Wind field data extrapolation method based on regression algorithm

Through the wind field data extrapolation method based on regression algorithm, the ground wind speed data of wind measurement lidar is used to establish the relationship between low-level and high-level wind speed, which solves the problem of insufficient external pushing accuracy of traditional methods on the wind turbine site, and achieves a higher level wind speed prediction accuracy and lower error.

CN118568682BActive Publication Date: 2024-11-15ZHUHAI GUANGHENG TECH CO LTD
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202411009541.0
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2024-07-26
Publication Date
2024-11-15
Estimated Expiration
2044-07-26

AI Technical Summary

Technical Problem

In the field wind resource evaluation of wind turbines, the traditional power law and logarithmic law extrapolation methods have large errors, especially in unstable atmospheric conditions, and the extrapolation accuracy of lidar data is insufficient.

Method used

The wind field data extrapolation method based on regression algorithm is used to obtain the ground wind speed data of wind measurement lidar as the training set, and the cross-correlation of wind speed data of different height layers is calculated, and linear regression, polynomial regression or random forest regression algorithm is used to establish Data extrapolation is performed on the relationship between the wind speed at the lower level and the wind speed at the higher level.

Benefits of technology

The accuracy of extrapolation of wind farm data is improved, especially when the data volume is sufficient, the random forest regression algorithm can significantly improve the prediction accuracy, reduce the average absolute error, and provide more accurate high-level wind speed prediction.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118568682B_ABST
    Figure CN118568682B_ABST
Patent Text Reader

Abstract

The present invention aims to provide a wind field data extrapolation method based on a regression algorithm with higher estimation accuracy. The method of the present invention comprises the following steps: a. Obtain a complete section of measurement data as a training set: Use a wind measuring laser radar as a wind field detection device, place it horizontally on the ground, measure wind speed data at different heights, and use the measured wind speed data at different heights as a training set; b. When the radar measurement height does not meet the design index, calculate the cross-correlation between the wind speed data at different heights. When the cross-correlation of two groups of wind speed data at different heights meets the set requirements, determine to use the low-level wind speed data to extrapolate the high-level wind speed data; c. Use a regression algorithm to establish the relationship between the wind speed at the low-level height and the wind speed at the high-level height, and predict the wind speed at the high-level height. The present invention is applied to the field of laser radar technology.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the field of laser radar technology, and in particular to a wind field data extrapolation method based on a regression algorithm. Background Art

[0002] The power generation of a wind turbine will increase with the increase of wind speed until it reaches the extreme value of power generation. In the range close to the ground, the wind speed will increase with the increase of height. Therefore, it is necessary to evaluate the wind resources on site before building a wind farm. Accurate wind resource assessment requires obtaining the wind speed at the height of the wind turbine hub. The cost of wind speed measurement increases significantly with the increase of height. Usually, mechanical wind measuring devices or Doppler laser wind radars and other wind measuring devices are installed on site when conducting wind resource assessments. Compared with the construction cost of mechanical wind measuring equipment on wind towers, laser radars are more convenient. Since wind laser radar is a weak signal detection device, it will be affected by external weather conditions. Therefore, when the effective data detection height does not meet the measurement requirements, the data needs to be extrapolated.

[0003] The traditional approach is to use the power law to extrapolate wind speed by establishing a meteorological model. The specific process of using the power law to calculate is as follows: Wind speed at height and wind shear index , it can be calculated Wind speed:

[0004] (14)

[0005] The wind shear index is calculated by measuring the average wind speed at different heights using the least squares fitting method. , the low-level wind field data can be used to extrapolate the high-level wind field. Here, the mean absolute error (MAE) is used to evaluate the level of the prediction results. The prediction results are shown in the attached figure. Figure 1 and Figure 2 .in, Figure 1 It is a scatter plot of measured data and power law calculation data; Figure 2 This is a graph showing the time-varying trend of measured data and power law calculation data. Figure 1 and Figure 2 It can be seen that due to the random changes in the atmosphere, using the traditional power law method to extrapolate high-level data from low-level data will produce large errors. Therefore, other schemes can be used to extrapolate the data.

