Multi-model cooperative wind farm scada data full-link autonomous processing method and system

Abstract

This application relates to a method and system for autonomous processing of wind farm SCADA data across the entire data chain under multi-model collaboration. Based on data processing methods such as nearest neighbor multi-objective weighted filling, adaptive quantile anomaly identification, and unbiased scale transformation, it covers the entire data processing workflow, including data acquisition, missing value filling, outlier identification, key variable screening, and scale standardization. This application establishes a feedback loop and iterative optimization mechanism, using a collaborative scheduling intelligent agent to uniformly schedule and manage multiple data processing processes, forming a standardized and unified data governance system. This improves the efficiency of wind farm SCADA data governance, reduces manual operation and maintenance costs, and provides a reliable data foundation for subsequent wind turbine status identification and fault early warning.

Description

Technical Field

[0001] This application relates to the field of wind farm data processing, specifically to a method and system for autonomous processing of wind farm SCADA data across the entire data chain under multi-model collaboration. Background Technology

[0002] Wind farm SCADA systems can continuously and in real-time collect multiple operating condition and state variables during wind turbine operation, such as wind speed, power, temperature, current, and voltage. This provides crucial support for wind turbine operation status monitoring and analysis, fault diagnosis, and risk warning. However, in actual industrial environments, due to uncontrollable factors such as sensor failures, environmental interference (e.g., extreme weather, electromagnetic signals, system vibration), and network fluctuations, the monitored variables inevitably suffer from various problems, including complete randomness, missing values, and outliers. This results in suboptimal data quality, severely hindering a comprehensive and accurate depiction of the system's internal state and posing a significant challenge to the accurate identification of wind turbine status risks.

[0003] Current data governance methods are mainly divided into two types: those based on statistical learning and those based on deep learning. However, these methods mainly suffer from the following problems: 1. Lack of a standardized data governance system covering the entire data lifecycle. Wind farm operation and maintenance data is characterized by massive volume, multiple sources, long time series, and strong nonlinearity. However, existing processing procedures generally lack standardized norms across the entire chain, from data acquisition, missing data imputation, anomaly cleaning, key variable extraction to scale transformation. The high degree of arbitrariness and poor reproducibility at each stage makes it difficult to ensure the consistency and comparability of data governance results across different wind farms and at different times. Therefore, a unified, standardized data governance system is urgently needed.

[0004] 2. Disconnected data processing links, static and isolated underlying models, and heavy reliance on manual intervention. Currently, the data processing models used in different data processing links (such as missing data imputation, anomaly detection, and feature engineering) operate independently, lacking a unified communication and coordination mechanism at the underlying level. This creates "data silos" in the governance process, hindering the effective flow of information between links. It relies heavily on manual experience for connection and adjustment, resulting in high processing costs and low efficiency, making it difficult to meet the real-time and automation requirements of large-scale wind farms. Therefore, an intelligent and autonomous method is needed to uniformly schedule and manage models across different links.

[0005] 3. Traditional statistical methods and deep learning models struggle to balance performance and cost. Traditional statistical methods (such as mean imputation, 3D modeling, etc.) are often ineffective in achieving a proper balance. While computationally lightweight and highly interpretable, traditional methods have limited effectiveness against complex noise and nonlinear missing patterns. Deep learning methods, on the other hand, can capture high-dimensional and complex features, but their large model parameters and high training costs make them difficult to deploy in real-world industrial edge or real-time monitoring scenarios. Summary of the Invention

[0006] To overcome at least one deficiency in the prior art, this application provides a method and system for autonomous processing of wind farm SCADA data across the entire link under multi-model collaboration.

[0007] Firstly, a method for autonomous processing of wind farm SCADA data across the entire data link under multi-model collaboration is provided, including: Step 1: Obtain wind turbine time-series data and perform normalization processing to obtain normalized wind turbine time-series data; wind turbine time-series data includes multi-dimensional monitoring variables; Step 2: Determine if there are any missing values ​​in the normalized wind turbine timing data. If yes, proceed to step 3; otherwise, proceed to step 4. Step 3: For the normalized wind turbine time series data, the missing data is imputed using the nearest neighbor multi-objective weighted missing data processing method to obtain the missing wind turbine time series data after missing data processing. Step 4: For the wind turbine time series data, an outlier identification method based on adaptive quantiles is used to identify and process the outlier data, and the wind turbine time series data after anomaly processing is obtained. Step 5: The key variable screening method for wind turbine status characterization based on the maximum mutual information coefficient is used to screen key variables that are strongly correlated with wind turbine operating status and fault risk from multidimensional monitoring variables after anomaly handling. Step 6: Determine if the number of key variables being filtered is greater than the set threshold. If yes, return to step 5 and filter the key variables again. If no, proceed to step 7. Step 7: The scale of each key variable is unified using the Bessel correction-based multidimensional variable scaling method for wind turbines to obtain standardized input features.

