2013 4th IEEE PES Innovative Smart Grid Technologies Europe (ISGT Europe), October 6-9, Copenhagen

The Effect of Mobility Forecasts for Stochastic

Charge Scheduling of Aggregated PEV
with Capacity Pricing Based on DSO’s Long Run Marginal Cost

Ilan Momber Tomás Gómez

Institute for Research in Technology (IIT) Comisión Nacional Energía (CNE)
Comillas Uni, C. Sta. Cruz de Marcenado 26 C. Alcalá 47, 28014 Madrid
28015 Madrid, Spain Spain
[email protected] [email protected]

Abstract—Plug-in Electric Vehicles (PEV) integration into supplier or marketer, and a large consumer with potential on-
innovative smart grids has been studied widely. Recently, special site generation [3]. If feeding back energy to the system is
attention is paid to the role of the aggregation agent, which might considered, too, the PEV retail problem shares characteristics
be responsible of controlling the charge schedules to its own
and the system’s benefit. Furthermore, an increasing awareness with wind power producers [4], conventional power producers
of the stochasticity involved in PEV charging, reflected in a under resource unavailability [5], [6], as well as with energy
number of recent publications, can be observed. Among others, storage system operators [7].
the PEV energy retail problem with interactions in day-ahead and 1) Stochastic PEV Charge Scheduling: There seems to be
balancing markets has been formulated from the aggregator’s an increasing awareness of the stochasticity involved in PEV
perspective, taking into account location dependent network
tariffs in the form of capacity prices for active power. However,
charging, which is reflected in a number of recent publica-
a numerical quantification of the benefit from accounting for tions. [8] assesses distribution system impact by considering
stochasticity has yet to be carried out. random processes in PEV charging times and battery states at
Therefore, this paper uses standard methodology to calculate starting time, by monte-carlo sampling. In an attempt to handle
stochastic programming quality metrics such as the value of the computational complexity [9] employs approximate dynamic
stochastic solution (VSS) and the expected value of private infor-
stochastic programming with receding horizon.
mation (EVPI) for an established PEV energy retail aggregator
model. Applied to a real medium voltage system with urban 2) Market Design for PEV Participation: Some authors
characteristics and realistic spatial PEV mobility, different levels added to the state of art by proposing market design changes,
of mobility forecasts are included as information at the first-stage such as alleviating symmetry in bidding upward and downward
here and now decisions. capacity, minimum bid sizes or availability requirements in
Index Terms—Plug-in Electric Vehicle (PEV) Aggregator, Op- ancillary service markets for capacity [10], [11], to enable
timal PEV Charging Schedules, Stochastic Programming Quality uni- or bi-directional PEV participation for the stabilization of
Metrics the system operation. In a review of control power in Europe
with particular focus on Germany, [12] proposes market design
measures regarding daily auctions, hourly intervals and condi-
I. I NTRODUCTION
tional bids, which would partly help demand side participation
ESEARCH on plug-in electric vehicles has received sub- from PEV aggregators as well.
R stantial attention across many disciplines. This is because
economic and environmental policy making has identified B. Main Contributions
the electrification of transportation systems as an option to Aligned with and building upon recent works, this paper
reduce greenhouse gas emissions, improve energy efficiency, formulates the PEV energy retail problem with interactions
enhance electric power system (EPS) operation as well as in day-ahead and balancing markets from the aggregator’s
facilitate renewable energy integration. Hence, goals for Plug- perspective, taking into account location dependent network
in Electric Vehicles (PEV) penetration levels have been set use-of-system (UoS) tariffs in the form of capacity prices for
by many governments. Additionally, the advent of a massive active power. Applied to a real medium voltage system with
electrification of vehicle propulsion may not only present an urban characteristics and realistic spatial PEV mobility, profit
opportunity for a more sustainable energy consumption model optimal charging schedules for the aggregator are found. In
in modern societies, it is also likely to have a marked impact on addition, this article specifically extends technical literature on
contractual relationships between EPS agents, some of which modeling methodologies for stochastic PEV charge scheduling
act on electricity markets. and its implications for modern EPSs. The extension is used
to calculate standard stochastic programming metrics for the
given case. At the same tame this study quantifies the effect
A. Background
of different mobility forecasts for aggregated PEV charge
As discussed in [1], the operational challenge of a PEV scheduling.
Aggregator presents a combination of the classic problems [4] analyzed the effect of certainty gained when trading in
of an electricity retailer [2], sometimes also referred to as shorter term adjustment markets on wind power producers. In

