0% found this document useful (0 votes)
733 views6 pages

Optimal Battery Energy Storage System (BESS) Charge Scheduling With Dynamic Programming

This document presents a dynamic programming algorithm for optimally scheduling charge and discharge of a battery energy storage system (BESS). The algorithm minimizes electricity costs while reducing battery stress to prolong life. Dynamic programming is used to determine the unique optimal charging curve, which is significantly different than conventional curves. The goal of the scheduling is to drive down electricity bills by charging when rates are low and discharging during high price peak periods, thereby shaving peaks and saving money compared to not using energy storage.

Uploaded by

api-3826450
Copyright
© Attribution Non-Commercial (BY-NC)
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
0% found this document useful (0 votes)
733 views6 pages

Optimal Battery Energy Storage System (BESS) Charge Scheduling With Dynamic Programming

This document presents a dynamic programming algorithm for optimally scheduling charge and discharge of a battery energy storage system (BESS). The algorithm minimizes electricity costs while reducing battery stress to prolong life. Dynamic programming is used to determine the unique optimal charging curve, which is significantly different than conventional curves. The goal of the scheduling is to drive down electricity bills by charging when rates are low and discharging during high price peak periods, thereby shaving peaks and saving money compared to not using energy storage.

Uploaded by

api-3826450
Copyright
© Attribution Non-Commercial (BY-NC)
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
You are on page 1/ 6

Optimal battery energy storage system (BESS)