[0008] In one embodiment, acquiring wind turbine time-series data and performing normalization processing includes: Based on the equipment configuration, measurement point distribution, and communication interfaces of each wind turbine unit in the wind farm, a multi-channel parallel acquisition channel is established for different wind turbine units, different subsystems, and different measurement point types. This channel synchronously acquires wind speed, wind direction, rotational speed, power, pitch angle, yaw angle, vibration, temperature, current, voltage, and status alarm signals to obtain wind turbine time-series data. The wind turbine time-series data is then uniformly synchronized, aligned to the time axis, structured, validated according to rules, and hierarchically aggregated to form standardized wind turbine time-series data.

[0009] In one embodiment, step 3 involves imputing missing data in the normalized wind turbine time-series data using a nearest-neighbor multi-objective weighted missing data processing method to obtain the missing wind turbine time-series data after missing data processing, including: The normalized wind turbine time series data is recorded as the target sample; Obtain a historical sample set, and search for candidate samples in the historical sample set that are similar to the operating conditions of the target sample to form a candidate sample set; Calculate the Euclidean distance between the target sample and each candidate sample, sort them in ascending order of Euclidean distance, and select the top K candidate samples to form the nearest neighbor sample set; The initial weights of each sample in the nearest neighbor set are determined based on the Euclidean distance and then normalized to obtain the normalized weights of each sample. The imputation values ​​for each missing variable are determined using the following formula:

[0010] in, For the target sample Filler values ​​for missing variables, For the nearest neighbor sample set, the first The first sample Observed values ​​of each variable For the nearest neighbor sample set, the first The normalized weights corresponding to each sample; Determine whether the imputed value of each missing variable is within the allowable range of the corresponding variable. If yes, the current imputed value is the final imputed value. If not, reconstruct the nearest neighbor sample set and determine the imputed value of the missing variable again.

[0011] In one embodiment, step 4 involves using an adaptive quantile-based outlier identification method to identify and process outlier data in the wind turbine time-series data, resulting in anomaly-processed wind turbine time-series data, including: Obtain a historical sample set, extract the historical observation sequence of each monitoring variable within a preset sliding time window, and calculate the lower quartile, upper quartile, and interquartile range of each monitoring variable based on the historical observation sequence; The adaptive quantile expansion coefficient of each monitoring variable is calculated based on the distribution skewness characterization of the historical observation sequence of each monitoring variable. Based on the lower quartile, upper quartile, interquartile range, and adaptive quantile expansion coefficient, the minor and extreme anomaly boundaries of each monitored variable are determined. The observed values ​​of each monitoring variable in the wind turbine time series data are compared with the boundaries of minor anomalies and extreme anomalies to determine whether the observed values ​​of the monitoring variables are abnormal data. Remove, replace, or mark abnormal data.

[0012] In one embodiment, the adaptive quantile expansion coefficient for each monitored variable is calculated using the following formula:

[0013] in, For the first The adaptive quantile expansion coefficient of each monitored variable, The basic quartile expansion coefficient For the first The skewness sensitivity adjustment coefficients corresponding to the monitored variables For the first The distribution skewness characterization of the historical observation sequence of a monitoring variable.

[0014] In one embodiment, the boundary of a minor anomaly is:

[0015]

[0016] in, For the first The upper bound of minor anomalies in the monitored variables. For the first The lower bound of minor anomalies in the monitored variables. For the first The upper quartiles of the monitored variables, For the first The lower quartiles of the monitored variables, For the first Interquartile range of the monitored variables For the first The adaptive quantile expansion coefficient of each monitored variable, The interquartile range coefficient; The extreme anomaly boundary is:

[0017]

[0018] in, For the first The upper bound of extreme outliers for each monitored variable. For the first The lower bound of extreme anomalies for each monitored variable.

