Real-Time Multiple Event Detection and Classification Using Moving Window PCA

Question

1. What have the authors contributed in "Real-time multiple event detection and classification using moving window pca" ?

2. What are the future works in "Real-time multiple event detection and classification using moving window pca" ?

3. Why is p1 required to construct the PCA model?

4. What is the common application of PCA?

Accepted Answer

This paper proposes a method for the detection and classification of multiple events in an electrical power system in real-time, namely ; islanding, high frequency events ( loss of load ) and low frequency events ( loss of generation ).

Accepted Answer

Future work will look at a number of enhancements to the MW-PCA based MEDC methodology to address these limitations.. Firstly, to improve classification accuracy and the ability to disaggregate multiple events in the power system, a multiple PCA model framework will be developed to enable power system operation during events to be modelled separately from normal operation.. Secondly, to improve robustness in terms of detection of islanding events where there is little frequency drift between the island and the rest of the power system voltage and phase angle information will be incorporated into the PCA models.

Accepted Answer

Due to the high correlation between frequency variables in a power system during normal operation only the first principal component, p1 is required to construct the PCA model.

Accepted Answer

The most common application of PCA is reducing the dimensionality of datasets, typically consisting of large numbers of correlated variables, with minimal information loss [5] in order to reveal any simplified structures that may underline them.

Accepted Answer

To reduce the potential for false alarms a threshold of 5 or more consecutive violations of the 99% confidence limits was defined as the trigger for the MEDC method.

Accepted Answer

An additional advantage that the PCA based approach offers is that, by fusing data from multiple sites, it enhances the signal to noise ratio and hence enables more robust event detection and classification.

Accepted Answer

It was4 found that for the power system studied in this work a window size of between 3 and 6 hours is able to achieve a false alarm rate of less than 1%.

Accepted Answer

Choosing a window size less than 3 hours gave too high a false alarm rate, whilst a window size above 6 hours was considered too computationally inefficient.

Accepted Answer

The classification index was able to classify both the low frequency and high frequency events due to the fact that the loss of load was substantially greater that the loss of generation and hence the event transitioned from a net loss of generation to a net loss of load event.

Accepted Answer

If the frequency variables are decomposed into subspaces corresponding to the different island groups, the frequency variables in the j-th subspace can be denoted as f (j) ∈ <1×mj , with the corresponding loadingFig.

Accepted Answer

4: Zoomed in section of case study 2, showing the dip in generation(generator 1) at 50 seconds, followed by the loss of Load A at 55 seconds.

Accepted Answer

this limitation is due to frequency being a global power system variable and is thus a limitation of all frequency based power system analysis techniques.

Accepted Answer

This allows the time-varying characteristics of the power system to be taken into consideration and facilitates real-time updating of the PCA model, leading to a greater situational awareness of the varying power system.

Accepted Answer

These events were caused by a short circuit fault in the north of Scotland resulting in a loss of generation, leading to the low frequency and islanding event observed at f5.

Accepted Answer

When w >> m, it can be more efficient for MW-PCA model updates to precompute the initial covariance matrix FTF offline, which has O(m2w) complexity.

Accepted Answer

The low frequency event was also detected in a timely fashion using all window sizes, however again the return to mains frequency detection was poor at smaller window sizes and improved as window size was increased.

Real-Time Multiple Event Detection and Classification Using Moving Window PCA

Chat with Paper

AI Agents for this Paper

Most frequently asked questions

1. What have the authors contributed in "Real-time multiple event detection and classification using moving window pca" ?

2. What are the future works in "Real-time multiple event detection and classification using moving window pca" ?

3. Why is p1 required to construct the PCA model?

4. What is the common application of PCA?

5. How many consecutive violations of the 99% confidence limits were defined as the trigger for the MEDC?

6. What is the advantage of using the PCA based approach?

7. How many hours can a system achieve a false alarm rate?

8. How many windows were considered too large?

9. What is the classification index able to classify?

10. What is the corresponding loading for the j-th subspace?

11. How many seconds of the zoomed in section of case study 2?

12. Why is frequency a limitation of power system analysis techniques?

13. What is the advantage of the MEDC method over the traditional PCA?

14. What caused the low frequency and islanding event at f5?

15. What is the way to calculate the initial covariance matrix?

16. What was the way to detect the low frequency event?

Figures

Citations

Real-Time Multiple Event Detection and Classification in Power System Using Signal Energy Transformations

Data-Driven Event Detection of Power Systems Based on Unequal-Interval Reduction of PMU Data and Local Outlier Factor

A Real Time Event Detection, Classification and Localization Using Synchrophasor Data

Data-Driven Coherency Identification for Generators Based on Spectral Clustering

Artificial neural network and phasor data-based islanding detection in smart grid

References

Principal Component Analysis

Principal component analysis

A User's Guide to Principal Components.

Fault Detection and Diagnosis in Industrial Systems

Final report on the August 14, 2003 blackout in the United States and Canada : causes and recommendations

Related Papers (5)

Principal Component Analysis of Wide-Area Phasor Measurements for Islanding Detection—A Geometric View

Dimensionality Reduction of Synchrophasor Data for Early Event Detection: Linearized Analysis

Supervisory Protection and Automated Event Diagnosis Using PMU Data

Real-Time Multiple Event Detection and Classification in Power System Using Signal Energy Transformations

Real-Time Identification of Dynamic Events in Power Systems Using PMU Data, and Potential Applications—Models, Promises, and Challenges