Wind power SCADA data online adaptive abnormal value detection method considering concept drift
By combining a physical constraint-based sliding window method and a power model with a Hampel filter for wind power SCADA data preprocessing, and utilizing the exponentially weighted moving average method and the KS test to detect conceptual drift, adaptive outlier detection of wind power SCADA data in wind farms is achieved. This solves the problems of insufficient detection accuracy and stability in existing technologies, and improves the operating efficiency and power generation expectations of wind farms.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- CHONGQING NORMAL UNIVERSITY
- Filing Date
- 2026-01-15
- Publication Date
- 2026-05-12
AI Technical Summary
Existing technologies struggle to effectively detect and identify outliers in wind power SCADA data, especially when dealing with concept drift, leading to frequent false alarms and missed reports, which impacts the operational efficiency and power generation expectations of wind farms.
Data preprocessing is performed using a physical constraint-based sliding window method, followed by initial screening using a power model and Hampel filter. Conceptual drift is detected using an exponentially weighted moving average method and the KS test, and model parameters are updated using a domain adaptation method based on the maximum mean difference to achieve adaptive outlier detection.
It achieves efficient and accurate outlier detection of wind power SCADA data under concept drift conditions, reduces computational complexity, improves detection stability and accuracy, and avoids misjudgments caused by model failure in traditional methods.
Smart Images
Figure CN122020195A_ABST
Abstract
Description
Technical Field
[0001] This invention belongs to the field of wind farm SCADA data detection technology, specifically relating to an online adaptive outlier detection method for wind power SCADA data that considers concept drift. Background Technology
[0002] In wind farm operation and management, SCADA systems are widely used for real-time monitoring and data acquisition to facilitate applications such as turbine status monitoring, fault diagnosis, wind power prediction, and power generation performance evaluation. Among the various SCADA data, wind power SCADA data is particularly important, as its quality directly affects the wind farm's operating efficiency and power generation expectations. However, due to complex operating conditions, wind power SCADA data often exhibits numerous outliers due to sensor failures, communication delays, and environmental interference. Therefore, how to efficiently and accurately detect and identify outliers in wind power SCADA data has become a key issue in the field of wind farm data processing.
[0003] Extensive research has been conducted both domestically and internationally on anomaly identification and detection technologies for wind power data, which can be mainly categorized into methods based on statistical analysis, machine learning, and image processing. Statistical methods divide historical wind power data into modules based on wind speed and then use statistical features to set relevant thresholds to determine the presence of outliers. This method relies heavily on historical data and is unsuitable for anomaly identification and detection in SCADA data streams. With the advancement of artificial intelligence, machine learning and image processing methods have also been widely applied to anomaly identification in wind power data. Since wind power anomaly data is typically unlabeled, the core idea of machine learning methods is to identify anomalies through density, with representative algorithms including Isolation Forest and Local Outlier Factor. These methods can capture complex patterns in the data through model training, but their main challenge lies in their poor adaptability to data changes, especially when facing the inevitable concept drift in wind power SCADA data, where algorithm performance significantly degrades. Image processing methods transform anomaly identification into an image segmentation problem, also relying heavily on historical data and are unsuitable for anomaly identification and detection in SCADA data streams. With the development of automation and intelligence in wind farms, the demand for online detection of anomalies in wind power SCADA data has increased dramatically. Meanwhile, wind power SCADA data can exhibit changes in statistical characteristics due to factors such as turbine aging, blade replacement, or changes in environmental characteristics; this is known as conceptual drift. It is important to emphasize that this drift can render anomaly detection thresholds ineffective, leading to frequent false negatives and false negatives. Therefore, anomaly detection in wind power SCADA data urgently needs to consider conceptual drift and online identification, which is particularly important for the intelligent development of wind farms, yet research in this area is scarce both domestically and internationally. Summary of the Invention
[0004] The purpose of this invention is to provide an online adaptive outlier detection method for wind power SCADA data that takes into account concept drift, in order to solve the problems existing in the background art.