[0019] In one embodiment, step 5 involves using a key variable screening method for wind turbine status characterization based on the maximum mutual information coefficient to screen the anomaly-handled wind turbine time-series data. This method selects key variables from the multidimensional monitoring variables that are strongly correlated with the wind turbine's operating status and fault risk, including: Calculate the maximum mutual information coefficient between each monitored variable and the wind turbine operating status characterization quantity; The maximum mutual information coefficients are sorted from largest to smallest, and the monitoring variables that are ranked first and have a maximum mutual information coefficient greater than a preset threshold are selected as key variables.

[0020] In one embodiment, step 7 involves scaling each key variable using a Bessel-corrected multidimensional variable scaling method for wind turbines to obtain standardized input features, including: Obtain a historical sample set and extract the historical observation sequences of each key variable based on the historical sample set; Based on the historical observation sequences of each key variable, and using the Bessel correction method, the sample standard deviation and sample mean of each key variable are determined; the sample standard deviation and sample mean are calculated using the following formulas:

[0021]

[0022] in, The standard deviation of the sample is 1. The first in the historical observation sequence The observed values ​​of key variables corresponding to each sample. For the sample size, The sample mean of the key variable; z-score standardization is performed on the observed values ​​of key variables based on the sample standard deviation to obtain the standardized values ​​of each key variable.

[0023] Secondly, a multi-model collaborative wind farm SCADA data end-to-end autonomous processing system is provided, including: a collaborative scheduling and closed-loop optimization unit, a data acquisition and standardization unit, a missing data processing unit, an outlier identification unit, a key variable screening unit, and a scale unification unit. The data acquisition and normalization unit is used to acquire wind turbine time-series data and perform normalization processing to obtain normalized wind turbine time-series data; the wind turbine time-series data includes multi-dimensional monitoring variables; The missing data processing unit is used to impute missing data in the normalized wind turbine time series data using a missing data processing method based on nearest neighbor multi-objective weighting, so as to obtain the missing wind turbine time series data after missing data processing. The outlier identification unit is used to identify and process outlier data in wind turbine time series data using an adaptive quantile-based wind turbine operation and maintenance data outlier identification method, and obtain outlier-processed wind turbine time series data. The key variable screening unit is used to screen key variables that are strongly correlated with the wind turbine operating status and failure risk from multidimensional monitoring variables by using a wind turbine state characterization key variable screening method based on the maximum mutual information coefficient for wind turbine time series data after anomaly handling. The scaling unit is used to scale each key variable using a Bessel correction-based multidimensional variable scaling method for wind turbines to obtain standardized input features. The collaborative scheduling and closed-loop optimization unit is used to determine whether there are missing values ​​in the normalized wind turbine time series data. If so, it enters the missing data processing unit; otherwise, it enters the outlier identification unit. It is also used to determine whether the number of key variables selected is greater than the set threshold. If so, it enters the key variable selection unit again; otherwise, it enters the scale unification unit.

[0024] In one embodiment, the collaborative scheduling and closed-loop optimization unit implements specific functions based on the collaborative scheduling agent.

[0025] Compared with existing technologies, this application has the following advantages: This application is based on data processing methods such as nearest neighbor multi-objective weighted filling, adaptive quantile anomaly identification, and unbiased scale transformation, covering the entire data processing process, including data acquisition, missing value filling, outlier identification, key variable screening, and scale standardization; This application establishes a feedback closed-loop and iterative optimization mechanism to uniformly schedule and manage multiple data processing processes, forming a standardized and unified data governance system, improving the efficiency of wind farm SCADA data governance, reducing manual operation and maintenance costs, and providing a reliable data foundation for subsequent wind turbine status identification and fault early warning. Attached Figure Description

[0026] This application can be better understood by referring to the description given below in conjunction with the accompanying drawings, which, together with the detailed description below, are incorporated in and form part of this specification. In the drawings: Figure 1 A flowchart of the end-to-end autonomous processing method for wind farm SCADA data under multi-model collaboration is shown. Figure 2 A schematic diagram of a wind farm SCADA data end-to-end autonomous processing system under multi-model collaboration is shown. Detailed Implementation

[0027] Exemplary embodiments of the present application will be described below with reference to the accompanying drawings. For clarity and brevity, not all features of the actual embodiments are described in the specification. However, it should be understood that many embodiment-specific decisions can be made in the development of any such actual embodiment to achieve the developer’s specific objectives, and these decisions may vary as the embodiments differ.