charge scheduling with dynamic programming

D.K. Maly
K.S. Kwan

Indexing terms: Buttery energy storage systems, D y m i c programming

Although battery storage is sometimes required at


Abstract: A dynamic programming algorithm for any cost for emergency backup, or storage for stand-
the optimal charge/discharge scheduling of BESS alone ‘alternative energy’ systems, a necessary condi-
energy storage is presented. It ensures the tion is often for BESS to save money, or at least to pay
minimisation of the electricity bill for a given for itself. As to the economy of BESS for end use, util-
battery capacity, while reducing stress on the ity pricing usually provides an incentive for load man-
battery and prolonging battery life. Optimal agement. Energy (kWh) savings are achieved by
scheduling of the battery charge state is in itself demanding utility power during cheaper off-peak times.
unique; the methods of multipass dynamic This energy is stored, and released into the customer
programming are used to accomplish this. load when required during expensive peak times. When
Maximum payoff for load redistribution and peak power (kW) limits are placed in the electricity
peak load shaving is determined while accounting contract, the electricity bill is reduced by ‘shaving’ the
for charging rate, battery voltage fluctuation and peaks. Stored energy is therefore released when peak
internal losses as a function of charge state. The power is demanded. An economic evaluation of a com-
optimal charging curve is significantly different plete BESS system is found in [l].
from the curve conventionally published for The payoff possible from the daily redistribution of
BESS. electricity through the meter is usually illustrated with
a figure similar to Fig. 1. There it is seen that the BESS
shaves the peak, and charges when generation capacity
is underutilised. Although this curve illustrates effec-
1 Introduction
tively the basic function of BESS, it should not by
Battery energy storage systems (BESS) are designed for any means be considered as an actual schedule for
a variety of reasons, namely: power factor correction, charging.
‘spinning’ power reserve, load frequency control, emer-
gency backup, peak shaving and economy. These
motives relate mainly to utility applications and are
briefly introduced as follows.
Power factor correction is basically the control of 5
reactive power, or VARs. Since battery voltage is DC, I \
VARs do not flow in the battery itself. Instead, the \
converter is controlled to operate as a reactive power
source. Since the battery is always ready to pump
energy into the power system when in service, it can be
counted as spinning reserve. Load frequency control
requires rapid power generation in response to sharp
changes in load, and so is similar to spinning reserve.
Batteries can also be applied in this case. Another daily cycle
application of batteries is for emergency backup. That Fig. 1 Busic loud redistribution with BESS
is, when the power system is disabled, or blacks out,
batteries can be used as an independent electricity gen- In engineering the BESS system, surprisingly little
eration source, and without any warm-up time neces- information exists in the literature for optimising either
sary. Finally, BESS can be used for economic reasons, the system components (most notably, batteries) or the
i.e. to save money. This is the subject of the paper and control and monitoring schemes. It is the purpose of
is a critical factor in consumer applications, as detailed this paper to describe such optimisation.
below. The following Section reviews BESS in the light of
optimisation requirements. Dynamic programming
0 IEE, 1995 (DP) is outlined in Section 3. Finally, practical con-
IEE Proceedings online no. 19951929 cerns and results are presented in Section 4.
Paper first received 3rd August 1994 and in revised form 23rd January
1995 2 Battery energy storage system
The authors are with Motorola Electronics Taiwan Ltd., Component
Products Division, 550 Chung-Hwa Road, Sec. 1, Chung-Li 320, Taiwan, As stated, the driving motivation for BESS depends on
Republic of China the specific application. Furthermore, the value of
IEE Proc.-Sci. Meas. Technol.. Vol. 142, No. 6. November 1995 453
power quality improvement or emergency back-up is tion is to drive the electricity bill down; however, the
difficult to quantify. For a treatment of the worth of battery is treated kindly in the process. That is, battery
the above see [2].In the following, the value of BESS is wear is reduced because it is coincident with internal
attributed entirely to its payback in reducing the elec- losses and wasted energy (money).
tricity bill. First, the nature of batteries is reviewed.
2.2 Evaluation functions
2.1 Battery characteristics The cost of electricity is generally made up of two
Batteries, as considered here, are chemical plants built parts: energy and power. Both may be priced higher
to convert electrical energy to chemical potential during utility peak generation to encourage end-use
energy, store the energy and convert it back to electri- management of energy. In Taiwan, pricing may divide
cal energy on demand. The major advantages of batter- the day into either two or three parts (see Appendix).
ies are very quick energy conversion and high energy The biperiod schedule is a subset of the triperiod sched-
density (kWh/m3 or JA). The major disadvantages are ule, so only the triperiod schedule need be treated here.
investment cost, caustic materials and limited lifetime.
Batteries are modelled as follows. The internal volt- 2.2, I Energy: The consumption cost of kWh is rela-
age is a characteristic of the compounds involved in the tively straightforward. The rate is low at off-peak
chemical reactions, and can be taken to be constant. (ROFF) and high at peak times (RPEAK). A third rate
