CN117851877A - Wind power blade fracture early warning method and device based on SCADA data association analysis and readable storage medium - Google Patents

Abstract

The present invention provides a wind turbine blade fracture warning method based on the association analysis of SCADA data. The method comprises: reducing the dimension of the SCADA data before and after the wind turbine blade fracture, so as to first screen out various parameters related to the operating state of the wind turbine blade. Data cleaning is performed on the original data corresponding to the screened parameters in the SCADA data under the normal operating state of the wind turbine blade, and the data corresponding to the screened parameters in the SCADA data after data cleaning are extracted, and the extracted data are divided into a training set and a verification set, and then the training set is used for NSET modeling. Historical data verification and further parameter screening are used to finally screen out various parameters related to the operating state of the wind turbine blade. The real-time observation data can be input to obtain the Euclidean distance curve according to the residual output of the model, so as to judge the operating status of the wind turbine blade at this time. The present invention fully explores the threshold of blade fracture warning based on the existing SCADA data, so as to achieve the purpose of online real-time monitoring.

Description

Wind turbine blade fracture warning method, device, and readable storage medium based on SCADA data correlation analysis

Technical Field

The present invention relates to the technical field of wind turbine blade monitoring, and in particular to a wind turbine blade fracture early warning method based on SCADA data association analysis, a computer device and a computer-readable storage medium.

Background technique

Wind turbine blades are key components that convert wind energy into mechanical energy and ultimately drive generators to generate electricity. Among the maintenance cost ratios of various components of wind power equipment, wind turbine blades account for the highest proportion of 29.55%. In addition, the value of wind turbine blades accounts for more than 20% of the entire unit. Wind turbine blades operate under harsh conditions. Although their design and test life is 20 to 25 years, experience shows that preventive maintenance is required to achieve this goal. With the continuous development of wind power generation and the increasing social electricity demand, the trend of large-scale development of wind turbines is becoming more and more obvious. As the length of blades increases, the risk of blade breakage also increases. The breakage of wind turbine blades may not only cause the burning of the wind turbine itself, but also affect other units due to the flying blade debris, causing huge economic losses and increasing safety hazards. Therefore, it is particularly important to monitor the breakage of wind turbine blades.

With the development trend of large-scale and large-scale wind turbines in my country, due to the huge differences in the airfoil, structure, material and shape of the blades of different types of wind turbines, it is difficult to unify and quantify the assessment standards of their damage degree, and the failure fracture threshold will also change dynamically when operating under different wind field conditions. Therefore, for real-time fracture monitoring of large wind turbine blades, the dynamic setting of the fracture threshold is a difficult problem. In addition, the integrated molding technology of wind turbine blades and intelligent sensors is not yet mature. If various sensors are installed in the internal cavity or outer wall skin of giant wind turbine blades, they are not only easy to fall or fail, but also the cost of the blade monitoring system cannot be controlled. Therefore, the wind turbine blade fracture warning method should consider the dynamic changes of the fracture alarm threshold and the cost issue at the same time to ensure that it is more universal and has a wider range of applications.

At present, most of the existing wind turbine monitoring systems at home and abroad use model-free artificial intelligence algorithms, which rely on a large amount of sample data and the accuracy of the model. The selection of modeling parameters is often without basis, and the calculation process lacks reliability analysis and verification. In particular, the blade monitoring system collects the existing SCADA system sensor signals, and there are data cleaning problems such as large data volume, large dimensions, and many redundant variables, which will lead to low efficiency of online monitoring, thereby reducing the accuracy of the model and making it difficult to monitor online in real time.

Summary of the invention

The present invention aims to solve at least one of the technical problems existing in the prior art or related art.

To this end, the first objective of the present invention is to propose a wind turbine blade fracture early warning method based on correlation analysis of SCADA data.

A second objective of the present invention is to provide a computer device.

A third objective of the present invention is to provide a computer-readable storage medium.

In order to achieve the above-mentioned purpose, the technical solution of the first aspect of the present invention provides a wind turbine blade fracture early warning method based on the correlation analysis of SCADA data, wherein the SCADA data is data collected by a sensor group in the SCADA system of the wind turbine set, and the various parameters included in the SCADA data are: wind turbine power, generator speed, rotor speed, wind turbine blade angle, grid-side current and wind speed; the early warning method comprises: obtaining a first group of SCADA data before and after the wind turbine blade fracture in a certain period of time; using chi-square verification to screen the first group of SCADA data, so as to use chi-square verification to screen out various parameters related to the operating state of the wind turbine blade; adding the wind speed parameter to the various parameters related to the operating state of the wind turbine blade screened out by using chi-square verification, so as to screen out various parameters related to the operating state of the wind turbine blade for the first time; obtaining The second group of SCADA data under the normal operating state of the wind turbine blades in a certain period of time is obtained; the original data corresponding to the various parameters related to the operating state of the wind turbine blades screened out for the first time in the second group of SCADA data are extracted; a wind speed-power scatter plot is fitted according to the wind speed data and the power data of the wind turbine generator set in the second group of SCADA data after data extraction; the wind speed-power scatter plot is cleaned by using the DBSCAN clustering algorithm; the second group of SCADA data after data cleaning is divided into a first training set and a first validation set; according to the first training set and the first validation set, a first NSET model under the normal operating state of the wind turbine blades is built; the original first group of SCADA data is input into the first NSET model, and the residual E between the predicted value output by the first NSET model and the corresponding observed value is calculated. _n , and fitting the first Euclidean distance curve corresponding to the first group of SCADA data according to the predicted value and the observed value, so as to observe that the first Euclidean distance curve begins to be in an upward trend after a certain position of the breakpoint delay; calculating the fault cumulative contribution rate e _i of each parameter related to the operating state of the wind turbine blade screened out using the chi-square validation according to the residual E _n ; removing the fault cumulative contribution rate e i of each parameter related to the operating state of the wind turbine blade screened out using the chi-square validation _i is less than a certain set value, and the wind speed parameter is added to the various parameters related to the operating state of the wind turbine blades after the removal, so as to screen out the various parameters related to the operating state of the wind turbine blades for the second time; extract the data corresponding to the various parameters related to the operating state of the wind turbine blades screened out for the second time in the second group of SCADA data after data cleaning, and divide the extracted data into a second training set and a second verification set; according to the second training set and the second verification set, establish a second NSET model under the normal operating state of the wind turbine blades; obtain the Euclidean distance curve corresponding to the second verification set, and use the maximum value of the Euclidean distance on the Euclidean distance curve corresponding to the second verification set as the first threshold; input the original first group of SCADA data into the second NSET model, and calculate the residual E _n between the predicted value output by the second NSET model and the corresponding observed value ', and fitting the second Euclidean distance curve corresponding to the first group of SCADA data according to the predicted value and the observed value, so as to observe that the second Euclidean distance curve begins to be in an upward trend at the position of the break point; obtaining the maximum value of the Euclidean distance on the second Euclidean distance curve before the break point as the second threshold value; obtaining the maximum value of the Euclidean distance on the second Euclidean distance curve after the break point as the third threshold value; obtaining the SCADA data collected by the sensor group in the SCADA system of the wind turbine in real time: inputting the SCADA data obtained in real time into the second NSET model, and calculating the residual E _n ' between the predicted value output by the second NSET model and the corresponding observed value, and fitting the third Euclidean distance curve corresponding to the SCADA data obtained in real time according to the predicted value and the observed value; according to the values of each Euclidean distance corresponding to the third Euclidean distance curve, as well as the first threshold value, the second threshold value and the third threshold value, the wind turbine blade of the wind turbine is given a fracture warning.

Preferably, the fracture warning of the wind turbine blades of the wind turbine set according to the value of the Euclidean distance on the third Euclidean distance curve, as well as the first threshold, the second threshold and the third threshold, specifically includes: determining the range of the values of each corresponding Euclidean distance on the third Euclidean distance curve; when the value of a certain Euclidean distance corresponding to the third Euclidean distance curve is greater than or equal to 0 and less than or equal to the first threshold, it is judged that the wind turbine blades of the wind turbine set are in a normal operating state; when the value of a certain Euclidean distance corresponding to the third Euclidean distance curve is greater than the first threshold and less than or equal to the second threshold, it is judged that the wind turbine blades of the wind turbine set are in an abnormal state; when the value of a certain Euclidean distance corresponding to the third Euclidean distance curve is greater than or equal to the third threshold, it is judged that the wind turbine blades of the wind turbine set are in a broken state.