[0028] It should also be noted that, in order to avoid obscuring this application with unnecessary details, only the device structure closely related to the solution of this application is shown in the accompanying drawings, while other details that are not closely related to this application are omitted.

[0029] It should be understood that this application is not limited to the described embodiments by virtue of the following description with reference to the accompanying drawings. In this document, embodiments may be combined with each other, features may be substituted or borrowed between different embodiments, and one or more features may be omitted in one embodiment, where feasible.

[0030] This application provides a method for autonomous processing of wind farm SCADA data across the entire data link under multi-model collaboration. Figure 1 A flowchart illustrating the end-to-end autonomous processing method for wind farm SCADA data under multi-model collaboration is shown. (See [link]) Figure 1 The method mainly includes the following steps: Step 1: Obtain wind turbine time series data and perform normalization processing to obtain normalized wind turbine time series data; wind turbine time series data includes multi-dimensional monitoring variables.

[0031] Here, based on the equipment composition, measurement point distribution, and communication interfaces of each wind turbine unit in the wind farm, a multi-channel parallel acquisition channel is established for different wind turbine units, different subsystems, and different measurement point types. This channel synchronously acquires wind speed, wind direction, rotational speed, power, pitch angle, yaw angle, vibration, temperature, current, voltage, and status alarm signals to obtain wind turbine time-series data. The wind turbine time-series data is then uniformly synchronized, aligned to the time axis, structured, validated according to rules, and hierarchically aggregated to form standardized wind turbine time-series data.

[0032] Specifically, the raw data uploaded from each acquisition channel is uniformly synchronized and aligned with the timeline, and an absolute timestamp is added to all measurement point data to eliminate clock drift and sampling timing deviation between different wind turbine units and different acquisition devices; the absolute timestamp is identified using a unified time base and supports conversion between Coordinated Universal Time and local time zones;

[0033] in, The aligned absolute timestamp; This is the original data collection timestamp; This is the time correction amount for clock drift and sampling timing deviation.

[0034] Based on a pre-defined unified data dictionary for wind turbine clusters, the raw data after time calibration is structured and encapsulated to form a unified data format. The unified data format includes at least the following fields: equipment identifier, unit identifier, timestamp, measurement point name, measurement point type, measurement value, unit, status bit, quality code, and data source identifier.

[0035] The structured and encapsulated data is validated using an expert rule base, which includes rules for data integrity, field validity, dimensional consistency, value range, status bit matching, and cross-variable logical consistency. Among these, the cross-variable logical consistency rules are used to screen the relationships between different measuring points based on the wind turbine's operating mechanism and maintenance experience, in order to identify abnormal uploads, field conflicts, duplicate records, and mismatched data.

[0036] in, For rule validation results This is the satisfaction flag for the h-th rule; h is the rule number; z is the total number of rules participating in the validation. Data that passes the rule verification is collected and indexed according to the hierarchy of "wind farm - turbine - subsystem - component - measuring point" to form a standardized format of multi-source heterogeneous operation and maintenance data of wind turbine units, and to provide standardized input for subsequent missing value filling, outlier identification, key variable screening and scale unification processing.

[0037] Step 2: Determine if there are any missing values ​​in the normalized wind turbine timing data. If yes, proceed to Step 3; otherwise, proceed to Step 4.

[0038] Here, a missing value detection function can be used to determine whether there are missing values ​​in the normalized wind turbine timing data.

[0039] Step 3: For the normalized wind turbine time series data, the missing data is imputed using a missing data processing method based on nearest neighbor multi-objective weighting to obtain the missing wind turbine time series data after missing data processing.

[0040] Specifically, firstly, the normalized wind turbine time-series data is recorded as the target sample; Then, the historical sample set is obtained, and candidate samples with similar operating conditions to the target sample are searched in the historical sample set to form a candidate sample set; Then, calculate the Euclidean distance between the target sample and each candidate sample, sort them in ascending order of Euclidean distance, and select the top K candidate samples to form the nearest neighbor sample set; Euclidean distance is expressed as:

[0041] in, For the target sample and the first Euclidean distance between candidate samples; For the target sample j Observations of non-missing variables; For the first In the candidate samples, the first jObservations of non-missing variables; This represents the total number of non-missing variables involved in the distance calculation.

