Enhancing Weather-Related Power Outage Prediction by Event Severity Classification

Abstract

The accuracy of machine learning-based power outage prediction models (OPMs) is sensitive to how well event severity is represented in their training datasets. Unbalanced or overly dispersed event severity can result in random errors in outage predictions and underestimation in severe events or overestimation in weak ones. To improve accuracy in the prediction of storm-caused power outages, we introduce a novel method called ''Conditioned OPM'' that divides an OPM training dataset into subsets of events representative of the predicted event's severity by calculating the quantile weight distance (QWD) between severe weather-related events in the dataset and the predicted event. Based on 102 storm events (including two hurricanes, Irene and Sandy), that have occurred since 2005 over Eversource Energy's Connecticut service territory, we quantified the weather differences among predicted events, which we classified into three groups of severity: low, moderate, and high. The Conditioned OPM creates a subset of the historical events based on their classified severity group and uses that subset as the training dataset to predict the power outages. The study shows that the accuracy of event severity classification was 0.76, and the mean absolute percentage error (MAPE) decreased by about 30%; this method was also tested on forecast events and exhibited a low (20%) MAPE.

Machine Learning-Based Error Modeling to Improve GPM IMERG Precipitation Product over the Brahmaputra River Basin

“…In terms of systematic and random error metrics, no significant differences were exhibited between two models (RF and NN). Generally, the machine learning-based model is expected not to capture very low and extremely high values successfully [14,93]. This is because the model accuracy is sensitive to sample size and the data representativeness in the training dataset [95,96].…”

“…A variable importance experiment was conducted by calculating the magnitude of the percentage increase in mean square error (%IncMSE) of the model [52,93]. Higher magnitudes of %IncMSE show higher importance of the input features for the error model.…”

The Effect of Lead-Time Weather Forecast Uncertainty on Outage Prediction Modeling

“…In 2020, Watson et al covered training data from five service territories (CT, EMA, WMA, NH, and UI), developed a clustering algorithm to summarize the different characteristics of the OPM grid cells, and used RF and BART models to develop an OPM demonstrating an MAPE of 58% to 63% and an NSE of 0.39 to 0.41 [39]. Because some service territories had fewer outage events in the training dataset than others, the inclusion in the training dataset of events from different service territories could benefit the ability of the OPM to predict future events [40,41]. Specifically, a regional OPM using data from different service territories and utilities allowed the study of information from the same or similar storm events in other areas.…”

Coordinated flexible energy and self‐healing management according to the multi‐agent system‐based restoration scheme in active distribution network

“…But, it is noted that this capability of EVs is not considered in different researches such as [12–32]. In the many types of research, the energy management and restoration process of ADN have been investigated, separately [12–40]. In other words, the coordination of zone agents and RESs, as well as FSs with the distribution system operator (DSO), has not been considered such as works in [12–40].…”

“…In contrast, this coordination can obtain the high advantage and optional to ADN from reliability, security, stability, flexibility, and operationally viewpoints. For example, it is forecasted that the load not supplied and the total number of switches operation at fault condition will be reduced if the restoration process is coordinated with FEM, in comparison with the cases including only the self‐healing method. Moreover, the restoration approach needs to calculate the network variables before fault occurrence, where research related to this approach such as [33–40] uses power flow analyses for this purpose. But, noted that the power flow method is not suitable for ADN that includes different sources and active loads; for this reason, it needs optimal operation formwork to obtain the network variables.…”