Learning of personalized strategies for the operation of rechargeable hybrid electric vehicles in a V2G context
Karol Lina Lopez
Christian Gagné (Supervisor)
Problem: With the advent of global warming and the depletion of certain non renewable resources, the movement towards vehicles based on electrical energy seems to be inevitable. This transition should have significant positive ecological impacts, given the resulting reduction in fuel consumption, but will also involve considerable economic costs, for citizens who will purchase these new types of vehicles, as well as for electrical energy producers who will be faced with a considerable increase in demand.
Motivation: The vehicle-to-grid (V2G) concept consists in exploiting electric vehicles, which will become significant electrical energy accumulators, so as to offset peak power demands. Such a system could favour the rapid acceptance of electric vehicles, by reducing the increase in demand on the electric grid, while allowing the electric vehicle owner to receive a return proportional to the energy contribution made during peak power demands. This could represent a significant economic incentive.
Approach: The present research project is devoted to the development of efficient strategies for the operation of electric vehicles in a V2G context. The objective will be to develop a decentralized model, in each electric vehicle, which aims to optimize decisions involving the inversion of electric flux between the standard mode (recharging the accumulator) and the inverse mode (where the accumulator provides energy to the electric network), depending on the needs of the electric grid. The decision model will be specific to each vehicle, according to observations made on the vehicle use history and the tendency for users to take risks (such as receiving greater returns while risking the complete discharging of the battery during unexpectedly long drives).

Since the physical infrastructure is not available at the moment, a model for the simulation of PHEV, Plugin Hybrid Electric Vehicles, will have to be developed, with models for the charging and use of electric energy by vehicles, and for the price of electricity as a function of the demand, as well as a model of several use patterns at varying time scales. This simulation model will then be used to test various artificial intelligence techniques such as reinforcement learning, multiagent systems, game theory, so as to establish automatic and personalized strategies to determine the optimum moments where each car should make the transition from a recharging mode to an electric network support mode, so as to maximize certain performance criteria.

Last modification: Nov 28 2011 10:37AM by kllop


©2002-. Computer Vision and Systems Laboratory. All rights reserved