[0042] Then, the initial weights of each sample in the nearest neighbor set are determined based on the Euclidean distance, and then normalized to obtain the normalized weights of each sample. The initial weights are expressed as:

[0043] in, For the nearest neighbor sample set, the first The initial weights corresponding to each sample; For the target sample and the nearest neighbor sample set, the first... Euclidean distance between samples.

[0044] The normalized weights are represented as:

[0045] in, For the nearest neighbor sample set, the first The normalized weights corresponding to each sample; For the nearest neighbor sample set, the first The initial weights corresponding to each sample; This represents the number of samples in the nearest neighbor set.

[0046] Then, determine the imputation values ​​for each missing variable using the following formula:

[0047] in, For the target sample Filler values ​​for missing variables, For the nearest neighbor sample set, the first The first sample Observed values ​​of each variable For the nearest neighbor sample set, the first The normalized weights corresponding to each sample; Then, determine whether the imputation values ​​of each missing variable are within the allowable range of the corresponding variable. If so, the current imputation value is the final imputation value. If not, reconstruct the nearest neighbor sample set and determine the imputation values ​​of the missing variables again.

[0048] Step 4: For the wind turbine time series data, an outlier identification method based on adaptive quantiles is used to identify and process the outlier data, and the wind turbine time series data after anomaly processing is obtained.

[0049] Specifically, first, a historical sample set is obtained, and the historical observation sequence of each monitoring variable is extracted within a preset sliding time window. Based on the historical observation sequence, the lower quartile, upper quartile, and interquartile range of each monitoring variable are calculated. The interquartile range is expressed as:

[0050] in, For the first Interquartile range of the monitored variables For the first The upper quartiles of the monitored variables, For the first The lower quartiles of each monitored variable.

[0051] Then, based on the distribution skewness characterization of the historical observation series of each monitored variable, the adaptive quantile expansion coefficient of each monitored variable is calculated using the following formula:

[0052] in, For the first The adaptive quantile expansion coefficient of each monitored variable, The basic quartile expansion coefficient For the first The skewness sensitivity adjustment coefficients corresponding to the monitored variables For the first The distribution skewness characterization of the historical observation sequence of a monitoring variable.

[0053] Then, based on the lower quartile, upper quartile, interquartile range, and adaptive quantile expansion coefficient, the minor and extreme anomaly boundaries of each monitored variable are determined. The boundary of minor anomalies is:

[0054]

[0055] in, For the first The upper bound of minor anomalies in the monitored variables. For the first The lower bound of minor anomalies in the monitored variables. For the first The upper quartiles of the monitored variables, For the first The lower quartiles of the monitored variables, For the first Interquartile range of the monitored variables For the first The adaptive quantile expansion coefficient of each monitored variable, The interquartile range coefficient; The extreme anomaly boundary is:

[0056]

[0057] in, For the first The upper bound of extreme outliers for each monitored variable. For the first The lower bound of extreme anomalies for each monitored variable.

[0058] Then, the observed values ​​of each monitoring variable in the wind turbine time series data are compared with the minor anomaly boundary and the extreme anomaly boundary to determine whether the observed values ​​of the monitoring variables are abnormal data. Here, when the observed value of the monitored variable is located at and When the observed value of the monitored variable is between [a certain value], it is considered a normal state; when the observed value of the monitored variable is between [a certain value], it is considered a normal state. and Between, or located and When the observed value of the monitored variable is between [a certain value], it is considered a minor abnormality; when the observed value of the monitored variable is greater than [a certain value], it is considered a minor abnormality. or less than When this occurs, it is determined to be an extreme abnormal state.

[0059] Then, abnormal data is removed, replaced, or marked.

[0060] Step 5: The key variable screening method for wind turbine status characterization based on the maximum mutual information coefficient is used to screen key variables that are strongly correlated with wind turbine operating status and fault risk from multidimensional monitoring variables after anomaly handling.

