|
Energy-Aware Opportunistic Charging and Energy
+ ?4 g. a7 c- B2 c- r! E6 |7 kDistribution for Sustainable Vehicular Edge and Fog
& c1 G, ?6 y: q: Y2 R# sNetworks
% K6 m) {' k3 d% M+ f$ Y3 @4 L6 u$ O- E5 ]0 C7 x, |& e5 I6 @
6 ^. r3 d* M1 {& \/ DThe fast-growing popularity of electric vehicles 1 e, K( v: [7 ?$ [" w" m1 A
(EVs) poses complex challenges for the existing power grid
0 T( A( r% z. Y' P* h7 Zinfrastructure to meet the high demands at peak charging hours. / g4 x+ x2 S5 U2 z$ v7 S( N! f
Discovering and transferring energy amongst EVs in mobile 9 F% D( w' k P; y6 T3 s2 ]
vehicular edges and fogs is expected to be an effective solution for 4 {9 F3 W$ \1 q* H
bringing energy closer to where the demand is and improving the
; P9 [+ V; ~" d$ ^$ r7 F6 zscalability and flexibility compared to traditional charging / m$ f6 n9 a5 @# X7 x1 Z
solutions. In this paper, we propose a fully-distributed energy
! M" ^% Y3 ^1 @$ c( u S/ ?aware opportunistic charging approach which enables distributed
$ R( h E, O8 b9 \4 B8 amulti-layer adaptive edge cloud platform for sustainable mobile 9 g+ f5 ]; v; t! }8 U- W
autonomous vehicular edges which host dynamic on-demand 8 g; }/ i, B( V. g0 f
virtual edge containers of on-demand services. We introduce a ( ~4 B( f8 L8 A! b k& X% [- f
novel Reinforcement Learning (Q-learning) based SmartCharge
; G, }( {, g# M4 q5 A5 p' J5 F, jalgorithm formulated as a finite Markov Decision Process. We
4 ?$ K9 U. M7 \/ u/ q# U( P: Ydefine multiple edge energy states, transitions and possible actions
8 ]( ^) u8 ^" i8 Z" ~of edge nodes in dynamic complex network environments which
/ [4 [$ H% K' F6 A, D0 {are adaptively resolved by multilayer real-time multidimensional
# _1 B0 v# `9 b3 b5 K$ O# Epredictive analytics. This allows SmartCharge edge nodes to more 9 N r- V. C& J2 z) X
accurately capture, predict and adapt to dynamic spatial-temporal
& s9 l& h+ {, C/ Qenergy supply and demand as well as mobility patterns when
- z( n- w9 p2 @* \energy peaks are expected. More specifically, SmartCharge edge
2 z$ U3 ]5 q* e6 P# ^+ @4 m5 @0 Xnodes are able to autonomously and collaboratively understand : Z' n2 J" a8 j/ J) K
when (how soon) and where the geo-temporal peaks are expected
! c) H) H- X- u% M7 P3 Uto happen, thus enable better local prediction and more accurate
' P/ m ?$ U* x5 [* _1 hglobal distribution of energy resources. We provide multi-criteria 9 o$ ?4 n! B. h7 y4 |" u6 w9 }9 \! ?
evaluation of SmartCharge against competitive protocols over
7 M9 j, l# g; J" A& ^6 h+ N/ p Dreal-world San Francisco Cab mobility traces and in the presence
3 D3 [! @; u( h; A8 Oof real-world users’ energy interest traces driven by Foursquare / _/ a# G/ _2 a( Z
San Francisco dataset. We show that SmartCharge successfully
) _- J* y r1 Z" }+ _( Y0 g6 _predicts and mitigates congestion in peak charging hours, reduces : Y; ~. A2 Q& L9 \0 g% f) C1 V
the waiting time between vehicles sending energy demand requests
9 P* T* l" R1 |, Kand being successfully charged as well as significantly reduces the
p9 e9 \" L4 W) f" ?/ \, ?total number of vehicles in need of energy.
! @8 D/ I' d' |( B9 Q" p7 n" E
, D9 v! S. i3 Q. p1 h8 w! ]+ u' Q/ V$ j' i) r: ^2 |4 S
4 c G3 D6 f5 I0 t7 n) y
! h& y- }( m4 b% o% l* W" ` |