978-1-4799-2984-9/13/$31.00 ©2013 IEEE

 2

principle, the closer to real time the producer can participate in

markets, the better the forecast (smaller forecasting error) and
∀v, h, n, ω : un,ω ≥ eRT, Y
v,h,n,ω , (2)
therefore the more profitable because imbalance settlements
are reduced. ∀v, h, n, ω : u0n,ω ≥ eRT, Z
v,h,n,ω , (3)
In this paper however, a different approach is used to X Xh i
∀h, ω : eRT, Y RT, Z
v,h,n,ω − ev,h,n,ω = (4)
to assess certainty gain. By including limited, indicative
v∈V n∈N
mobility forecasts as information at the first-stage here and    
now decisions, impacts on typical stochastic programming eD,
h,ω
Y
− eD,
h,ω
Z
+ eB+ B−
h,ω − eh,ω ,
quality metrics can be observed. The standard methodology
∀v, h, n, ω : eRT, Z RT, Y n
v,h,n,ω +ev,h,n,ω ≤[αv,h,ω ·A v,h,ω ] P n (5)
of calculating the value of the stochastic solution (VSS) and X h 
the expected value of private information (EVPI) is followed ∀v, h, ω : eSOC SOC
v,h,ω = ev,h−1,ω + eRT, Y
v,h,n,ω η Y
v (6)
and results presented.  n∈N
i
− eRT, Z Z
v,h,n,ω /ηv − ρv,h,ω ,
C. Paper Organization ∀v, h, ω : eSOC
v,h,ω ≤ E v , (7)
The subsequent sections of this paper are organized as ∀v : eSOC
v,0 = ιSOC
v , eSOC
v,|H| = φSOC
v , (8)
follows: Section II provides the mathematical formulation where (2) and (3) refer to the maximum network capacity
of the proposed scheduling problem. Section III defines an used, (4) is the market side energy balance, (5) is the charge
illustrative case study in terms of network and market data, as availability constraint, (6) is the client side, vehicle based,
well as mobility scenarios. Section IV presents the numerical inter-temporal energy state-of-charge balance (SOC), (7) is the
results of the case study and finally, a concluding discussion is upper state-of-charge limit and (8) set the initial and final state
provided in section V. This paper must be seen as a companion of charge conditions. In [1] all constraints are derived in detail.
of [1]. Hence, to avoid redundancy, the detailed nomenclature
to the symbols used throughout this paper is provided in the
appendix of [1]. B. Information Constraints
The above formulation represents the information uncon-
strained problem. Now different degrees of knowledge regard-
II. P ROBLEM F ORMULATION ing second stage realizations can be included. The common
constraint alternatives for non-anticipativity (9) and offer-
The here used problem formulation of a PEV aggregator
curves (10) are:
interacting with electricity markets as well as incorporating
network prices was first presented in [1]. To avoid repetition ∀h, ∀ω, ω 0 : eD, Y D, Y D, Z D, Z
h,ω =eh,ω 0 , eh,ω =eh,ω 0 , (9)
only algebraic expressions are shown and not every symbol is
∀h, ∀ω, ω 0 : λD D
eD, Z D, Z D, Y D, Y


h,ω ≤λh,ω 0 , h,ω ≤eh,ω 0 , eh,ω 0 ≤eh,ω . (10)
re-introduced. For a detailed explanation of all involved terms,
please refer to the original publication. Note that offer-curves (10) are a relaxation of non-
anticipativity constraints (9). Hence when including both, (10)
is overruled. Besides these common constraints, this paper
A. Stochastic Self-Scheduling of a PEV Aggregator specifically analyzes the effects of including either one of the
following constraints, all pertaining to different aspects of the
The self-scheduling problem in (1) seeks the risk neutral
PEV mobility behavior.
maximization of expected profits from scenario weighted day-
ahead transactions ΠD B
 
