbine power, the error comparison is shown in Table 2.

[0135] Table 2 Comparison of calculation errors of submarine cable power loss using collector power and wind turbine power

[0136] MAE / MW RMSE / MW Collector power 0.18 0.23 Fan power 0.24 0.31

[0137] The results show that using the power of the collector line to calculate the power loss of the submarine cable is more accurate. Finally, the wind turbine power is summed up and the adjustable capacity is obtained after subtracting various losses. The results are as follows: Figure 5 shown.

[0138] The step-by-step mapping method proposed in the present invention is compared with the currently used method of directly mapping the adjustable capacity using the fan power. The error comparison is shown in Table 3.

[0139] Table 3 Comparison of adjustable capacity error between the method proposed in this invention and direct mapping of wind turbine power

[0140]

[0141] The above error comparison shows that the error of the evaluation method provided by the present invention is effectively reduced, which verifies the effectiveness and superiority of the provided method over the prior art. The method provided by the present invention uses the power of the collector line to calculate the power loss of the submarine cable, and establishes a mapping of the wind turbine power to the grid connection point power based on a neural network, and finally realizes the use of the real-time maximum power generation capacity of the wind turbine to evaluate the real-time maximum adjustable capacity of the wind farm, providing a basis for grid dispatch and promoting offshore wind power consumption; this method can effectively improve the accuracy of adjustable capacity estimation, thereby improving the AGC regulation performance of offshore wind farms and promoting the coordinated development of offshore wind power and power grids. The error is greatly reduced, which verifies the effectiveness and superiority of the proposed method.

