See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/266584432

Optimal Control of a Distribution System with a Virtual Power Plant

Article · January 2004

CITATIONS READS

49 1,105

3 authors, including:

Roberto Caldon Roberto Turri

University of Padova University of Padova
92 PUBLICATIONS 960 CITATIONS 124 PUBLICATIONS 891 CITATIONS

SEE PROFILE SEE PROFILE

Some of the authors of this publication are also working on these related projects:

uncontrolled islanding of MV and LV networks in presence of relevant amount of DER View project

Novel methodologies for the analysis and control of low voltage actiive distribution network View project

All content following this page was uploaded by Roberto Turri on 13 January 2015.

The user has requested enhancement of the downloaded file.

278 Bulk Power System Dynamics and Control - VI, August 22-27, 2004, Cortina d’Ampezzo, Italy

Optimal Control of a Distribution System with a Virtual Power Plant

Roberto Caldon, Andrea Rossi Patria and Roberto Turri
Department of Electrical Engineering – University of Padova, Italy

1. Introduction compliance with the network constrains. In order to apply the

procedure, however, the DSO needs constantly updated
Changes in environmental policies and liberalisation of energy information regarding local generation and network state
markets lead towards an ever greater level of integration conditions.
between natural gas/thermal energy and electrical energy
distribution services. The availability of new technologies
makes it suitable the idea of combining small environmentally 2. Optimisation Problem
compatible generators located close to the loads supplying both
thermal and electrical power [1,2]. The primary objective of optimal operation can vary. For
example, economic optimisation can either aim at minimising
In this perspective, a number of Distributed Generators (DG) the costs of producing energy and supplying it to the loads or
embedded in the distribution grid might be aggregated into a so maximising the profits of a virtual utility owner.
called Virtual Power Plant (VPP) having the capability of
selling both thermal and electrical energy to neighbouring In this perspective, further new uncertainties, as regards power
customers [3,4]. flows and voltage levels, will affect distribution system
operational conditions since each producer manages its own
The main aim of a VPP would thus be to co-ordinate the plant in an independent manner. The operational optimisation
different energy resources in order to satisfy the customer is focused on providing, besides minimum overall production
requirements with the least generation cost, which means: costs, the ability to optimise other quantities such as reliability
ƒ to exploit, at any time, all the renewable resources and power quality.
which are stochastically available (wind energy, solar
energy, etc.). The management problem is here formulated as a non-linear
ƒ when needed, to integrate the generation with fossil constrained optimisation problem where the objective function
resources giving priority to the most economically is the total cost of supplying active, reactive and thermal power
efficient units. to the loads, constrained by voltage level and overload of lines
and transformers. The VPP would directly dispatch the
This implies a continuous variation, during the day, of the generated active power, whereas the DSO would be
amount and location within the grid of the generated power responsible for the safe and reliable operation of the network.
which may affect the operation of the grid as regards voltage
regulation and line loadings.
3. Cost Functions and Network Components
Within the framework of a distribution system a VPP would
thus have to manage the combined production of electric and In this section the cost models of the various network elements
thermal energy as well as other dispersed sources, if present, is described. The overall network model includes all passive
with the respect of the network constrains and optimising the and active components, regardless whether or not they belong
profit of the dispersed generation on the basis of the generation to the VPP, since they affect the system operational conditions.
costs of the various types of DG, the main grid energy
exchange costs and the Distribution System Operator (DSO) 3.1 Electric production
wheeling costs.
An electric generator can produce both active and reactive
In this paper a cost based optimisation procedure is proposed power, depending upon its capability curve. The cost function
for harmonising the concurrent operation of DSO and VPP associated with the active power production is given by the
which, although acting in an independent manner, can be co- following well known quadratic function:
ordinated by means of suitable economic signals.
C ( Pg ) = α ⋅ Pg 2 + β ⋅ Pg + γ (3.1.1)
The present approach for an optimal control of a distribution
system undermeans that a sort of local market is established.
The DSO optimisation procedure provides to the VPP nodal where C (C.U./h – Currency Unit per hour) is the hourly cost,
nodal economic indicators of active and reactive energy. These Pg (MW) is the generated active power and α (C.U./MW2h), β
signals are side results of the DSO optimal network control and (C.U./MWh) and γ (C.U./h) are coefficients depending on the
depend upon main grid supply prices, distribution losses and technology (e.g. fuel cost, efficiency, etc.). The production is
network ancillary services requirements. On the basis of such limited between a maximum and a minimum values, i.e.:
information, the VPP is able to optimise its production in
Bulk Power System Dynamics and Control - VI, August 22-27, 2004, Cortina d’Ampezzo, Italy 279