[0006] In addition, the method of empirical extrapolation can also use the logarithmic law:

[0007] (15)

[0008] in, is the wind speed, where is the Karman constant (usually 0.4), is the height from the ground, is the friction speed, is the roughness length, It is the momentum stability function that adjusts the wind profile according to the atmospheric stability. L is the Monin–Obukhov length, which is often used to determine the stability of neutral gases:

[0009] (16)

[0010] in is the acceleration due to gravity, is the surface temperature, is the kinetic heat flux, is the specific heat of the atmosphere, is the atmospheric density.

[0011] From the above calculation process, it can be seen that the power law calculation performs well under stable atmospheric conditions of neutral gas, but cannot be applied under unstable conditions, so it has limitations. Summary of the invention

[0012] The technical problem to be solved by the present invention is to overcome the deficiencies of the prior art and provide a wind field data extrapolation method based on a regression algorithm with higher estimation accuracy.

[0013] The technical solution adopted by the present invention is that the method of the present invention comprises the following steps:

[0014] a. Obtain a complete set of measurement data as a training set: Use a wind laser radar as a wind field detection device, place it horizontally on the ground, and measure wind speed data at different heights. The wind speed data at different heights is used as a training set;

[0015] b. When the radar measurement height does not meet the design index, calculate the cross-correlation between wind speed data at different altitudes. When the cross-correlation between two sets of wind speed data at different altitudes meets the set requirements, determine to use the low-level wind speed data to extrapolate the high-level wind speed data;

[0016] c. Use regression algorithm to establish the relationship between wind speed at low altitude and wind speed at high altitude, and predict the wind speed at high altitude.

[0017] Furthermore, in step b, the specific steps of judging whether the correlation between two sets of wind speed data at different heights meets the set requirements are as follows: assuming that a set of low-level wind speed data is x i , a set of high-level wind speed data is y i , two sets of datax i and y i The cross-correlation function is:

[0018] (1),

[0019] Where n is the number of height layers, i Take an integer from 1 to n,

[0020] When the correlation is close to 1, it means that the two sets of data are highly correlated, so the low-level wind speed data is used. x i Extrapolating high-level wind speed data y i .

[0021] Furthermore, in step c, the regression algorithm is a linear regression algorithm, a polynomial regression algorithm or a random forest regression algorithm.

[0022] In addition, in step c, when a linear regression algorithm is used, the specific calculation process is as follows:

[0023] Create a linear regression function:

[0024] (2),

[0025] in, is the linear function intercept, is the coefficient value at different heights, is the measured value of the corresponding height,

[0026] The relationship between the extrapolated data and the measured data and the corresponding parameters is:

[0027] (3),

[0028] Using the least squares method to calculate the coefficients, formula (3) is similar to:

[0029] Y=AX (4),

[0030] Then there is

[0031] (5),

[0032] in, , , ,

[0033] in Y is the wind speed data to be extrapolated,

[0034] The data set measured in step a is randomly divided into a training set and a test set, and then the data is trained and tested using linear regression prediction to evaluate the model performance. After training with m sets of measured data, the coefficient X is obtained as the linear regression extrapolation parameter.

[0035] Specifically, when the data set measured in step a is randomly divided into a training set and a test set, 80% of the data is used as a training set and 20% of the data is used as a test set.

[0036] In addition, in step c, when the polynomial regression algorithm is used, the specific calculation process is as follows:

[0037] Create a polynomial regression function:

[0038] (6),

[0039] in, w i is the first-order function coefficient, u i is the coefficient of the second-order function, v i is the coefficient of the third-order function, is the coefficient of the n-th order function;

[0040] Referring to formulas (3) to (5), the least square method is used to calculate the parameters corresponding to formula (6). The calculated X coefficient is used as an extrapolation for combining the measured low-level data to calculate the wind speed at the corresponding height.

[0041] Furthermore, in step c, when the random forest regression algorithm is used, the specific calculation process is as follows:

[0042] Assuming that the random forest consists of m regression trees, the predicted value can be calculated by the following formula:

[0043] (7),

[0044] in, For the test sample The predicted value of y i For the The predicted value of a regression tree, is the weight of the i-th regression tree;

[0045] In random forests, the instantaneous loss is used to represent the absolute difference between the predicted value and the true value:

[0046] (8),

[0047] The square error is used as the target loss function:

[0048] (9),

[0049] in is the true value of the tth test sample;

[0050] When the test sample falls into a leaf node of the tree, the weight of the leaf node is adjusted in the direction of reducing the target loss function value, so that the value of the loss function decreases rapidly. Represents the incremental change of weight relative to the previous one, then the formula of weight iteration process is:

[0051] (10),

[0052] in is the learning rate of the leaf node in the i-th decision tree, and the update formula of the leaf weight can be iteratively calculated as follows:

[0053] (11),

[0054] After obtaining the weights, the wind speed at the corresponding height is calculated by extrapolating the measured low-level data.

