Sie sind auf Seite 1von 5

1

Chebyshev Based Continuous Time Power System


Operation Approach
Marcelo Matus, Nicolás Cáceres, Sebastián Püschel-Løvengreen, and Rodrigo Moreno, Member, IEEE

Abstract—There is a growing interest to increase temporal There is a consensus that higher temporal resolution is
resolution in power system optimization models in order to needed to accurately represent the variability of renewable
improve representation of intermittent renewable generation and generation, and measure its impacts on costs, thermal gen-
thus capture more variability and costs. This increase in temporal
resolution, however, presents important modeling challenges since eration unit commitment and cycling, demand for flexibility
the optimization problem complexity grows with the higher services, wind/solar spillage, requirements for new infrastruc-
number of variables and constraints needed to represent further ture, etc. [5]–[11]. The main concern is that a higher temporal
(discrete) time periods. In this context, this paper proposes a mod- resolution will significantly increase the size of optimization
eling approach in continuous time, based on a pseudo-spectral models, reducing the ability of current methods to find op-
representation of Chebyshev polynomials that can be applied to a
variety of power system optimization problems. In particular, this timum solutions in reasonable timescales and with limited
paper illustrates a specific application of the proposed modeling computational resources [11], [12]. This problem, so far,
approach to determine continuous time operation of a small- has been approached by developing advanced simplification
scale power system, highlighting its main advantages such as the techniques and more efficient algorithms (e.g. optimization-
continuous temporal resolution, straightforward implementation based and heuristic-based algorithms that can take advantage
of differential and integral constraints needed to properly model
flexibility, and reduction of computational burden. Comparison of parallel computing) [12]–[14].
between continuous and discrete time power system operation In this context, the key contribution of this paper is to
solutions is also presented. propose a novel optimization model that can determine power
Index Terms—Chebyshev polynomials, Economic dispatch, system operation in continuous time where parameters (e.g.
Power system operation, Power system economics. demand, wind speed) and variables (e.g. generation units’ out-
puts) are functions represented by finite series of Chebyshev
polynomials [15]. The proposed representation allows to fit
I. I NTRODUCTION
outputs of wind and solar plants with significant accuracy
The current paradigm to model power system operation is and limited amount of information (i.e. number of Chebyshev
through discrete time optimization problems that can deter- coefficients), reducing computational burden, increasing the
mine outputs and/or commitments of generating units (and up quality of the solution, and overcoming the problems observed
to some extent demand) during a given time period. Temporal for discrete time power system operation models. Chebyshev
resolution within that time period will depend on the model’s representation also allows the treatment of integrals and deriva-
application: hourly resolution has been historically acceptable tives (including ramp rate limits) in a straightforward fashion,
for electricity market modeling and unit commitment prob- which is critical in optimization models that assess effects of
lems [1], while a less refined representation where several high penetration of renewables such as that discussed in [16].
hours are clustered or grouped in representative “blocks” Although this paper illustrates a first application of this
has been used for mid and long-term modeling, commonly technique on a specific continuous time generation dispatch
observed in problems such as hydrothermal coordination (e.g. model which controls a microgrid that powers a small town
SDDP [2], [3]) and infrastructure planning [4]. in Chile [17], it can be certainly expanded and applied on
Lately, governments’ interest to integrate large amounts of a full unit commitment problem and further power system
renewable generation has caused a need for developing more optimization applications related to operation and investment.
advanced decision support tools that can properly capture This paper is structured as follows: Section II explains the
the intermittent nature of renewables such as wind and solar Chebyshev pseudo-spectral representation used and the pro-
power. In this context, an active research area is the devel- posed continuous time generation dispatch model; Section III
opment of modeling approaches that can deal with refined shows our results obtained for the optimum operation of a
temporal resolution (e.g. less than an hour for operation small-scale microgrid; and Section IV explains our conclu-
problems and hourly for planning problems) and thus capture sions and further work.
more variability [5]–[11].
II. C HEBYSHEV P SEUDO - SPECTRAL R EPRESENTATION
M. Matus, N. Cáceres, S. Püschel-Løvengreen, and R. Moreno are with
the Energy Center of the Faculty of Mathematical and Physical Sciences, The Chebyshev pseudo-spectral representation of a given
School of Engineering (CMM, ISCI, DIE), Universidad de Chile. R. Moreno continuous function f (t) in the [−1, 1] domain is constructed
is also with Imperial College London. The authors are grateful for the financial by using the Chebyshev polynomials, Ti (t), defined by
support of Conicyt (through grants Conicyt/Fondecyt/Iniciación/11130612
and Conicyt/Fondap/15110019) (emails: mmatus@centroenergia.cl; rmoren-
ovieyra@ing.uchile.cl) Ti (t) = cos(i arccos(t)), (1)