Pg min ≤ Pg ≤ Pg max (3.1.2) the CHP. The production is limited between a maximum and a
Qg min ≤ Qg ≤ Qg max (3.1.3) minimum values, i.e.:

Pg 2 + Qg 2 ≤ Sg max (3.1.4) Hcmin ≤ Hc ≤ Hcmax (3.3.3)

where Sgmax (MW) is the apparent nominal power of the 3.4 DSO Electric power exchanged with the HV grid
device, namely the size of the machine.
Equation (3.1.4) defines a limit of the capability curve of a New scenarios of deregulated market of electricity will give to
generator. If a generator is connected to the grid through an utilities an hourly spot price for both active and reactive power
inverter, eq. (3.1.2), (3.1.3) and (3.1.4) define completely its exchanged with the HV transmission grid. The cost due to
capability curve, while in case of a rotating machine two more purchasing active and reactive power from the grid is given by:
equations are needed:
C HV ( PHV , QHV ) = ρ P ,HV ⋅ PHV + ρ Q ,HV ⋅ QHV (3.4.1)
Pg + (Qg − Qg min ) ≤ (Qg max − Qg min )
2 2
(3.1.5)
where ρP,HV (C.U./MWh) and ρQ, HV (C.U./kVArh) are the
(Qg − Qg min ) 2 hourly spot prices for active and reactive power, while PHV
Pg − ≤ − ΔPg (3.1.6)
(MW) and QHV (MVAr) represent the imported power from the
2 ⋅ ΔPg
HV grid. Assuming the hypothesis of perfect competition, there
is no reason for suggesting different prices for purchasing
where ΔPgj (MW) is a prudential margin allowable before rather than selling energy, and thus if P HV and/or QHV are
reaching the steady state limit of stability. exported to the grid the CHV will become negative, actually
giving a revenue instead of a cost.
3.2 Heat production
The exchange with the HV grid is limited by the size of the
A heat generator is able to produce thermal power only. The transformer of the HV/MV main substation. Defining SHV the
cost function associated with the heat production can be nominal apparent power of the machine the constrain is thus
considered simply proportional to the heat flow, i.e.: given by:

C ( Hg ) = α ⋅ Hg (3.2.1) 2 2
PHV + Q HV ≤ S HV (3.4.2)

where C (C.U./h) is the hourly cost, Hg (MW) is the heat flow

3.5 Distribution network and components
and α is a coefficient depending on the technology (e.g. fuel
cost, efficiency, etc.). The production is limited between a
The optimisation procedure accounts for the distribution grid
maximum value and a minimum value, so gives:
interconnecting loads and generators. Each bus is characterised
by a specified permitted voltage range, whereas each line by a
Hg min ≤ Hg ≤ Hg max (3.2.2) specified ampacity.

3.3 CHP production The loads and generators can be connected to any network bus
and they can either belong or not to the VPP. The loads can be
A CHP generator can produce both heat and electric power, the electric loads, with an active and reactive power demand,
two variables being strictly correlated. As it is usually the case, and/or thermal loads with specified heat demand.
it is here assumed that the independent variable is the heat
flow, thus the active power is given by: DSO control of the primary substation on-load tap changer
transformers, as well as control of the shunt capacitor banks
Pc ( Hc) = ηc ⋅ Hc (3.3.1) connected to the grid, is integrated in the algorithm.