[0005] To achieve the above-mentioned technical objectives, the technical solution adopted by the present invention is as follows: An online adaptive outlier detection method for wind power SCADA data considering concept drift includes the following steps: S100, Data Acquisition and Preprocessing: The actual wind power and wind speed time-series data of the wind turbine are collected in real time through the SCADA system. The time series data is preprocessed using a sliding window method based on physical constraints, specifically including constructing a data window and performing a reasonable selection. S200, Initial screening of outliers: The prior wind power sequence is obtained by using the wind power of a pre-established power model computer group; Based on the actual wind power and the prior wind power, a residual sequence is constructed; By introducing the input wind speed as a scaling factor, the residual sequence is scaled to obtain a scaled residual sequence. The scaled residual sequence is processed by a Hampel filter, and outliers are identified and marked according to a preset threshold. S300, Concept Drift Detection: The scaled residual sequence is dynamically monitored using an exponentially weighted moving average method. When the EWMA value exceeds the control limit UCL, a concept drift candidate event is triggered; The KS test is used to test the difference between the current residual distribution and the historical reference distribution. If the difference exceeds a preset threshold, concept drift is confirmed. S400, Model Adaptive Update: After determining that concept drift has occurred, the parameters of the power model are updated based on the domain adaptation method of maximum mean difference.
[0006] As a further limitation of the technical solution of the present invention, step S100, "preprocessing the time series data using a sliding window method based on physical constraints," specifically includes the following steps: S110. Construct a sliding data window of length N to build the actual wind power data stream; S120. Based on physical constraints, perform reasonable filtering on the data within the window, eliminating data exceeding the rated power. And data less than 0.
[0007] As a further limitation of the technical solution of the present invention, the step S200 of "processing the scaled residual sequence through a Hampel filter and identifying and marking outliers according to a preset threshold" specifically means: Calculate the scaled residual sequence median The estimated standard deviation of each data point in the scaled residual sequence is calculated by using the absolute value of the difference between each data point in the scaled residual sequence and the median. Based on a preset threshold, determine whether the actual wind power data collected by the SCADA system is an anomaly.
[0008] As a further limitation of the technical solution of the present invention, the step S300 of "dynamically monitoring the scaled residual sequence using the exponentially weighted moving average method" specifically means: S310. Calculate the EWMA value at time t. ; S320, Set control limit UCL; S330, when there are M consecutive time intervals When all exceed the control limit (UCL), it is determined to be a concept drift candidate event; S340. Use the KS test to test the difference between the current residual distribution and the historical reference distribution; S350. If the difference exceeds a preset threshold, concept drift is confirmed to have occurred, specifically: When the value exceeds a preset threshold, it is determined that the actual wind power data collected by the SCADA system has undergone conceptual drift. When the value is less than a preset threshold, it is determined that the actual wind power data collected by the SCADA system has not undergone conceptual drift.
[0009] The present invention has the following beneficial effects: ① By introducing prior knowledge based on the power model to detect outliers in wind power SCADA data, the computational complexity can be significantly reduced while ensuring detection accuracy, thus achieving efficient online outlier detection of wind power SCADA data.
[0010] ② By using a wind speed scaling factor to detect outliers in wind power SCADA data, the heteroscedasticity problem caused by wind speed range differences in wind power SCADA data can be alleviated, further improving the stability and accuracy of outlier detection.
[0011] ③ A concept drift detection and adaptive update mechanism was designed, which can identify changes in the distribution of wind power SCADA data in real time and adaptively update the power model parameters according to the changes. This allows the present invention to avoid the problem of misjudging outliers caused by model failure in traditional methods when facing factors such as unit aging and environmental changes. Attached Figure Description
[0012] The present invention can be further illustrated by the non-limiting embodiments given in the accompanying drawings.
[0013] Figure 1 This is a simplified flowchart illustrating the process of this invention. Figure 2 This is a flowchart of the wind power SCADA data outlier detection based on prior wind power and wind speed scaling factor of the present invention. Figure 3 This is a flowchart illustrating the concept of drift detection and adaptive updating in this invention. Detailed Implementation
[0014] To enable those skilled in the art to better understand the present invention, the technical solution of the present invention will be further described below in conjunction with the accompanying drawings and embodiments.
[0015] Example: like Figures 1 to 3 The method for online adaptive outlier detection of wind power SCADA data considering concept drift, as shown, includes the following steps: S100, Data Acquisition and Preprocessing: The actual wind power and wind speed time-series data of the wind turbine are collected in real time through the SCADA system. A sliding window method based on physical constraints is used to preprocess time series data, specifically including constructing a data window and performing a reasonable selection.
[0016] In this step, "preprocessing time series data using a sliding window method based on physical constraints" specifically includes the following steps: S110. Construct a sliding data window of length N to build the actual wind power data stream, represented as: ; in, For the actual wind power data collected by the SCADA system at time t, if the historical data is insufficient to construct a window of length N, a copy-filling mechanism is used to expand the window data; S120. Based on physical constraints, perform reasonable filtering on the data within the window, eliminating data exceeding the rated power. And for data less than 0, the judgment rules are as follows: ; in, An anomaly label indicating the actual wind power at time t. This indicates that the actual wind power at that moment is within the normal range. This indicates that the actual wind power at that moment is abnormal data. It is determined based on the unit model.