[0055] The method also includes the step of evaluating the predicted data, specifically:

[0056] Using the specific data of 35m, 40m, 60m, and 100m, extrapolating the data at a height of 140m, and using the training set to calculate the corresponding parameters of the three regression algorithms X and leaf weight ’ , the specific calculation formulas are shown in formula (2), formula (6) and formula (7). Substituting the measured data of the lower layer into the higher layer data can be calculated. The difference between the measured data and the predicted data is used to determine the coefficient Or mean absolute error MAE to evaluate the accuracy of the algorithm; and MAE are:

[0057] Coefficient of determination for:

[0058] (12),

[0059] in, represents the actual observed value, represents the predicted value, is the average of the true observed values, is the sample size;

[0060] The mean absolute error MAE is:

[0061] (13);

[0062] in, represents the actual observed value, Represents the predicted value, and N is the number of samples. When the predicted value is completely consistent with the true value, it is equal to , that is, a perfect model; otherwise, the larger the error, the larger the MAE.

[0063] The beneficial effects of the present invention are as follows: in the method of the present invention, the newly proposed wind field data extrapolation algorithm has higher accuracy than using the traditional meteorological model to perform data extrapolation analysis; among the several regression algorithms used, linear regression is used to extrapolate the data when the data is small, and when the amount of sample data is large enough, a polynomial or random forest regression algorithm can be used to improve the accuracy of the predicted data. BRIEF DESCRIPTION OF THE DRAWINGS

[0064] Figure 1 It is a scatter plot of measured data and power law calculated data in the prior art;

[0065] Figure 2 It is a trend chart of measured data and power law calculated data over time in the prior art;

[0066] Figure 3 It is a scatter plot of measured data and linear regression predicted data in the method of the present invention;

[0067] Figure 4 It is a trend chart of measured data and linear regression predicted data changing over time in the method of the present invention;

[0068] Figure 5 It is a scatter plot of measured data and polynomial regression predicted data in the method of the present invention;

[0069] Figure 6 It is a time-varying trend graph of measured data and polynomial regression predicted data in the method of the present invention;

[0070] Figure 7 It is a simplified diagram of the random forest regression algorithm in the method of the present invention;

[0071] Figure 8 It is a scatter plot of measured data and data predicted by random forest regression algorithm in the method of the present invention;

[0072] Fig. 9 It is a trend chart of measured data and data predicted by random forest regression algorithm over time in the method of the present invention. DETAILED DESCRIPTION

[0073] like Figures 3 to 9 As shown, the method of the present invention comprises the following steps:

[0074] a. Obtain a complete set of measurement data as a training set: Use a wind laser radar as a wind field detection device, place it horizontally on the ground, and measure wind speed data at different heights. The wind speed data at different heights is used as a training set;

[0075] b. When the radar measurement height does not meet the design index, calculate the cross-correlation between wind speed data at different altitudes. When the cross-correlation between two sets of wind speed data at different altitudes meets the set requirements, determine to use the low-level wind speed data to extrapolate the high-level wind speed data;

[0076] c. Use regression algorithm to establish the relationship between wind speed at low altitude and wind speed at high altitude, and predict the wind speed at high altitude.

[0077] In step b, the specific steps of judging whether the correlation between two sets of wind speed data at different heights meets the set requirements are as follows: assuming that a set of low-level wind speed data is x i , a set of high-level wind speed data is y i , two sets of data x i and y i The cross-correlation function is:

[0078] (1),

[0079] Where n is the number of height layers, i Take an integer from 1 to n,

[0080] When the correlation is close to 1, it means that the two sets of data are highly correlated, so the low-level wind speed data is used. x i Extrapolating high-level wind speed data y i .

[0081] The regression algorithm is one of a linear regression algorithm, a polynomial regression algorithm or a random forest regression algorithm.