Preferably, the parameters related to the operating state of the wind turbine blades screened out using the chi-square verification are: wind turbine power, generator speed, rotor speed, wind turbine blade angle, first grid-side current, second grid-side current and third grid-side current; the parameters related to the operating state of the wind turbine blades screened out for the first time are: wind speed, wind turbine power, generator speed, rotor speed, wind turbine blade angle, first grid-side current, second grid-side current and third grid-side current;

The expression of the residual E _n is:

_En = _xn - _xn ^*

Wherein, x _n is the observed value of the nth parameter in the first group of SCADA data; x _n ^* is the predicted value corresponding to the nth parameter in the first group of SCADA data; En _is the difference between the observed value of the nth parameter and the predicted value corresponding to the nth parameter; x ₁ to x ₇ are the parameters related to the operating state of the wind turbine blade screened out using the chi-square validation; and the _calculation of the fault cumulative contribution rate e i of the parameters related to the operating state of the wind turbine blade screened out using the chi-square validation according to the residual _En specifically includes:

Calculating the number of failures e _c of each parameter related to the wind turbine blade operation state selected by using chi-square validation according to the residual E _n ;

The expression of the fault number e _c is:

Let a = number of observation vector groups/number of observation quantity groups input each time, and _calculate the cumulative fault contribution rate e _{i of each parameter related to the wind turbine blade operation state screened by using chi-square validation according to the number of faults e c} ;

The expression of the cumulative contribution rate of faults e _i is:

And the removing of the parameters whose cumulative fault contribution rate e _i is less than a certain set value from the various parameters related to the operating state of the wind turbine blades screened out by using the chi-square verification, and adding the wind speed parameter to the various parameters related to the operating state of the wind turbine blades after the removal, so as to screen out the various parameters related to the operating state of the wind turbine blades for the second time, specifically includes: removing the parameters whose cumulative fault contribution rate e _i is less than a certain set value from x ₁ to x ₇ , and adding the wind speed parameter to the various parameters related to the operating state of the wind turbine blades after the removal, so as to screen out the various parameters related to the operating state of the wind turbine blades for the second time; wherein the various parameters related to the operating state of the wind turbine blades screened out for the second time are: wind speed, wind turbine power, generator speed, rotor speed and wind turbine blade angle.

Preferably, establishing a first NSET model of a wind turbine blade in a normal operating state according to the first training set and the first validation set specifically includes: obtaining an original NSET model, inputting the first training set into the original NSET model, so as to obtain a trained NSET model;

The overall observation matrix of the wind turbine is represented as an n×b-sized M _n×b , and the expression of M _n×b is:

Wherein, n is the time state, b is the number of observed variables at each time; the row vector of the matrix Mn _×b is _Xi = [ _xi ( _t1 ) _xi ( _t2 ) ... _xi ( _tb )], and the row vector of the matrix Mn _×b is all the observed values of a given observation parameter _Xi within a certain observation time period; the column vector of the matrix Mn _×b is X( _tj ) = [ _x1 ( _tj ) _x2 ( _tj ) ... _xb ( _tj )] ^T , and the column vector of the matrix Mn _×b is the observed values of all observed parameters at time _tj ; the parameters of a period of time are selected from the Mn _×b and recorded as the historical observation matrix K, and the historical observation matrix K is the health status of each observation parameter, then the expression of the historical observation matrix K is:

A portion of state data is selected from the historical observation matrix K, and the selected portion of state data is used to form a process memory matrix D _n . The process matrix D _n can be expressed as:

The observation matrix X _obs and D _n corresponding to the observation matrix X _obs are input into the expression , obtain a prediction output matrix X _est ; set the first validation set as an observation matrix X _obs ; input the first validation set into the trained NSET model; the NSET model after inputting the first validation set outputs a corresponding prediction output matrix X _est ; calculate the residual between the observation matrix X _obs corresponding to the first validation set and the corresponding prediction output matrix X _est , and fit the Euclidean distance curve corresponding to the first validation set according to the observation matrix X _obs corresponding to the first validation set and the corresponding prediction output matrix X _est to establish a first NSET model in a normal operating state of the wind turbine blade; and the second NSET model under normal operating state of the wind turbine blade is established according to the second training set and the second validation set, specifically including: obtaining an original NSET model, inputting the second training set into the original NSET model to obtain a trained NSET model; setting the second validation set as an observation matrix X _obs ; inputting the second validation set into the trained NSET model; the NSET model after inputting the second validation set outputs a corresponding prediction output matrix X _est ; calculate the residual between the observation matrix X _obs corresponding to the second validation set and the corresponding prediction output matrix X _est , and fit the Euclidean distance curve corresponding to the first validation set according to the observation matrix X _obs corresponding to the second validation set and the corresponding prediction output matrix X est _est is used to fit the Euclidean distance curve corresponding to the second validation set to establish a second NSET model under normal operating conditions of the wind turbine blade.

Preferably, the selecting a part of state data from the historical observation matrix K and using the selected part of state data to form a process memory matrix _Dn specifically includes: setting each observation vector of the historical observation matrix K to be composed of n variables; for each of the n variables, dividing the interval [0,1] into h equal parts, searching for a number of observation vectors X(1) X(2)...X(k) from the historical observation matrix K with a step size of 1/h and adding them to the process memory matrix _Dn ;

For each of the n variables, the interval [0, 1] is equally divided into h parts, and a number of observation vectors X(1)X(2)...X(k) are found from the historical observation matrix K with a step size of 1/h and added to the process memory matrix D _n , specifically including: setting i=1; wherein i is a positive integer; executing A=1/h*i; wherein h is a positive integer; setting k=1; wherein k is a positive integer; judging whether |X(k)-A| is less than δ; wherein δ is a positive number; when |X(k)-A| is less than δ, adding X(k) to the process memory matrix D n _n ; when |X(k)-A| is greater than or equal to δ, determine whether k is greater than M; wherein M is the number of columns of the historical observation matrix K; when k is less than or equal to M, execute k=k+1, and return to the step of determining whether |X(k)-A| is less than δ; when k is greater than M, determine whether i is greater than h; when i is less than or equal to h, execute i=i+1, and return to the step of executing A=1/h*i; when i is greater than h, execution ends.

Preferably, the method of using chi-square validation to screen the first group of SCADA data to screen out various parameters related to the operating state of the wind turbine blades by using chi-square validation specifically includes: establishing an original hypothesis H0, wherein the original hypothesis H0 is that the operating state of the wind turbine blades is independent of various parameters in the first group of SCADA data; using the data of the operating state of the wind turbine blades as the first variable threshold; using a certain parameter in the first group of SCADA data as the second variable threshold each time; recording respectively the actual value of the number of data of the first variable threshold when the wind turbine blades are normal as a, the actual value of the number of data of the wind turbine blades when the wind turbine blades are faulty as b, and the actual value of the total number of data of the wind turbine blades as a+b; recording respectively the actual value of the number of data of the yth parameter in the SCADA data when the wind turbine blades are normal as _cy , the actual value of the number of data of the wind turbine blades when the wind turbine blades are faulty as _dy , and the actual value of the total number of data of the wind turbine blades as _cy + _dy , wherein y is a positive integer; recording respectively the actual value of the total number of data of the first variable threshold and the second variable threshold when the wind turbine blades are normal as a+ _cy , and the actual value of the total number of data of the two under the condition of wind turbine blade failure is b+ _dy , and the actual value of the total number of data of the two wind turbine blades is a+b+ _cy + _dy ; the theoretical value of the number of data of the first variable threshold under normal wind turbine blade is (a+b)×(a+ _cy )/(a+b+ _cy + _dy ), the theoretical value of the number of data under wind turbine blade failure is (a+b)×(b+ _dy )/(a+ _b + _cy + _dy ), and the theoretical value of the total number of data of wind turbine blade is a+b; the theoretical value of the number of data of the second variable threshold under normal wind turbine blade is (cy+ _dy )×(a+ _{cy )} /(a+b+ _cy + _dy ), and the theoretical value of the number of data of wind turbine blade failure is ( _cy +dy)×(b+ _dy )/(a+b+ _cy + _dy ), and the theoretical value of the total number of data of wind turbine blades is _cy + _dy ; the degree of freedom is set to 1; the chi-square value is calculated according to the chi-square value calculation formula, the chi-square value is used to measure the difference between each of the actual values and each of the theoretical values, and the chi-square value calculation formula is: χ ² =∑(AT) ² /T

Wherein, χ ² is the chi-square value, A is each actual value, and T is each theoretical value; according to the degree of freedom and chi-square value table, find the corresponding P value, and the P value is the probability of committing a first-type true rejection error; select a parameter in the first group of SCADA data corresponding to the P value less than 0.05 as the parameter related to the operating state of the wind turbine blade, so as to screen out the various parameters related to the operating state of the wind turbine blade in turn.

