Acquiring Better Load Estimates by Combining Anomaly and Change Point Detection in Power Grid Time-series Measurements

TL;DR

A novel methodology for automatic anomaly and switch event filtering to improve load estimation in power grid systems by leveraging unsupervised methods with supervised optimization that prioritizes interpretability while ensuring robust and generalizable performance on unseen data.

Abstract

In this paper we present novel methodology for automatic anomaly and switch event filtering to improve load estimation in power grid systems. By leveraging unsupervised methods with supervised optimization, our approach prioritizes interpretability while ensuring robust and generalizable performance on unseen data. Through experimentation, a combination of binary segmentation for change point detection and statistical process control for anomaly detection emerges as the most effective strategy, specifically when ensembled in a novel sequential manner. Results indicate the clear wasted potential when filtering is not applied. The automatic load estimation is also fairly accurate, with approximately 90% of estimates falling within a 10% error margin, with only a single significant failure in both the minimum and maximum load estimates across 60 measurements in the test set. Our methodology's interpretability makes it particularly suitable for critical infrastructure planning, thereby enhancing decision-making processes.

Figures (10)

• Figure 1: An illustration of a switch event. The diamonds numbered 1 and 2 indicate two primary substations. The squares named A-F indicate secondary substations. The circle named X indicates a switch. Solid lines indicates connections between substations, and dashed lines indicate connections to other substations outside of the figure. The primary substation and the secondary substations powered by it are colored light blue or light yellow for primary substations 1 and 2 respectively. When a switch event occurs, for example due to cable failure between secondary substations D and E, power will be supplied from primary substation 1 rather than 2. This is indicated by the switch X color change from red to green. This leads to a temporary increase in apparent power measured at 1 and a decrease at 2, as illustrated with the apparent power measurements as a function of time at the bottom of the figure.
• Figure 2: A plot of the measured load ($S$) and the bottom-up load ($B$) as measured or estimated over the entire year for station 005. The S measurement is visualized in blue, and the bottom-up load is visualized in orange. The minimum and maximum load estimates are shown by the dashed lines. The load limit of the primary substation is shown by the dotted line. The green and blue areas indicate the unused and redundant capacity, these are fictitious and only shown for illustrative purposes.
• Figure 3: A plot of the measured load ($S$) and the bottom-up load ($B$) as measured or estimated over the entire year for station 010. The S measurement is visualized in blue, and the bottom-up load is visualized in orange. The minimum and maximum load estimates are shown by the black dashed lines. The load limit of the primary substation is shown by the dotted line. The true minimum and maximum load limits are shown by the red dashed lines. The capacity that would be incorrectly included in the estimate is shown by the opaque red boxes.
• Figure 4: Histogram of the length of the events and anomalies over all datasets. Note that the y-axis is log-scaled due to the frequency of short events. A year typically consists out of 35040 15-minute interval measurements.
• Figure 5: Plot of the one-sided threshold optimization procedure. The F$1.5$ score, on the y-axis, as a function of the threshold, on the x-axis, is shown for all four distinct segment length categories, as well as their average. The red vertical line indicates the selected threshold which maximizes the F$1.5$ score on the average.