[0082] like Figure 3 and Figure 4 As shown, when the linear regression algorithm is used, the specific calculation process is as follows:

[0083] Create a linear regression function:

[0084] (2),

[0085] in, is the linear function intercept, is the coefficient value at different heights, is the measured value of the corresponding height,

[0086] The relationship between the extrapolated data and the measured data and the corresponding parameters is:

[0087] (3),

[0088] Using the least squares method to calculate the coefficients, formula (3) is similar to:

[0089] Y=AX (4),

[0090] Then there is

[0091] (5),

[0092] in, , , ,

[0093] in Y is the wind speed data to be extrapolated,

[0094] The data set measured in step a is randomly divided into a training set and a test set. In this embodiment, 80% of the data is used as a training set and 20% of the data is used as a test set. Then, the data is trained and tested using linear regression prediction to evaluate the model performance. After training with m sets of measured data, the coefficient X is obtained as a linear regression extrapolation parameter.

[0095] like Figure 5 and Figure 6 As shown, when the polynomial regression algorithm is used, the specific calculation process is as follows: Establish a polynomial regression function:

[0096] (6),

[0097] in, w i is the first-order function coefficient, u i is the coefficient of the second-order function, v i is the coefficient of the third-order function, is the coefficient of the n-th order function;

[0098] Referring to formulas (3) to (5), the least square method is used to calculate the parameters corresponding to formula (6). The calculated X coefficient is used as an extrapolation for combining the measured low-level data to calculate the wind speed at the corresponding height.

[0099] like Figures 7 to 9 As shown, when the random forest regression algorithm is used, the specific calculation process is as follows:

[0100] Classification and Regression Tree (CART) is a typical decision tree algorithm. It provides a basic algorithm framework for machine learning classification or regression prediction modeling. It can be used for both discrete variable classification tasks and continuous numerical variable regression tasks. The random forest for processing regression problems can also be referred to as regression forest. Assume that the random forest consists of m regression trees, and the prediction value can be calculated by the following formula:

[0101] (7),

[0102] in, For the test sample The predicted value of y i For the The predicted value of a regression tree, is the weight of the i-th regression tree;

[0103] In random forests, the instantaneous loss is used to represent the absolute difference between the predicted value and the true value:

[0104] (8),

[0105] The square error is used as the target loss function:

[0106] (9),

[0107] in is the true value of the tth test sample;

[0108] When the test sample falls into a leaf node of the tree, the weight of the leaf node is adjusted in the direction of reducing the target loss function value, so that the value of the loss function decreases rapidly. Represents the incremental change of weight relative to the previous one, then the formula of weight iteration process is:

[0109] (10),

[0110] in is the learning rate of the leaf node in the i-th decision tree, and the update formula of the leaf weight can be iteratively calculated as follows:

[0111] (11),

[0112] After obtaining the weights, the wind speed at the corresponding height is calculated by extrapolating the measured low-level data.

[0113] like Figure 4 , Figure 6 and Fig. 9 As shown, the method also includes the step of evaluating the above-mentioned prediction data, specifically:

[0114] Using the specific data of 35m, 40m, 60m, and 100m, extrapolating the data at a height of 140m, and using the training set to calculate the corresponding parameters of the three regression algorithms X and leaf weight ’ , the specific calculation formulas are shown in formula (2), formula (6) and formula (7). Substituting the measured data of the lower layer into the higher layer data can be calculated. The difference between the measured data and the predicted data is used to determine the coefficient Or mean absolute error MAE to evaluate the accuracy of the algorithm; and MAE are:

[0115] Coefficient of determination for:

[0116] (12),

[0117] in, represents the actual observed value, represents the predicted value, is the average of the true observed values, is the sample size;

[0118] The mean absolute error MAE is:

[0119] (13);

[0120] in, represents the actual observed value, Represents the predicted value, and N is the number of samples. When the predicted value is completely consistent with the true value, it is equal to , that is, a perfect model; otherwise, the larger the error, the larger the MAE.

[0121] Finally, it should be emphasized that the above is only a preferred embodiment of the present invention and is not intended to limit the present invention. For those skilled in the art, the present invention may have various changes and modifications (such as the number of laser radar beams, the pointing angle of the beam). Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the scope of protection of the present invention.

