Using molecular simulation for adsorbent screening is computationally expensive and thus prohibitive to materials discovery. Machine learning (ML) algorithms trained on fundamental material properties can potentially provide quick accurate methods purposes. Prior efforts have focused structural descriptors use with ML. In this work, the of chemical descriptors, in addition was introduced adsorption analysis. Evaluation coupled various ML algorithms, including decision tree, Poisson...

This paper introduces new developments in an outage prediction model (OPM) for electric distribution network the Northeastern United States and assesses their significance to OPM performance. The uses regression tree models fed by numerical weather outputs, spatially distributed information on soil, vegetation, utility assets, historical power data forecast number spatial of outages across grid. New modules introduced hereby consist 1) a storm classifier based variables; 2) multimodel...

Current human population growth along Earth's coasts is on a collision path with anticipated consequences of increasing natural and anthropogenic induced coastal hazards. Using recently-available ambient, dasymetric data, we developed methods to estimate annual continental global populations from (2000-2018) measured horizontally the shoreline inward. We found: (1) large concentrations in relatively small bands regions coast (~ 2 billion within 50 km ~ 1 10 km); (2) higher rates than inland...

This article compares two nonparametric tree‐based models, quantile regression forests (QRF) and Bayesian additive trees (BART), for predicting storm outages on an electric distribution network in Connecticut, USA. We evaluated point estimates prediction intervals of outage predictions both models using high‐resolution weather, infrastructure, land use data 89 events (including hurricanes, blizzards, thunderstorms). found that spatially BART predicted more accurate than QRF. However, QRF...

Natural and anthropogenic hazards are frequently responsible for disaster events, leading to damaged physical infrastructure, which can result in loss of electrical power affected locations. Remotely-sensed, nighttime satellite imagery from the Suomi National Polar-orbiting Partnership (Suomi-NPP) Visible Infrared Imaging Radiometer Suite (VIIRS) Day/Night Band (DNB) monitor outages disaster-affected areas through identification missing city lights. When combined with locally-relevant...

A growing number of electricity utilities use machine learning-based outage prediction models (OPMs) to predict the impact storms on their networks for sustainable management. The accuracy OPM predictions is sensitive sample size and event severity representativeness in training dataset, extent which has not yet been quantified. This study devised a randomized out-of-sample validation experiment quantify an OPM’s uncertainty different sizes representativeness. showed random error decreasing...

Abstract Hurricane Sandy (2012, referred to as Current Sandy) was among the most devastating storms impact Connecticut’s overhead electric distribution network, resulting in over 15 000 outage locations that affected more than 500 customers. In this paper, severity of tree-caused outages Connecticut is estimated under future-climate simulations, each exhibiting strengthened winds and heavier rain accumulation study area from large-scale thermodynamic changes atmosphere track year ~2100...

Thunderstorms are complex weather phenomena that cause substantial power outages in a short period. This makes thunderstorm outage prediction challenging using eventwise models (OPMs), which summarize the storm dynamics over entire course of into limited number parameters. We developed new, temporally sensitive framework designed for to learn hourly thunderstorm-caused directly from forecasts. Validation several built on this hour-by-hour and comparison with baseline model show abilities...

Nighttime light (NTL) imagery has been widely used to predict economic impacts, electricity consumption, and pollution. In this research, we evaluated nighttime the county-level population in United States using deep learning methods. Rather than condensing brightness information into a single statistic (e.g., mean annual per county), research aims exploit temporal nature of NTL imagery. Monthly composites VIIRS were analyzed for each county create quantiles brightness. These provided as an...

In the United States, weather-related power outages cost economy tens of billions annually, and there has been an upward trend in billion-dollar disasters over last two decades. Thus, it is growing importance to be able predict understand local resilience. However, many outage prediction models rely on utility infrastructure data, which can difficult obtain when a study domain covers territories. This demonstrates gradient-boosting machine-learning driven by utility-agnostic non-proprietary...

Extreme weather can cause severe damage and widespread power outages across utility service areas. The restoration process be long costly emergency managers may have limited computational resources to optimize the process. This study takes an agent based modeling (ABM) approach storm recovery in Connecticut. ABM is able replicate past recoveries test future case scenarios. We found that parameters such as number of outages, repair time range crews working substantially impact estimated...

The outage prediction model (OPM) is a weather-related machine learning-based power model, which has been developed at the University of Connecticut for many years and recently grown to cover three states five utility service territories. This large heterogeneous domain supported by dataset hundreds storm events. presents opportunity investigate effect spatial organisation training structure on performance, identify potential weaknesses in modelling approach, evaluate generalisability...

Abstract Electric distribution utilities have an obligation to inform the public and government regulators about when they expect complete service restoration after a major storm. In this study, we explore methods for calculating estimated time of (ETR) from weather impacts, defined as it will take 99.5% customers be restored. Actual data Storm Irene (2011), October Nor’easter (2011) Hurricane Sandy (2012) within Eversource Energy-Connecticut territory were used calibrate test methods;...

Emergency managers at electric distribution utilities benefit from tools that help estimate the impacts of weather on their infrastructure networks when communicating with customers and regulators. In this paper, we adopted deep learning methods - Long Short-Term Memory (LSTM) Multilayer Perceptron (MLP) to model a time series county-level customer outages for across ten counties in New York State. We utilized hourly data (i.e. wind, temperature, precipitation, gust, wind direction) State...

Power outage restoration following extreme storms is a complicated process that couples engineering processes and human decisions. Emergency managers typically rely on past experiences have limited access to computer simulations aid in decision-making. Climate scientists predict although hurricane frequency may decrease, the intensity of increase. Increased damage from hurricanes will result new challenges emergency not experience solving. Our study uses agent-based modeling (ABM) determine...

More and more frequently, electric utility emergency response personnel are required to manage the impact of severe weather events on distribution networks. In US, economic losses associated with extreme estimated between $20 billion $55 annually. Spatiotemporal modeling customer outages from data can mitigate personal adverse by reducing downtimes increasing confidence in providers during a power outage event. this paper, we consider problem forecasting integrating distributed temporal...

This study aims to develop models for predicting hourly energy demand in the State of Connecticut, USA from 2011 2021 using machine learning algorithms inputted with airport weather stations' data Automated Surface Observing System (ASOS), ISO New England (ISO-NE). We built and evaluated nine different model experiments each algorithm hour day addressing patterns, variations between workdays weekends, COVID-19 impacts. Error metrics analysis results highlighted GBR demonstrated better...