Claims

1. A method for evaluating the adjustable capacity of offshore wind farms considering the calculation of submarine cable line losses. It is characterized in that The method comprises the following steps: 1) Input the offshore wind farm wind turbine SCADA historical data and the offshore wind farm operation and maintenance system historical data; 2) Spectral clustering of wind turbines based on offshore wind farm wind turbine SCADA historical data; 3) Establish a mapping from the cluster power of wind turbines to the power of offshore wind farm collector lines; 4) Establish a mapping from offshore wind farm collector power to submarine cable power loss; 5) Establish an adjustable capacity assessment model based on offshore wind farm collector power and submarine cable power loss; 6) Evaluate the adjustable capacity of wind farms based on real-time SCADA information of offshore wind farm wind turbines; In step 1), the input wind turbine SCADA historical data includes the historical active power P of all wind turbines in the offshore wind farm. wl ; The input offshore wind farm operation and maintenance system historical data includes offshore collector power P gl , power P at the head end of the submarine cable sl , power at the end of the submarine cable P ml 、Offshore booster station power P hl , Power of onshore centralized control center P ll ; The input time of historical data is 30 days, the time interval is 15 minutes, and the time dimension of various time series data is T = 30 × 24 × 4 = 2880; historical active power P wl The dimension is a×T, where a is the number of wind turbines; the offshore collector power P gl The dimension is b×T, where b is the number of collector wires; the power at the head end of the submarine cable is P sl , power at the end of the submarine cable P ml , offshore booster station power P hl and the power of the onshore centralized control center P ll The dimensions are all 1×T; In the step 2), the method for spectral clustering of wind turbines based on the offshore wind farm wind turbine SCADA historical data is as follows: 201) For any power sequence vector combination of wind turbine i and wind turbine j in an offshore wind farm, calculate its Pearson correlation coefficient and denote it as p ij , thus constructing the similarity matrix P: 202) Based on the similarity matrix P, calculate the degree matrix H: The matrix is ​​a diagonal matrix, and the element h i is the sum of the elements in the i-th row of the matrix P; 204) Construct the Laplace matrix L and normalize it: L=HP(3) L′=H -0.5 LH -0.5 (4) 205) Determine the number of spectral clusters m, find the first m minimum eigenvalues ​​and corresponding eigenvectors of L′, normalize the eigenvectors and construct a new matrix U a×m ; 206) For the matrix U a×m The row vectors of are clustered using K-means to obtain the division of wind turbines into m clusters C 1 ,C 2 ,…C m , realize spectral clustering of wind turbines; In step 3), the mapping method of establishing the wind turbine cluster power to the offshore wind farm collector power is as follows: 301) Based on the cluster division results in the spectral clustering of wind turbines, the historical sequence Z of the total power of wind turbines in m clusters is calculated: Among them, the i-th row vector of Z represents the historical time series of the total power of wind turbines in the i-th cluster; 302) Based on BP neural network, a mapping from wind turbine cluster power to collector power is established: The 2880 column vectors of the total power history sequence Z of the wind turbines in the m clusters are used as the input of the BP neural network training set. gl The 2880 column vectors are used as the output of the BP neural network training set. The number of input and output samples of the neural network training is 2880. In the BP neural network, the input vector is X = (x 1 ,x 2 ,…,x n ) T ; The hidden layer output is Y = (y 1 ,y 2 ,…,y m ) T , the output layer output is O = (o 1 ,o 2 ,…o l ) T , the expected value is D=(d 1 ,d 2 ,…,d l ) T ,ω ij ,ω jk are the weights from the input layer to the hidden layer and the weights from the hidden layer to the output layer respectively; Calculate the hidden layer output Calculate the output layer output In the above two equations, f(·) is the activation function of the BP neural network, and the output error E of the network is as follows: Substituting equations (6) and (7) into error expression (8), we get equation (9): From formula (9), we can see that the output error E of the network is a weight value ω ij ,ω jk The function adjusts the weights to reduce the output error of the network. When the output error E of the network is less than the expected value or the number of training times reaches the standard, the network training ends. The weights are updated along the negative gradient direction of the network's output error to reduce the error. η is the learning rate, ranging from 0 to 1; the network weights are iteratively updated according to the increments obtained in the above formula. oh ij (n+1)=ω ij (n)+Give ij (13) oh jk (n+1)=ω jk (n)+Give jk (14) After the update is completed, the information is forwarded again until the number of training times is greater than the target value or the error is less than the target value. Based on the above BP neural network training process, a mapping from wind turbine cluster power to collector line power is established: P G =F 1 (P Z )(15) Among them, F 1 represents the trained neural network model, P Z represents the clustered power of a group of wind turbines, P G represents the corresponding collector power; In step 4), a mapping method for establishing the power loss of the offshore wind farm collector line to the power loss of the submarine cable is as follows: 401) Calculate the historical data of submarine cable power loss: P xl =P sl -P ml (16) Among them, P xl is the historical data of submarine cable power loss, with a dimension of 1×b, where b is the number of collector wires; 402) Take the power of the offshore wind farm collector line P gl As input, the submarine cable power loss P xl For output, based on the BP neural network described in step 302), a mapping of the offshore wind farm collector power to the submarine cable power loss is established: P X =F 2 (P G )(17) Among them, P G is a group of collector power, P X represents the corresponding submarine cable power loss; In step 5), an adjustable capacity assessment model based on the offshore wind farm collector power and submarine cable power loss is established as follows: P K =sum(P G )-F 2 (P G )-ave(P hl +P ll )(18) Where sum() represents vector summation, and ave() represents vector mean; In step 6), the model for evaluating the adjustable capacity of the wind farm based on the real-time SCADA information of the wind turbines in the offshore wind farm is as follows: 601) reads the real-time maximum power generation capacity of the wind turbine in the offshore wind farm wind turbine SCADA real-time information, and based on the spectral clustering results of the m clusters of wind turbines in step 206), calculates the sum of the maximum power generation capacity of the wind turbines in the m clusters to form an m-dimensional vector, denoted as P ZS ; 602) The calculation formula for the real-time adjustable capacity of an offshore wind farm is: P KS =sum(F 1 (P ZS ))-F 2 (F 1 (P ZS ))-ave(P hl +P ll )(19) Among them, P KS Indicates the real-time maximum adjustable capacity of the wind farm.