Wind turbine generator operation and maintenance optimization decision-making system based on big data

Abstract

The invention belongs to the technical field of wind turbine generator management, and discloses a big data-based wind turbine generator operation and maintenance optimization decision-making system. Comprising a data acquisition module used for acquiring data of a wind turbine generator; the data preprocessing module is used for preprocessing the data of the wind turbine generator to obtain preprocessed data of the wind turbine generator; the security maintenance module is used for performing vulnerability maintenance on the wind turbine generator based on the preprocessed wind turbine generator data to obtain security wind turbine generator data; the cost control module is used for processing the safe wind turbine generator data to obtain an optimal operation parameter combination; applying the optimal operation parameter combination to the wind turbine generator; according to the method, the data of the wind turbine generator is accurately processed, meanwhile, the vulnerability of the wind turbine generator is accurately identified, the safety of the wind turbine generator is improved, and the operation efficiency is improved on the basis of reducing the operation cost and the operation energy consumption.

Description

Technical Field

[0001] The present invention relates to the technical field of wind turbine management, and more specifically, to a wind turbine operation and maintenance optimization decision system based on big data. Background Art

[0002] The patent application with the publication number CN118134456A discloses an artificial intelligence-based wind turbine operation and maintenance management system and method thereof, the system includes an operation and maintenance center, an early warning platform, a personnel allocation module, a unit equipment monitoring module and a data transmission platform, the unit equipment monitoring module is used to obtain wind turbine operation data; the personnel allocation module is used to obtain operation and maintenance workshop personnel data, and generate management information based on the personnel data. By obtaining wind turbine operation data and operation and maintenance workshop personnel data, fault detection is performed based on the unit operation data of the multi-component system of the wind turbine, early warning information is generated based on the fault detection results, and maintenance information is generated through the early warning information of the wind turbine and the management information of the personnel data, the wind turbine fault can be quickly and effectively identified and judged, and the operation and maintenance efficiency of the wind turbine can be improved.

[0003] In the field of existing technologies, data collection for wind turbines often faces the situation of missing values ​​and noisy data in the data. Traditional preprocessing methods cannot efficiently process missing values ​​and noisy data, such as using linear interpolation algorithms to fill missing values ​​and using high-pass or low-pass filters to process noisy data. For security vulnerabilities in wind turbines, the existing technical field often relies on manual inspection or simple threshold judgment to identify security vulnerabilities, which is ineffective and easy to miss potential security vulnerabilities. Existing wind turbine equipment has high operating energy consumption and insufficient performance optimization, resulting in a significant increase in operating costs and reduced economic benefits.

[0004] In view of this, the present invention proposes a wind turbine operation and maintenance optimization decision system based on big data to solve the above problems. Summary of the invention

[0005] In order to overcome the above-mentioned defects of the prior art and to achieve the above-mentioned purpose, the present invention provides the following technical solution: a wind turbine operation and maintenance optimization decision system based on big data, comprising:

[0006] Data acquisition module, used to collect data of wind turbines;

[0007] A data preprocessing module is used to preprocess the data of the wind turbine generator set to obtain preprocessed wind turbine generator set data;

[0008] The safety maintenance module performs vulnerability maintenance on the wind turbines based on the pre-processed wind turbine data to obtain safe wind turbine data;

[0009] The cost control module processes the safety wind turbine data to obtain an optimal operating parameter combination; applies the optimal operating parameter combination to the wind turbine; and each module is connected via wired and / or wireless means.

[0010] Furthermore, the data of the wind turbine set includes time-series operation data, image data and log data; the time-series operation data is collected by the wind turbine set's sensor, the image data is collected by the imaging device in the workplace, and the log data is obtained by querying the wind turbine set's database.

[0011] Furthermore, the method of preprocessing the data of the wind turbine generator set includes:

[0012] Process the time series operation data to obtain denoised time series data; process the image data to obtain enhanced image data; process the log data to obtain accurate log data; integrate the denoised time series data, enhanced image data and accurate log data to obtain pre-processed wind turbine data;

[0013] The method of processing the time series operation data includes:

[0014] The missing values ​​of the time series operation data are processed to obtain complete time series data; the complete time series data are denoised to obtain denoised time series data.

[0015] Furthermore, the method of processing missing values ​​of time series operation data includes:

[0016] Construct a missing value prediction model and use the RNN neural network model as the basic framework of the missing value prediction model;

[0017] Collect complete historical time series operation data as training labels for the missing value prediction model; randomly select XX moments from the complete historical time series operation data, and delete the values ​​corresponding to the moments to obtain the missing value operation data set, and use the missing value operation data set as the training set;

[0018] Use the missing value running data set to train the missing value prediction model; define the prediction loss function of the missing value prediction model, calculate the function value of the prediction loss function, until the function value of the prediction loss function no longer decreases, fix the parameters of the missing value prediction model at this time, and obtain the trained missing value prediction model;

[0019] The trained missing value prediction model is used to process the missing values ​​of the time series operation data to obtain complete time series data.

[0020] Furthermore, the method of denoising the complete temporal data includes:

[0021] Convert the complete time series data into frequency domain data; preset a frequency threshold, extract the part of the frequency domain data that is greater than the frequency threshold based on the preset frequency threshold, and obtain high-frequency noise data, and the remaining part of the frequency domain data is normal frequency domain data; construct an adaptive filter, and optimize the performance parameters of the adaptive filter; use the adaptive filter to filter the high-frequency noise data to obtain denoised frequency domain data; combine the denoised frequency domain data with the normal frequency domain data to obtain filtered frequency domain data; convert the filtered frequency domain data into time series data to obtain denoised time series data;

[0022] The calculation formula for converting the complete time series data into frequency domain data is:

[0023] Wherein, P represents frequency domain data; Φ represents wavelet transform; T(n) represents the value corresponding to the nth moment in the complete time series data; N represents the window size; W represents the window function; m represents the starting position of the window; f represents the frequency component; I represents the imaginary unit;

[0024] High frequency noise data Here, FP represents a preset frequency threshold.

[0025] Furthermore, the method of optimizing the performance parameters of the adaptive filter includes:

[0026] Initialize adaptive filter performance parameters;

[0027] Define the performance index of the adaptive filter J(i) = E[e 2 [HP(i)]]; where E represents the mathematical expectation; e[HP(i)] represents the error output at the i-th moment in the high-frequency noise data;

[0028] e[HP(i)]=D(i)-ω(i)×HP(i); where D(i) represents the value corresponding to the i-th moment in the denoised frequency domain data of the desired output; ω(i) represents the iterative weight of the adaptive filter; HP(i) represents the value corresponding to the i-th moment in the high-frequency noise data;

[0029] The iterative weights of the adaptive filter are updated; the error output of the high-frequency noise data is recalculated based on the updated iterative weights of the adaptive filter, and a new performance index of the adaptive filter is calculated based on the error output; the performance parameters of the adaptive filter are updated by dynamically adjusting the new performance index of the adaptive filter, and when the new performance index of the adaptive filter is less than or equal to a preset performance index threshold, the performance parameters of the adaptive filter are fixed at this time to obtain the optimal performance parameters of the adaptive filter;

[0030] The calculation formula for updating the iterative weights of the adaptive filter is:

[0031] Wherein, ω(j+1) represents the iterative weight of the adaptive filter updated for the j+1th time; ω(j) represents the iterative weight of the adaptive filter updated for the jth time; α represents the smoothing factor; e avg [HI] represents the average error output of historical high-frequency noise data;

[0032] The calculation formula for updating the performance parameters of the adaptive filter is:

[0033] Wherein, F(i+1) represents the performance parameter of the adaptive filter at the i+1th moment; F(i) represents the performance parameter of the adaptive filter at the i-th moment; μ represents the step size parameter; Represents the partial derivative of the new performance indicator newJ(i) with respect to the adaptive filter performance parameter F(i) at the i-th moment.