978-1-4673-8040-9/15/$31.00 ©2015 IEEE


2

PV generation
and by approximating the function with a polynomial series
20
in the form

PV Power [kW]
10

N
 0
Discretized PV generation
pN (c, t) = ci Ti (t), c = {ci }, i = 0 . . . N (2) −10
64 degree Polynomial reconstruction
64 degree Chebyshev reconstruction
i=0
−20
0 10 20 30 40 50 60 70 80 90
where the coefficients, c = {ci }, can be understood as Period [15−minute step]

the projection of the function f (t) in the Chebyshev space.


Fig. 1. PV power production fit using Chebyshev and canonical polynomial
Usually the coefficients are computed in order for the resulting basis (N = 64).
series pN (c, t) to approximate the function f (t) in a standard
distance minimization problem

min ||f (t) − pN (c, t)||. (3) min ε+ + ε−


c c
s.t. Ac = p
The selection of the functional norm || · || is arbitrary, but p − y = ε+ − ε−
in general the norm L2 is used. Watson [18] discussed the use (7)
p≥0
of L1 and Lp norms, which allows to linearize (3) and use the
following formulation ε+ ≥ 0
ε− ≥ 0
+ −
min ε + ε
c This formulation has been successfully tested for different
s.t. Ac = p power profiles and has been proved more accurate than a
p − y = ε+ − ε− (4) regular polynomial fit. Fig. 1 shows the approximation of
+
ε ≥0 the power profile of a photovoltaic power plant (PV) using
both Chebyshev and the canonical basis (ti ). This figure also
ε− ≥ 0 shows that the Chebyshev approximation is significantly more
accurate (especially close to the ends of the interval), and it
where the Chebyshev coefficients c are variables of the prob-
properly fits the original data, even at nighttime when power
lem, and the rest is defined as:
production is equal to zero.

c = (ci ), i = 0 . . . N
x = (xj ), j = 0 . . . M A. Continuous Time Power System Operation Problem
 
jπ The power system operation problem refers to the need
xj = cos
M to operate different power plants (thermal and renewable
y = (f (xj )) generation) such that power demand is met at a minimum
(5) cost. The following equations (8) show a simple continuous
p = (pj )
⎛ ⎞ time formulation of the power system operation problem in the
T0 (x0 ) T1 (x0 ) ... TN (x0 ) [−1, 1] domain, where the operating cost of K power plants
⎜ T0 (x1 ) T1 (x1 ) ... TN (x1 ) ⎟
⎜ ⎟ is minimized in a given horizon subject to meeting demand.
A=⎜ .. .. .. .. ⎟
⎝ . . . . ⎠
K
 1
T0 (xM ) T1 (xM ) ... TN (xM )
min αk gk (t)dt
c −1
By construction, p holds the Chebyshev approximation of the k=1
K
function f (t) at the Chebyshev extreme points x [19], i.e.
s.t. gk (t) = d(t) (8)
k=1
N

f (xj ) ≈ pj = ci Ti (xj ), j = 0...M (6) 0 ≤ gk (t) ≤ Gmax
k (t)
i=0 t ∈ [−1, 1]

Starting with the general formulation (4), additional con- In this problem:
straints can be added to the Chebyshev approximation. For • d(t) is the power demand function over period [−1, 1],
example, a “one-sided” approximation may be obtained if • gk (t) is the power generation function of power plant k,
either ε+ or ε− are constrained to be zero [20], [21]. • Gmax
k (t) is the maximum power or capacity of power
Following this idea, the Chebyshev non-negative approxi- plant k,
mation, which is relevant for power and energy functions, can • αk is the generation cost of power plant k,
be formulated as follows: • K is the total number of power plants available.
3

Wind generation profile


If every power generation function and the demand function 4

Descretized wind profile


are represented in the Chebyshev space, the continuous time 192 degree Chebyshev reconstruction

Wind Power [kW]


3

power system operation problem can be expressed as:


2

K
 1
min αk γ T g k
c 0
k=1 0 20 40 60 80 100 120 140 160 180
Period [15 minutes step]
K (9) PV generation profile
s.t. gk = d 20
Discretized PV profile
k=1 192 degree Chebyshev reconstruction

Gmax

PV Power [kW]
15
0 ≤ Agk ≤ k
10

Where
5
• d = (dj ) is the vector of Chebyshev coefficients of the
0
demand function, 0 20 40 60 80 100
Period [15 minutes step]
120 140 160 180