ω , imbalance settlements Πω , client side
∀h, ∀ω, ω 0 :
P D P D
τv,ω = τv,ω0 , (11)
retail to PEV Πω and network use-of-system prices κU
C
ω
oS
: v∈V v∈V

eD, Y =eD, Y D, Z D, Z
h,ω 0 , eh,ω =eh,ω 0 .
  h,ω 
Maximize E ΠTω otal 0
∀h, ∀ω, ω :
P
αv,h,ω =
P
αv,h,ω ,
0 (12)
 
P  D  v∈V v∈V
B C U oS
P
= πω Πh,ω + Πh,ω + Πh,ω − κω , (1)
ω∈Ω h∈H eD,
h,ω
Y
=eD, Y
h,ω 0 , e D, Z D, Z
h,ω =eh,ω 0 .
 
0 P P
where h  i ∀h, ∀ω, ω : ρv,h,ω = ρv,h,ω0 , (13)
ΠDh,ω = eD, Z D, Y D
h,ω τ − eh,ω /τ λh,ω ,
v∈V v∈V
h
B− − B+ +
i eD, Y D, Y D, Z D, Z
h,ω =eh,ω 0 , eh,ω =eh,ω 0 .
ΠBh,ω = e λ
h.ω h,ω − e h,ω h,ω ,
λ
By including one of (11)-(13) at a time, different levels of
h
i
eRT, Y
RT, Z
ΠC Y Y
−ev,h,n,ω γ Z +ϑZ
P P
h,ω = v,h,n,ω γ +ϑ information are available to the decision making at the first
v∈VP n∈N
and κω = n∈N un,ω − u0n,ω CnU oS .


stage of day-ahead market involvement. Equation (11) would
allow these decisions to be taken with the knowledge of
This objective function is subject to the following con- a coarse indicative forecast in the form of τv,ω . With the
straints: help of this forecast the program can distinguish between
 3

scenarios of high and low mobility, measured by how many 1.5

h − ̺ h + 1[p.u.]
vehicles travel in the given setting. Equation (12) would make
knowledge of a detailed indicative forecast available, in the 1
B1

−
B2
form of αv,h,ω . Thus it would allow to distinguish between

̺+
scenarios of high and low unavailability, measured by the 0.5
0 4 8 12 Hour [h] 16 20 24
connection and disconnection of each vehicle in the respective
hours. And finally, (13) would permit a perfect aggregated Figure 1. Two Sub-Scenarios for Time Dependent Balancing Prices as Ratios
mobility forecast in the form of ρv,h,ω . With the help of this to Day-Ahead Market Prices. Scenario B1: Upward Balancing Prices increase
with hour h of the day. Scenario B2: Vice Versa
forecast the program could distinguish between scenarios of
high and low unavailability and consumption, measured by
the SOC reductions during travel. The disconnection of each 2) Client Side Prices: On the client side, the sales price
vehicle in the same hour would be implied automatically. This component attributable to wholesale, excluding network tar-
would leave the program only with second stage uncertainty iff components, is γ Y = 0.065 e/kWh. The per energy
regarding the balancing market prices. unit basis component representing the network use-of-system
prices ϑY = γ Y = 0.065 e/kWh is set such that both
components make up the same amount in the final customer
C. Quality Metrics bill. Discharging is turned off in the uni-directional case, hence
γ Z becomes irrelevant.
To motivate the use of the stochastic programming frame-
work [13], pp. 48-57, including various sources of uncertainty
[1], the following quality metrics are studied. Let zsto be the B. PEV Data - Unavailability Scenarios
value of the objective function with binding non-anticipativity Suppose a small fleet of 5 vehicles, which are made up
constraints, then the following holds: of Plug-in Hybrid Electric Vehicles (PHEVs) only, whose
Expected Value of Perfect Information (EVPI) - how much characteristics are mainly based on [11], [14]. The PEV
would one be willing to spend for acquiring a perfect forecast: mobility is depicted in Tab. I, II, III and IV, which show τv,ω ,
zEVPI = zsto −zpf , where zpf (perfect forecast) stands for the as well as αv,h,ω overlaying ρv,h,ω (some zero columns are
objective function value with fully relaxed non-anticipativity not shown for the sake of readability). Mobility sub-scenarios
constraints. M1, M2, M3 and M4 are equiprobable (πM 1..4 = 0.25). Initial
Value of the Stochastic Solution (VSS) - how much is as well as target SOCs are, in both mobility sub-scenarios at
it worth to know the distribution of the stochastic inputs: ιSOC
v = 8 kWh and φSOC v = 8 kWh, respectively.
zVSS = zsto −zdet , where zdet is the objective function value
when first-stage decision variables are fixed at the optimal IV. R ESULTS
solution of the problem in which stochastic inputs are replaced The objective of this section is to assess the effects of differ-
by expected values. ent mobility forecasts on typical quality metrics of stochastic
programming. Therefore, we first describe the used simulation
procedure, i.e. the order and nature of different optimization
runs.1