where Hc (MW) is the heat flow, Pc (MW) is the cogenerated

4. VPP Optimisation Algorithm
electric power and ηc (MWel / MWth) is the ratio between
electric and thermal efficiencies. Since the controlled variable The VPP has a direct control on both its generators and loads.
is the heat flow produced the cost function, which includes It is assumed that the information and the data are always
both thermal and electric production, is similar to the heat available, and also that a peer-to-peer communication is
generators cost function: available from/to the DSO. This data communication makes it
possible for the DSO to exploit the (distributed) reactive power
C ( Hc ) = α ⋅ Hc (3.3.2) resources, and also for the VPP to exchange power with the
HV grid wheeled throughout the DSO’s lines. Moreover, it is
In this work it is assumed that the CHP is driven at unity power assumed that the DSO will add on the HV spot prices an extra-
factor (cosφ=1), so that no reactive power can be supplied by cost, accounting for the network losses; this increment is just
the wheeling cost.
280 Bulk Power System Dynamics and Control - VI, August 22-27, 2004, Cortina d’Ampezzo, Italy

The input variables needed to completely describe the state of transformer and on the available shunt compensators. Since
the system are availability of the generators, the production of such actions interfere with, and are in turn affected by, the VPP
renewable resources and the price of the active and reactive decisions an algorithm is here proposed based on the solution
power imported from the grid; these prices are computed and of a further optimisation problem, which is aimed at
then communicated by the DSO. Then, on the basis of the harmonising the operations of both DSO and VPP in order to
actual load demands, the algorithm produces the output achieve the desired regulation with minimum wheeling costs.
variables, which are the set-points for the operations of the
generators (electric, thermal and CHP). Furthermore, the In practice, the DSO would perform a sort of virtual active and
availability of reactive power for each generator is reactive power dispatching of the distributed generators by
communicated to the DSO. solving an optimisation problem aimed at minimising the
network losses which correspond to minimise the cost of the
The chosen VPP optimisation strategy is to minimise the short active and reactive power imported from the HV grid, since the
term variable production cost, which may be expressed by the distributed power generation and demand are fixed boundary
following objective function: conditions. In order to do so, however, the DSO needs
complete knowledge of the system state (generated and
f = ∑ [C j ( Pg j ) + C j ( Hg j ) + C j ( Hc j )] + demanded active and reactive power at any busbar) and cost
functions of the VPP generators. The objective function is thus
j∈JVPP

∑ [ρ P, j ⋅ ( Pd j − Pg j ) + ρ Q, j ⋅ (Qd j − Qg j )]
simply given by:
+ (4.1)
j∈JVPP f = C HV ( PHV , Q HV ) (5.1)

where the first term represents the fuel cost of the generators This similarity with the VPP objective function makes it
that are running at a Pgj, Hgj and Hcj, and JVPP is the set of all possible for the two algorithms to operate together, and co-
buses/devices belonging to the Virtual Power Plant. The ordinately.
second term represents the cost of the active and reactive
power imported from the HV grid incremented by the wheeling There are two kinds of constraints that have to be met. Firstly,
cost. Pdj, Qdj and Hdj. are the load power demands, while the there are the power flow balance equations, at each bus of the
cost coefficients ρP,j and ρQ,j are regularly computed and network, for both active and reactive power, i.e. for a generic
communicated by the DSO (as described in the next section). bus h of a n-buses network:
The constraints functions for the VPP concern the electric n
energy balance and the thermal energy balance, between VPP Pg h − Pd h − Vh ⋅ ∑ Yhk ⋅ Vk ⋅ cos(ϑh − ϑk − γ hk ) = 0 (5.2)
generators and loads. These functions are respectively: k =1

∑ [Pg j + Pc j − Pd j ] + ΔPSL = 0
n
(4.2) Qg h − Qd h − Vh ⋅ ∑ Yhk ⋅ Vk ⋅ sin (ϑh − ϑ k − γ hk ) = 0 (5.3)
j∈J VPP k =1

