An Online Learning Framework for Energy-Efficient Navigation of Electric Vehicles
Thought: energy efficiency is a problem for only EVs. What’s the special point?
Contributions:
1. a Bayesian approach to model the energy consumption at road segments
2. an online learning framework and investigate several exploration strategies such as Thompson Sampling and Upper Confidence Bound
3. extend our online learning framework to multi-agent setting
4. real-world experiments on Luxembourg SUMO Traffic dataset
Thompson Sampling (TS)
also called posterior sampling and probability matching
Upper Confidence Bound (UCB)
another approach used widely for exploration-exploitation trade-off
used by AlphaZero
energy consumption Model
model the non-negative edge weights by (conjugate) Log-Gaussian likelihood and prior distributions
this model is compared with the normal Gaussian model
Online learning
start from estimation of model parameters
solve the problem with multi-armed bandit problem
return the actual travel time to update parameters
Note:
The energy consumption model is useful
The dataset is potential to use.
点击咨询 