The most common batteries are lead-acid; they are (half-peak) priced between these two is scheduled early
cheaper and less caustic than many other batteries. The morning, lunchtime and evening. Optimised paths usu-
internal voltage of lead-acid batteries is 2.00V. ally command the battery either to rest or charge in the
The terminal voltage, however, is a nonlinear func- middle-rate (half-peak) times, and are subtracted from
tion of curreat and stored charge. A reliable equation the savings:
relating terminal voltage ( v ) to internal voltage (Eo) and $savings = ~ ( R P E A -K R O F F ) ( Q m a a : - Qmzn)
charge state (q), based on experimental testing, is found
in [3]: - (half-peak costs)
where q is the effective efficiency.
+
‘U = BO r’(q)z + Aexp(-Bq/Q,,,) (1) In fact, it is shown below that the BESS is at rest
The last term is neglected here, and: during half-peak times, except for rapid charging at
~ ’ ( q=) R + K Q m a z / q discharging (2a) noontime to cash in on the afternoon peak rates.
~ ’ ( q=) R + KQmas/(Qmaz - 4 ) charging (2b) 2.2.2 Power: In Taiwan the power contract sets peak
where R, K, A and B are constants of the battery mate- power limits in each of the pricing periods. The
rials, Q,, is the rated storage capacity, q < Q,, is the monthly bill is based on these preset limits and the
charge storage state and Qmu, - q = depth of discharge. peak power demand for each period, respectively. If the
Far lead-acid batteries the constants are R = -0.20Q limits are crossed a penalty is imposed (see the Appen-
and K = 0.251R. The constant A is neglected (thus B is dix for details). The month is characterised by a typical
irrelevant). Note that the constant resistance (R) is neg- day’s load curve, scaled to have the month’s peak in
ative, which is counterintuitive. However, K R in the each pricing period.
second term of eqn. 2, so r’(q) is always positive.
Eqn. 2 are the basis of battery modelling. Power 3 Dynamic programming
flow at the terminals for terminal current ( i j is
*
P = vi = Eo2 r’(q)i2 (3)
The theory of DP may be considered as an alternative
to calculus in solving for minimum cost in daily energy
PLOSS
= PSTORBDi- storage cycles. An advantage of
Thus, from eqn. 3 the stored energy is proportional to includes avoiding differentiation, a
the integral of current; the power loss (PLoss) depends ing continuous functions. This is useful in the present
on the square of the current as well as the nonlinear context for limiting conditions; for example, battery
r’(q).
The characteristic eqn. 1 has limitations. The charge held between hard limits:
(4) should be held less than rated (Q,,), and greater
than a minimum fraction of rated (Qmin= 0.30 Q, is
used below). Qmzn < &man (4b)
The terminal voltage must also be held between lim- -Imuz < < Imaz (44
its for eqn. 1 to be valid. Not only is this limitation respectively, where v = terminal voltage, q = stored
necessary for theory but also for practice. Extreme charge and i = current.
voltages cause various secondary (and detrimental) Another advantage of DP is that nonlinear equations
effects to get out of hand, for example, plate deteriora- (such as battery voltagekurrent curves) are easily han-
tion and hydrolysis. Hard limits of 0 . 8 5 < ~ < 1.2pu
~ V dled. They are simply CalGukited at each interval based
are used here (lpu = 2.00V for lead-acid batteries). on the state. Perhaps the greatest advantage of DP is
The charging rate must be limited to protect the bat- that, quite unlike calculus, it is fundamentally compati-
tery and operate the converter safely. A hard limit for ble with digital computers. That is, the program is
maximum current is thus programmed. However, this implemented by calculating over discrete time intervals
limit was never reached because the voltage limits numerous times with nested loops.
(which are more fundamental to battery operation)
automatically prevented large current flow. 3.1 Background
All of these limiting conditions are included in the Groundbreaking work with DP was performed at Bell
optimisation below. The evaluation functions are Labs in the 1950s; a good fundamental reference is
detailed in the following subsection. The basic motiva- Bellman and Dreyfus [4,5]. Their work was geared spe-
454 IEE Proc -Sei Meas Technol, Vol 142, No d, November 1995
cifically to exploit the inherent characteristics of the Energy rates andhowerare given in the Appendix. An
then newly emerging digital computer. initial pass is run for a period of 6t = 113day = 8h and
The theorem of optimality is the basis of DP: ‘An 6q = NAQl3. This initial case is shown in Fig. 3 with
optimal policy must contain only optimal sub-policies’ the step size set to its largest: step = 32 (6t = 8h). The
IS]. That is, when optimising a state at a given time (a optimal path from the first iteration is then traced
subpolicy) the only information necessary is the previ- backwards from the end point (k = 96), similar to the
ous states and the cost of moving to the new state. Fur- brute force approach in the preceding subsection. If the
thermore, an optimised subpolicy must be part of the optimal point at each k is not qk, then qk is updated to
overall optimal policy. In practice, this limits memory the optimal point, and the calculation reiterated. This
space requirements because the only information that reiteration continues until the guide path qk converges
needs to be recorded for a given state is its optimal pre- on itself.
vious state, or path. The limiting cost of DP is thus optimal path
computer time.
The basic idea behind DP is seen in Fig. 2. The path
from each point at time k to each point at (k + 1) is
calculated. The path with the least cost is recorded.
When all paths are calculated, the optimal path is thus
traced backwards from the final point.
AT AT
..
I
Umin )-----L