• gk is the vector of Chebyshev coefficients of power plant


k, Fig. 2. 48-hour wind and PV profiles
max
• Gk is the vector of maximum power generation of
power plant k evaluated at extreme points x,
• and γ is the “integration vector”, which allows evaluation

of the integral of a given Chebyshev series in terms of ⎨j if i = 0 and j − i is even
its coefficients D = (Dij ) , Dij = 2j if j > i and j − i is even (13)

0 otherwise
1 1 N
 N
 1 Hence, one can introduce additional ramp rate limits to the
f (t)dt = ci Ti (t)dt = ci Ti (t)dt formulation (9), as follows:
−1 −1 i=0 i=0 −1
(10) K
N
 
= ci γi = γ c T
min αk γ T g k
c
i=0 k=1
K
where [22]
s.t. gk = d (14)
⎧ k=1
1 0 ⎨ if i is odd 0 ≤ Agk ≤ Gmax
k
γi = Ti (t)dt = 2 (11)
−1 ⎩− if i is even Rkmin ≤ ADgk ≤ Rkmax
(i + 1)(i − 1)
where Rkmin and Rkmax correspond to the minimum and
Note that while the optimization problem associated with maximum ramp rate (constant) limits of power plan k.
(8) presents a number of variables equal to (M + 1)K and Computing integrals in the Chebyshev space is also straight-
a number of constraints equal to M + 1 + 2K(M + 1), the forward, and this can be used to model state of charge of
optimization problem associated with (9) presents a number of BESS (Battery Energy Storage System) and reservoir systems,
variables equal to (N +1)K and a number of constraints equal cycling costs of thermal plants, etc.
to N + 1 + 2K(M + 1). Hence, size of optimization problem
associated with the proposed Chebyshev formulation could
III. C ASE STUDY
be significantly smaller than the discrete time formulation,
if N can be reduced without compromising the accuracy of The purpose of the following case study is to illustrate how
Chebyshev approximations of functions gk (t). Chebyshev representation works on determining the operation
of a small-scale power system with three generators and
one BESS (this is an existing microgrid in Chile [17]). The
B. Differentiation and ramp rate limits operation is determined for a 48-hour horizon with a 15-minute
The explicit use of Chebyshev series allows to compute temporal resolution (i.e. 192 periods).
derivatives in a straightforward fashion. For a given function The set of parameters for the thermal unit is presented in
f (t), if c corresponds to the Chebyshev approximation coef- Table I. The value of lost load is equal to 100 [$/kWh]. The
ficients, the derivative function f  (t) can be approximate by other generating units are a photovoltaic power plant (PV) and
the coefficients c , where a wind power plant, whose expected availability profiles are
those presented in Fig. 2. The parameters for the BESS are
c = Dc (12) presented in Table II.
Fig. 3 shows one discrete time dispatch solution (3a) and
and D is the differentiation matrix in the Chebyshev space, two continuous time dispatch solutions (3b and 3c), demon-
given by [23] strating that our approach can successfully solve power system
4