Preferably, the DBSCAN clustering algorithm is used to clean the wind speed-power scatter plot, specifically including: taking all points in the wind speed-power scatter plot as sample data S, and marking each data point in the sample data S as unprocessed; assigning initial values to Eps and Minpts, wherein Eps is the neighborhood distance threshold of a certain data point p in the sample data S, and Minpts is the minimum number of data points in the neighborhood with a radius of Eps of a certain data point p in the sample data S; setting the neighborhood with a radius of Eps of the certain data point p to N _Eps (p); clustering the high-density area formed by Eps and Minpts; the clustering of the high-density area formed by Eps and Minpts specifically includes: determining whether the certain data point p has been added to a cluster or has been listed as noise; if the certain data point p has been added to a cluster or has been listed as noise, the classification ends; if the certain data point p has not been added to a cluster and has not been listed as noise, determining whether there are at least Minpts objects in N _Eps (p); if N If there are at least Minpts objects in _Eps (p), a new cluster U is constructed and the data point p is added to U; if the number of objects in N _Eps (p) is less than Minpts, the data point p is listed as a boundary point or noise.

Preferably, the Eps is set to 4.5, and the Minpts is set to 18.5.

The technical solution of the second aspect of the present invention also provides a computer device, which includes a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, the steps of the wind turbine blade fracture warning method based on correlation analysis of SCADA data as in any of the above technical solutions are implemented.

The technical solution of the third aspect of the present invention also provides a computer-readable storage medium on which a computer program is stored. When the computer program is executed by a processor, the steps of the wind turbine blade fracture warning method based on correlation analysis of SCADA data as in any of the above technical solutions are implemented.

Beneficial effects of the present invention:

1. The wind turbine blade fracture early warning method based on correlation analysis of SCADA data provided by the present invention adopts some technical means to address the existing technical gaps, thus breaking through the technical difficulty that wind turbine blades cannot use existing SCADA data to monitor blade fractures in real time. Specifically, the data collected by the sensor group in the SCADA system of the wind turbine unit contains many redundant variables, which will increase the modeling time and lead to a decrease in efficiency, thereby reducing the accuracy of the model. In response to the above problems of large SCADA data volume and large dimensionality, the present invention proposes a chi-square verification method to perform preliminary screening of SCADA data related to the blade operating status, thereby achieving dimensionality reduction processing of SCADA data and improving the computational efficiency of the monitoring system.

2. For the null values, singular points, power limit data and noise data in the SCADA data, the DBSCAN algorithm is used to cluster the wind speed-power scatter plot, and the interruption data such as shutdown data, power limit data, noise, etc. are removed for data cleaning.

3. Most of the previous monitoring systems rely on the accuracy of sample data and models. The screening of modeling parameters often has no basis, and the calculation process lacks reliability analysis and verification. The present invention first performs a preliminary quantitative screening of the modeling parameters using chi-square verification, establishes the NSET model under normal service status as a semi-supervised model, inputs the known wind turbine blade fracture data set into the model, calculates the residuals between the observed values and the predicted values and the overall Euclidean distance curve to verify the modeling effect, and analyzes the failure contribution rate of each parameter to further screen the parameters related to the SCADA system and the blades, making the screening and verification of the modeling parameters more reliable and accurate.

4. In order to solve the problem that blade damage of different degrees is difficult to quantify, the present invention proposes to establish a semi-supervised learning model using the SCADA data of the normal operating status of wind turbine blades, and adopts the data fusion method to fuse the wind speed, power, generator speed, rotor speed, and blade angle. The overall data fusion result of the blade in the normal state is used as a comparison parameter for judging blade fracture, and the Euclidean distance between the real-time monitoring data of the blade and the normal state is used as a quantitative indicator of blade fracture.

5. In order to solve the problem of dynamically setting the blade fracture warning threshold, the present invention is based on the existing SCADA system historical data of the wind turbine set. Without installing other sensors, the SCADA data before and after the wind turbine blade fracture in a certain period of time is input into the model. It is proposed to use the SCADA data before and after the fracture of the blade to compare and analyze the Euclidean distance curve before the wind turbine blade fracture, obtain the maximum Euclidean distance before the model fracture, and obtain the wind turbine blade fracture warning threshold, thereby updating the fracture threshold of the blade operation service status in real time.

In summary, the present invention proposes to associate and quantify the sensor signals in the blade and SCADA system for preliminary screening through chi-square verification, and then use fault data to further verify and screen the input parameters of the model, and finally use the screened SCADA parameters to establish an NSET model under the normal state of the wind turbine blade as a semi-supervised model to monitor the operation status of the blade, and use the known wind turbine blade fracture data set to verify the modeling results and set the warning threshold. The threshold of the blade fracture warning can be fully mined based on the existing SCADA data, so as to achieve the purpose of online real-time and efficient monitoring. The wind turbine blade fracture warning method based on the association analysis of SCADA data provided by the present invention can realize online real-time fracture warning of various types of wind turbine blades, is more universal, can effectively avoid huge economic losses and safety hazards, and has broad prospects for promotion and application.

Additional aspects and advantages of the invention will become apparent from the following description, or may be learned by practice of the invention.

BRIEF DESCRIPTION OF THE DRAWINGS

FIG1 shows a schematic flow chart of a wind turbine blade fracture early warning method based on correlation analysis of SCADA data according to an embodiment of the present invention;

FIG. 2 shows a schematic flow chart of parameter screening and modeling of SCADA data according to an embodiment of the present invention;

FIG3 shows a Euclidean distance curve graph of the training model output corresponding to the unscreened parameters according to an embodiment of the present invention;

FIG4 shows a Euclidean distance curve graph of a training model output corresponding to parameters selected by using chi-square validation and DBSCAN clustering algorithm according to an embodiment of the present invention;

FIG5 shows a Euclidean distance curve diagram of the final model output of an embodiment of the present invention;

FIG6 shows a schematic flow chart of a process memory matrix construction procedure according to an embodiment of the present invention;

FIG7 shows a schematic flow chart of real-time monitoring of wind turbine blade fracture according to an embodiment of the present invention;

FIG8 shows a Euclidean distance curve graph output when a verification set in the data corresponding to the finally screened parameters is input into the finally established NSET model under normal working conditions of the blade according to an embodiment of the present invention;

FIG9 shows a distribution diagram of wind turbine blade warning thresholds according to an embodiment of the present invention;

FIG10 shows a Euclidean distance curve graph output when the original SCADA data before and after the wind turbine blade fracture in January is input into the NSET model under the normal working condition of the blade finally established according to an embodiment of the present invention;

FIG11 shows a Euclidean distance curve graph output when the original SCADA data before and after the wind turbine blade fracture in February is input into the NSET model under the normal working condition of the blade finally established according to an embodiment of the present invention;

FIG12 shows a Euclidean distance curve graph output when the original SCADA data before and after the wind turbine blade fracture from March 1 to March 10 is input into the NSET model under the normal working condition of the blade finally established according to an embodiment of the present invention;

FIG13 shows a Euclidean distance curve graph output when the original SCADA data before and after the wind turbine blade fracture from March 10 to March 14 is input into the NSET model under the normal working condition of the blade finally established according to an embodiment of the present invention;

FIG. 14 shows a schematic block diagram of a computer device according to an embodiment of the present invention.

Detailed ways

In order to more clearly understand the above-mentioned purpose, features and advantages of the present invention, the present invention is further described in detail below in conjunction with the accompanying drawings and specific embodiments. It should be noted that the embodiments of the present application and the features in the embodiments can be combined with each other without conflict.

In the following description, many specific details are set forth to facilitate a full understanding of the present invention. However, the present invention may also be implemented in other ways different from those described herein. Therefore, the protection scope of the present invention is not limited to the specific embodiments disclosed below.

FIG1 shows a schematic flow chart of a wind turbine blade fracture early warning method based on SCADA data association analysis according to an embodiment of the present invention. As shown in FIG1 , by reducing the dimension of the historical SCADA data before and after the wind turbine blade fracture, the various sensor parameters related to the wind turbine blade operating state are screened out for the first time. The original data corresponding to the screened parameters in the SCADA data of the wind turbine blade in the normal operating state for a period of time are cleaned, and the data corresponding to the screened sensor parameters in the SCADA data of the wind turbine blade in the normal operating state after data cleaning are extracted, and the extracted data are divided into a training set and a verification set, and then the training set is used for NSET modeling (NSET modeling) to establish an NSET model (establishing a normal state model) under the normal operating state of the wind turbine blade. And historical data verification and further parameter screening are used to finally screen out various parameters related to the operating state of the wind turbine blade. Further, the input real-time related observation data can be used to obtain the overall Euclidean distance curve and the fitting curve of the predicted value and the observed value of each parameter according to the model output residual, so as to judge the operating status of the wind turbine blade at this time, and further judge whether the blade has the risk of fracture, and notify the wind farm staff if it exceeds the warning threshold.