[0034] Furthermore, the method of processing the image data includes:

[0035] Performing color space conversion on each image in the image data to obtain brightness image data; using an image segmentation algorithm to divide each image in the brightness image data into NN pixel blocks; calculating the brightness mean based on the brightness value of each pixel block in each image in the brightness image data; constructing a brightness suppression function and a darkness enhancement function based on the brightness mean; presetting a brightness interval, using the brightness suppression function to process the pixel blocks in each image of the brightness image data whose brightness value is greater than the maximum value of the brightness interval to obtain brightness suppressed image data; using the darkness enhancement function to process the pixel blocks in each image of the brightness suppressed image data whose brightness value is less than the minimum value of the brightness interval to obtain darkness enhanced image data; combining the brightness suppression function and the darkness enhancement function to obtain a global brightness function; using the global brightness function to process all images in the darkness enhanced image data to obtain brightness enhanced image data; calculating the contrast of each image in the brightness enhanced image data, and adjusting the contrast based on the contrast of each image in the brightness enhanced image data to obtain adjusted image data; performing color space conversion on the adjusted image data to obtain enhanced image data;

[0036] The calculation formula of the brightness suppression function is:

[0037] Among them, B light represents the brightness suppression function; b represents the brightness adjustment coefficient; L avg represents the brightness mean; L0 represents the brightness value of the pixel block whose brightness value is greater than the maximum value of the brightness interval in each image of the brightness image data;

[0038] Darkness enhancement function B dark =(1+b)×L1+b×L avg; Wherein, L1 represents the brightness value of the pixel block whose brightness value is less than the minimum value of the brightness interval in each image of the brightness suppression image data;

[0039] Global brightness function B = L2 × [c × B dark +(1-c)×B light ]; where c represents the scale adjustment coefficient; L2 represents the brightness value of any pixel block in all images of the dark enhanced image data;

[0040] Brightness enhances the contrast of each image in the image data Among them, L3 ma x represents the maximum brightness value of any image in the brightness enhanced image data; L3 min Indicates the minimum brightness value of any image in the brightness enhanced image data;

[0041] The contrast adjustment calculation formula is:

[0042] DB after =λ(DB-DB avg ), where DB after Represents the contrast of any image in the adjusted image data; DB avg represents the average contrast of all images in the brightness enhanced image data; λ represents the contrast adjustment parameter.

[0043] Furthermore, the methods of performing vulnerability maintenance on the wind turbine generator set based on the pre-processed wind turbine generator set data include:

[0044] By scanning all images in the enhanced image data, equipment faults existing in the wind turbine equipment structure are determined, and the fault type and fault location are marked to obtain marked image data; the marked image data is sent to the maintenance department of the wind turbine, and the maintenance department of the wind turbine performs vulnerability maintenance on the wind turbine;

[0045] Process the denoised time series data to obtain repaired time series data; process the accurate log data to obtain safe log data; integrate the repaired time series data and the safe log data to obtain safe wind turbine data;

[0046] Methods for processing denoised time series data include:

[0047] Build a machine learning model to process the denoised time series data, determine the fault type and mark the data corresponding to the fault type to obtain the fault time series data; perform targeted maintenance on the wind turbines based on the fault time series data; collect the time series data of the wind turbines after maintenance to obtain the time series data after repair;

[0048] Collect historical fault time series data as a training set for the machine learning model; determine the fault type in the historical fault time series data by querying a preset wind turbine fault database, and use the fault type as a training label;

[0049] Adjust the parameters in the machine learning model to obtain the optimal parameter combination and apply the optimal parameter combination to the machine learning model; use historical fault timing data to train the machine learning model; define the loss function of the machine learning model, calculate the function value of the loss function of the machine learning model until the function value of the loss function no longer decreases, and obtain a trained machine learning model.

[0050] Furthermore, the methods for adjusting the parameters in the machine learning model include:

[0051] Initialize the particle population parameters, which include the number of iterations, population size, individual particle positions and coordination coefficients; randomly generate NUM parameter combinations as the initial particle population, and each particle in the initial particle population represents a parameter combination;

[0052] Define the fitness function and calculate the fitness function value of each individual particle in the initial particle population. Arrange the fitness function values ​​of each individual particle in the initial particle population in descending order from large to small, select the first three individual particles as the leading particles, and the other particles as follower particles;

[0053] Calculate the initial optimal position based on the individual position of the leading particle Among them, ler1 represents the individual position of the leading particle No. 1, ler2 represents the individual position of the leading particle No. 2, and ler3 represents the individual position of the leading particle No. 3; ω1 represents the weight of the leading particle No. 1, ω2 represents the weight of the leading particle No. 2, and ω3 represents the weight of the leading particle No. 3;

[0054] Synergy coefficient Where, t represents the number of iterations; TI represents the maximum number of iterations; l2 represents a second-class random constant;

[0055] The iteration is divided into three stages based on the value of the synergy coefficient Q;

[0056] When the synergy coefficient Q is in the range of (1, 2], it is a one-stage iteration. In each one-stage iteration, the fitness values ​​of all individual particles are recalculated, three leading particles are reselected based on the fitness values ​​of all individual particles, and the initial optimal position is updated based on the individual particle positions of the three reselected leading particles, and the distance between the individual particle position of each particle and the optimal position is calculated; the position of each individual particle is adjusted based on the distance between the individual particle position of each particle and the optimal position;

[0057] The distance between the individual particle position and the optimal position Wherein, Loc(t) represents the optimal position at the tth iteration; Y(t) represents the individual position of any particle at the tth iteration; l1 represents a type of random constant;

[0058] The calculation formula for adjusting the position of each individual particle based on the distance between the individual position of each particle and the optimal position is:

[0059] Y(t+1)=Loc(t)-Q×dis(t); where Y(t+1) represents the individual particle position of any particle at the t+1th iteration;

[0060] When the cooperation coefficient Q is in the range of (0, 1], it is a two-stage iteration. The process of reselecting the leading particle in the first-stage iteration is repeated, the distance between each leading particle and other particles is calculated in each two-stage iteration, and the individual particle position of each leading particle is updated; and the individual particle position of each particle is adjusted based on the individual particle position of each leading particle;

[0061] The calculation formula of the distance between each leading particle and other particles in each second-stage iteration is:

[0062] Among them, dis NO1 represents the distance between the leading particle NO1 and other particles; Y NO1 (t) represents the individual particle position of the leading particle NO1 at the tth iteration; l3 is a three-type random constant; dis NO2 represents the distance between the leading particle NO2 and other particles; Y NO2 (t) represents the individual particle position of the leading particle NO2 at the tth iteration; l4 is a four-type random constant; dis NO3 represents the distance between the leading particle NO3 and other particles; Y NO3 (t) represents the individual particle position of the leading particle NO3 at the tth iteration; l5 is a five-category random constant;

[0063] The calculation formula for updating the position of each leading particle is:

[0064] Among them, Y NO1 (t+1) represents the individual particle position of the leading particle NO1 at the t+1th iteration; Q1 represents the first-class synergy coefficient of the second stage; Y NO2 (t+1) represents the individual particle position of the leading particle NO2 at the t+1th iteration; Q2 represents the second-stage second-type synergy coefficient; Y NO3(t+1) represents the individual particle position of the leading particle NO3 at the t+1th iteration; Q3 represents the three-type synergy coefficient of the second stage;

[0065] The ways to adjust the individual particle position of each particle include:

[0066] Calculate each particle's individual target position based on each leader particle's individual position All particles move closer to the individual target position, and use the individual target position as the individual position of all particles in the next iteration;

[0067] When the synergy coefficient Q infinitely tends to 0, if the maximum number of iterations is reached, the particle with the largest fitness function value at this time is taken as the optimal particle individual, that is, the optimal parameter combination.

[0068] Furthermore, the method of processing the safe wind turbine data includes:

[0069] Collect wind turbine operating cost data and wind turbine operating loss data, process safe wind turbine data, wind turbine operating cost data and wind turbine operating loss data, and obtain the optimal operating parameter combination.

[0070] The technical effects and advantages of the wind turbine operation and maintenance optimization decision system based on big data of the present invention are as follows:

[0071] By collecting wind turbine data, preprocessing the wind turbine data and maintaining the wind turbines, and building a cost control module to optimize the resources of the wind turbines, a wind turbine operation and maintenance optimization decision-making system based on big data is realized; compared with existing experience, the wind turbine data is preprocessed more accurately, and higher quality wind turbine data is obtained, which indirectly improves the accuracy and convenience of subsequent operations; by building a model to detect vulnerabilities in wind turbine data, and optimizing the model parameters, the effect of vulnerability detection is better, and the accuracy and work efficiency of vulnerability detection are improved; according to the different detected vulnerabilities, corresponding methods are used for maintenance to improve the safety of wind turbines; by collecting the wind turbine data after maintenance and the cost data and loss data of the wind turbines, the wind turbines are optimized, the operating cost and operating energy consumption are reduced, and the operating efficiency is improved. BRIEF DESCRIPTION OF THE DRAWINGS

[0072] Figure 1 A schematic diagram of a wind turbine operation and maintenance optimization decision system based on big data according to the present invention;

[0073] Figure 2 It is a schematic diagram of a wind turbine operation and maintenance optimization decision-making method based on big data according to the present invention. DETAILED DESCRIPTION

[0074] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0075] Embodiment 1;

[0076] See also Figure 1 As shown, the wind turbine operation and maintenance optimization decision system based on big data described in this embodiment includes:

[0077] Data acquisition module, used to collect data of wind turbines;

[0078] A data preprocessing module is used to preprocess the data of the wind turbine generator set to obtain preprocessed wind turbine generator set data;

[0079] The safety maintenance module performs vulnerability maintenance on the wind turbines based on the pre-processed wind turbine data to obtain safe wind turbine data;

[0080] The cost control module processes the safety wind turbine data to obtain an optimal operating parameter combination; applies the optimal operating parameter combination to the wind turbine; and each module is connected via wired and / or wireless means.

[0081] The data of the wind turbine include time-series operation data, image data and log data; the time-series operation data is collected through the sensors of the wind turbine (the time-series operation data refers to the data in the wind turbine operation data that can change over time, and each data point in the data has a timestamp corresponding to it; for example, the vibration, pressure and speed data of the wind turbine), and the image data is collected through the imaging equipment in the workplace (for example, the image of the external structure of the wind turbine taken by the camera is used to observe whether there is damage or corrosion; for example, the temperature distribution on the surface of the wind turbine equipment is captured by thermal imaging equipment to identify whether there is overheating or poor cooling), and the log data is obtained by querying the database of the wind turbine (indicating text type data such as operation logs, maintenance records and fault reports stored in the wind turbine database).

[0082] The methods for preprocessing the data of wind turbines include:

[0083] The time series operation data is processed to obtain denoised time series data; the image data is processed to obtain enhanced image data; the log data is processed to obtain accurate log data (for example, each wind turbine operation log in the log data is processed using an NLP natural language processing algorithm, and the log data is processed into sentences); the denoised time series data, enhanced image data and accurate log data are integrated to obtain pre-processed wind turbine data.

[0084] The methods for processing time series operation data include:

[0085] In a general working environment, due to equipment failure, network delay or environmental interference, the sensors of wind turbines may have missing values ​​and noisy data during the data collection process. By filling the missing values ​​and denoising the noise data, higher quality data can be obtained, thus improving the efficiency and effectiveness of subsequent processing.

[0086] The missing values ​​of the time series operation data are processed to obtain complete time series data; the complete time series data are denoised to obtain denoised time series data.

[0087] Ways to handle missing values ​​for time series operation data include:

[0088] Construct a missing value prediction model and use the RNN neural network model as the basic framework of the missing value prediction model.

[0089] Collect historical complete time series operation data (extract a section of data without missing values ​​from the time series operation data of a certain period of history, that is, the historical complete time series operation data) as the training label of the missing value prediction model; randomly select XX moments from the historical complete time series operation data (XX is a positive integer, which is less than the number of all moments in the historical complete time series data), and delete the value corresponding to the moment to obtain the missing value operation data set, and use the missing value operation data set as the training set.

[0090] The missing value prediction model is trained by running the data set with missing values; a prediction loss function (such as a mean square error function) of the missing value prediction model is defined, and the function value of the prediction loss function is calculated until the function value of the prediction loss function no longer decreases, and the parameters of the missing value prediction model are fixed at this time to obtain a trained missing value prediction model.

[0091] The trained missing value prediction model is used to process the missing values ​​of the time series operation data to obtain complete time series data; the missing value prediction model is used to process the time series operation data to obtain the predicted values ​​of the missing values ​​in the time series operation data, and the predicted values ​​are used to fill the missing values.

[0092] Since the time series operation data has a strong time dependence, and the RNN model is very suitable for capturing the time dependence in the data, and the operating environment of the wind turbine may change over time, the RNN model can adapt to the changes in the operating environment of the wind turbine by learning from historical data, ensuring that it can provide more accurate prediction results in different time periods; by constructing an RNN neural network model to predict and fill in missing values, the time series operation data can be more complete; complete data can help to fully reflect the actual operating status of the wind turbine and improve the accuracy of the data.

[0093] Methods for denoising complete time series data include:

[0094] The existence of noise data may distort the true trend of time series data and affect the quality and accuracy of the data. In addition, there is often a large amount of redundant and useless information in the noise data. If it is retained, the subsequent data processing will be inefficient and inaccurate.