6t

0 32 64 96
k
Fig.3 Initial puss of multbuss dynamic p r o g r m i n g

k k+l k+2 At convergence, the time and charge intervals are


Fig.2 Brute force dynamic programming: every possible path is culm- halved, and the iterations begun anew:
lated, for an instant k
step = step12 (7a)
In the BESS application, stored charge is chosen as = s t e p x AQ
Sq (7b)
the state. Thus, when a minimum charge is chosen as
the start and end points of the daily cycle, the optimal S t = step x AT (74
path calculated is the charging schedule with the least The second pass (step = 16, 61 = 4h) is shown in Fig. 4.
cost. Energy and peak power costs are used as the eval- By choosing both K = N = 96, and the initial step size
uation criteria; these and the battery characteristics are = 32, the step size is successively halved until step = 1.
all accounted for as described in the preceding Section. In practice, this approach cuts the calculation time by
two orders of magnitude down to about 3s.
3.2 Multipass dynamic programming
Note that the limiting conditions in eqn. 4, contrary to Q max
s
the calculus approach, reduce the number of calcula-
tions required. Even so, the number of iterations will
soon become overwhelming with larger problems (or c
with smaller resolution). In the BESS example, the
Q
U
upper limit of loop calculations is KN2, where:
K = total number of intervals = 96
Qmin
for one day if AT = 15 min intervals, and
N = number of charge increments = (Qma2- Qmzn)/AQ
where AQ is the charge increment. I
0 16 32 48 64 80 96
In applications of DP with larger dimensions, even k
with higher-speed computers, the program takes hours Fig.4 Second puss of multipass dynamic programming
or even days for completion, This is simply unaccepta-
ble, and the algorithm must be improved. A method of 4 Implementation
reducing the calculation time required by the brute
force iteration is known as multipass DP [6]. The programming procedure described above was
For the multipass, at each instant k, only three implemented in C-language. Some of the practical
charge states are taken: ‘guide’ charge qk and its two adaptations and lessons learned are seen below. Exam-
neighbours ples of optimal load levelling and peak shaving are
*
4 = (4k S q ) (5) included. Note that they are generally different from
Fig. 1, and perhaps not intuitively expected.
where 6q = step x A Q and step is the step size discussed
in detail below.
The evaluation functions at each step are: 4.1 Program description
fenes.gy(k) = f e n e r g y ( J C - 1) + Rate(k)kWSt
The program’s inner loop is the multipass DP. At the
@a)
heart of the nested loops is the cost function and a
f p o w e r ( k ) = f[kW,,, (off-peak), kWm,, (hdf-peak), search for the minimum cost. Conditions on the proc-
kW,,, (peak)] (6b) ess are conveniently included in the appropriate spots.
IEE Proc -Sei. Meas Technol, Vol. 142, No 6, November 1995 455
At each pass, the best path for the present iteration is derive a hard, benchmark figure for battery payback.
traced backwards, the ‘guide’ path updated and the The results of a sound, customised pricing of BESS
step size halved if at convergence. When the guide path hardware can be compared to the generally applicable
converges at the smallest step size, the multipass is savings for the electricity consumer calculated below.
complete and the optimal cost returned to the outer Payoff can be found by rephrasing the question. In
loop. other words, what must BESS investment cost be to
All quantities except prices are expressed per unit. pay for itself in a given time? Five years is chosen to
The bases are: represent battery and system lifetime. The savings, nor-
malised to battery size, are thus
powerBASE = day’s maximum kW (sa)
timeBASE = 1 h
voltsBASE = utility service nominal volts
(8b)
(8c)
[(
PmerSavzngs
month
EnergySavzngs
) (
month
+
La f etzme
XBESS
capacity
11
and all other BASES are derived from these. For (11)
example, This is the break-even cost for BESS. That is, BESS
chargeBASE= powerBASE x timeBASE/voltsBASE (9) will pay for itself in the assumed lifetime if its invest-
ment cost/capacity is lower than that calculated in
is the charge BASE ( A h) eqn. 11. The break-even costs found for two represent-
ative examples are given in the following Section.
4.2 Practical considerations
The following points are noted as of practical concern. 4.4 Discussion
(1) Step size. For large batteries (5 - 10h x peak kW) The charge scheduling and energy redistribution from
the smallest integer step size may still be too large. In applying BESS are shown in Figs. 5 and 6 for a capac-
this case, after the smallest time step 6t = 1 x AT is ity of 2.0pu and in Figs. 7 and 8 for a capacity of
reached, 6q is further scaled down: 0.lpu. The larger capacity is discussed first because its
effect is more obvious. The load is the shape of this
6q = SCALE x 6q SCALE < 1 (10) island’s aggregate scaled down to 200 kW peak.
in further iterations. This makes for a smooth charge 3
schedule when 6q becomes small enough. .B 2.51
(2) Error handling was included as follows. If one of the energy cost
conditions, e.g. eqn. 4, was violated the charge state energy and
was simply ignored. However, if all three charges (qk,
. 6q) were in violation the guide qk was pushed
qIc f
towards the previous &step) and the present ‘k‘ loop 0
5
p.t’
1