[0017] S200, Initial screening of outliers: S210. Obtain the prior wind power sequence through the wind power of the pre-established power model computer group; The power model is expressed as follows: ; in, This represents the prior wind power obtained from the power model. For multi-parameter power calculation formulas, neural networks, and other parameter-based power models, For model parameters, Input parameters for the power model, including wind speed; The prior wind power sequence obtained from the wind power of the power model computer group is expressed as: ; S220. Based on the actual wind power and the prior wind power, construct the residual sequence; expressed as: ; S230. Introducing the input wind speed as a scaling factor, the residual sequence is scaled to obtain the scaled residual sequence, expressed as: ; in, This represents the unit's input wind speed at time t; S240. The scaled residual sequence is processed using a Hampel filter, and outliers are identified and marked according to a preset threshold. Specifically: Calculate the scaled residual sequence median Using the absolute values of the differences between each data point in the scaled residual series and the median, we can then calculate the estimated standard deviation of each data point in the scaled residual series relative to the median, expressed as: ; S250. Determine whether the actual wind power data collected by the SCADA system is an outlier according to the following rules: ; in, Indicates a preset threshold, when This indicates that the actual wind power at that moment is within the normal range. This indicates that the actual wind power at that moment is abnormal data.
[0018] S300, Concept Drift Detection: The scaled residual sequence is dynamically monitored using an exponentially weighted moving average method. When the EWMA value exceeds the control limit UCL, a concept drift candidate event is triggered; The KS test is used to test the difference between the current residual distribution and the historical reference distribution. If the difference exceeds a preset threshold, concept drift is confirmed. The specific process for this step is as follows: S310. Calculate the EWMA value at time t. , is represented as: ; in, Smoothing coefficient and ; S320, Set control limit UCL, as follows: ; in, The target average value, The standard deviation of the data. To control the width factor; S330, The conditions for triggering the concept drift candidate event are: When M consecutive moments When all exceed the control limit (UCL), it is determined to be a concept drift candidate event; S340. Use the KS test to test the difference between the current residual distribution and the historical reference distribution, specifically: Calculate the cumulative distribution function of the current window. and the cumulative distribution function of the reference window The maximum absolute distance D is expressed as: ; S350. If the difference exceeds a preset threshold, concept drift is confirmed to have occurred, specifically: When D is greater than the preset threshold At that time, it was determined that the actual wind power data collected by the SCADA system had undergone conceptual drift; When D is less than the preset threshold At that time, it was determined that the actual wind power data collected by the SCADA system did not experience conceptual drift.
[0019] S400, Model Adaptive Update: After determining that concept drift has occurred, the parameters of the power model are updated based on the domain adaptation method of maximum mean difference.
[0020] In this step, the parameter update rule is expressed as follows: ; in, These are the weighting coefficients. It is a loss function used to measure the difference between the historical distribution P and the new distribution Q. It is defined as the distance between the mean embeddings in the Hilbert space of the reproducing kernel, and is expressed as: ; in, For kernel function The induced feature map, H is the regenerating kernel Hilbert space.
[0021] For a finite sample Its empirical estimate is expressed as: .
[0022] The above embodiments are merely illustrative of the principles and effects of the present invention and are not intended to limit the invention. Any person skilled in the art can modify or alter the above embodiments without departing from the spirit and scope of the present invention. Therefore, all equivalent modifications or alterations made by those skilled in the art without departing from the spirit and technical concept disclosed in the present invention should still be covered by the claims of the present invention.
Claims
1. A method for online adaptive outlier detection of wind power SCADA data considering concept drift, characterized in that: Includes the following steps: S100, Data Acquisition and Preprocessing: The actual wind power and wind speed time-series data of the wind turbine are collected in real time through the SCADA system. The time series data is preprocessed using a sliding window method based on physical constraints, specifically including constructing a data window and performing a reasonable selection. S200, Initial screening of outliers: The prior wind power sequence is obtained by using the wind power of a pre-established power model computer group; Based on the actual wind power and the prior wind power, a residual sequence is constructed; By introducing the input wind speed as a scaling factor, the residual sequence is scaled to obtain a scaled residual sequence. The scaled residual sequence is processed by a Hampel filter, and outliers are identified and marked according to a preset threshold. S300, Concept Drift Detection: The scaled residual sequence is dynamically monitored using an exponentially weighted moving average method. When the EWMA value exceeds the control limit UCL, a concept drift candidate event is triggered; The KS test is used to test the difference between the current residual distribution and the historical reference distribution. If the difference exceeds a preset threshold, concept drift is confirmed. S400, Model Adaptive Update: After determining that concept drift has occurred, the parameters of the power model are updated based on the domain adaptation method of maximum mean difference.