[0095] Convert the complete time series data into frequency domain data; preset a frequency threshold, and extract the part of the frequency domain data that is greater than the frequency threshold based on the preset frequency threshold (noise usually appears as some abnormal peaks in the spectrum, most of which are located in the high-frequency part) to obtain high-frequency noise data, and the rest of the frequency domain data is normal frequency domain data; construct an adaptive filter and optimize the performance parameters of the adaptive filter; use the adaptive filter to filter the high-frequency noise data to obtain denoised frequency domain data; combine the denoised frequency domain data with the normal frequency domain data to obtain filtered frequency domain data; convert the filtered frequency domain data into time series data to obtain denoised time series data.

[0096] When time series data is in the time domain, noise and useful signals are mixed together and difficult to distinguish. After converting time series data into frequency domain data, since noise data usually appears as high-frequency noise, it is possible to intuitively separate noise and useful signals by observing the frequency, and then process the noise more specifically.

[0097] The calculation formula for converting complete time series data into frequency domain data is:

[0098] Wherein, P represents frequency domain data; Φ represents wavelet transform; T(n) represents the value corresponding to the nth moment in the complete time series data; N represents the window size (used to determine the length of the complete time series data segment that can be processed at each time point); W represents the window function (common window functions include rectangular window and Gaussian window, etc., which are used to reduce the diffusion of the spectrum between different frequencies); m represents the starting position of the window (indicates the position of the first data point in the window); f represents the frequency component (refers to the expression of the complete time series data in the frequency domain, which describes the attribute information such as amplitude and phase of the complete time series data in the frequency domain); I represents the imaginary unit;

[0099] High frequency noise data Here, FP represents a preset frequency threshold.

[0100] Ways to optimize adaptive filter performance parameters include:

[0101] Initialize the adaptive filter performance parameters.

[0102] Define the performance index of the adaptive filter J(i) = E[e 2 [HP(i)]]; where E represents the mathematical expectation; e[HP(i)] represents the error output at the i-th moment in the high-frequency noise data;

[0103] e[HP(i)]=D(i)-ω(i)×HP(i); wherein D(i) represents the value corresponding to the i-th moment in the denoised frequency domain data of the desired output; ω(i) represents the iterative weight of the adaptive filter; HP(i) represents the value corresponding to the i-th moment in the high-frequency noise data.

[0104] The iterative weights of the adaptive filter are updated; the error output of the high-frequency noise data is recalculated based on the updated iterative weights of the adaptive filter, and a new performance index of the adaptive filter is calculated based on the error output; the performance parameters of the adaptive filter are updated by dynamically adjusting the new performance index of the adaptive filter, and when the new performance index of the adaptive filter is less than or equal to a preset performance index threshold, the performance parameters of the adaptive filter are fixed at this time to obtain the optimal performance parameters of the adaptive filter.

[0105] Optimizing the parameters of the adaptive filter can make the adaptive filter remove high-frequency noise more accurately while retaining useful signals; the dynamic adjustment of the parameters ensures that the adaptive filter can achieve the best denoising effect in different types of noise environments; for general filters, if the parameters are not set properly, it is easy to cause overfitting or underfitting. By optimizing the parameters, a balance point is found so that the filter can remove noise and retain useful signals; the optimized adaptive filter has stronger adaptability and automation level. By continuously learning the noise characteristics in the processed data, the parameters are gradually improved, and the parameters are flexibly adjusted according to actual needs, so that the adaptive filter is more robust and always ensures that the adaptive filter is in the best working state.

[0106] The calculation formula for updating the iterative weights of the adaptive filter is:

[0107] Wherein, ω(j+1) represents the iterative weight of the adaptive filter updated for the j+1th time; ω(j) represents the iterative weight of the adaptive filter updated for the jth time; α represents the smoothing factor (α∈(1,2), the smoothing factor is used to control the influence of the average error output of the historical high-frequency noise data); e avg [HI] represents the average error output of historical high-frequency noise data (collecting complete historical time series data of a certain period of time in history, and converting the complete historical time series data into historical frequency domain data; extracting historical high-frequency noise data from the historical frequency domain data, and calculating the average error output of all moments in the historical high-frequency noise data).

[0108] This formula shows that the update of the iterative weights of the adaptive filter depends on the influence of historical errors and smoothing factors; by introducing the average error output of historical high-frequency noise data, the adaptive filter is prevented from being overly dependent on the error output at the current moment, ensuring that the adaptive filter can adapt to long-term changing trends; the smoothing factor controls the degree of influence of historical errors, so that the adaptive filter can find a balance between fast convergence and stable performance, thereby improving the stability of the adaptive filter.

[0109] The calculation formula for updating the performance parameters of the adaptive filter is:

[0110] Wherein, F(i+1) represents the performance parameter of the adaptive filter at the i+1th moment; F(i) represents the performance parameter of the adaptive filter at the ith moment; μ represents the step size parameter (μ is a constant; the step size parameter is used to control the adjustment amplitude of the update, and the selection of the step size parameter has an important influence on the convergence speed, stability and performance of the adaptive filter; the appropriate step size parameter is the key to ensure the normal operation of the adaptive filter); Represents the partial derivative of the new performance indicator newJ(i) with respect to the adaptive filter performance parameter F(i) at the i-th moment.

[0111] This formula uses the gradient descent method, which calculates the partial derivative of the performance indicator with respect to the performance parameter, calculates the gradient direction of the performance indicator, and then adjusts the performance parameter according to the gradient direction and size to make the performance indicator gradually smaller, and finally minimizes the performance indicator.

[0112] The methods for processing image data include:

[0113] The goal of processing image data is to adjust the brightness and contrast of the image to make the image clearer and the fault location of the wind turbine more obvious, which is convenient for subsequent maintenance;

[0114] Perform color space conversion on each image in the image data (convert the image from RGB color space to HSV color space, in HSV color space, each color is represented by hue, saturation and brightness value) to obtain brightness image data; use image segmentation algorithm (such as super pixel segmentation algorithm) to segment each image in the brightness image data into NN pixel blocks (NN∈(0,+∞) and NN is an integer); calculate the brightness mean based on the brightness value of each pixel block of each image in the brightness image data (one of the basic units in HSV color space); construct brightness suppression function and darkness enhancement function based on brightness mean; preset brightness interval (such as interval (L low ,L high ), where L low and L high are all integers greater than 0), using a brightness suppression function to process pixel blocks whose brightness values ​​in each image of the brightness image data are greater than a maximum value of a brightness interval, to obtain brightness suppressed image data; using a darkness enhancement function to process pixel blocks whose brightness values ​​in each image of the brightness suppressed image data are less than a minimum value of a brightness interval, to obtain darkness enhanced image data; combining the brightness suppression function and the darkness enhancement function to obtain a global brightness function; using the global brightness function to process all images in the darkness enhanced image data, to obtain brightness enhanced image data; calculating the contrast of each image in the brightness enhanced image data, and performing contrast adjustment based on the contrast of each image in the brightness enhanced image data, to obtain adjusted image data; performing color space conversion on the adjusted image data, to obtain enhanced image data.