[0061] Specifically, the maximum mutual information coefficient between each monitoring variable and the wind turbine operating status characterization quantity is calculated; the maximum mutual information coefficients are sorted from largest to smallest, and the monitoring variables that are ranked first and whose maximum mutual information coefficients are greater than a preset threshold are selected as key variables.

[0062] Step 6: Determine if the number of key variables being filtered is greater than the set threshold. If yes, return to step 5 and filter the key variables again. If no, proceed to step 7.

[0063] Step 7: The scale of each key variable is unified using the Bessel correction-based multidimensional variable scaling method for wind turbines to obtain standardized input features.

[0064] Specifically, first, a historical sample set is obtained, and then historical observation sequences of each key variable are extracted based on the historical sample set; Then, based on the historical observation sequences of each key variable, and using the Bessel correction method, the sample standard deviation of each key variable is determined; the sample standard deviation is calculated using the following formula:

[0065]

[0066] in, The standard deviation of the sample is 1. The first in the historical observation sequence The observed values ​​of key variables corresponding to each sample. For the sample size, The sample mean of the key variable; Then, the observed values ​​of the key variables are standardized by z-score based on the sample standard deviation to obtain the standardized values ​​of each key variable.

[0067] z-score standardization uses the following formula:

[0068] Where z is the standardized value of the key variable. These are the observed values ​​of the key variables.

[0069] This embodiment utilizes data processing methods such as nearest neighbor multi-objective weighted filling, adaptive quantile anomaly identification, and unbiased scale transformation to cover the entire data processing workflow, including data acquisition, missing value filling, outlier identification, key variable screening, and scale standardization. It establishes a feedback loop and iterative optimization mechanism to uniformly schedule and manage multiple data processing processes, forming a standardized and unified data governance system. This improves the efficiency of wind farm SCADA data governance, reduces manual operation and maintenance costs, and provides a reliable data foundation for subsequent wind turbine status identification and fault early warning.

[0070] This application also provides a wind farm SCADA data end-to-end autonomous processing system under multi-model collaboration. Figure 2 This diagram illustrates a wind farm SCADA data end-to-end autonomous processing system under multi-model collaboration. (See attached diagram) Figure 2 The system includes: The system includes a collaborative scheduling and closed-loop optimization unit, a data acquisition and standardization unit, a missing data processing unit, an outlier identification unit, a key variable screening unit, and a scale unification unit. The data acquisition and normalization unit is used to acquire wind turbine time-series data and perform normalization processing to obtain normalized wind turbine time-series data; the wind turbine time-series data includes multi-dimensional monitoring variables; The missing data processing unit is used to impute missing data in the normalized wind turbine time series data using a missing data processing method based on nearest neighbor multi-objective weighting, so as to obtain the missing wind turbine time series data after missing data processing. The outlier identification unit is used to identify and process outlier data in wind turbine time series data using an adaptive quantile-based wind turbine operation and maintenance data outlier identification method, and obtain outlier-processed wind turbine time series data. The key variable screening unit is used to screen key variables that are strongly correlated with the wind turbine operating status and failure risk from multidimensional monitoring variables by using a wind turbine state characterization key variable screening method based on the maximum mutual information coefficient for wind turbine time series data after anomaly handling. The scaling unit is used to scale each key variable using a Bessel correction-based multidimensional variable scaling method for wind turbines to obtain standardized input features. The collaborative scheduling and closed-loop optimization unit is used to determine whether there are missing values ​​in the normalized wind turbine time series data. If so, it enters the missing data processing unit; otherwise, it enters the outlier identification unit. It is also used to determine whether the number of key variables selected is greater than the set threshold. If so, it enters the key variable selection unit again; otherwise, it enters the scale unification unit.

[0071] The system in this embodiment performs autonomous scheduling and collaborative management of other units through a collaborative scheduling and closed-loop optimization unit. That is, it performs autonomous collaborative scheduling of multiple data processing small models based on a collaborative scheduling intelligent agent, forming a wind farm multi-dimensional operation and maintenance data intelligent autonomous full-link governance system under feedback closed loop and iterative optimization.

[0072] The wind farm SCADA data end-to-end autonomous processing system under multi-model collaboration in this embodiment has the same inventive concept as the wind farm SCADA data end-to-end autonomous processing method under multi-model collaboration described above. Therefore, the specific implementation of this system can be found in the embodiment section of the wind farm SCADA data end-to-end autonomous processing method under multi-model collaboration described above, and its technical effects correspond to the technical effects of the above method, so they will not be repeated here.