UL

energy only optimised
/

started afresh. 0 0400 0800 1200 1600 2000


I

2400
(3) A n outer loop was placed around the DP Ioop to feed time,o’clock
it increasing battery sizes. The maximum kW shaving Fig.5 Chmge scheduling and energy cost for a large battery
and kWh savings for each battery size can be found. Battery slze = 2.0 pu = 400 kwh
The resulting reduction in the electricity bill for a pro-
posed battery size is thus determined, or the maximum 1.6
($ savings)/(BESS capacity) can be found, for example. 1.41 nenergy only optimised

4.3 Payback time


The critical question for BESS is: How long does it
take to pay for itself? Unfortunately, the question of ..
BESS investment cost would be challenging even for a 0.2
separate paper exclusively dedicated to it. That is, it 0 0400 0800 1200 1600 2000 2400
depends on several nonlinearly interrelated parameters, time, o‘clock
which change yearly with improved technology, e.g. Fig. 6 Load curvesfor a large battery: originaland optimised with BESS
battery cost, battery lifetime, converter cost, floorspace Battery size = 2.0 pu = 400 k w h
cost and maintenance in addition to customer load and
utility pricing. A general treatment of these costs would The charge level is shown with the energy pricing in
be even more confusing than the rate structure of elec- Fig. 5. The charge is chosen as the ‘state’ for DP. The
tricity on Taiwan. For example, battery cost is a func- battery is seen to be charged at the quickest rate when
tion of both energy storage and power rate capacity. near minimum charge; it is discharged at the quickest
Converter cost is a nonlinear function characterised rate when near maximum charge. This takes advantage
mainly by ampacity. Floorspace is a widely variable of the nonlinear effective resistance r’(q), and maxim-
cost depending on environment (floorspace is a pri- ises efficiency as seen from eqns. 1 and 3.
mary factor, yet an ill-defined cost, for island states). Note that, for end-user optimisation, actually it is
The authors have succeeded in generalising the cost irrelevant when the maximum peak power and mini-
analysis independent of (normalised to) the actual mag- mum power occur within a given pricing period. Since
nitude of the customer electricity demand. This was the customer is basically interested in saving money,
done by removing the actual battery cost from the the late evening relative peak is in fact increased. In
analysis, as the pricing of BESS is a nontrivial under- this case it should be of no harm to the utility because
taking, well beyond the scope of this paper. Unfortu- of its general night-time generation capacity surplus.
nately, the results are not expressed in familiar ‘years (For utility BESS, or where additional constraints are
to payback’ terms. However, the analysis allows one to imposed on peak kW, the tradeoff for battery efficiency
456 IEE Proc -Sei Meas Technol, Vol 142,No 6, November 1995
and placement in the time of the peak power demand is Again, the payback is very small relative to the power
calculated.) contract savings. Yet the break-even cost with power
A 'trough levelling' as in Fig. 1 is nonoptimal and optimisation is larger than the NT$10,000/kWh esti-
misleading because of the unnecessarily high charging mate, so the system will break-even in less than five
rate, especially timing it when the battery is not near years. These results show the economic incentive for
minimum storage. When charging is optimised, the bat- BESS installation of a relatively small battery.
tery is more evenly charged, as is Fig. 5; the charge rate A note of caution must be added here. The optimised
is slowed as the charge approaches full capacity. The power curves are almost indistinguishable from the
battery remains idle in the morning 'half-peak' (07:30 - load curve in Fig. 8. Hence, it is difficult to forecast
1O:OO) after slowly charging all night. It takes a booster future kilowatt levels for the power contract, which is
charge at the noon half-peak (12:OO - 13:OO) to dis- signed annually.