[0115] Converting image data to HSV color space can better separate the brightness information of the image, facilitate brightness calculation, and facilitate subsequent adjustments; segmenting the image makes it easier to capture fault features at different levels; processing the brightness value of each pixel block separately enhances local details, highlights the bright and dark details of the image, and makes the overall image clearer; by adjusting the image contrast, the details in the image are enhanced to make the fault features more obvious and ensure that the fault type can be accurately identified; constructing a global brightness function can avoid both over-enhancement and over-suppression of certain areas, so that all areas in the image are appropriately enhanced, ensuring the overall quality of the image.

[0116] The calculation formula of the brightness suppression function is:

[0117] Among them, B light represents the brightness suppression function; b represents the brightness adjustment coefficient (b∈(1,2), which is used to adjust the suppression effect of the brightness suppression function and the enhancement effect of the darkness enhancement function); L avg represents the brightness mean; L0 represents the brightness value of the pixel block whose brightness value is greater than the maximum value of the brightness interval in each image of the brightness image data.

[0118] The brightness suppression function is used to prevent overexposure of areas with high brightness in the image, thereby avoiding loss of details in the image.

[0119] Darkness enhancement function B dark =(1+b)×L1+b×L avg ; Wherein, L1 represents the brightness value of the pixel block whose brightness value is less than the minimum value of the brightness interval in each image of the brightness suppressed image data.

[0120] The darkness enhancement function is used to enhance the brightness value of the area with too low brightness, so as to make the details of the area with too low brightness clearer and facilitate the discovery of hidden fault features.

[0121] Global brightness function B = L2 × [c × B dark +(1-c)×B light ]; wherein c represents the scale adjustment coefficient (c∈(0,1)); L2 represents the brightness value of any pixel block in all images of the darkness-enhanced image data.

[0122] The global brightness function linearly combines the brightness suppression function and the darkness enhancement function, adjusts the proportion of the brightness suppression function and the darkness enhancement function by changing the value of the proportional adjustment coefficient, finds a balance between brightness suppression and darkness enhancement, and comprehensively adjusts the brightness of different areas of the image.

[0123] Brightness enhances the contrast of each image in the image data Among them, L3max Indicates the maximum brightness value of any image in the brightness enhanced image data; L3 min Indicates the minimum brightness value of any image in the brightness enhanced image data.

[0124] The contrast adjustment calculation formula is:

[0125] DB after =λ(DB-DB avg ), where DB after Represents the contrast of any image in the adjusted image data; DB avg represents the average contrast of all images in the brightness enhanced image data; λ represents the contrast adjustment parameter (λ∈(1,1.5)).

[0126] By comparing the contrast of each image with the average contrast and changing the contrast according to the contrast adjustment coefficient, the contrast of the image can be effectively enhanced, making the fault features more obvious and improving the efficiency of subsequent maintenance.

[0127] Methods for performing vulnerability maintenance on wind turbines based on pre-processed wind turbine data include:

[0128] By scanning all images in the enhanced image data, equipment faults existing in the wind turbine equipment structure (such as the casing, control panel, and connecting lines of the wind turbine equipment) are determined, and the fault type (such as casing corrosion, line aging, control panel damage, and uneven temperature distribution causing the cooling device to stop operating) and the fault location are marked to obtain marked image data; the marked image data is sent to the maintenance department of the wind turbine set, and the maintenance department of the wind turbine set performs vulnerability maintenance on the wind turbine set.

[0129] The denoised time series data is processed to obtain repaired time series data; the accurate log data is processed to obtain safe log data; a random forest model is constructed to process the accurate log data, and the error records or malicious characters in the accurate log data are identified and deleted; historical log data containing error records (such as records containing the key character ERROR) are collected, and the historical log data is used as a training set for the language recognition model; feature words related to the error records are obtained by querying the wind turbine database, and these feature words are used as training labels for the random forest model; the random forest model is trained using the historical log data; the loss function of the random forest model is defined (such as the cross entropy loss function), and the function value of the loss function is calculated until the function value of the loss function is no longer reduced, and the parameters are fixed to obtain a trained random forest model; the repaired time series data and the safe log data are integrated to obtain safe wind turbine data.

[0130] Methods for processing denoised time series data include:

[0131] Construct a machine learning model (such as an XGBoost classification model) to process the denoised time series data, determine the fault type and mark the data corresponding to the fault type to obtain the fault time series data; perform targeted maintenance on the wind turbines based on the fault time series data (such as deleting the marked data in the fault time series data or taking different methods to handle different fault types); collect the time series data of the wind turbines after maintenance to obtain the time series data after repair.

[0132] Collect historical fault time series data as the training set of the machine learning model; determine the fault type in the historical fault time series data by querying the preset wind turbine fault database, and use the fault type as the training label.

[0133] Adjust the parameters in the machine learning model to obtain the optimal parameter combination and apply the optimal parameter combination to the machine learning model; use historical fault timing data to train the machine learning model; define the loss function of the machine learning model (such as a polynomial logistic regression function), calculate the function value of the loss function of the machine learning model until the function value of the loss function no longer decreases, and obtain a trained machine learning model.

[0134] Ways to adjust parameters in machine learning models include:

[0135] The parameters of the machine learning model have a significant impact on the model performance. Different parameter configurations are often required for different data sets and different application scenarios. It is difficult to find the optimal parameters by manual adjustment. Therefore, an optimization algorithm is used to globally search for the optimal parameter combination in the parameter space.

[0136] Initialize the particle population parameters, which include the number of iterations, population size, individual particle positions and coordination coefficients; randomly generate NUM parameter combinations (NUM is a positive integer, which represents the population size) as the initial particle population, and each particle in the initial particle population represents a parameter combination.

[0137] Define the fitness function (such as classification accuracy function and F1 score function), and calculate the fitness function value of each individual particle in the initial particle population, arrange the fitness function values ​​of each individual particle in the initial particle population in descending order from large to small, select the first three individual particles as the leading particles, and the other particles as follower particles.

[0138] The leading particle is considered to be the particle closest to the optimal position, and the follower particles approach the optimal position by following the guidance of the leading particle; the follower particles continuously adjust their positions according to the leadership of the leading particle, helping the algorithm to explore the solution space more widely, and gradually narrow the range of the optimal position, and finally lock the optimal position.

[0139] Calculate the initial optimal position based on the individual position of the leading particle Among them, ler1 represents the individual position of the leading particle No. 1, ler2 represents the individual position of the leading particle No. 2, and ler3 represents the individual position of the leading particle No. 3; ω1 represents the weight of the leading particle No. 1, ω2 represents the weight of the leading particle No. 2, and ω3 represents the weight of the leading particle No. 3.

[0140] Synergy coefficient Where t represents the number of iterations; TI represents the maximum number of iterations; l2 represents a second-class random constant (l2∈(0,1)); and the synergy coefficient Q decreases as the number of iterations increases.

