Structural statistical learning of heterogeneous preferences for smart energy choices with a case study on coordinated electric vehicle charging

Cornell University, Ithaca NY

Investigators

Abstract

This project focuses on creating simple and efficient computer algorithms capable of learning from real-world data to develop user-friendly tools to guide optimal choices about electric charging technologies. The significance lies in users adapting to increasingly complex energy environments, adopting more efficient options and behaviors, and gaining more control over energy use. The ultimate objective is to simplify decision-making, making it easier for users to manage energy consumption in optimal ways. Advanced statistical techniques are employed, utilizing highly efficient computer algorithms that will ensure speed, flexibility, interpretability, and real-world policy relevance. The proposed system adapts and learns charging preferences in real-time, providing energy management recommendations in the context of coordinated electric charging while considering factors such as demand, pricing, and grid stability. The methodical modular Markov chain fast and scalable sampler of the proposed incentive-compatible recommender system is designed by integrating: an expansion of Ultimate Pólya-Gamma data augmentation for multi-index choice models to create conjugacy where it does not exist, amortized and non-factorized variational inference for efficiency, Choquet aggregation for complex nonparametric tradeoffs across service features (relaxing linear compensatory behavior), Bayes endogeneity controls, Bayesian optimization for tuning hyperparameters and for integration with a system equilibrium algorithm, and flexible semiparametric representation of heterogeneity in preferences and motives between and within strategic agents. The research team implements the project through an actual residential program for coordinated scheduling of electric charging, such as avoiding in-force price endogeneity and constraints via self-selection price discrimination through discrete targeted bundles. Smart technology is employed to gather information, enabling the provision of personalized suggestions to users. The impact of the project extends beyond individual assistance, contributing to the overall efficiency of energy systems and providing valuable tools for energy-related organizations, utilities, and policymakers. This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.