2. The method for online adaptive outlier detection of wind power SCADA data considering concept drift according to claim 1, characterized in that: The step S100, "preprocessing the time series data using a sliding window method based on physical constraints," specifically includes the following steps: S110. Construct a sliding data window of length N to build the actual wind power data stream, represented as: ； in, For the actual wind power data collected by the SCADA system at time t, if the historical data is insufficient to construct a window of length N, a copy-filling mechanism is used to expand the window data; S120. Based on physical constraints, perform reasonable filtering on the data within the window, eliminating data exceeding the rated power. And data less than 0.
3. The online adaptive outlier detection method for wind power SCADA data considering concept drift according to claim 2, characterized in that: The power model in step S200 is expressed as follows: ； in, This represents the prior wind power obtained from the power model. For multi-parameter power calculation formulas, neural networks, and other parameter-based power models, For model parameters, Input parameters for the power model, including wind speed.
4. The online adaptive outlier detection method for wind power SCADA data considering concept drift according to claim 3, characterized in that: The specific steps S200 are as follows: Includes the following steps: S210. Using the wind power from the power model computer group, the prior wind power sequence is obtained, expressed as: ； S220. Based on the actual wind power and the prior wind power, a residual sequence is constructed, expressed as follows: ； S230. Introduce the input wind speed as a scaling factor to scale the residual sequence, obtaining a scaled residual sequence, as follows: ； in, This represents the unit's input wind speed at time t.
5. The online adaptive outlier detection method for wind power SCADA data considering concept drift according to claim 4, characterized in that: In step S200, "processing the scaled residual sequence using a Hampel filter and identifying and marking outliers according to a preset threshold" specifically means: S240, Calculate the scaled residual sequence median Using the absolute values of the differences between each data point in the scaled residual series and the median, we can then calculate the estimated standard deviation of each data point in the scaled residual series relative to the median, expressed as: ； S250. Determine whether the actual wind power data collected by the SCADA system is an outlier according to the following rules: ； in, Indicates a preset threshold, when This indicates that the actual wind power at that moment is within the normal range. This indicates that the actual wind power at that moment is abnormal data.
6. The online adaptive outlier detection method for wind power SCADA data considering concept drift according to claim 5, characterized in that: The step S300, "dynamically monitoring the scaled residual sequence using the exponentially weighted moving average method," specifically refers to: S310. Calculate the EWMA value at time t. , represented as: ； in, Smoothing coefficient and ; S320, Set control limit UCL, as follows: ； in, The target average value, The standard deviation of the data. To control the width factor; S330, The conditions for triggering the concept drift candidate event are: When M consecutive moments When all exceed the control limit (UCL), it is determined to be a concept drift candidate event; S340. Use the KS test to test the difference between the current residual distribution and the historical reference distribution, specifically: Calculate the cumulative distribution function of the current window. and the cumulative distribution function of the reference window The maximum absolute distance D is expressed as: ； S350. If the difference exceeds a preset threshold, concept drift is confirmed to have occurred, specifically: When D is greater than the preset threshold At that time, it was determined that the actual wind power data collected by the SCADA system had undergone conceptual drift; When D is less than the preset threshold At that time, it was determined that the actual wind power data collected by the SCADA system did not experience conceptual drift.
7. The online adaptive outlier detection method for wind power SCADA data considering concept drift according to claim 6, characterized in that: The parameter update rule for the "domain adaptation method based on maximum mean difference" in step S400 is expressed as follows: ； in, These are the weighting coefficients. It is a loss function used to measure the difference between the historical distribution P and the new distribution Q. It is defined as the distance between the mean embeddings in the Hilbert space of the reproducing kernel, and is expressed as: ； in, For kernel function The induced feature map, H is the regenerating kernel Hilbert space.
8. The online adaptive outlier detection method for wind power SCADA data considering concept drift according to claim 7, characterized in that: For a finite sample Its empirical estimate is expressed as: 。