∑ [Hg j + Hc j − Hd j ] + ΔH SL ≥ 0 (4.3) The second kind of constraints regards the allowable voltage
j∈J VPP levels and the line loadings, that represent for the optimisation
algorithm another set of inequality equations:
where ΔPSL and ΔHSL are slack variables; the former takes into
account the active power exchanged with the MV grid and the Vmin h ≤ Vh ≤ Vmax h (5.4)
active power losses, while the latter represents the heat flow I l ≤ I max l (5.5)
losses, due to heat flow wheeling, which depend on the
characteristics and configuration of the heat network. In addition, the tap-changer positions of the primary substation
transformers and mechanically switched capacitors are further
optimisation parameters characterised by their own bounds.
5. DSO Regulation Algorithm
To be noted that here Pgh and Qgh regarding the production of
The DSO has always to satisfy active and reactive power the VPP are fixed values, while they are optimisation
demands at each bus. It has to manage the network in such a parameters in the VPP optimisation algorithm. Furthermore, as
way to maintain voltage levels inside the permitted range of a result of the optimisation, the present algorithm provides an
variation, and also it has to avoid the emerging congestions, estimation of the nodal prices for active and reactive grid
due to abnormal power flows into the lines which may arise, exchanged power ρP,j and ρQ,j required by the VPP optimisation
for instance, by uncontrollable variations of load demands and algorithm (corresponding, with the assumptions adopted, to the
DG productions. These productions, such as wind energy and Lagrangian multipliers [5]).
solar energy, have a stochastic distribution during the day and
could overload some lines. The VPP is free to adequate its production on the basis of such
At present the DSO may perform regulating actions on the signals, in which case the wheeling costs would be minimised.
network only by acting on the main substation OLTC
Bulk Power System Dynamics and Control - VI, August 22-27, 2004, Cortina d’Ampezzo, Italy 281

The overall optimisation procedure is thus accomplished by Table 1 VPP generators cost data
the iterative application of both DSO and VPP optimisation Bus
Pmin Pmax α β γ
algorithms, where the common variables are alternatively (MW) (MW) (C.U./MW2h) (C.U./MWh) (C.U./h)
3 0 3 3.1 91 1000
treated as optimisation parameters in one algorithm and fixed 4 0 4 1.9 89 1000
boundary conditions in the other. The criterion of termination 5 0 4 2.1 87 1000
is the minimisation of the price values variation of the energy 6 0 3 2.9 93 1000
in the network.
6.1 VPP optimisation
6. Examples of application
As a base case, let us consider first the VPP operating to supply
The procedure has been tested on different configurations of its loads without co-ordination with the DSO, i.e. without
distribution systems with embedded generation, part of which considering the nodal economic indicators, accounting for the
aggregated into a VPP, and it has demonstrated the ability to grid exchange power costs only.
give consistent and reliable results in compliance with all the
network physical constraints and bounds. The overall VPP hourly electric loads demand shown in Figure
2 has been assumed as input data for the optimisation
In the following, a case study example carried out on the algorithm. As a result, the procedure generates the 24-hour
system depicted in fig. 1 is presented. The MV network VPP generators power production dispatching and grid
consists of 5 feeders, 6 dispersed generators (4 of them exchanged power reported in fig. 3, whereas the corresponding
aggregated into a VPP) and 21 loads (5 of them belonging to incremental costs of each production unit ( λ P=δC(P)/ δP ) and
the VPP). VPP generators cost data are listed in table 1, while the overall electrical energy marginal cost are shown in fig. 4.
an HV Grid price for electric energy of 120 C.U./MWh has
been assumed as base unit for comparison with the marginal As it was expected, as long as each generator is producing
costs of the VPP units and with the economic indicators within its limits the unit commitment sets the production in
(Lagrangian Multipliers). order to have all the units running with the same marginal
costs, otherwise the algorithm gives priority to the unit with the
For the sake of clarity, in this example only the electric lowest marginal cost.
network has been considered; further applications of the VPP
optimisation algorithm on systems including thermal Pd (M W h)
production units and CHP plants are available in [6]. 16
It is assumed that the DSO have direct control of the HV/MV
14
substation OLTC transformer and the mechanically switched
shunt capacitors located on bus-bar no. 2 and 7 respectively. 12

10

HV Grid 8
1
6

C2 2
2
0
a b c d e 0 5 10 15 20 25
Hour of the day
8 26 27
15 Figure 2 - Electrical energy demanded by the VPP loads
14 11
17 10 28 29
(MW)
29 4.5

