CN-122020195-A - Wind power SCADA data online self-adaptive outlier detection method considering conceptual drift

CN122020195ACN 122020195 ACN122020195 ACN 122020195ACN-122020195-A

Abstract

The invention belongs to the technical field of SCADA data detection of a wind power plant, and particularly relates to an online self-adaptive abnormal value detection method of wind power SCADA data considering conceptual drift, comprising the following steps of S100, acquiring actual wind power in real time through an SCADA system; preprocessing and reasonably screening, S200, obtaining a priori wind power sequence through wind power of a power model computer unit, constructing a residual sequence, introducing input wind speed as a scaling factor to obtain a scaled residual sequence, processing the scaled residual sequence, marking an abnormal value according to a preset threshold value, S300, dynamically monitoring the scaled residual sequence by adopting an exponential weighted moving average method, triggering a concept drift candidate event when the EWMA value exceeds a control limit UCL, carrying out difference test on current residual distribution and historical reference distribution by adopting KS test, confirming that concept drift occurs if the difference exceeds the preset threshold value, and updating parameters of a power model after judging that the concept drift occurs.

Inventors

ZHANG FANGHONG
PEI FAN
WANG YU
ZHANG YONGLIANG

Assignees

重庆师范大学

Dates

Publication Date: 20260512
Application Date: 20260115

Claims (8)

1. The wind power SCADA data online self-adaptive outlier detection method considering conceptual drift is characterized by comprising the following steps of: s100, data acquisition and pretreatment: Acquiring actual wind power and wind speed time sequence data of the wind turbine generator in real time through an SCADA system; preprocessing the time sequence data by adopting a sliding window method based on physical constraint, and specifically comprising the steps of constructing a data window and carrying out rationality screening; S200, performing outlier primary screening: The wind power of the set is calculated through a pre-established power model, and a priori wind power sequence is obtained; Constructing a residual sequence based on the actual wind power and the priori wind power; introducing an input wind speed as a scaling factor, and scaling the residual sequence to obtain a scaled residual sequence; processing the scaled residual sequence through Hampel filters, and identifying and marking abnormal values according to a preset threshold; S300, concept drift detection: Dynamically monitoring the scaled residual sequence by adopting an exponential weighted moving average method; triggering concept drift candidate events when the EWMA value exceeds a control limit UCL; performing difference detection on the current residual error distribution and the historical reference distribution by adopting KS detection, and if the difference exceeds a preset threshold value, confirming that conceptual drift occurs; S400, model self-adaptive updating: And after judging that the concept drift occurs, updating parameters of the power model based on a domain adaptation method of the maximum mean value difference.
2. The method for online adaptive outlier detection of wind power SCADA data considering conceptual drift according to claim 1, wherein the step S100 of preprocessing the time series data by adopting a sliding window method based on physical constraints specifically comprises the following steps: S110, constructing a sliding data window with the length of N, and constructing an actual wind power data stream, wherein the actual wind power data stream is expressed as: ; Wherein, the If the history data is insufficient to construct a window with the length of N, the window data is expanded by adopting a copying filling mechanism; s120, rationality screening is carried out on the data in the window according to physical constraint, and exceeding rated power is removed And data less than 0.
3. The method for online adaptive outlier detection of wind power SCADA data taking into account conceptual drift according to claim 2, wherein the power model in step S200 is expressed as: ; Wherein, the Representing a priori wind power derived from the power model, For a multi-parameter power calculation formula, a neural network and other power models based on parameters, As a parameter of the model, it is possible to provide, Parameters are input for a power model comprising wind speed.
4. The method for online adaptive outlier detection of wind power SCADA data considering conceptual drift according to claim 3, wherein the step S200 specifically comprises the following steps: S210, obtaining a priori wind power sequence through wind power of a power model computer unit, wherein the priori wind power sequence is expressed as: ; s220, constructing a residual sequence based on the actual wind power and the prior wind power, wherein the residual sequence is expressed as follows: ; S230, introducing an input wind speed as a scaling factor, and scaling the residual sequence to obtain a scaled residual sequence, wherein the scaled residual sequence is expressed as: ; Wherein, the And the input wind speed of the unit at the time t is shown.
5. The method for online adaptive outlier detection of wind power SCADA data considering conceptual drift according to claim 4, wherein the step S200 of processing the scaled residual sequence through Hampel filter and identifying and marking outliers according to a preset threshold is specifically as follows: S240, calculating the scaled residual sequence Median of (2) Calculating an estimated standard deviation of each data and the median of the scaled residual sequence by using the absolute value of the difference between each data and the median of the scaled residual sequence, wherein the estimated standard deviation is expressed as: ; s250, judging whether actual wind power data acquired by the SCADA system is an abnormal value according to the following rule: ; Wherein, the Representing a preset threshold, when Indicating that the actual wind power at that time is normal data, The actual wind power at this time is shown as abnormal data.
6. The method for online adaptive outlier detection of wind power SCADA data considering conceptual drift according to claim 5, wherein the step S300 of dynamically monitoring the scaled residual sequence by using an exponential weighted moving average method is specifically: s310, calculating EWMA value at t time Expressed as: ; Wherein, the Is a smooth coefficient and ; S320, a control limit UCL is set, expressed as: ; Wherein, the As a target average value of the values, As the standard deviation of the data, To control the width factor; s330, triggering concept drift candidate events is as follows: when M times are consecutive When the concept drift candidate events are beyond the control limit UCL, judging the concept drift candidate events; s340, performing difference test on the current residual error distribution and the historical reference distribution by adopting KS test, wherein the difference test is specifically as follows: calculating cumulative distribution function of current window And a cumulative distribution function of the reference window Is set to a maximum absolute distance D of (a), expressed as: ; S350, if the difference exceeds a preset threshold, confirming that the concept drift occurs, specifically: when D is greater than a preset threshold When the actual wind power data acquired by the SCADA system is judged to have conceptual drift; When D is smaller than the preset threshold And when the actual wind power data acquired by the SCADA system is judged not to have conceptual drift.
7. The method for online adaptive outlier detection of wind power SCADA data considering conceptual drift according to claim 6, wherein the parameter update rule of "domain adaptation method based on maximum mean value difference" in step S400 is expressed as: ; Wherein, the As the weight coefficient of the light-emitting diode, Is a loss function for measuring the difference between the history distribution P and the new distribution Q, defined as the distance of mean embedding in the regenerated kernel hilbert space, expressed as: ; Wherein, the As a kernel function The induced feature map, H, is the regenerated nuclear hilbert space.
8. The method for online adaptive outlier detection of wind power SCADA data having concept drift taken into account of claim 7, wherein for a limited sample The empirical estimation form is expressed as: 。