--nr-L-
charge more in the afternoon. Finally, it discharges
completely before the evening half-peak (17:OO - 22:30)
and remains idle until it can recharge at off-peak rates.
Energy optimisation (Fig. 6) is used for normal (non- energy cost
maximum power) days until the kW peak demand
approaches its limit. Note that Fig. 1 serves as a poor
;$
rl-
1.51
. I energy and power optimised I
guide for charge scheduling, and the new method pre- I
sented here saves money and decreases battery deterio-
'U "I
nerav
*, onlv
, ooti- .
ration. 0 0400 0800 1200 1600 2000 2400
Power optimisation is also included in Fig. 6. First, time, o'clock
the contract is optimised for the curve, which is taken Fig.7 Charge scheduling and energy cost for a smaller battery
as characteristic (maximum kilowatts in each period) Battery size = 0.1 pu = 20 kWh (Charge scale x 0.1)
for this month. The night-time charging is nearly
unchanged. 1.6
Power optimisation redistributes the demand curve
significantly, but reduces the maximum peak-period
kilowatts by surprisingly little (-0.1 pu). However, it is
3
1.41
1.2 rod curve

in fact optimal because the noontime peak is limited.


The tradeoff is seen for morning discharge (1O:OO - 0.4
12:OO) and noon recharge (12:OO - 13:OO).
It is perhaps counterintuitive that the battery was not 0 0400 08001200 1600 2000 2400
pushed as hard when power optimisation was included. time, o'clock
However, it cannot be recharged very much during the Fig.8 Load curves for a smaller battery: original and optimised with
noon hour because power optimisation also limits the BESS
Battery size = 0.1 pu = 20 k w h
noontime peak.
The savings with and without power optimisation are
included in Table 1. First, note that with the utility The main points are summarised as follows. Sizing
pricing on Taiwan (see Appendix), energy savings are and charge scheduling of BESS depends on utility
only about 5% of power savings; consequently, BESS rates. Smaller batteries give faster payback times; larger
should be designed primarily for limiting monthly batteries are necessary for reliable load forecasts. These
power contract levels. two opposing trends must be balanced.
Secondly, the energy savings are higher when maxi- In closing, note that battery voltage fluctuation, non-
mum power levels are not optimised. The 10% differ- linear charge-dependent internal losses, charge rate lim-
ence is a result of high-current charging in the noon its, energy pricing and peak kilowatt limits have all
hour when the noontime kilowatt peak is not con- been accounted for in the optimisation. Even so, the
strained. particular charging path is still highly dependent on the
Thirdly, the break-even cost is compared to the original load which is continuously demanded.
actual BESS system cost for the consumer. The latter is
estimated at roughly NT$10,000/kWh [l]. Thus, it is 5 Conclusion
seen that there is no chance for payback in this case,
even if the system lifetime were 10 - 20 years. A dynamic programming method for minimising the
Charge and power levels for a capacity of 0.lpu are electricity bill with BESS has been presented. Although
shown in Fig. 7. The general trends are the same as the 'seat-of-the-pants' approach of 'shave the peaks, fill
those for the larger battery. In this case, however, the the troughs' explains the general idea of BESS, it was
break-even cost is much higher, as seen in Table 1. shown to be a poor guide if applied to actual charge
Table 1: Savings and break-even investment cost for BESS

5-year
Battery size Power Energy savings Power savings
(pu) optimised (NT$/kWh/mo) (NT$/kWh/rno) break-even
( NT$/kWh)
0.1 No 2.55 - 153
0.1 Yes 2.40 642 14800
2.0 No 1.58 - 94.5
2.0 Yes 1.44 74.6 1800
NT$I = US$0.0375 = f0.0250