TABLE I
T HERMAL U NIT PARAMETERS with mid and longer-term optimization problems associated
with large-scale power systems.
Parameter Value Furthermore, we are currently developing further applica-
Pmax [kW] 120 tions of this approach on two types of power system operation
Cost [$/kWh] 5 problems: unit commitment and hydrothermal coordination
Rampup/down [kW/h] 8 with high penetration of renewable generation. The prelimi-
nary results confirm that the approach is promising in terms
TABLE II of capturing renewables’ variability and its effects on optimum
BESS PARAMETERS operational decisions.
Parameter Value Parameter Value
P chargemax [kW] 10 P dischgmax [kW] 10 R EFERENCES
ηch [pu] 0.9 ηdis [pu] 0.85 [1] M. Carrión and J. Arroyo, “A computationally efficient mixed-integer
Emin [kWh] 23.4 Emax [kWh] 117 linear formulation for the thermal unit commitment problem,” Power
Eini [kWh] 117 Systems, IEEE Transactions on, vol. 21, no. 3, pp. 1371–1378, Aug
2006.
[2] M. V. F. Pereira and L. M. V. G. Pinto, “Multi-stage stochastic
optimization applied to energy planning,” Mathematical Programming,
vol. 52, pp. 359–375, 1991.
operation problems. Fig. 3a and 3b show solutions for the same [3] M. E. P. Maceira, D. D. J. Penna, A. C. G. de Melo, L. Moraes,
operating conditions and although the size of the optimization and V. S. Duarte, “Ten years of application of stochastic dual dynamic
problems are similar (the discrete time problem has a temporal programming in official and agent studies in Brazil - description of the
newave program,” in Proc. 16th Power Systems Computation Confer-
resolution of 192 15-min periods while the continuous time ence, Glasgow, UK, 2008.
problem has 192 Chebyshev coefficients), our continuous time [4] Q. Chen, C. Kang, Q. Xia, and J. Zhong, “Power generation expansion
dispatch solution presents a lower total cost (-0.5%). In fact, planning model towards low-carbon economy and its application in
China,” Power Systems, IEEE Transactions on, vol. 25, no. 2, pp. 1117–
we demonstrated in this case that it is possible to reduce 1125, May 2010.
the number of Chebyshev coefficients (especially in smooth [5] B. Ummels, M. Gibescu, E. Pelgrum, W. Kling, and A. Brand, “Impacts
functions such as demand) by circa 30% without lessening of wind power on thermal generation unit commitment and dispatch,”
Energy Conversion, IEEE Transactions on, vol. 22, no. 1, pp. 44–51,
quality of the continuous time solution. March 2007.
Following the fundamentals introduced in Section II, we [6] M. R. Milligan and N. R. E. L. (U.S.), Large-Scale Wind Integration
represented cycling cost of the thermal unit in the con- Studies in the United States: Preliminary Results; Preprint. Golden,
Colorado: National Renewable Energy Laboratory, Sep 2009. [Online].
tinuous time optimization problem. Fig. 3c illustrates how, Available: http://nla.gov.au/nla.cat-vn4765600
especially between 18h and 31h, operation of the thermal [7] E. Ela, M. Milligan, B. Parsons, D. Lew, and D. Corbus, “The evolution
unit is steadier due to representation of the cycling cost. of wind power integration studies: Past, present, and future,” in Power
Energy Society General Meeting, IEEE, Calgary, Canada, July 2009,
Interestingly, continuous time operation of the thermal unit pp. 1–8.
shown in Fig. 3c is similar to that obtained by a traditional unit [8] C. Philbrick, “Wind integration and the need for advanced decision
commitment formulation, albeit the latter needs supplementary support tools,” in Power and Energy Society General Meeting, IEEE,
Detroit, USA, July 2011, pp. 1–7.
binary variables. [9] E. Ela, M. Milligan, and M. O’Malley, “A flexible power system
Similarly to adding cycling cost used to obtain results in operations simulation model for assessing wind integration,” in Power
Fig. 3c, further modeling aspects can be considered to repre- and Energy Society General Meeting, IEEE, Detroit, USA, July 2011,
pp. 1–8.
sent more accurately the level of flexibility of power system [10] J. Deane, G. Drayton, and B. O. Gallachóir, “The impact of sub-
operation, given the straightforward treatment of derivatives hourly modelling in power systems with significant levels of renewable
and integrals in the proposed Chebyshev optimization ap- generation,” Applied Energy, vol. 113, pp. 152–158, 2014.
[11] H. Pandzzic, Y. Dvorkin, Y. Wang, T. Qiu, and D. Kirschen, “Effect
proach. of time resolution on unit commitment decisions in systems with high
wind penetration,” in Power Energy Society General Meeting, IEEE,
Washington, USA, July 2014, pp. 1–5.
IV. C ONCLUSIONS [12] B. Palmintier and M. Webster, “Heterogeneous unit clustering for
efficient operational flexibility modeling,” Power Systems, IEEE Trans-
We have developed a novel approach to solve power system actions on, vol. 29, no. 3, pp. 1089–1098, May 2014.
optimization problems in continuous time, which is based on [13] S. Kamalinia, M. Shahidehpour, and A. Khodaei, “Security-constrained
expansion planning of fast-response units for wind integration,” Electric
a pseudo-spectral representation of Chebyshev polynomials. Power Systems Research, vol. 81, no. 1, pp. 107–116, 2011.
Through this approach, we can improve representation of inter- [14] A. Pina, C. A. Silva, and P. F. ao, “High-resolution modeling framework
mittent renewable generation and thus capture more variability for planning electricity systems with high penetration of renewables,”
Applied Energy, vol. 112, pp. 215–223, 2013.
and cost, increasing the quality of optimization solutions and [15] C. W. Clenshaw, “Polynomial approximations to elementary functions,”
overcoming the problems observed in discrete time models. Mathematical Tables and Other Aids to Computation, vol. 8, no. 47, pp.
Our proposed approach also allowed treatment of integrals and 143–147, 1954. [Online]. Available: http://www.jstor.org/stable/2001927
[16] P. Denholm and M. Hand, “Grid flexibility and storage required to
derivatives (including ramp rate limits) in a straightforward achieve very high penetration of variable renewable electricity,” Energy
fashion, which is a key aspect to properly model flexibility and Policy, vol. 39, no. 3, pp. 1817–1830, 2011.
assess effects of renewables. We illustrated the above proper- [17] R. Palma-Behnke, C. Benavides, F. Lanas, B. Severino, L. Reyes,
J. Llanos, and D. Saez, “A microgrid energy management system based
ties of the proposed approach on a specific small-scale power on the rolling horizon strategy,” Smart Grid, IEEE Transactions on,
system operation problem, albeit it can be expanded to deal vol. 4, no. 2, pp. 996–1006, June 2013.
5