III. C ASE S TUDY D ESCRIPTION A. Simulation Procedure

In this section the proposed model is applied to a numerical With the model defined in section II, the simulation pro-
example, which is very similar to the one found in the cedure consists of six runs with the following characteristics.
companion [1]. If not explicitly specified otherwise, case study The unconstrained base case does not take into account any of
settings from [1] are taken. A stylized uni-directional, i.e. load the information restrictions for stochastic programming. The
only, case with a 24 hour time horizon, is defined in terms of five remaining runs are as follows:
network data, day-ahead and balancing market price scenarios, 1 - Full Non-Anticipativity: all non-anticipativity constraints
as well as mobility (unavailability scenarios and locations), in are fully binding. Hence, in each hour the program cannot
the subsections underneath. differentiate between different scenarios, such that only one
optimal energy quantity for the day-ahead market schedule is
found. Only information constraint (9) is binding.
A. Stylized Price Data 2 - Full Offer-curve Constraints: the non-anticipativity con-
straints are relaxed to find multiple day-ahead pairs of price
1) Balancing Market Price Scenarios : Presented in Fig. 1, and optimal energy quantity. Only information constraint (10)
−
the imbalance price settlements, %+ h and %h , express the ratio
1 All calculations were performed running M AT L AB © for handling input
of real-time price to each hourly day-ahead market price. There
and output data. For comparison, the optimization problem was formulated
exist different probabilities to have increasing (πB1 = 0.6) or
and solved in GAMS© B UILD 24.0.1 employing the CPLEX™ 12.5.0
decreasing(πB2 = 0.4) positive imbalance, and the amount of solver on a 64-bit MS Windows© 7 machine with 8.00 GB RAM and an
imbalance represented in the prices are time dependent. Intel© Core™ i7-3770 CPU clocked at 3.4 GHz.
 4

Table I Table III

M OBILITY S UB -S CENARIO 1 - αv,h,M 1 :  , ρv,h,M 1 : |1.52| [ K W H ], τv.ω M OBILITY S UB -S CENARIO 3 - αv,h,M 3 :  , ρv,h,M 3 : |1.52| [ K W H ],τv.ω
7 8 9 10 11 12 13 14 15 16 17 18 τv,ω 7 8 9 10 11 12 13 14 15 16 17 18 τv,ω
1 1.52 1.86 3.95 2.46 2.45 1 1 0.52 0.86 3.95 0.46 0.45 1

Vehicles
Vehicles

2 3.95 1.44 3.95 3.95 0.1 1.68 2.26 1 2 3.95 0.44 3.95 3.95 0.1 0.68 0.26 1
3 0.95 2.45 2.14 2.86 1 3 0.95 0.45 0.14 0.86 1
4 2.26 2.45 2.68 2.22 1 4 0.26 0.45 0.68 0.22 1
5 2.14 2.95 2.14 2.22 1 5 0.14 0.95 0.14 0.22 1
Hours h ∈ H [h] Hours h ∈ H [h]

Table II Table IV
M OBILITY S UB -S CENARIO 2 - αv,h,M 2 :  ,ρv,h,M 2 : |0.95| [ K W H ],τv.ω M OBILITY S UB -S CENARIO 4 - αv,h,M 4 :  ,ρv,h,M 4 : |0.95| [ K W H ],τv.ω
7 8 9 10 11 12 13 14 15 16 17 18 τv,ω 7 8 9 10 11 12 13 14 15 16 17 18 τv,ω