[0073] The above descriptions are merely various embodiments of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. A method for autonomous processing of wind farm SCADA data in full link under multi-model cooperation, characterized in that, include: Step 1: Obtain wind turbine timing data and perform normalization processing to obtain normalized wind turbine timing data; The wind turbine time-series data includes multi-dimensional monitoring variables; Step 2: Determine whether there are missing values ​​in the normalized wind turbine timing data. If yes, proceed to step 3; otherwise, proceed to step 4. Step 3: For the normalized wind turbine time series data, the missing data is imputed using a missing data processing method based on nearest neighbor multi-objective weighting to obtain the missing wind turbine time series data after missing data processing. Step 4: For the wind turbine time series data, an outlier identification method based on adaptive quantiles is used to identify and process the outlier data, and the wind turbine time series data after anomaly processing is obtained. Step 5: The wind turbine time series data after anomaly processing is processed using a key variable screening method for wind turbine status characterization based on the maximum mutual information coefficient to screen out key variables that are strongly correlated with wind turbine operating status and fault risk from multidimensional monitoring variables. Step 6: Determine if the number of key variables being filtered is greater than the set threshold. If yes, return to step 5 and filter the key variables again. If no, proceed to step 7. Step 7: The scale of each key variable is unified using the Bessel correction-based multidimensional variable scaling method for wind turbines to obtain standardized input features.

2. The method of claim 1, wherein, Acquire wind turbine time-series data and perform normalization processing, including: Based on the equipment configuration, measurement point distribution, and communication interfaces of each wind turbine unit in the wind farm, a multi-channel parallel acquisition channel is established for different wind turbine units, different subsystems, and different measurement point types. This channel synchronously acquires wind speed, wind direction, rotational speed, power, pitch angle, yaw angle, vibration, temperature, current, voltage, and status alarm signals to obtain wind turbine time-series data. The wind turbine time-series data is then uniformly synchronized, aligned to the time axis, structured, validated according to rules, and hierarchically aggregated to form standardized wind turbine time-series data.

3. The method of claim 1, wherein, Step 3: For the normalized wind turbine time series data, perform missing data imputation using a nearest neighbor multi-objective weighted missing data processing method to obtain missing wind turbine time series data, including: The normalized wind turbine time series data is recorded as the target sample; Obtain a historical sample set, and search the historical sample set for candidate samples that are similar to the operating conditions of the target sample to form a candidate sample set; Calculate the Euclidean distance between the target sample and each candidate sample, sort them in ascending order of Euclidean distance, and select the top K candidate samples to form the nearest neighbor sample set; The initial weights of each sample in the nearest neighbor sample set are determined based on the Euclidean distance and then normalized to obtain the normalized weights of each sample. The imputation values ​​for each missing variable are determined using the following formula: wherein, is a fill value for the i-th missing variable in the target sample, is an observed value of the j-th variable of the i-th sample in the nearest neighbor sample set, is a fill value for the i-th missing variable in the target sample, is an observed value of the j-th variable of the i-th sample in the nearest neighbor sample set, is an observed value of the j-th variable of the i-th sample in the nearest neighbor sample set, is an observed value of the j-th variable of the i-th sample in the nearest neighbor sample set, is an observed value of the j-th variable of the i-th sample in the nearest neighbor sample set. Determine whether the imputed value of each missing variable is within the allowable range of the corresponding variable. If yes, the current imputed value is the final imputed value. If not, reconstruct the nearest neighbor sample set and determine the imputed value of the missing variable again.

4. The method of claim 1, wherein, Step 4: For the wind turbine time-series data, an outlier identification method based on adaptive quantiles is used to identify and process the outlier data, resulting in the anomaly-processed wind turbine time-series data, including: Obtain a historical sample set, extract the historical observation sequence of each monitoring variable within a preset sliding time window, and calculate the lower quartile, upper quartile, and interquartile range of each monitoring variable based on the historical observation sequence; The adaptive quantile expansion coefficient of each monitoring variable is calculated based on the distribution skewness characterization of the historical observation sequence of each monitoring variable. Based on the lower quartile, upper quartile, interquartile range, and adaptive quantile expansion coefficient, the minor and extreme anomaly boundaries of each monitored variable are determined. The observed values ​​of each monitoring variable in the wind turbine time series data are compared with the boundaries of minor anomalies and extreme anomalies to determine whether the observed values ​​of the monitoring variables are abnormal data. The abnormal data is removed, replaced, or marked.