[0141] The iteration is divided into three stages based on the value of the coordination coefficient Q. In the first stage, all particles disperse in the entire solution space to find the optimal position. Each particle adjusts its position according to the currently known optimal position, gradually reducing the distance from the optimal position. At this time, the value of the coordination coefficient is large, allowing all particles to search over a large range to avoid falling into the local optimum. In the second stage, the leading particle guides other follower particles to approach the optimal position based on the current optimal position. All particles will update their positions based on the individual position of the leading particle at this time. At this time, the value of the coordination coefficient gradually decreases, entering the precise search stage. In the third stage, the value of the coordination coefficient is close to 0, and all particles focus on fine-tuning their individual positions and finally lock in the optimal solution.

[0142] When the cooperation coefficient Q is in the range of (1, 2], it is a one-stage iteration. In each one-stage iteration, the fitness values ​​of all individual particles are recalculated, three leading particles are reselected based on the fitness values ​​of all individual particles, and the initial optimal position is updated based on the individual particle positions of the three reselected leading particles. The distance between the individual particle position of each particle and the optimal position is calculated; the position of each individual particle is adjusted based on the distance between the individual particle position of each particle and the optimal position.

[0143] The distance between the individual particle position and the optimal position Among them, Loc(t) represents the optimal position at the tth iteration; Y(t) represents the individual particle position of any particle at the tth iteration; l1 represents a type of random constant (l1∈(0,1)).

[0144] The calculation formula for adjusting the position of each individual particle based on the distance between the individual position of each particle and the optimal position is:

[0145] Y(t+1)=Loc(t)-Q×dis(t); wherein Y(t+1) represents the individual particle position of any particle at the t+1th iteration.

[0146] When the cooperation coefficient Q is in the range of (0, 1], it is a two-stage iteration. The process of reselecting the leading particle during the first-stage iteration is repeated, the distance between each leading particle and other particles is calculated during each two-stage iteration, and the individual particle position of each leading particle is updated; and the individual particle position of each particle is adjusted based on the individual particle position of each leading particle.

[0147] The calculation formula of the distance between each leading particle and other particles in each second-stage iteration is:

[0148] Among them, dis NO1 represents the distance between the leading particle NO1 and other particles; Y NO1 (t) represents the individual particle position of the leading particle NO1 at the tth iteration; l3 is a three-type random constant (l3∈(0,1)); dis NO2 represents the distance between the leading particle NO2 and other particles; Y NO2 (t) represents the individual particle position of the leading particle NO2 at the tth iteration; l4 is a four-type random constant (l4∈(0,1)); dis NO3 represents the distance between the leading particle NO3 and other particles; Y NO3 (t) represents the individual particle position of the leading particle NO3 at the tth iteration; l5 is a five-category random constant (l5∈(0, 1)).

[0149] The calculation formula for updating the position of each leading particle is:

[0150] Among them, Y NO1 (t+1) represents the individual particle position of the leading particle NO1 at the t+1th iteration; Q1 represents the first-class synergy coefficient of the second stage (Q1∈(0,1]); Y NO2 (t+1) represents the individual particle position of the leading particle NO2 at the t+1th iteration; Q2 represents the second-stage second-type synergy coefficient (Q2∈(0,1]); Y NO3 (t+1) represents the individual particle position of the leading particle NO3 at the t+1th iteration; Q3 represents the two-stage three-category synergy coefficient (Q3∈(0,1]).

[0151] The ways to adjust the individual particle position of each particle include:

[0152] Calculate each particle's individual target position based on each leader particle's individual position All particles move closer to the individual target position, and the individual target position is used as the individual position of all particles in the next iteration.

[0153] When the synergy coefficient Q infinitely tends to 0, if the maximum number of iterations is reached, the particle with the largest fitness function value at this time is taken as the optimal particle individual, that is, the optimal parameter combination.

[0154] Using optimization algorithms to adjust the parameters of machine learning models reduces the process of manual intervention, speeds up parameter adjustment, enables the model to adapt to complex data environments, enhances the model's robustness and autonomous learning capabilities, and helps improve the accuracy and work efficiency of machine learning models.

[0155] The methods for processing the safe wind turbine data include:

[0156] Collect wind turbine operating cost data and wind turbine operating loss data, process safe wind turbine data, wind turbine operating cost data and wind turbine operating loss data, and obtain the optimal operating parameter combination.

[0157] The operating cost data of wind turbines include data such as the maintenance costs, labor maintenance costs and equipment costs of wind turbines; the operating loss data of wind turbines include data such as energy loss rate, equipment energy consumption and electrical consumption.

[0158] A multi-objective optimization mathematical model is constructed, and the objective functions of the multi-objective optimization mathematical model are defined as maximizing the working efficiency of the wind turbine, minimizing the operating cost of the wind turbine, and minimizing the operating energy consumption of the wind turbine; a multi-objective genetic algorithm (such as the NSGA-II algorithm) is used to process the safe wind turbine data, the wind turbine operating cost data, and the wind turbine operating loss data; the population is initialized, and each individual in the population represents a decision variable (such as speed, cost, output voltage, and other parameters that can affect the operating state of the wind turbine); a multi-objective fitness function is defined, and the function value of the multi-objective fitness function of each individual is calculated; some individuals are selected from the population for crossover and mutation operations to generate a new offspring population; the crossover and mutation operations are repeated, and the iteration is stopped when the maximum number of iterations is reached, and individuals whose fitness function values ​​are greater than the preset fitness threshold are extracted to form the Pareto frontier; based on the actual operating conditions of the wind turbine, decision variables are selected from the Pareto frontier to form the optimal parameter combination.

[0159] This embodiment collects wind turbine data, preprocesses the wind turbine data and maintains the wind turbine, and builds a cost control module to optimize the resources of the wind turbine, thereby realizing a wind turbine operation and maintenance optimization decision system based on big data; compared with existing experience, the wind turbine data is preprocessed more accurately, and higher quality wind turbine data is obtained, which indirectly improves the accuracy and convenience of subsequent operations; the loopholes in the wind turbine data are detected by building a model, and the model parameters are optimized at the same time, so that the effect of vulnerability detection is better, and the accuracy and work efficiency of vulnerability detection are improved; corresponding methods are used for maintenance of different detected loopholes, thereby improving the safety of the wind turbine; the wind turbine is optimized by collecting the wind turbine data after maintenance and the cost data and loss data of the wind turbine, thereby reducing the operating cost and operating energy consumption and improving the operating efficiency.

[0160] Embodiment 2;

[0161] See also Figure 2 As shown, the part not described in detail in this embodiment is described in Example 1, which provides a wind turbine operation and maintenance optimization decision method based on big data, including:

[0162] S1. Collect data of wind turbines;

[0163] S2. Preprocessing the data of the wind turbine generator set to obtain preprocessed wind turbine generator set data;

[0164] S3. Perform vulnerability maintenance on the wind turbine generator set based on the pre-processed wind turbine generator set data to obtain safe wind turbine generator set data;

[0165] S4. Process the safety wind turbine data to obtain an optimal operating parameter combination; and apply the optimal operating parameter combination to the wind turbine.

[0166] Embodiment 3;

[0167] This embodiment discloses an electronic device, including 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 operation mode of the wind turbine operation and maintenance optimization decision-making method based on big data provided above is implemented.