Vehicles
Vehicles

1 0 1 0
2 2.95 2.68 2.14 2.14 1 2 0.95 0.68 0.14 0.14 1
3,4,5 0 3,4,5 0
Hours h ∈ H [h] Hours h ∈ H [h]
Table V
C ASE S TUDY P ROBLEM S UMMARY
Obj. Fn. Value CPU Time Total Iterations Equations Non-Zeros Real Variables EVPI EVPI [%] VSS VSS [%]
Base Run 0.59 80 15 0.343 1866 152 137 554 041 106 536 -
Run 1 0.52 37 77 0.359 2036 178 633 607 033 106 536 0.07 42 38 14.17 0.03 268 6.24
Run 2 0.52 37 78 0.328 2047 169 417 588 601 106 536 0.07 42 37 14.17 0.03 269 6.24
Run 3 0.59 73 25 0.344 2119 164 809 579 385 106 536 0.00 06 90 0.12 0.10 267 17.19
Run 4 0.53 25 84 0.328 2018 172 297 594 361 106 536 0.06 54 31 12.29 0.04 149 7.79
Run 5 0.53 28 82 0.343 2039 169 705 589 177 106 536 0.06 51 33 12.22 0.04 179 7.84

is binding. the base run optimum of the objective function - amounts to

3 - Non-Anticipativity with Coarse Indicative Forecast: similar a remarkable 14.17%. The VSS is similarly significant with
to Run 1, but with knowledge of τv,ω . Only information 0.03 268 and 6.24%, respectively.
constraint (11) is binding. These results give rise to the main finding of this study:
4 - Non-Anticipativity with Detailed Indicative Forecast: sim- both quality metrics indicate a necessity to use stochastic
ilar to Run 3, but with aggregated information about αv,h,ω . programming for the given problem. This suggests that with
Only information constraint (12) is binding. the combined uncertainty of market prices and fleet mobil-
5 - Non-Anticipativity only Regarding Imbalance Prices: sim- ity, finding optimal day-ahead market schedules requires the
ilar to Run 4, but with perfect aggregated mobility forecast. stochastic programming framework.
Only information constraint (13) is binding. 2) Full Offer-Curve Constraints: Run 2 shows very little
As a problem summary, Tab. V gives an overview of the case difference to the previous one. It appears that even though
study results and shows the size of the problem in the different problem size decreases eminently, the solution shows almost
runs. On a high level valid for all runs, it can be observed no improvements at an objective function value of 0.52 37 78.
that the problem is solved in very little time, i.e. roughly one Hence, EVPI and VSS are the same as in Run 1. The reason
third of a second. Solver iterations do not significantly surpass for this is assumed to lie in the structure of the day-ahead price
the 2k mark. Furthermore, the linear characteristic of the scenarios and the aggregated availability and consumption of
formulation accords with the absence of any binary and integer the vehicles. Since all three day-ahead sub-scenarios follow
variables. However, since wait-and-see decisions for state-of the same price profile only at different levels, the cost signals
charge as well as charging and discharging are represented on giving preference to one hour over another are the same.
a detailed vehicle basis, the amount of real variables, with ca. Also, each day-ahead market sub-scenario has the same sub-
100k, is remarkably high for such a small case study. Adding branches of mobility and balancing uncertainty, thus making
to the problem of dimensionality is the inclusion of distribution it unprofitable to distinguish between price levels.
network use-of-system prices for each node. However, the 3) Non-Anticipativity with Coarse Indicative Forecast: Run
linear nature with its polynomial complexity spurs hope that 3 provides the most unintuitive finding. Even though only
bigger instances of this problem can be solved in reasonable enough information is provided to differentiate between sub-
time and effort. scenarios of distinguishing mobility in terms of how many
vehicles travel, a close to unconstrained perfect-information
The base run sets the benchmark to the result of which all optimum is found at 0.59 73 25. Thus the EVPI is close to
following runs are measured against. The optimal objective zero with 0.00 06 90 or 0.12%. This means that in the given set
function value is found at 0.59 80 15. Since it is the least up it is very valuable to know whether there is more or less
constraint of all runs, it exhibits the smallest problem instance. travel. This is even true although for instance consumption
1) Full Non-Anticipativity: Run 1 comes along with the values still differ very much among scenarios with same
biggest problem instance (178 633 equations and 607 033 non- travel. Nevertheless, the VSS of 0.10 267 or 17.19% still
zeros) compared to all others. It stands for the typical two- clearly indicates that scheduling according to expected values
stage stochastic problem formulation in which no information of involved stochastic processes would lead to a significantly
about future outcomes are available at the first stage. Because worse outcome.
the feasibly region is so constraint, it is not surprising that it 4) Non-Anticipativity with Detailed Indicative Forecast:
shows the lowest objective function value with 0.52 37 77. It Run 4 takes second rank in terms of constraints and non-zeros.
has hence an EVPI of 0.07 42 38, which - put in relation to Accordingly the objective function value amounts to 0.53 25
 5