Specifically, SCADA data is data collected by a sensor group in a SCADA system of a wind turbine generator set, and the SCADA data includes various parameters: power of the wind turbine generator set, generator speed, rotor speed, wind turbine blade angle, grid-side current and wind speed; the wind turbine blade fracture early warning method based on correlation analysis of SCADA data includes: obtaining a first group of SCADA data before and after the wind turbine blade fracture in a certain period of time; using chi-square verification to screen the first group of SCADA data to screen out various parameters related to the operating status of the wind turbine blade using chi-square verification; adding the wind speed parameter to the various parameters related to the operating status of the wind turbine blade screened out using chi-square verification to screen out various parameters related to the operating status of the wind turbine blade for the first time; obtaining the normal operating status of the wind turbine blade in a certain period of time. The second group of SCADA data under normal operation is extracted from the second group of SCADA data; the original data corresponding to the various parameters related to the operation state of the wind turbine blades that are first screened out are extracted from the second group of SCADA data; a wind speed-power scatter plot is fitted according to the wind speed data and the power data of the wind turbine generator set in the second group of SCADA data after data extraction; the wind speed-power scatter plot is cleaned by using the DBSCAN clustering algorithm; the second group of SCADA data after data cleaning is divided into a first training set and a first validation set; according to the first training set and the first validation set, a first NSET model under normal operation of the wind turbine blades is built; the original first group of SCADA data is input into the first NSET model, and the residual E between the predicted value output by the first NSET model and the corresponding observed value is calculated. _n , and fitting the first Euclidean distance curve corresponding to the first group of SCADA data according to the predicted value and the observed value, so as to observe that the first Euclidean distance curve begins to be in an upward trend after a certain position of the breakpoint delay; calculating the fault cumulative contribution rate e _i of each parameter related to the operating state of the wind turbine blade screened out using the chi-square validation according to the residual E _n ; removing the fault cumulative contribution rate e i of each parameter related to the operating state of the wind turbine blade screened out using the chi-square validation _i is less than a certain set value, and the wind speed parameter is added to the various parameters related to the operating state of the wind turbine blades after the removal, so as to screen out the various parameters related to the operating state of the wind turbine blades for the second time; extract the data corresponding to the various parameters related to the operating state of the wind turbine blades screened out for the second time in the second group of SCADA data after data cleaning, and divide the extracted data into a second training set and a second verification set; according to the second training set and the second verification set, establish a second NSET model under the normal operating state of the wind turbine blades; obtain the Euclidean distance curve corresponding to the second verification set, and use the maximum value of the Euclidean distance on the Euclidean distance curve corresponding to the second verification set as the first threshold; input the original first group of SCADA data into the second NSET model, and calculate the residual E _n between the predicted value output by the second NSET model and the corresponding observed value ', and fitting the second Euclidean distance curve corresponding to the first group of SCADA data according to the predicted value and the observed value, so as to observe that the second Euclidean distance curve begins to be in an upward trend at the position of the break point; obtaining the maximum value of the Euclidean distance on the second Euclidean distance curve before the break point as the second threshold value; obtaining the maximum value of the Euclidean distance on the second Euclidean distance curve after the break point as the third threshold value; obtaining the SCADA data collected by the sensor group in the SCADA system of the wind turbine in real time: inputting the SCADA data obtained in real time into the second NSET model, and calculating the residual E _n ' between the predicted value output by the second NSET model and the corresponding observed value, and fitting the third Euclidean distance curve corresponding to the SCADA data obtained in real time according to the predicted value and the observed value; according to the values of each Euclidean distance corresponding to the third Euclidean distance curve, as well as the first threshold value, the second threshold value and the third threshold value, the wind turbine blade of the wind turbine is given a fracture warning.

In this embodiment, the data collected by the sensor group in the SCADA system of the wind turbine set at the beginning include: data corresponding to parameters such as wind turbine set power, generator speed, rotor speed, wind turbine blade angle, grid-side current and wind speed.

In a specific embodiment, the "certain period of time" in the first set of SCADA data before and after the wind turbine blade breaks is obtained. The certain period of time can be one minute, the timestamp is 0.02s, and there are 3000 data in total. The second NSET model is the NSET model under the normal working condition of the blade finally established.

In one embodiment of the present invention, the fracture warning of the wind turbine blades of the wind turbine set according to the value of the Euclidean distance on the third Euclidean distance curve, as well as the first threshold, the second threshold, and the third threshold, specifically includes: determining the range of the values of each corresponding Euclidean distance on the third Euclidean distance curve; when the value of a certain Euclidean distance corresponding to the third Euclidean distance curve is greater than or equal to 0 and less than or equal to the first threshold, it is judged that the wind turbine blades of the wind turbine set are in a normal operating state; when the value of a certain Euclidean distance corresponding to the third Euclidean distance curve is greater than the first threshold and less than or equal to the second threshold, it is judged that the wind turbine blades of the wind turbine set are in an abnormal state; when the value of a certain Euclidean distance corresponding to the third Euclidean distance curve is greater than or equal to the third threshold, it is judged that the wind turbine blades of the wind turbine set are in a broken state.

In one embodiment of the present invention, the parameters related to the operating state of the wind turbine blades screened out using chi-square verification are: wind turbine power, generator speed, rotor speed, wind turbine blade angle, first grid-side current, second grid-side current and third grid-side current; the parameters related to the operating state of the wind turbine blades screened out for the first time are: wind speed, wind turbine power, generator speed, rotor speed, wind turbine blade angle, first grid-side current, second grid-side current and third grid-side current;

The expression of the residual E _n is:

_En = _xn - _xn ^*

Wherein, x _n is the observed value of the nth parameter in the first group of SCADA data; x _n ^* is the predicted value corresponding to the nth parameter in the first group of SCADA data; En _is the difference between the observed value of the nth parameter and the predicted value corresponding to the nth parameter; x ₁ to x ₇ are the parameters related to the operating state of the wind turbine blade screened out using the chi-square validation; and the _calculation of the fault cumulative contribution rate e i of the parameters related to the operating state of the wind turbine blade screened out using the chi-square validation according to the residual _En specifically includes:

Calculating the number of failures e _c of each parameter related to the wind turbine blade operation state screened out using chi-square validation according to the residual E _n ;

The expression of the fault number e _c is:

Let a = number of observation vector groups/number of observation quantity groups input each time, and _calculate the cumulative fault contribution rate e _{i of each parameter related to the wind turbine blade operation state screened by using chi-square validation according to the number of faults e c} ;

The expression of the cumulative contribution rate of faults e _i is:

And the removing of the parameters whose cumulative fault contribution rate e _i is less than a certain set value from the various parameters related to the operating state of the wind turbine blades screened out by using the chi-square validation, and adding the wind speed parameter to the various parameters related to the operating state of the wind turbine blades after the removal, so as to screen out the various parameters related to the operating state of the wind turbine blades for the second time, specifically includes: removing the parameters whose cumulative fault contribution rate e _i is less than a certain set value from x ₁ to x ₇ , and adding the wind speed parameter to the various parameters related to the operating state of the wind turbine blades after the removal, so as to screen out the various parameters related to the operating state of the wind turbine blades for the second time; wherein the various parameters related to the operating state of the wind turbine blades screened out for the second time are: wind speed, wind turbine power, generator speed, rotor speed and wind turbine blade angle.