Claims

1. A wind field data extrapolation method based on regression algorithm, It is characterized in that The method comprises the following steps, a. Obtain a complete set of measurement data as a training set: Use a wind laser radar as a wind field detection device, place it horizontally on the ground, and measure wind speed data at different heights. The wind speed data at different heights is used as a training set; b. When the radar measurement height does not meet the design index, calculate the cross-correlation between wind speed data at different altitudes. When the cross-correlation between two sets of wind speed data at different altitudes meets the set requirements, determine to use the low-level wind speed data to extrapolate the high-level wind speed data; c. Use regression algorithm to establish the relationship between wind speed at low altitude and wind speed at high altitude, and predict the wind speed at high altitude; In step b, the specific steps of judging whether the correlation between two sets of wind speed data at different heights meets the set requirements are as follows: assuming that a set of low-level wind speed data is x i , a set of high-level wind speed data is y i , two sets of data x i and y i The cross-correlation function is: (1), Where n is the number of height layers, i Take an integer from 1 to n, When the correlation is close to 1, it means that the two sets of data are highly correlated, so the low-level wind speed data is used. x i Extrapolating high-level wind speed data y i ; In step c, the regression algorithm is a linear regression algorithm, a polynomial regression algorithm or a random forest regression algorithm; In step c, when a linear regression algorithm is used, the specific calculation process is as follows: Create a linear regression function: (2), in, is the linear function intercept, is the coefficient value at different heights, is the measured value of the corresponding height, The relationship between the extrapolated data and the measured data and the corresponding parameters is: (3), Using the least squares method to calculate the coefficients, formula (3) is similar to: Y=AX (4), Then there is (5), in, , , , in Y is the wind speed data to be extrapolated, The data set measured in step a is randomly divided into a training set and a test set, and then the data is trained and tested using linear regression prediction to evaluate the model performance. After training with m sets of measured data, the coefficient X is obtained as the linear regression extrapolation parameter.

2. The wind field data extrapolation method based on regression algorithm according to claim 1, It is characterized in that In step c, when the polynomial regression algorithm is used, the specific calculation process is as follows: Create a polynomial regression function: (6), in, w i is the first-order function coefficient, u i is the coefficient of the second-order function, v i is the coefficient of the third-order function, is the coefficient of the n-th order function; Referring to formulas (3) to (5), the least square method is used to calculate the parameters corresponding to formula (6). The calculated X coefficient is used as an extrapolation for combining the measured low-level data to calculate the wind speed at the corresponding height.

3. The wind field data extrapolation method based on regression algorithm according to claim 2, It is characterized in that In step c, when the random forest regression algorithm is used, the specific calculation process is as follows: Assuming that the random forest consists of m regression trees, the predicted value can be calculated by the following formula: (7), in, For the test sample The predicted value of y i For the The predicted value of a regression tree, is the weight of the i-th regression tree; In random forests, the instantaneous loss is used to represent the absolute difference between the predicted value and the true value: (8), The square error is used as the target loss function: (9), in is the true value of the tth test sample; When the test sample falls into a leaf node of the tree, the weight of the leaf node is adjusted in the direction of reducing the target loss function value, so that the value of the loss function decreases rapidly. Represents the incremental change of weight relative to the previous one, then the formula of weight iteration process is: (10), in is the learning rate of the leaf node in the i-th decision tree, and the update formula of the leaf weight can be iteratively calculated as follows: (11), After obtaining the weights, the wind speed at the corresponding height is calculated by extrapolating the measured low-level data.

4. The wind field data extrapolation method based on regression algorithm according to claim 3, It is characterized in that The method also includes the step of evaluating the predicted data, specifically: Using the specific data of 35m, 40m, 60m, and 100m, extrapolating the data at a height of 140m, and using the training set to calculate the corresponding parameters of the three regression algorithms X and leaf weight ’ , the specific calculation formulas are shown in formula (2), formula (6) and formula (7). Substituting the measured data of the lower layer into the higher layer data can be calculated. The difference between the measured data and the predicted data is used to determine the coefficient Or mean absolute error MAE to evaluate the accuracy of the algorithm; and MAE are: Coefficient of determination for: (12), in, represents the actual observed value, represents the predicted value, is the average of the true observed values, is the sample size; The mean absolute error MAE is: (13); in, represents the actual observed value, Represents the predicted value, N is the number of samples, and when the predicted value is completely consistent with the true value, it is equal to , that is, a perfect model; otherwise, the larger the error, the larger the MAE.

5. The wind field data extrapolation method based on regression algorithm according to claim 1, It is characterized in that When the data set measured in step a is randomly divided into a training set and a test set, 80% of the data is used as a training set and 20% of the data is used as a test set.