Deep Learning-Based Weather-Related Power Outage Prediction with Socio-Economic and Power Infrastructure Data

TL;DR

The paper tackles hourly outage probability forecasting at the census-tract level under weather conditions by fusing ASOS weather features, weather-station proximity, socio-economic distributions, and power-infrastructure counts. It introduces two MLP architectures—conditional and unconditional—trained with either an exponential loss ($\beta=20$) or a weighted cross-entropy loss ($w=500$), and evaluates them on data from 1,102 tracts within the DTE service area. A key finding is that socio-economic and infrastructure features meaningfully reduce prediction error, with the conditional model excelling under cross-entropy and the unconditional model performing best with exponential loss; ablation confirms the value of additional contextual features. The work supports targeted outage response and resilience planning, and points to future directions such as incorporating LSTM-based temporal modeling and expanding analyses to larger geographic areas to address data imbalance.

Abstract

This paper presents a deep learning-based approach for hourly power outage probability prediction within census tracts encompassing a utility company's service territory. Two distinct deep learning models, conditional Multi-Layer Perceptron (MLP) and unconditional MLP, were developed to forecast power outage probabilities, leveraging a rich array of input features gathered from publicly available sources including weather data, weather station locations, power infrastructure maps, socio-economic and demographic statistics, and power outage records. Given a one-hour-ahead weather forecast, the models predict the power outage probability for each census tract, taking into account both the weather prediction and the location's characteristics. The deep learning models employed different loss functions to optimize prediction performance. Our experimental results underscore the significance of socio-economic factors in enhancing the accuracy of power outage predictions at the census tract level.

Figures (7)

• Figure 1: The average power outage duration in hours per capita for each census tract.
• Figure 2: Histogram of the causes of the power outage events.
• Figure 3: Power infrastructure components in the studied area.
• Figure 4: Structure of the conditional model, d_out=1 for the exponential loss and d_out=2 for the weighed cross entropy loss
• Figure 5: Structure of the unconditional model, d_out=1 for the exponential loss and d_out=2 for the weighed cross entropy loss