In one embodiment of the present invention, establishing a first NSET model of a wind turbine blade in a normal operating state according to the first training set and the first validation set specifically includes: obtaining an original NSET model, inputting the first training set into the original NSET model, so as to obtain a trained NSET model;

The overall observation matrix of the wind turbine is represented as an n×b-sized M _n×b , and the expression of M _n×b is:

Wherein, n is the time state, b is the number of observed variables at each time; the row vector of the matrix Mn _×b is _Xi = [ _xi ( _t1 ) _xi ( _t2 ) ... _xi ( _tb )], and the row vector of the matrix Mn _×b is all the observed values of a given observation parameter _Xi within a certain observation time period; the column vector of the matrix Mn _×b is X( _tj ) = [ _x1 ( _tj ) _x2 ( _tj ) ... _xb ( _tj )] ^T , and the column vector of the matrix Mn _×b is the observed values of all observed parameters at time _tj ; the parameters of a period of time selected from the Mn _×b are recorded as the historical observation matrix K, and the historical observation matrix K is the health status of each observation parameter, then the expression of the historical observation matrix K is:

A portion of state data is selected from the historical observation matrix K, and the selected portion of state data is used to form a process memory matrix D _n . The process matrix D _n can be expressed as:

The observation matrix X _obs and D _n corresponding to the observation matrix X _obs are input into the expression , obtain a prediction output matrix X _est ; set the first validation set as an observation matrix X _obs ; input the first validation set into the trained NSET model; the NSET model after inputting the first validation set outputs a corresponding prediction output matrix X _est ; calculate the residual between the observation matrix X _obs corresponding to the first validation set and the corresponding prediction output matrix X _est , and fit the Euclidean distance curve corresponding to the first validation set according to the observation matrix X _obs corresponding to the first validation set and the corresponding prediction output matrix X _est to establish a first NSET model in a normal operating state of the wind turbine blade; and the second NSET model under normal operating state of the wind turbine blade is established according to the second training set and the second validation set, specifically including: obtaining an original NSET model, inputting the second training set into the original NSET model to obtain a trained NSET model; setting the second validation set as an observation matrix X _obs ; inputting the second validation set into the trained NSET model; the NSET model after inputting the second validation set outputs a corresponding prediction output matrix X _est ; calculate the residual between the observation matrix X _obs corresponding to the second validation set and the corresponding prediction output matrix X _est , and fit the Euclidean distance curve corresponding to the first validation set according to the observation matrix X _obs corresponding to the second validation set and the corresponding prediction output matrix X est _est is used to fit the Euclidean distance curve corresponding to the second validation set to establish a second NSET model under normal operating conditions of the wind turbine blade.

In one embodiment of the present invention, the method of using chi-square validation to screen the first group of SCADA data to screen out various parameters related to the operating state of the wind turbine blades specifically includes: establishing an original hypothesis H0, wherein the original hypothesis H0 is that the operating state of the wind turbine blades is independent of various parameters in the first group of SCADA data; using the data of the operating state of the wind turbine blades as the first variable threshold; using a certain parameter in the first group of SCADA data as the second variable threshold each time; recording the actual value of the number of data of the first variable threshold when the wind turbine blades are normal as a, the actual value of the number of data of the wind turbine blades when the wind turbine blades are faulty as b, and the actual value of the total number of data of the wind turbine blades as a+b; recording the actual value of the number of data of the yth parameter in the SCADA data when the wind turbine blades are normal as _cy , the actual value of the number of data of the wind turbine blades when the wind turbine blades are faulty as _dy , and the actual value of the total number of data of the wind turbine blades as _cy + _dy , wherein y is a positive integer; recording the actual value of the total number of data of the first variable threshold and the second variable threshold when the wind turbine blades are normal as a+ _cy , and the actual value of the total number of data of the two under the condition of wind turbine blade failure is b+ _dy , and the actual value of the total number of data of the two wind turbine blades is a+b+ _cy + _dy ; the theoretical value of the number of data of the first variable threshold under normal wind turbine blade is (a+b)×(a+ _cy )/(a+b+ _cy + _dy ), the theoretical value of the number of data under wind turbine blade failure is (a+b)×(b+ _dy )/(a+ _b + _cy + _dy ), and the theoretical value of the total number of data of wind turbine blade is a+b; the theoretical value of the number of data of the second variable threshold under normal wind turbine blade is (cy+ _dy )×(a+ _{cy )} /(a+b+ _cy + _dy ), and the theoretical value of the number of data of wind turbine blade failure is ( _cy +dy)×(b+ _dy )/(a+b+ _cy + _dy ), and the theoretical value of the total number of data of wind turbine blades is _cy + _dy ; the degree of freedom is set to 1; the chi-square value is calculated according to the chi-square value calculation formula, the chi-square value is used to measure the difference between each of the actual values and each of the theoretical values, and the chi-square value calculation formula is: χ ² =∑(AT) ² /T

Wherein, χ ² is the chi-square value, A is each actual value, and T is each theoretical value; according to the degree of freedom and chi-square value table, find the corresponding P value, and the P value is the probability of committing a first-type true rejection error; select a parameter in the first group of SCADA data corresponding to the P value less than 0.05 as the parameter related to the operating state of the wind turbine blade, so as to screen out the various parameters related to the operating state of the wind turbine blade in turn.

In one embodiment of the present invention, the DBSCAN clustering algorithm is used to clean the wind speed-power scatter plot, specifically including: taking all points in the wind speed-power scatter plot as sample data S, and marking each data point in the sample data S as unprocessed; assigning initial values to Eps and Minpts, wherein Eps is the neighborhood distance threshold of a certain data point p in the sample data S, and Minpts is the minimum number of data points in the neighborhood with a radius of Eps of a certain data point p in the sample data S; setting the neighborhood with a radius of Eps of the certain data point p to N _Eps (p); clustering the high-density area formed by Eps and Minpts; the clustering of the high-density area formed by Eps and Minpts specifically includes: judging whether the certain data point p has been added to a cluster or has been listed as noise; if the certain data point p has been added to a cluster or has been listed as noise, the classification is terminated; if the certain data point p has not been added to a cluster and has not been listed as noise, judging N _Eps (p) contains at least Minpts objects; if there are at least Minpts objects in N _Eps (p), a new cluster U is constructed and the data point p is added to U; if the number of objects in N _Eps (p) is less than Minpts, the data point p is listed as a boundary point or noise.

In this embodiment, a circle with point p as the center and Eps as the radius is formed. Whether the number of all points in this circle is greater than Minpts, if it is satisfied, it is retained, if not, it is regarded as noise. N _Eps (p) refers to traversing N points, that is, all points. The parameter setting of the DBSCAN clustering algorithm is relatively complex and requires joint repeated adjustment. In a specific embodiment, after normalizing the data, Eps is set to 4.5 and Minpts is set to 18.5 to remove noise.

FIG2 shows a schematic flow chart of parameter screening and modeling of SCADA data in one embodiment of the present invention. As shown in FIG2, firstly, the SCADA data before and after the wind turbine blade fracture in a certain period of time is preliminarily screened using chi-square verification, and the number of overall blade failure data and the number of abnormal parameter data in SCADA are recorded when the wind turbine blade fails, and recorded in the following table:

Calculate the theoretical value based on the actual value and record it in the table:

Assumption H0: The operating status of wind turbine blades is independent of other parameters; calculate the test statistic to measure the difference between the actual value and the theoretical value, and use the chi-square value calculation formula to obtain the chi-square value; look up the table based on the degrees of freedom and chi-square value to find the corresponding P value, and decide whether to reject or accept the null hypothesis H0, and preliminarily screen out the SCADA parameters related to the operating status of the blades; select parameters with a P value less than 0.05 as parameters related to the operation of wind turbine blades, that is, reject the null hypothesis, and there is a correlation between the operating status of wind turbine blades and other parameters.

In a specific embodiment, the number of normal and abnormal SCADA parameters before and after the blade breaks is counted, and the parameters related to the blade operating status obtained by filtering through the above algorithm are shown in the following table. After filtering the original dozens of parameters, the parameters related to the wind turbine blade operating status screened out using chi-square verification are: power, generator speed, rotor speed, blade angle, grid-side current 1, grid-side current 2, grid-side current 3, a total of 7 parameters, realizing dimensionality reduction processing of the data.

Furthermore, as shown in Figure 2, because the chi-square validation did not obtain the relationship between wind speed and blade status, in order to classify the working conditions of SCADA data, the wind speed parameter was added. The DBSCAN clustering algorithm was used to clean the wind speed-power scatter plot to remove null values, singular points, power-limited data, etc. As shown in Figure 3, the Euclidean distance of the model established before data screening after the blade broke was 0.42, and the Euclidean distance of the model established after data screening after the blade broke was 8.24 as shown in Figure 4, which was significantly better than before data processing. The results show that the chi-square validation and DBSCAN data cleaning process for SCADA data improves the accuracy of the model and solves the problems of long modeling time and low accuracy caused by SCADA data redundancy and noise.

Furthermore, as shown in Figure 2, the SCADA data of wind turbine blades under normal operating conditions that were initially screened out were divided into training set and verification set according to a ratio of 7:3 to build the NSET model; the process memory matrix Dn of each scheme was obtained, where n is a positive integer; the key to whether the NSET modeling is successful is whether the process memory matrix is successfully constructed.