5 4
16
4 18 3.5
DG5
DG4
3 3
13 20 19
DG4 6 2.5 DG6
DG6 2
DG3 DG5
VPP 1.5
22
9 31 30 1 DG3
0.5 Grid
0
24 23 C7 7 21 12 -0.5
0 5 10 15 20 25
Hour of the Day

Figure 3 - 24-hour VPP power production and exchanged

Figure 1 - One-line diagram of case study network power with the Grid
282 Bulk Power System Dynamics and Control - VI, August 22-27, 2004, Cortina d’Ampezzo, Italy

Incremental costs of VPP electrical energy : λ P (C.U./MWh ) B U S V O L T A G E A M P L I T U D E ( p. u. )

125

1.05
120
Overall marginal cost

115 1
DG3 (III)
110 (II)
0.95
DG5
105 DG6
0.9 (I)
100
DG4
# 2 18 5 18 20 19 #
NETWORK BUSES
95
0 5 10 15 20 25
Hour of the Day

Figure 5 Bus voltage profile along feeder c

Figure 4 VPP generators incremental costs and overall (peak load conditions)
marginal cost of electrical energy

Pg (MW)
6.2 Co-ordinated DSO-VPP optimisation
50 Scenarios
(I)
The above VPP operation may result, in the absence of DSO 40 (II)
intervention, in unacceptable voltage levels or congestions in (III)
the distribution network. The DSO is thus called to provide the 30

necessary regulating actions in order to maintain the network 20

state within acceptable levels.
10

In order to demonstrate the effectiveness of the co-ordinated 0

regulation, in the following the results of the procedure applied HV Grid DG3 DG4 DG5 DG6
at two specific network load conditions (respectively peak load
at 9:00 am and minimum load at 3:00 am) are reported.
Qg (M VAr)
Three different scenarios are compared, namely
I) VPP optimisation without DSO regulation
50 Scenarios
II) VPP optimisation with independent DSO regulation (I)
III) Co-ordinated DSO-VPP optimisation 40 (II)
(III)
Peak load condition (h 9.00 am) 30

20
Scenario I - Under peak load conditions, in the absence of any
DSO regulation, the distribution network bus voltages result 10
rather depressed, as can be seen in fig. 5 curve reporting the
0
bus voltage profile along feeder c (the other feeders have HV Grid C2 DG3 DG4 DG5 DG6 C7
similar profiles).

In this case it has been assumed that the VPP generators are
running at cosϕ=1 and all the required reactive power is Figure 6 Active and reactive power production
imported from the HV grid. (peak load conditions)

Scenario II - Application of the DSO optimisation algorithm

enables to determine the necessary actions to be undertaken by Scenario III - The co-ordinated DSO-VPP regulation further
the DSO in order to restore network voltage within acceptable improve the network bus voltages (see curve III in fir. 5). This
levels (see curve II in fig. 5) by adjusting the OLTC is accomplished by a participation of the VPP generators to the
transformer tap position and the shunt capacitors C2 and C7. reactive power production attained by following the indications
of the nodal economic signals (Lagrangian multipliers)
As a consequence, the HV grid imported reactive power is generated by the DSO optimisation procedure. This results in a
reduced (see fig. 6). It can be noted that the VPP units continue reduction of the reactive power imported from the HV grid.
to generate the same active power as in scenario I.
Bulk Power System Dynamics and Control - VI, August 22-27, 2004, Cortina d’Ampezzo, Italy 283

The nodal economic indicators of the “active” nodes of interest B U S V O L T A G E A M P L I T U D E ( p. u. )

for the three scenarios are reported in fig. 7 below.
1.05 (I)
ρP (C.U./MWh) (II)
1 (III)
140 Scenarios
(I)
135 (II) 0.95
(III)
130
0.9
125
# 2 18 5 18 20 19 #
120 NETWORK BUSES

115
HVgrid DG3 DG4 DG5 DG6
Figure 8 Bus voltage profile along feeder c
(minimum load conditions)
Pg (MW)
ρQ (C.U./MVArh)

Scenarios 50 Scenarios
40 (I)
(I)
40 (II)
35 (II)
(III) 30
(III)
30
20
25
10
20
0
15 HV Grid DG3 DG4 DG5 DG6
HVgrid C2 DG3 DG4 DG5 DG6 C7