84, which implies an EVPI of 0.06 54 31, 12.29% respectively. R EFERENCES

Similar to the previous run, here the program is allowed to [1] I. Momber, T. Gómez, and L. Söder, “PEV fleet scheduling with
find a schedule that also alternates with the scenario counter. electricity market and grid signals”, in International Conference on
the European Energy Market (EEM), vol. 10th, Stockholm, Sweden:
Thus it is possible to distinguish between scenarios of different IEEE, May 2013.
unavailability, even though the amount of consumption that [2] M. Carrion, J. Arroyo, and A. Conejo, “A bilevel stochastic program-
lies behind this disconnection is unknown to the program. ming approach for retailer futures market trading”, Power Systems,
IEEE Transactions on, vol. 24, no. 3, pp. 1446 –1456, Aug. 2009.
The VSS of 0.04 149 or 7.79% still suggests that scheduling [3] M. Carrion, A. Philpott, A. Conejo, and J. Arroyo, “A stochastic
according to expected values of involved stochastic processes programming approach to electric energy procurement for large con-
would lead to a significantly worse outcome. sumers”, Power Systems, IEEE Transactions on, vol. 22, no. 2, pp. 744
–754, May 2007.
5) Non-Anticipativity only Regarding Imbalance Prices: [4] J. Morales, A. Conejo, and J. Perez-Ruiz, “Short-term trading for a
Finally, Run 5 resolves the uncertainty regarding mobility, wind power producer”, Power Systems, IEEE Transactions on, vol.
which leads to a slightly better result than the previous run 25, no. 1, pp. 554 –564, Feb. 2010.
[5] S. Pineda-Morente, “Medium-term electricity trading for risk-averse
with the objective function value at optimum 0.53 28 82, power producers via stochastic programming”, Phd. Thesis, Univer-
which implies an EVPI of 0.06 54 31, 12.22%. The VSS of sidad de Castilla-la Mancha, Department of Electrical Engineering,
0.04 179 or 7.84% still suggests that scheduling according to 2011.
[6] L. Garces and A. Conejo, “Weekly self-scheduling, forward con-
expected values of involved stochastic processes would lead tracting, and offering strategy for a producer”, Power Systems, IEEE
to a significantly worse outcome. Transactions on, vol. 25, no. 2, pp. 657 –666, May 2010.
[7] G. Bathurst and G. Strbac, “Value of combining energy storage and
wind in short-term energy and balancing markets”, Electric Power
V. C ONCLUSION Systems Research, vol. 67, no. 1, pp. 1–8, Oct. 2003.
[8] T. Tran-Quoc, X. Le Pivert, M. Saheli, and O. Beaude, “Stochastic
This paper has assessed the effect of different mobility approach to assess impacts of electric vehicles on the distribution net-
forecasts for stochastic charge scheduling of aggregated PEV work”, in 2012 3rd IEEE PES International Conference and Exhibition
fleets. An existing model, which integrates network capacity on Innovative Smart Grid Technologies (ISGT Europe), 2012, pp. 1–8.
[9] F. Ruelens, S. Vandael, W. Leterme, B. J. Claessens, M. Hommelberg,
pricing based on DSO’s long run marginal cost in the aggre- T. Holvoet, and R. Belmans, “Demand side management of electric
gator’s objective function was extended for this purpose. This vehicles with uncertainty on arrival and departure times”, in 2012
extension included a number of different non-anticipativity 3rd IEEE PES International Conference and Exhibition on Innovative
Smart Grid Technologies (ISGT Europe), 2012, pp. 1–8.
constraints. These were used to show the impact on optimal [10] C. Sandels, U. Franke, N. Ingvar, L. Nordstrom, and R. Hamren, “Ve-
scheduling and operational decisions for a retailer executing hicle to grid monte carlo simulations for optimal aggregator strategies”,
transactions in two consecutive markets under uncertainty in in 2010 International Conference on Power System Technology, IEEE,
Oct. 2010, pp. 1–8.