The overall observation matrix of the wind turbine is represented as an n×b-sized M _n×b , and the expression of M _n×b is:

Wherein, n is the time state, b is the number of observed variables at each time; the row vector of the matrix Mn _×b is _Xi = [ _xi ( _t1 ) _xi ( _t2 ) ... _xi ( _tb )], and the row vector of the matrix Mn _×b is all the observed values of a given observation parameter _Xi within a certain observation time period; the column vector of the matrix Mn _×b is X( _tj ) = [ _x1 ( _tj ) _x2 ( _tj ) ... _xb ( _tj )] ^T , and the column vector of the matrix Mn _×b is the observed values of all observed parameters at time _tj ; the parameters of a period of time selected from the Mn _×b are recorded as the historical observation matrix K, and the historical observation matrix K is the health status of each observation parameter, then the expression of the historical observation matrix K is:

A portion of state data is selected from the historical observation matrix K, and the selected portion of state data is used to form a process memory matrix D _n . The process matrix D _n can be expressed as:

The observation matrix X _obs and D _n corresponding to the observation matrix X _obs are input into the expression In the above equation, we get the prediction output matrix _Xest ;

The seven training sets of the SCADA data of wind turbine blades in normal operation that were initially screened were set as the observation matrix _Xobs ; the three validation sets were input into the trained NSET model; the NSET model after inputting the validation set output the corresponding prediction output matrix _Xest ; the corresponding Euclidean distance curve was fitted according to the observation matrix _Xobs and the prediction output matrix _Xest .

Furthermore, the Euclidean distance curve is used to verify the validity of the model. The Euclidean distance can reflect the absolute value of individual differences. For example, if the coordinates of two points are ( _xi , _yi ) and ( _xj , _yj ), then the Euclidean distance is The fracture data (i.e., SCADA data before and after the wind turbine blade fracture for one minute, with a timestamp of 0.02s, a total of 3000 data) are input to verify the validity of the model. As shown in Figure 4, the 3000 sample data before and after the blade fracture are input into the blade normal state model established by NSET. The blade fractures at the 1500th sample point. After the data is input into the normal model, the Euclidean distance curve rises significantly after the 1500 sample points. The model can diagnose and monitor blade fractures.

Furthermore, due to the limitations of the chi-square validation, the break point of the Euclidean distance curve that should have been at 1500 samples has changed and has been delayed to 1664 samples, so further screening is needed. Continuing to analyze the residual E _n = x _n -x _n ^* between the observation matrix and the prediction output matrix can determine the abnormal parameters. Among them, x _n is the observed value of the nth parameter in the SCADA data before and after the wind turbine blade breaks; x _n ^* is the predicted value corresponding to the nth parameter in the SCADA data before and after the wind turbine blade breaks; En _is the difference between the observed value of the nth parameter and the predicted value corresponding to the nth parameter; x ₁ to x ₇ are the various parameters related to the operating status of the wind turbine blade screened out using the chi-square validation;

Calculating the number of failures e _c of each parameter related to the wind turbine blade operation state screened out using chi-square validation according to the residual E _n ;

The expression of the number of failures e _c is:

Let a = number of observation vector groups/number of observation quantity groups input each time, and _calculate the cumulative fault contribution rate e _{i of each parameter related to the wind turbine blade operation state screened by using chi-square validation according to the number of faults e c} ;

The expression of the cumulative contribution rate of failure ei is:

The parameters whose cumulative fault contribution rate e _i is less than a certain set value among x ₁ to x ₇ are removed, and the wind speed parameter is added to the parameters related to the operating state of the wind turbine blades after the removal, that is, the parameters with low fault contribution rate are removed, and the parameters with higher sensitivity to blade breakage are left as the final modeling parameters, so as to finally screen out the parameters related to the operating state of the wind turbine blades; wherein the parameters related to the operating state of the wind turbine blades finally screened out are: wind speed, wind turbine power, generator speed, rotor speed and wind turbine blade angle.

Input the blade fracture data into the model (i.e., SCADA data before and after the wind turbine blade fracture for one minute, with a timestamp of 0.02s, and a total of 3000 data), and the overall Euclidean distance result is shown in Figure 5. Comparing Figures 4 and 5, it can be seen that the Euclidean distance curve in Figure 5 has a large fluctuation before the fracture, and begins to change at 1500 sample points, which is more in line with the actual working conditions and more accurate. The reliability analysis is used to screen the parameters based on the reliability of the modeling parameters, solving the problem of modeling parameter reliability.

FIG6 shows a schematic flow chart of a process memory matrix construction procedure of an embodiment of the present invention. A portion of state data is selected from the historical observation matrix K, and the selected portion of state data is used to form a process memory matrix _Dn . The process memory matrix D is constructed so that its internal k observation vectors X(1) X(2)...X(k) can cover the normal working space of the device as much as possible.

The method of selecting a part of state data from the historical observation matrix K and forming a process memory matrix _Dn with the selected part of state data specifically includes: setting each observation vector of the historical observation matrix K to be composed of n variables; for each of the n variables, dividing the interval [0, 1] into h equal parts, and finding a number of observation vectors X(1) X(2) ... X(k) from the historical observation matrix K with a step size of 1/h and adding them to the process memory matrix _Dn ;

As shown in FIG6 , for each of the n variables, the interval [0, 1] is equally divided into h parts, and a number of observation vectors X(1)X(2)...X(k) are found from the historical observation matrix K with a step size of 1/h and added to the process memory matrix D _n , specifically including: setting i=1; wherein i is a positive integer; executing A=1/h*i; wherein h is a positive integer; setting k=1; wherein k is a positive integer; determining whether |X(k)-A| is less than δ; wherein δ is a positive number; when |X(k)-A| is less than δ, adding X(k) to the process memory matrix D n _n ; when |X(k)-A| is greater than or equal to δ, determine whether k is greater than M; wherein M is the number of columns of the historical observation matrix K; when k is less than or equal to M, execute k=k+1, and return to the step of determining whether |X(k)-A| is less than δ; when k is greater than M, determine whether i is greater than h; when i is less than or equal to h, execute i=i+1, and return to the step of executing A=1/h*i; when i is greater than h, execution ends.

In a specific embodiment, δ is 0.001.

Each observation vector of the normal working space of the equipment is composed of n variables, and its observation values have been normalized. For each variable, [0,1] is divided into h equal parts, and several observation vectors are found from the set K with a step size of 1/h and added to the matrix D. In the figure, δ is a small positive number. For the remaining n-1 variables, the same process as the diagram is used to select observation vectors from the set K with a step size of 1/h and add them to D. This method is used to construct a process memory matrix, which can select the historical records corresponding to the different measurement values of the n variables that make up the observation vector into the matrix D, so that it can better cover the normal working space of the equipment.

FIG7 shows a schematic flow chart of real-time monitoring of wind turbine blade fracture according to an embodiment of the present invention. As shown in FIG7 , the cleaned data corresponding to the finally screened parameters are divided into a training set and a validation set according to a ratio of 7:3 for a second NSET modeling, and the training set and the validation set are used as the best model for data fusion of the screened parameters, and the validation set is input to obtain the maximum normal Euclidean distance threshold and recorded; the fracture data is input to obtain the blade fracture Euclidean distance and recorded. The real-time SCADA parameters are input to compare the real-time data with the Euclidean distance curve of the historical model to determine whether the blade fracture warning threshold is exceeded, and a graded blade fracture warning is performed.

In this embodiment, as shown in Figure 8, the maximum normal Euclidean distance threshold of the model is 0.13. As shown in Figure 5, the Euclidean distance curve rises sharply at the blade break, and the Euclidean distance value after the blade break is 8.24, which quantifies the blade damage and solves the problem of quantifying different degrees of blade damage.

FIG5 shows a distribution diagram of blade warning threshold division according to an embodiment of the present invention. As shown in FIG5, the data of several months before and after the blade fracture is used as input and input into the NSET model under the normal working condition of the blade that is finally established, and the fracture threshold is divided by the change of Euclidean distance. Since the warning thresholds obtained by units of different models and blades of different lengths are not all the same, it is necessary to compare the Euclidean distance curve before the target monitoring wind turbine blade fracture to obtain the best warning threshold. Specifically, the Euclidean distance between 0 and the maximum value of the verification set is used as the normal range of blade operation status, the value between the maximum value of the verification set and the maximum value of the Euclidean distance before the blade fracture is used as the range of blade status inspection, and the maximum value of the Euclidean distance before the blade fracture is used as the basis for emergency inspection of the blade, and the Euclidean distance value after the blade fracture is used as the signal that the blade has fractured. There may be differences in the warning thresholds of different models. In the specific embodiment, the data from January to March 12 are input into the best model. Because the blade fractured on March 12, the main analysis is placed on the time before the fracture. As shown in Figures 5, 8, 9, 10 to 13, the maximum threshold of the Euclidean distance when the blade is running normally is 0.13, the maximum Euclidean distance before the blade breaks is 2.21, and the Euclidean distance after the blade breaks is 8.24. Therefore, the Euclidean distance of 2.21 can be used as the blade break warning threshold. When the Euclidean distance is between 0.13 and 2.21, the blade needs to be checked at an appropriate time. When the Euclidean distance reaches 2.21, the machine should be stopped for inspection in time, which solves the problem of setting the blade break warning threshold.