System dispatch − Discrete time


20

BESS stored energy [kWh]


Thermal 100
PV
10 Wind
Power [kW]

50
Demand
BESS
kW+
0 BESS 0
kW−
BESS
kWh

−10
0 20 40 60 80 100 120 140 160 180 200
Period [15−minute step]

(a) Discrete time

System dispatch − Continuous time (Chebyshev representation)


20
100

BESS stored energy [kWh]


Thermal
PV
10
Power [kW]

Wind 50
Demand
BESS
kW+
0 BESS 0
kW−
BESS
kWh

−10
0 5 10 15 20 25 30 35 40 45 50
Time (hours)

(b) Continuous time (Chebyshev representation)

System dispatch − Continuous time (Chebyshev representation) with cycling cost


20
Thermal

BESS stored energy [kWh]


100
PV
10 Wind
Power [kW]

Dem 50
BESSkW+
0 BESSkW− 0
BESSkWh

−10
0 5 10 15 20 25 30 35 40 45 50
Time (hours)

(c) Continuous time (Chebyshev representation) with cycling cost


Fig. 3. Simulation results

[18] G. Watson, “The calculation of best restricted approximations,” SIAM Nicolás Cáceres Lagos holds the B.Sc. degree in electrical engineering
Journal on Numerical Analysis, vol. 11, no. 4, pp. 693–699, 1974. from the Universidad de Chile, Santiago. Currently he works at the Energy
[19] J. Boyd, Chebyshev and Fourier Spectral Methods: Second Revised Center of the Faculty of Mathematical and Physical Sciences, Universidad de
Edition, ser. Dover Books on Mathematics. Dover Publications, 2001. Chile, Santiago. His main research interests cover power system optimization,
[20] G. A. Watson, “The calculation of best linear one-sided lp approxi- including short and long term operation of hydrothermal power systems.
mations,” Mathematics of Computation, vol. 27, no. 123, pp. 607–620,
1973.
[21] N. Abdelmalek, “Computer program for the discrete linear restricted
chebyshev approximation,” Journal of Computational and Applied Math-
ematics, vol. 7, no. 2, pp. 141–150, 1981. Sebastián Püschel-Løvengreen holds the B.Sc. and M.Sc. degrees in electri-
[22] W. H. Press, S. A. Teukolsky, W. T. Vetterling, and B. P. Flannery, cal engineering from the Universidad de Chile, Santiago. Currently he works
Numerical Recipes in C (2Nd Ed.): The Art of Scientific Computing. at the Energy Center of the Faculty of Mathematical and Physical Sciences,
New York, NY, USA: Cambridge University Press, 1992. Universidad de Chile, Santiago. His main research interests comprehend long
[23] D. Trif, “Matrix based operatorial approach to differential and integral term operation of hydrothermal power systems and electricity markets.
problems,” in MATLAB - A Ubiquitous Tool for the Practical
Engineer, C. M. Ionescu, Ed. InTech, 2011. [Online]. Available:
http://www.intechopen.com/books/matlab-a-ubiquitous-tool-for-the-
practical-engineer/matrix-based-operatorial-approach-to-differential-
and-integral-problems Rodrigo Moreno (M’05) received the B.Sc. and M.Sc. degrees from Pontificia
Universidad Católica de Chile and the Ph.D. degree from Imperial College
London, U.K. He is currently an Assistant Professor at Universidad de Chile
and a Research Associate at Imperial College London. His research interests
are: power systems optimization, reliability and economics; renewable energy;
and the smart grid.

Marcelo Matus received his B.Sc. and M.Sc. degrees in Electrical Engineer-
ing from the Pontificia Universidad Católica de Chile and a Ph.D. degree
from the University of Tucson, Arizona. He is currently a research associate
at the Energy Center of the Faculty of Mathematical and Physical Sciences,
Universidad de Chile. His research fields are planning and operational research
of electrical systems and complex systems.

Das könnte Ihnen auch gefallen