Predicting Public Health Impacts of Electricity Usage

Yejia Liu, Zhifeng Wu, Pengfei Li, Shaolei Ren

Keywords: Public Health Impact, Electricity Generation, Health-informed Computing

TL;DR: We build a domain-specific AI model to predict health impacts of electricity usage.

Abstract: Electricity consumption impacts public health due to pollutant emissions from fossil fuel power plants. While stricter regulations have reduced emissions, fossil fuels remain a dominant energy source, necessitating advanced methods to quantify and mitigate these societal health effects. Here, we present a domain specific AI model, $\texttt{HealthPredictor}$, an end-to-end pipeline that links electricity usage to public health outcomes. Our system integrates three key components: a fuel mix predictor that forecasts energy source contributions, an air quality converter that models pollutant emissions and dispersion, and a health impact assessor that translates environmental changes into monetary health costs. We demonstrate that our health-driven optimization approach achieves significantly lower prediction errors compared to fuel mix-driven methods across multiple U.S. regions. Through a case study on electric vehicle charging schedules, we show the public health benefit of our approach in providing actionable insights about electricity usage for users. This work thus demonstrates how AI model can be explicitly designed to optimize for public health and societal well-being. Our datasets and code will be released upon publication of our paper.

Submission Number: 125