[0168] Since the electronic device introduced in this embodiment is an electronic device used to implement a method for optimizing wind turbine operation and maintenance decision-making based on big data in the embodiment of this application, based on the method for optimizing wind turbine operation and maintenance decision-making based on big data introduced in the embodiment of this application, the technical personnel of this field can understand the specific implementation of the electronic device of this embodiment and its various variations, so how the electronic device implements the method in the embodiment of this application is not introduced in detail here. As long as the technical personnel of this field implement the electronic device used in the method for optimizing wind turbine operation and maintenance decision-making based on big data in the embodiment of this application, it belongs to the scope of protection of this application.

[0169] The above formulas are all dimensionless and numerical calculations. The formula is a formula for the most recent real situation obtained by collecting a large amount of data and performing software simulation. The preset parameters and thresholds in the formula are set by technicians in this field according to actual conditions.

[0170] The above is only a preferred embodiment of the present invention, and the protection scope of the present invention is not limited to the above embodiments. All technical solutions under the concept of the present invention belong to the protection scope of the present invention. It should be pointed out that for ordinary technical users in this technical field, some improvements and modifications without departing from the principle of the present invention should also be regarded as the protection scope of the present invention.

Claims

1. A wind turbine operation and maintenance optimization decision system based on big data, characterized in that: include: Data acquisition module, used to collect data of wind turbines; A data preprocessing module is used to preprocess the data of the wind turbine generator set to obtain preprocessed wind turbine generator set data; The safety maintenance module performs vulnerability maintenance on the wind turbines based on the pre-processed wind turbine data to obtain safe wind turbine data; The cost control module processes the safety wind turbine data to obtain an optimal operating parameter combination; applies the optimal operating parameter combination to the wind turbine; and each module is connected via wired and / or wireless means.

2. The wind turbine operation and maintenance optimization decision system based on big data according to claim 1 is characterized in that: The data of the wind turbine generator set includes time-series operation data, image data and log data; the time-series operation data is collected by the wind turbine generator set's sensor, the image data is collected by the imaging device in the workplace, and the log data is obtained by querying the wind turbine generator set's database.

3. The wind turbine operation and maintenance optimization decision system based on big data according to claim 2 is characterized in that: The method of preprocessing the data of the wind turbine generator set includes: Process the time series operation data to obtain denoised time series data; process the image data to obtain enhanced image data; process the log data to obtain accurate log data; integrate the denoised time series data, enhanced image data and accurate log data to obtain pre-processed wind turbine data; The method of processing the time series operation data includes: The missing values ​​of the time series operation data are processed to obtain complete time series data; the complete time series data are denoised to obtain denoised time series data.

4. The wind turbine operation and maintenance optimization decision system based on big data according to claim 3 is characterized in that: The method of processing missing values ​​of time series operation data includes: Construct a missing value prediction model and use the RNN neural network model as the basic framework of the missing value prediction model; Collect complete historical time series operation data as training labels for the missing value prediction model; randomly select XX moments from the complete historical time series operation data, and delete the values ​​corresponding to the moments to obtain the missing value operation data set, and use the missing value operation data set as the training set; Use the missing value running data set to train the missing value prediction model; define the prediction loss function of the missing value prediction model, calculate the function value of the prediction loss function, until the function value of the prediction loss function no longer decreases, fix the parameters of the missing value prediction model at this time, and obtain the trained missing value prediction model; The trained missing value prediction model is used to process the missing values ​​of the time series operation data to obtain complete time series data.

5. The wind turbine operation and maintenance optimization decision system based on big data according to claim 4 is characterized in that: The method of denoising the complete temporal data includes: Convert the complete time series data into frequency domain data; preset a frequency threshold, extract the part of the frequency domain data that is greater than the frequency threshold based on the preset frequency threshold, and obtain high-frequency noise data, and the remaining part of the frequency domain data is normal frequency domain data; construct an adaptive filter, and optimize the performance parameters of the adaptive filter; use the adaptive filter to filter the high-frequency noise data to obtain denoised frequency domain data; combine the denoised frequency domain data with the normal frequency domain data to obtain filtered frequency domain data; convert the filtered frequency domain data into time series data to obtain denoised time series data; The calculation formula for converting the complete time series data into frequency domain data is: Wherein, P represents frequency domain data; Φ represents wavelet transform; T(n) represents the value corresponding to the nth moment in the complete time series data; N represents the window size; W represents the window function; m represents the starting position of the window; f represents the frequency component; I represents the imaginary unit; High frequency noise data Here, FP represents a preset frequency threshold.

6. The wind turbine operation and maintenance optimization decision system based on big data according to claim 5 is characterized in that: The method of optimizing the performance parameters of the adaptive filter includes: Initialize adaptive filter performance parameters; Define the performance index of the adaptive filter J(i) = E[e 2 [HP(i)]]; where E represents the mathematical expectation; e[HP(i)] represents the error output at the i-th moment in the high-frequency noise data; e[HP(i)]=D(i)-ω(i)×HP(i); where D(i) represents the value corresponding to the i-th moment in the denoised frequency domain data of the desired output; ω(i) represents the iterative weight of the adaptive filter; HP(i) represents the value corresponding to the i-th moment in the high-frequency noise data; The iterative weights of the adaptive filter are updated; the error output of the high-frequency noise data is recalculated based on the updated iterative weights of the adaptive filter, and a new performance index of the adaptive filter is calculated based on the error output; the performance parameters of the adaptive filter are updated by dynamically adjusting the new performance index of the adaptive filter, and when the new performance index of the adaptive filter is less than or equal to a preset performance index threshold, the performance parameters of the adaptive filter are fixed at this time to obtain the optimal performance parameters of the adaptive filter; The calculation formula for updating the iterative weights of the adaptive filter is: Wherein, ω(j+1) represents the iterative weight of the adaptive filter updated for the j+1th time; ω(j) represents the iterative weight of the adaptive filter updated for the jth time; α represents the smoothing factor; e avg [HI] represents the average error output of historical high-frequency noise data; The calculation formula for updating the performance parameters of the adaptive filter is: Wherein, F(i+1) represents the performance parameter of the adaptive filter at the i+1th moment; F(i) represents the performance parameter of the adaptive filter at the ith moment; μ represents the step size parameter; Represents the partial derivative of the new performance indicator newJ(i) with respect to the adaptive filter performance parameter F(i) at the i-th moment.