inputs regarding market prices and mobility. [11] D. Dallinger, D. Krampe, and M. Wietschel, “Vehicle-to-grid regula-
The computation of typical quality metrics of stochastic tion reserves based on a dynamic simulation of mobility behavior”,
programming metrics, expected value of perfect information Smart Grid, IEEE Transactions on, vol. 2, no. 2, pp. 302 –313, Jun.
2011.
and value of the stochastic solution, has quantified the above [12] L. Hirth and I. Ziegenhagen, “Control power and variable renewables
named effects of different mobility forecasts. Furthermore - a glimpse at german data”, in International Conference on the
it has detected the implications of the different sources of European Energy Market (EEM), vol. 10th, Stockholm, Sweden: IEEE,
May 2013.
uncertainty: market prices, mobility, availability and consump- [13] A. J. Conejo, M. Carrión, and J. M. Morales, Decision Making
tion. Overall, the results are well aligned with the observed Under Uncertainty in Electricity Markets, ser. International Series in
tendency in literature to account for uncertainty and support Operations Research & Management Science. Boston, MA: Springer
US, 2010, vol. 153.
the use of stochastic optimization frameworks. [14] B. Dietz, K. Ahlert, A. Schuller, and C. Weinhardt, “Economic
Future work is manifold. This paper merely includes a benchmark of charging strategies for battery electric vehicles”, in
stylized case study that shows the functionality of the proposed PowerTech, 2011 IEEE Trondheim, Jun. 2011, pp. 1 –8.
PEV aggregator model as well as advantages of the stochastic
optimization framework. In the future, it could be interesting B IOGRAPHIES
Ilan Momber received the degree of Business & Industrial Engineering
to apply this approach to a case with realistic mobility, big from the Karlsruhe Institute of Technology (KIT), Germany in 2010. He is
scale, real, medium to low voltage networks and using time currently pursuing the SETS Joint Doctorate at UPCO and KTH. Previously,
series models for generating market price (day-ahead and he worked at the Fraunhofer Institute for Systems and Innovation Research
(FhG ISI) in Karlsruhe, Germany, visited the Lawrence Berkeley National
balancing) scenarios. Furthermore, a risk aversion measure Laboratory, US and the University College London (UCL), UK. His interests
could be included. include regulation, technical and economic modeling of power systems with
particular focus on the integration of plug-in electric vehicles and distributed
generation.
ACKNOWLEDGMENT
The Erasmus Mundus Joint Doctorate on Sustainable Energy Technologies Tomás Gómez San Román (Senior Member) received the Ph.D. degree
and Strategies (SETS) funded by the European Commission’s Directorate- in industrial engineering from the Universidad Politécnica, Madrid, Spain, in
General for Education & Culture. The authors would like to express gratitude 1989. He is a Professor of electrical engineering at the Engineering School of
towards all partner institutions delivering the joint degree: Universidad Pontif- Universidad Pontificia Comillas (UPCo) in Madrid. He has broad industrial
icia Comillas (UPCO), Spain, Royal Institute of Technology (KTH), Sweden, experience in joint research projects in the field of electric power systems. His
and Delft University of Technology (TUDelft), The Netherlands. areas of interest are the operation and planning of transmission and distribution
Dedicated thanks are expressed towards Afzal Siddiqui, who has gener- systems, power quality assessment and regulation, as well as economic and
ously regulatory issues in the electric power sector. Since May 2011 he is on
. contributed to earlier versions of this model with valuable discussions. extended leave of absence acting as commissioner at the Spanish regulator:
National Energy Commission (CNE).
—