Description

Wind power SCADA data online self-adaptive outlier detection method considering conceptual drift Technical Field The invention belongs to the technical field of SCADA data detection of wind power plants, and particularly relates to an online self-adaptive abnormal value detection method for wind power SCADA data considering conceptual drift. Background In wind farm operation and management, the SCADA system is widely applied to real-time monitoring and data real-time acquisition so as to perform fan unit state monitoring, fault diagnosis, wind power prediction, power generation performance evaluation and assessment and other applications. Among the numerous SCADA data, wind power SCADA data is particularly important, and its data quality directly affects the operational efficiency of the wind farm and the power generation expectations. However, under complex working conditions, the wind power SCADA data usually causes a large number of abnormal values due to factors such as sensor faults, communication delays, environmental interference, and the like. Therefore, how to efficiently and accurately detect and detect abnormal values in wind power SCADA data has become a key problem in the field of wind farm data processing. Aiming at the wind power abnormal data identification and detection technology of wind power generation, a great deal of research work is carried out at home and abroad, and the method can be mainly divided into identification methods based on statistical analysis, machine learning and image processing. The statistical method is that a great amount of historical wind power is divided according to wind speed bins, and then a relevant threshold value is set by using statistical characteristics to judge whether abnormal values exist. The method relies on a large amount of historical data and is not suitable for identifying and detecting abnormal data of the SCADA data flow. With the artificial intelligence method, the machine learning and image processing method is also widely applied to wind power anomaly data identification. Since wind power anomaly data is typically unlabeled, the core idea of machine learning methods is to identify through density, representative algorithms include isolated forests, local outliers, and the like. These methods can capture complex patterns in the data by training the model, but their main challenge is that the adaptation to data changes is poor, especially in the face of unavoidable conceptual drift in wind power SCADA data, where the performance of the algorithm can be significantly degraded. The method based on image processing is to convert outlier identification into image segmentation problem, and also depends on a large amount of historical data, so that the method is not suitable for identifying and detecting the outlier of SCADA data flow. Along with the development of automation and intellectualization of wind farms, the demand for online detection of abnormal values of wind power SCADA data has increased dramatically. Meanwhile, wind power SCADA data can change data statistics characteristics due to factors such as unit aging, blade replacement or environmental characteristic change, namely the wind power SCADA data shows conceptual drift. It is emphasized that this drift can cause the anomaly detection threshold to fail, resulting in false alarms and false alarms. Therefore, the detection of abnormal values of wind power SCADA data is urgent to consider the problems of concept drift and online identification, which is particularly important for the intelligent development of wind power plants, but researchers at home and abroad are few. Disclosure of Invention The invention aims to provide an online adaptive outlier detection method for wind power SCADA data considering conceptual drift, which is used for solving the problems in the background technology. In order to achieve the technical purpose, the invention adopts the following technical scheme: A wind power SCADA data online self-adaptive outlier detection method considering conceptual drift comprises the following steps: s100, data acquisition and pretreatment: Acquiring actual wind power and wind speed time sequence data of the wind turbine generator in real time through an SCADA system; preprocessing the time sequence data by adopting a sliding window method based on physical constraint, and specifically comprising the steps of constructing a data window and carrying out rationality screening; S200, performing outlier primary screening: The wind power of the set is calculated through a pre-established power model, and a priori wind power sequence is obtained; Constructing a residual sequence based on the actual wind power and the priori wind power; introducing an input wind speed as a scaling factor, and scaling the residual sequence to obtain a scaled residual sequence; processing the scaled residual sequence through Hampel filters, and identifying and marking abnormal values according to