As shown in FIG. 14 , a computer device 1400 includes: a memory 1402 , a processor 1404 , and a computer program stored in the memory 1402 and executable on the processor 1404 . When the processor 1404 executes the computer program, the steps of the method in any of the above embodiments are implemented.

The computer device 1400 provided by the present invention, when the processor 1404 executes the computer program, associates and quantifies the blades with the sensor signals in the SCADA system through chi-square verification for preliminary screening, and then uses the fault data to further verify and screen the input parameters of the model. Finally, the screened SCADA parameters are used to establish an NSET model of a wind turbine blade in a normal state as a semi-supervised model to monitor the blade operation state. The modeling results are verified using a known wind turbine blade fracture data set and an early warning threshold is set. The threshold of the blade fracture early warning can be fully mined based on the existing SCADA data, thereby achieving the purpose of online real-time and efficient monitoring.

The present invention also proposes a computer-readable storage medium having a computer program stored thereon. When the computer program is executed by a processor, the steps of the frequency closed-loop control method for an oil beam type pumping unit as described in any of the above embodiments are implemented.

The computer-readable storage medium provided by the present invention, when the computer program is executed by the processor, performs preliminary screening of the association quantification of blades and sensor signals in the SCADA system through chi-square verification, and then uses fault data to further verify and screen the input parameters of the model, and finally uses the screened SCADA parameters to establish an NSET model of a wind turbine blade in a normal state as a semi-supervised model to monitor the blade operation state, and uses a known wind turbine blade fracture data set to verify the modeling results and set the warning threshold, so that the threshold of the blade fracture warning can be fully mined on the basis of the existing SCADA data, thereby achieving the purpose of online real-time and efficient monitoring.

The above description 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 modifications and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims (10)