7. A wind turbine operation and maintenance optimization decision system based on big data according to claim 6, characterized in that: The method of processing the image data includes: Performing color space conversion on each image in the image data to obtain brightness image data; using an image segmentation algorithm to divide each image in the brightness image data into NN pixel blocks; calculating the brightness mean based on the brightness value of each pixel block in each image in the brightness image data; constructing a brightness suppression function and a darkness enhancement function based on the brightness mean; presetting a brightness interval, using the brightness suppression function to process the pixel blocks in each image of the brightness image data whose brightness value is greater than the maximum value of the brightness interval to obtain brightness suppressed image data; using the darkness enhancement function to process the pixel blocks in each image of the brightness suppressed image data whose brightness value is less than the minimum value of the brightness interval to obtain darkness enhanced image data; combining the brightness suppression function and the darkness enhancement function to obtain a global brightness function; using the global brightness function to process all images in the darkness enhanced image data to obtain brightness enhanced image data; calculating the contrast of each image in the brightness enhanced image data, and adjusting the contrast based on the contrast of each image in the brightness enhanced image data to obtain adjusted image data; performing color space conversion on the adjusted image data to obtain enhanced image data; The calculation formula of the brightness suppression function is: Among them, B light represents the brightness suppression function; b represents the brightness adjustment coefficient; L avg represents the brightness mean; L0 represents the brightness value of the pixel block whose brightness value is greater than the maximum value of the brightness interval in each image of the brightness image data; Darkness enhancement function B dark =(1+b)×L1+b×L avg ; Wherein, L1 represents the brightness value of the pixel block whose brightness value is less than the minimum value of the brightness interval in each image of the brightness suppression image data; Global brightness function B = L2 × [c × B dark +(1-c)×B light ]; where c represents the scale adjustment coefficient; L2 represents the brightness value of any pixel block in all images of the dark enhanced image data; Brightness enhances the contrast of each image in the image data Among them, L3 max Indicates the maximum brightness value of any image in the brightness enhanced image data; L3 min Indicates the minimum brightness value of any image in the brightness enhanced image data; The contrast adjustment formula is: DB after =λ(DB-DB avg ), where DB after Represents the contrast of any image in the adjusted image data; DB avg represents the average contrast of all images in the brightness enhanced image data; λ represents the contrast adjustment parameter.

8. The wind turbine operation and maintenance optimization decision system based on big data according to claim 7 is characterized in that: Methods for performing vulnerability maintenance on wind turbines based on pre-processed wind turbine data include: By scanning all images in the enhanced image data, equipment faults existing in the wind turbine equipment structure are determined, and the fault type and fault location are marked to obtain marked image data; the marked image data is sent to the maintenance department of the wind turbine, and the maintenance department of the wind turbine performs vulnerability maintenance on the wind turbine; Process the denoised time series data to obtain repaired time series data; process the accurate log data to obtain safe log data; integrate the repaired time series data and the safe log data to obtain safe wind turbine data; Methods for processing denoised time series data include: Build a machine learning model to process the denoised time series data, determine the fault type and mark the data corresponding to the fault type to obtain the fault time series data; perform targeted maintenance on the wind turbines based on the fault time series data; collect the time series data of the wind turbines after maintenance to obtain the time series data after repair; Collect historical fault time series data as a training set for the machine learning model; determine the fault type in the historical fault time series data by querying a preset wind turbine fault database, and use the fault type as a training label; Adjust the parameters in the machine learning model to obtain the optimal parameter combination and apply the optimal parameter combination to the machine learning model; use historical fault timing data to train the machine learning model; define the loss function of the machine learning model, calculate the function value of the loss function of the machine learning model until the function value of the loss function no longer decreases, and obtain a trained machine learning model.

9. The wind turbine operation and maintenance optimization decision system based on big data according to claim 8 is characterized in that: Ways to adjust parameters in machine learning models include: Initialize the particle population parameters, which include the number of iterations, population size, individual particle positions and coordination coefficients; randomly generate NUM parameter combinations as the initial particle population, and each particle in the initial particle population represents a parameter combination; Define the fitness function and calculate the fitness function value of each individual particle in the initial particle population. Arrange the fitness function values ​​of each individual particle in the initial particle population in descending order from large to small, select the first three individual particles as the leading particles, and the other particles as follower particles; Calculate the initial optimal position based on the individual position of the leading particle [ω1×ler1+ω2×ler2+ω3×ler3]; where ler1 represents the individual position of the leading particle No. 1, ler2 represents the individual position of the leading particle No. 2, and ler3 represents the individual position of the leading particle No. 3; ω1 represents the weight of the leading particle No. 1, ω2 represents the weight of the leading particle No. 2, and ω3 represents the weight of the leading particle No. 3; Synergy coefficient Where, t represents the number of iterations; TI represents the maximum number of iterations; l2 represents a second-class random constant; The iteration is divided into three stages based on the value of the synergy coefficient Q; When the synergy coefficient Q is in the range of (1, 2], it is a one-stage iteration. In each one-stage iteration, the fitness values ​​of all individual particles are recalculated, three leading particles are reselected based on the fitness values ​​of all individual particles, and the initial optimal position is updated based on the individual particle positions of the three reselected leading particles, and the distance between the individual particle position of each particle and the optimal position is calculated; the position of each individual particle is adjusted based on the distance between the individual particle position of each particle and the optimal position; The distance between the individual particle position and the optimal position Wherein, Loc(t) represents the optimal position at the tth iteration; Y(t) represents the individual position of any particle at the tth iteration; l1 represents a type of random constant; The calculation formula for adjusting the position of each individual particle based on the distance between the individual position of each particle and the optimal position is: Y(t+1)=Loc(t)--Q×dis(t); where Y(t+1) represents the individual particle position of any particle at the t+1th iteration; When the cooperation coefficient Q is in the range of (0, 1], it is a two-stage iteration. The process of reselecting the leading particle in the first-stage iteration is repeated, the distance between each leading particle and other particles is calculated in each two-stage iteration, and the individual particle position of each leading particle is updated; and the individual particle position of each particle is adjusted based on the individual particle position of each leading particle; The calculation formula of the distance between each leading particle and other particles in each second-stage iteration is: Among them, dis NO1 represents the distance between the leading particle NO1 and other particles; Y NO1 (t) represents the individual particle position of the leading particle NO1 at the tth iteration; l3 is a three-type random constant; dis NO2 represents the distance between the leading particle NO2 and other particles; Y NO2 (t) represents the individual particle position of the leading particle NO2 at the tth iteration; l4 is a four-type random constant; dis NO3 represents the distance between the leading particle NO3 and other particles; Y NO3 (t) represents the individual particle position of the leading particle NO3 at the tth iteration; l5 is a five-category random constant; The calculation formula for updating the position of each leading particle is: Among them, Y NO1 (t+1) represents the individual particle position of the leading particle NO1 at the t+1th iteration; Q1 represents the first-class synergy coefficient of the second stage; Y NO2 (t+1) represents the individual particle position of the leading particle NO2 at the t+1th iteration; Q2 represents the second-stage second-type synergy coefficient; Y NO3 (t+1) represents the individual particle position of the leading particle NO3 at the t+1th iteration; Q3 represents the three-type synergy coefficient of the second stage; The ways to adjust the individual particle position of each particle include: Calculate each particle's individual target position based on each leader particle's individual position All particles move closer to the individual target position, and use the individual target position as the individual position of all particles in the next iteration; When the synergy coefficient Q infinitely tends to 0, if the maximum number of iterations is reached, the particle with the largest fitness function value at this time is taken as the optimal particle individual, that is, the optimal parameter combination.

10. A wind turbine operation and maintenance optimization decision system based on big data according to claim 9, characterized in that: The methods for processing the safe wind turbine data include: Collect wind turbine operating cost data and wind turbine operating loss data, process safe wind turbine data, wind turbine operating cost data and wind turbine operating loss data, and obtain the optimal operating parameter combination.

Images