Figure 7 Nodal economic indicators of the “active” network Qg (MV Ar)

nodes computed by the DSO (peak load condition)

25 Scenarios
(I)
20 (II)
Minimum load condition (h 3.00 am) (III)
15
Figs. 8 to 10 show the results, similar to the above, of the
10
comparison between the three different scenarios considering
the network under minimum load conditions. 5

In this case the network voltages without DSO regulation 0

HV Grid C2 DG3 DG4 DG5 DG6 C7
resulted unacceptably high, as can be seen in fig. 8 (curve I).
Again, it can be seen that the co-ordinated DSO-VPP
regulation results in a better voltage profile with an overall
lower energy cost. Figure 9 Active and reactive power production
(minimum load conditions)
It should be noted that, following the nodal economic
indicators, the VPP has substantially modified the active power
production of its generating units.
ρP (C.U./MWh)

140 Scenarios
(I)
135 (II)
(III)
130

125

120

115
HVgrid DG3 DG4 DG5 DG6
284 Bulk Power System Dynamics and Control - VI, August 22-27, 2004, Cortina d’Ampezzo, Italy

ρQ (C.U./MVArh)

Scenarios [2] M. W. Davis, “Distributed Resource Electric Power

40
(I) Systems Offer Significant Advantages Over Central
35 (II) Station Generation and T&D Power Systems”, IEEE
(III) Power Engineering Society Summer Meeting, Piscataway
30
(NJ – USA), September 2002.
25
[3] E. Hendschin, F. Uphaus, Th. Wiesner, “The Integrated
20 Service Network as a Vision of the Future Distribution
Systems”, International Symposium on Distributed
15
HVgrid C2 DG3 DG4 DG5 DG6 C7 Generation: Power System and Market Aspects, Royal
Institute of Technology, Stockholm (SVE), 11-13 June
2001.
Figure 10 Nodal economic indicators of the [4] A. Stothert, O. Fritz, M. Sutter, “Optimal Operation of a
“active” network nodes computed by the DSO Virtual Utility”, International Symposium on Distributed
(minimum load conditions) Generation: Power System and Market Aspects, Royal
Institute of Technology, Stockholm (SVE), 11-13 June
2001.
[5] F. C. Schweppe, M. C. Caramanis, R. D. Tabors, R. E.
7. Conclusions Bohn, Spot Pricing of Electricity, Boston (MA – USA),
Dordrecht (NL), London (UK), Kluwer Academic
Even in a liberalised electric energy market, the independent Publishers, 1988.
actors have to operate on the same physical network and their
actions, which necessarily influence each other, should be [6] R. Caldon, A. Rossi Patria, R. Turri, “Optimisation
coordinated. algorithm for a Virtual Power Plant operation”, Proc. 39th
International Universities Power Engineering
This work is an attempt to provide a suitable framework for Conference, UPEC2004, Bristol (UK), sept. 2004
harmonising (on a voluntary basis) the operations of different [7] J. Mariyappan, J. Gregerson, “Advanced in Distributed
subjects, DSO and independent producers aggregated into a Energy Control and Communication Systems”,
VPP, which should be able to operate in an autonomous International Symposium on Distributed Generation:
manner, both taking decisions and making profit although Power System and Market Aspects, Royal Institute of
complying with the required power quality levels and physical Technology, Stockholm (SVE), 11-13 June 2001.
network constraints.

The approach is based on cost function models of the system

components, accounting for the network constrains, and an
effort has been made to keep the highest generalisation in order
to include, in principle, any kind of cost, objective function and
type of constrain.

The proposed algorithm, although of not immediate application

since it requires future development of sophisticated and
reliable control and communication systems [7], would
provide the independent producers useful economic signals for
carrying out their own management complying with the
network requirement, without affecting their decision making
freedom.

8. References

[1] H. L. Willis, W. G. Scott, Distributed Power Generation

Planning and Evaluation, Amsterdam (NL), Marcel Deker
Edition, 2000.

View publication stats