5. The method as described in claim 4, characterized in that, The adaptive quantile expansion coefficient for each monitored variable is calculated using the following formula: in, For the first The adaptive quantile expansion coefficient of each monitored variable, The basic quartile expansion coefficient For the first The skewness sensitivity adjustment coefficients corresponding to the monitored variables For the first The distribution skewness characterization of the historical observation sequence of a monitoring variable.

6. The method as described in claim 4, characterized in that, The boundary of the slight anomaly is: in, For the first The upper bound of minor anomalies in the monitored variables. For the first The lower bound of minor anomalies in the monitored variables. For the first The upper quartiles of the monitored variables, For the first The lower quartiles of the monitored variables, For the first Interquartile range of the monitored variables For the first The adaptive quantile expansion coefficient of each monitored variable, The interquartile range coefficient; The extreme anomaly boundary is: in, For the first The upper bound of extreme outliers for each monitored variable. For the first The lower bound of extreme anomalies for each monitored variable.

7. The method as described in claim 1, characterized in that, Step 5: The anomaly-processed wind turbine time-series data is processed using a key variable screening method based on the maximum mutual information coefficient to identify key variables strongly correlated with wind turbine operating status and fault risk from multi-dimensional monitoring variables. These variables include: Calculate the maximum mutual information coefficient between each monitored variable and the wind turbine operating status characterization quantity; The maximum mutual information coefficients are sorted from largest to smallest, and the monitoring variables that are ranked first and have a maximum mutual information coefficient greater than a preset threshold are selected as key variables.

8. The method as described in claim 1, characterized in that, Step 7: Scale the key variables using a Bessel correction-based multidimensional variable scaling method for wind turbines to obtain standardized input features, including: Obtain a historical sample set and extract the historical observation sequences of each key variable based on the historical sample set; Based on the historical observation sequences of each key variable, and using the Bessel correction method, the sample standard deviation and sample mean of each key variable are determined; the sample standard deviation and sample mean are calculated using the following formulas: in, The standard deviation of the sample is 1. The first in the historical observation sequence The observed values ​​of key variables corresponding to each sample. For the sample size, The sample mean of the key variable; z-score standardization is performed on the observed values ​​of key variables based on the sample standard deviation to obtain the standardized values ​​of each key variable.

9. A wind farm SCADA data end-to-end autonomous processing system under multi-model collaboration, characterized in that, include: The system includes a collaborative scheduling and closed-loop optimization unit, a data acquisition and standardization unit, a missing data processing unit, an outlier identification unit, a key variable screening unit, and a scale unification unit. The data acquisition and normalization unit is used to acquire wind turbine time-series data and perform normalization processing to obtain normalized wind turbine time-series data; the wind turbine time-series data includes multi-dimensional monitoring variables; The missing data processing unit is used to perform missing data interpolation on the normalized wind turbine time series data using a missing data processing method based on nearest neighbor multi-objective weighting, to obtain missing wind turbine time series data after missing data processing. The outlier identification unit is used to identify and process the wind turbine time series data using an outlier identification method based on adaptive quantiles, so as to obtain the wind turbine time series data after anomaly processing. The key variable screening unit is used to screen the key variables of the wind turbine status characterization based on the maximum mutual information coefficient for the wind turbine time series data after anomaly processing, and to screen out key variables that are strongly correlated with the wind turbine operating status and fault risk from multidimensional monitoring variables. The scaling unit is used to scale each key variable using a Bessel correction-based multidimensional variable scaling method for wind turbines to obtain standardized input features. The collaborative scheduling and closed-loop optimization unit is used to determine whether there are missing values ​​in the normalized wind turbine time series data. If so, it enters the missing data processing unit; if not, it enters the outlier identification unit. It is also used to determine whether the number of key variables being screened is greater than a set threshold. If so, it enters the key variable screening unit again; if not, it enters the scale unification unit.

10. The system as described in claim 9, characterized in that, The collaborative scheduling and closed-loop optimization unit implements specific functions based on the collaborative scheduling agent.

Images