MDN-MPC: Learning EV Charging Behavior With Mixture Density Networks for Controlling PV Charging Stations
Abstract
Limited prior knowledge and uncertainty of electric vehicle (EV) charging behavior present significant challenges for effective EV charging control. This study presents a novel framework for joint prediction and control of EV charging by integrating mixture density networks (MDNs) with model predictive control (MPC). The MDN-MPC framework uses MDNs to stochastically model EV charging behavior as a set of probability distributions. These models are learned from historical EV transaction data using an autoregressive distribution estimation (ADE) approach and are integrated into a closed-loop MPC controller. The proposed control framework is evaluated through a case study at the University of Twente, Netherlands, demonstrating its capability to manage uncertainties in system dynamics, PV generation, and EV charging behavior, while achieving user satisfaction and operational profitability.
Fields of Study & Subjects
Electric Vehicles and Infrastructure
Smart Grid Energy Management
Advanced Battery Technologies Research
Publication details
- Journal / source
- IEEE Transactions on Transportation Electrification
- Publisher
- Institute of Electrical and Electronics Engineers (IEEE)
- Volume
- 12
- Issue
- 2
- Pages
- 3525–3535
- Publication date
- 19 January 2026
- Language
- EN