IEE Proc.-Sci. Meas. Techno[., Vol. 142, No. 6,November 199s 451
Table 2: Power level pricing as a function of both time-of-day and
ti me-of-year
Biperiod pricing Triperiod pricing
Electricity rates
Effective IJune, 1993 Summer Non- Summer Non-
months* summer months* summer
Power contract rates Base rate 213 159 213 159
(NT$/kW per month Half-peak - - 159 159
Night 42.6 31.8 42.6 31.8
Energy rates Peak rate 1.89 2.95
(NT$/kWh) Half-peak - 1.78
Night 0.77 0.70
Biperiod and triperiod rates are shown for high-voltage.
* Summer months are June -September

scheduling. The method can be applied to maximising The base charge is


energy efficiency over a charge cycle, minimising kilo-
watt peaks, or a compromise of both.
$BASE = JCWOFFCOFF~ W H P C +
H~ +
PW P K C P K
where CpER is the contract price for peak power in
Internal losses and the nonlinear terminal voltage pricing period PER = (OFF, HP, PK) for off-peak,
fluctuation were accounted for. The program calcu-
half-peak and peak times, respectively. The penalty
lated the maximum payback for a battery size, and can
charge is based on period limits. The relation between
thus help in sizing the battery to a specific application.
contract settings (kW,) and period limits (limit,) is
seen in Fig. 9. The penalty charge is
6 Acknowledgment
$PENALTY = sum{p(maz ~ W P BZRz m , iCp~~)}
Support for this work from the Taiwan Ministry of over the pricing periods PER, where
Economics Energy Commission is gratefully acknowl-
edged. p(rnaz kW, limzt) =
if max kW < lzmzt

1
0
7 References
2CPER(maX kw’ - limit)
1 ‘Design and costs for a generic 10-MW utility lead-acid battery if lzmzt < rnax k W < l.llzmzt
energy storage plant’, Electric Power Research Institute, EPRI
AP-5845, Project 2123-6, Final Report, June 1988 C P E Rx[ ~
O.llzmit + 3(mas kW’ - l.llzmzt)]
2 SUBRAMANIAM,R.K., BILLINTON,R., and WACKER,G.: if max kW > l.llimzt
Understanding industrial losses resulting from electric service
interruptions’, IEEE Trans. Ind. Appl., 1993, 29, (l), pp. 238-244 where modified max k W ’ is calculated as:
3 SHEPHERD,C.M.: ‘Design of primary and secondary cells 11.
An equation describing battery discharge’, J. Electrochem. Soc., max kWbK = max kWpK
1965, DD. 657-664
4 BELLMAN,R.E.: ‘Dynamic programming’ (Pnnceton University maz kWkp = max kWHp
Press, Princeton, N.J., 1957)
5 BELLMAN,R.E., and DREYFUS,S.E.: ‘Applied dynamic pro-
- max(0, max ~ W P-KlimitpK}
gramming’ (Princeton University Press, Princeton, N.J., 1962) max JCW&,, = max kWoFF
6 YANG,J.S., and CHEN,N.M.: ‘Short term hydrothermal coordi-
nation using multi-pass dynamic programming’, IEEE Trans., - maz(0,max
kWpK - ZimitpK}
1989, PWRS-I, (3), pp. 1050-1056
7 ‘Electricity pricing schedule’, Taiwan Power Company, Taipei, - maz(0,max kWfip - IzmotHp]
Taiwan (Effective 1 June 1993 (in Chinese))
8 KWAN,K.S., and MALY,D.K.: ‘Optimal contract selection for The total power cost ishower= $BASE 3- $PENALTY.
electrical power customers’, Power Engineers Conference, Chung- A contract optimiser is nested in the inner loop [8],
Li, Taiwan, 17-18 December 1993
Since only a single month is considered in this paper,
8 Appendix the optimiser automatically selects a contract without
penalty.
The energy and power rate schedule is shown in June-September only
Table 2 [7] for biperiod and triperiod pricing of high
voltage service ( 3 . 3 - 22.8kV). The kWh charge is cal-
culated every 15 min. The rate is a function of time and
season. Currently the currency exchange is about US$1
= NT$27.
The peak kilowatt charge is calculated every month
depending on that month’s peak power in each period.
The biperiod schedule may be considered as a subset of
the triperiod schedule. The latter is described below.
The total kilowatt charge is made up of base charge I+
0 3
, I

6 9
, + ,
12
I
15 18 21
I
24
($BASE) and penalty ($PENALTY). It is generally
hour (daily)
optimal to pay a penalty for a few months to reduce Fi .9 Trperiod power contract lmits as a function of time (weekdays
$BASE. B
on Y )

458 IEE Proc.-Sei. Meas. Technol., Vol. 142, No. 6, November 1995

You might also like