1. A wind turbine blade fracture early warning method based on correlation analysis of SCADA data, wherein the SCADA data is data collected by a sensor group in a SCADA system of a wind turbine generator set, and the various parameters included in the SCADA data are: wind turbine generator set power, generator speed, rotor speed, wind turbine blade angle, grid-side current and wind speed; characterized in that the early warning method comprises: Obtain the first set of SCADA data before and after the wind turbine blade breaks over a certain period of time; Using chi-square validation to screen the first group of SCADA data, so as to screen out various parameters related to the operating status of the wind turbine blades using chi-square validation; The wind speed parameter is added to various parameters related to the operating state of wind turbine blades screened out using chi-square validation, so as to screen out various parameters related to the operating state of wind turbine blades for the first time; Obtain a second set of SCADA data of wind turbine blades in normal operating state for a certain period of time; Extracting original data corresponding to each parameter related to the operating state of the wind turbine blade screened out for the first time from the second group of SCADA data; According to the wind speed data and the power data of the wind turbine generator set in the second group of SCADA data after data extraction, a wind speed-power scatter plot is fitted; The DBSCAN clustering algorithm is used to perform data cleaning on the wind speed-power scatter plot; The second group of SCADA data after data cleaning is divided into a first training set and a first validation set; According to the first training set and the first validation set, a first NSET model is constructed under normal operating conditions of the wind turbine blade; Inputting the original first set of SCADA data into the first NSET model, calculating the residual E _n between the predicted value output by the first NSET model and the corresponding observed value, and fitting the first Euclidean distance curve corresponding to the first set of SCADA data according to the predicted value and the observed value, so as to observe that the first Euclidean distance curve begins to be in an upward trend after a certain position of the breakpoint delay; Calculating the cumulative fault contribution rate e _i of each parameter related to the wind turbine blade operation state screened out using chi-square validation according to the residual E _n ; Removing the parameters whose cumulative contribution rate e of failure _is less than a certain set value from among the parameters related to the operating state of the wind turbine blades screened out by using the chi-square verification, and adding the wind speed parameter to the parameters related to the operating state of the wind turbine blades after the removal, so as to screen out the parameters related to the operating state of the wind turbine blades for the second time; Extracting data corresponding to the parameters related to the operating state of the wind turbine blades screened out for the second time from the second group of SCADA data after data cleaning, and dividing the extracted data into a second training set and a second validation set; According to the second training set and the second validation set, a second NSET model is established under the normal operating state of the wind turbine blade; Obtaining a Euclidean distance curve corresponding to the second validation set, and taking the maximum value of the Euclidean distance on the Euclidean distance curve corresponding to the second validation set as a first threshold; Inputting the original first group of SCADA data into the second NSET model, calculating the residual E _n ' between the predicted value output by the second NSET model and the corresponding observed value, and fitting the second Euclidean distance curve corresponding to the first group of SCADA data according to the predicted value and the observed value, so as to observe that the second Euclidean distance curve starts to be in an upward trend at the position of the breakpoint; obtaining a maximum value of the Euclidean distance on the second Euclidean distance curve before the breaking point as a second threshold; obtaining a maximum value of the Euclidean distance on the second Euclidean distance curve after the breakpoint as a third threshold; Real-time acquisition of SCADA data collected by the sensor group in the SCADA system of the wind turbine: Inputting the SCADA data acquired in real time into the second NSET model, calculating the residual E _n ″ between the predicted value output by the second NSET model and the corresponding observed value, and fitting a third Euclidean distance curve corresponding to the SCADA data acquired in real time according to the predicted value and the observed value; According to the values of the corresponding Euclidean distances on the third Euclidean distance curve, as well as the first threshold, the second threshold, and the third threshold, a fracture warning is issued for the wind turbine blades of the wind turbine generator set. 2. The wind turbine blade fracture early warning method based on association analysis of SCADA data according to claim 1 is characterized in that the wind turbine blade fracture early warning of the wind turbine generator set is performed according to the value of the Euclidean distance on the third Euclidean distance curve, and the first threshold, the second threshold, and the third threshold, specifically comprising: Determine the range of the values of each corresponding Euclidean distance on the third Euclidean distance curve; When the value of a certain Euclidean distance corresponding to the third Euclidean distance curve is greater than or equal to 0 and less than or equal to the first threshold, it is judged that the wind turbine blades of the wind turbine generator set are in a normal operating state; When the value of a certain Euclidean distance corresponding to the third Euclidean distance curve is greater than the first threshold value and less than or equal to the second threshold value, it is determined that the wind turbine blade of the wind turbine generator set is in an abnormal state; When the value of a certain Euclidean distance corresponding to the third Euclidean distance curve is greater than or equal to the third threshold, it is determined that the wind turbine blade of the wind turbine generator set is in a broken state. 3. The wind turbine blade fracture early warning method based on correlation analysis of SCADA data according to claim 1 is characterized in that: The parameters related to the operation status of wind turbine blades screened out by using chi-square verification are: wind turbine power, generator speed, rotor speed, wind turbine blade angle, first grid-side current, second grid-side current and third grid-side current; The first screened out parameters related to the operating state of the wind turbine blades are: wind speed, wind turbine power, generator speed, rotor speed, wind turbine blade angle, first grid-side current, second grid-side current and third grid-side current; The expression of the residual E _n is: _En = _xn - _xn ^* Wherein, _xn is the observed value of the nth parameter in the first set of SCADA data; _xn ^* is the predicted value corresponding to the nth parameter in the first set of SCADA data; En _is the difference between the observed value of the nth parameter and the predicted value corresponding to the nth parameter; _x1 to _x7 are the parameters related to the operating state of the wind turbine blade screened using the chi-square validation; and The calculating, according to the residual E _{n ,} the fault cumulative contribution rate e _i of each parameter related to the wind turbine blade operation state screened out by using the chi-square validation specifically includes: Calculating the number of failures e _c of each parameter related to the wind turbine blade operation state selected by using chi-square validation according to the residual E _n ; The expression of the fault number e _c is: Let a = number of observation vector groups/number of observation quantity groups input each time, and _calculate the cumulative fault contribution rate e _{i of each parameter related to the wind turbine blade operation state screened by using chi-square validation according to the number of faults e c} ; The expression of the cumulative contribution rate of faults e _i is: And the removing of the parameters whose cumulative contribution rate of faults e _i is less than a certain set value from the various parameters related to the operating state of the wind turbine blades screened out by using the chi-square verification, and adding the wind speed parameter to the various parameters related to the operating state of the wind turbine blades after the removal, so as to screen out the various parameters related to the operating state of the wind turbine blades for the second time, specifically includes: The parameters _x1 to _x7 whose cumulative fault contribution rate e _i is less than a set value are removed, and the wind speed parameter is added to the parameters related to the operating state of the wind turbine blades after the removal, so as to screen out the parameters related to the operating state of the wind turbine blades for the second time; wherein the parameters related to the operating state of the wind turbine blades screened out for the second time are: wind speed, wind turbine power, generator speed, rotor speed and wind turbine blade angle. 4. The wind turbine blade fracture early warning method based on association analysis of SCADA data according to any one of claims 1 to 3, characterized in that the first NSET model under the normal operating state of the wind turbine blade is established according to the first training set and the first validation set, specifically comprising: Obtaining an original NSET model, and inputting the first training set into the original NSET model to obtain a trained NSET model; The overall observation matrix of the wind turbine is represented as an n×b-sized M _n×b , and the expression of M _n×b is: Where n is the time state, b is the number of observed variables at each time; the row vector of the matrix Mn _×b is _Xi = [ _xi ( _t1 ) _xi ( _t2 ) ... _xi ( _tb )], and the row vector of the matrix Mn _×b is all the observed values of a given observation parameter _Xi in a certain observation time period; the column vector of the matrix Mn _×b is X( _tj ) = [ _x1 ( _tj ) _x2 ( _tj ) ... _xb ( _tj )] ^T , and the column vector of the matrix Mn _×b is the observed values of all observation parameters at time _tj ; The parameters of a period of time are selected from the M _n×b and recorded as the historical observation matrix K. The historical observation matrix K is the health status of each observation parameter. The expression of the historical observation matrix K is: A portion of state data is selected from the historical observation matrix K, and the selected portion of state data is used to form a process memory matrix D _n . The process matrix D _n can be expressed as: The observation matrix X _obs and D _n corresponding to the observation matrix X _obs are input into the expression In the above equation, we get the prediction output matrix _Xest ; The first validation set is set as the observation matrix X _obs ; Input the first validation set into the trained NSET model; The prediction output matrix _Xest corresponding to the output of the NSET model after inputting the first validation set; Calculating the residual between the observation matrix X _obs corresponding to the first verification set and the corresponding prediction output matrix X _est , and fitting the Euclidean distance curve corresponding to the first verification set according to the observation matrix X _obs corresponding to the first verification set and the corresponding prediction output matrix X _est , so as to establish a first NSET model under normal operating conditions of the wind turbine blade; and The step of establishing a second NSET model under normal operation of the wind turbine blade according to the second training set and the second validation set specifically includes: Obtaining an original NSET model, and inputting the second training set into the original NSET model to obtain a trained NSET model; The second validation set is set as the observation matrix X _obs ; Input the second validation set into the trained NSET model; The prediction output matrix _Xest corresponding to the output of the NSET model after inputting the second validation set; The residual between the observation matrix X _obs corresponding to the second verification set and the corresponding prediction output matrix X _est is calculated, and the Euclidean distance curve corresponding to the second verification set is fitted according to the observation matrix X _obs corresponding to the second verification set and the corresponding prediction output matrix X _est , so as to establish a second NSET model under normal operating state of the wind turbine blade. 5. The wind turbine blade fracture early warning method based on SCADA data correlation analysis according to claim 4 is characterized in that the selecting a part of state data from the historical observation matrix K and using the selected part of state data to form a process memory matrix D _n specifically includes: Set each observation vector of the historical observation matrix K to consist of n variables; For each of the n variables, the interval [0, 1] is divided into h equal parts, and a number of observation vectors X(1) X(2) ... X(k) are found from the historical observation matrix K with a step size of 1/h and added to the process memory matrix _Dn ; For each of the n variables, the interval [0, 1] is divided into h equal parts, and a plurality of observation vectors X(1) X(2) ... X(k) are found from the historical observation matrix K with a step size of 1/h and added to the process memory matrix D _n , specifically including: Set i=1; where i is a positive integer; Execute A=1/h*i; where h is a positive integer; Set k=1; where k is a positive integer; Determine whether |X(k)-A| is less than δ; where δ is a positive number; When |X(k)-A| is less than δ, add X(k) to the process memory matrix Dn; When |X(k)-A| is greater than or equal to δ, determine whether k is greater than M; wherein M is the number of columns of the historical observation matrix K; When k is less than or equal to M, execute k=k+1, and return to the step of determining whether |X(k)-A| is less than δ; When k is greater than M, determine whether i is greater than h; When i is less than or equal to h, execute i=i+1, and return to the step of executing A=1/h*i; When i is greater than h, execution ends. 6. The wind turbine blade fracture early warning method based on correlation analysis of SCADA data according to claim 1 is characterized in that the first set of SCADA data is screened using chi-square validation to screen out various parameters related to the operating status of the wind turbine blade using chi-square validation, specifically including: Establishing an original hypothesis H0, wherein the original hypothesis H0 is that the operating state of the wind turbine blade is independent of each parameter in the first set of SCADA data; Using the data of the wind turbine blade operating status as the first variable threshold; Each time, a parameter in the first set of SCADA data is used as a second variable threshold; Record respectively the actual value of the number of data of the first variable threshold when the wind turbine blade is normal as a, the actual value of the number of data when the wind turbine blade is faulty as b, and the actual value of the total number of data of the wind turbine blade as a+b; Recording respectively the actual value of the number of data of the yth parameter in the SCADA data when the wind turbine blade is normal as _cy , the actual value of the number of data when the wind turbine blade is faulty as _dy , and the actual value of the total number of data of the wind turbine blade as _cy + _dy , wherein y is a positive integer; Record respectively the actual value of the total number of data of the first variable threshold and the second variable threshold when the wind turbine blade is normal as a+c _y , the actual value of the total number of data of the first variable threshold and the second variable threshold when the wind turbine blade is faulty as b+d _y , and the actual value of the total number of data of the wind turbine blade as a+b+c _y +d _y ; The theoretical value of the number of data of the first variable threshold when the wind turbine blade is normal is calculated to be (a+b)×(a+c _y )/(a+b+c _y + _dy ), the theoretical value of the number of data when the wind turbine blade is faulty is calculated to be (a+b)×(b+d _y )/(a+b+c _y + _dy ), and the theoretical value of the total number of data of the wind turbine blade is a+b; The theoretical value of the number of data of the second variable threshold when the wind turbine blade is normal is calculated to be ( _cy + _dy )×(a+ _cy )/(a+b+ _cy + _dy ), the theoretical value of the number of data when the wind turbine blade is faulty is calculated to be ( _cy + _dy )×(b+ _dy )/(a+b+ _cy + _dy ), and the theoretical value of the total number of data of the wind turbine blade is _cy + _dy ; Set the degrees of freedom to 1; The chi-square value is calculated according to the chi-square value calculation formula, and the chi-square value is used to measure the difference between each of the actual values and each of the theoretical values. The chi-square value calculation formula is: χ ² = ∑(AT) ² /T Wherein, χ ² is the chi-square value, A is each of the actual values, and T is each of the theoretical values; According to the table of degrees of freedom and chi-square values, find the corresponding P value, which is the probability of making a first type of true rejection error; A parameter in the first group of SCADA data corresponding to a P value less than 0.05 is selected as a parameter related to the operating state of the wind turbine blade, so as to sequentially screen out various parameters related to the operating state of the wind turbine blade. 7. The wind turbine blade fracture early warning method based on association analysis of SCADA data according to claim 1 is characterized in that the DBSCAN clustering algorithm is used to clean the wind speed-power scatter plot, specifically comprising: Taking all points in the wind speed-power scatter plot as sample data S, and marking each data point in the sample data S as an unprocessed state; Assign initial values to Eps and Minpts, where Eps is the neighborhood distance threshold of a data point p in the sample data S, and Minpts is the minimum number of data points in the neighborhood of a data point p in the sample data S with a radius of Eps; The neighborhood of the data point p with a radius of Eps is set to N _Eps (p); Cluster the high-density areas formed by Eps and Minpts; The clustering of the high-density areas formed by Eps and Minpts specifically includes: Determine whether the data point p has been added to a cluster or has been classified as noise; If the data point p has been added to a cluster or has been classified as noise, the classification ends; If the data point p is not added to a cluster and is not classified as noise, then determine whether there are at least Minpts objects in N _Eps (p); If there are at least Minpts objects in N _Eps (p), a new cluster U is constructed and the data point p is added to U; If the number of objects in N _Eps (p) is less than Minpts, the data point p is listed as a boundary point or noise. 8. The wind turbine blade fracture early warning method based on correlation analysis of SCADA data according to claim 7, characterized in that the Eps is set to 4.5, and the Minpts is set to 18.5. 9. A computer device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the steps of the wind turbine blade fracture early warning method based on correlation analysis of SCADA data as described in any one of claims 1 to 8 when executing the computer program. 10. A computer-readable storage medium having a computer program stored thereon, wherein when the computer program is executed by a processor, the steps of the wind turbine blade fracture early warning method based on correlation analysis of SCADA data as described in any one of claims 1 to 8 are implemented.