You are on page 1of 4

CIRED 18th International Conference on Electricity Distribution Turin, 6-9 June 2005

A NOVEL METHODOLOGY FOR MULTI-YEAR PLANNING


OF LARGE-SCALE DISTRIBUTION NETWORKS

Zeljko N. Popovic
EPS Electric Power Industry of Serbia, Serbia&Montenegro
zeljko.popovic @su.ev.co.yu
Dragan S. Popovic
School of Technical Sciences, University of Novi Sad, Serbia&Montenegro
Vojin Dj. Kerleta
School of Technical Sciences, University of Novi Sad, Serbia&Montenegro

optimization), they do not provide any assurance that the best


Summary - This paper presents a new optimization procedure solution will be found and do not handle constraints well.
for multi-year distribution network planning based on
dynamic programming, local network concept and mixed In this paper is proposed a new decomposition procedure for
integer model. By introducing principles of dynamic multi-year planning of large-scale distribution networks
programming and local network concept the multi-year based on dynamic programming, local network concept and
planning problem of real-size networks is divided into mixed integer model. The proposed procedure is developed in
sequences of sub problems with significantly reduced order to overcome above-mentioned problems.
dimensions, enabling application of complex mixed integer
programming model in each sub problem. The results Based on the principle of dynamic programming the multi-
obtained on several real-life distribution networks have year planning problem is embedded within a family of similar
shown that suggested procedure enables more efficient use of one-stage problems. Furthermore, according to the local
existing capacities and reduction of future expansion costs. network concept the whole network at each stage is divided
into a number of local networks (subproblems). In this way
dimension of the overall problem is drastically reduced
1. INTRODUCTION enabling application of sophisticated and time-consuming
mixed integer model at each subproblem. This model
minimize capital costs of future design, i.e. determines
Deregulation and competitive marketplace enforced optimal reinforcements of existing feeders as well as location
completely new conditions in power industry. Even though a and size of new feeders (decision variables) taking into
distribution company may still operate within regulatory account switching (load transfer) capabilities and operational
framework, it is faced with financial pressures and noticeable constraints in each local network (subproblem). Thus, the
decrease in available capital budgets. In such conditions optimal one-year network design is obtained by sequential
distribution planning process becomes business-driven (rather solving of a number of defined (obtained) local networks
than engineering driven) with the aim of reducing capital while the optimal multi-year planning scenario is obtained
costs of future enhancements and changes in distribution by sequential solving of a number of obtained one-stage
networks. In this sense, there is a strict demand for efficient problems.
planning tools that enable maximal utilization of existing
capacities in distribution network, i.e. finding a way to work A software package, based on proposed procedure, has been
around operational limitations [1]. developed and tested on several real-life distribution
networks. Obtained results prove that suggested procedure is
The main difficulty in developing of a high quality planning a powerful planning tool which can significantly improve
tool is large dimension of distribution networks and the fact utilization of existing capacities and minimize future
that this problem is highly constrained. Necessity to include expansion costs.
dynamic in solving planning problems makes it even more
complex. For solving planning problems several approaches
are used: 1) optimization methods [2-8]; 2) heuristic methods 2. PLANNING PROCEDURE
[9-12]; 3) artificial intelligence (AI) based methods [13,14].
The main feature of optimization methods is the convergence
to the global optimal solution. However, these methods can Multi-year distribution network planning problem can be
hardly be applied to real-size networks because of stated as an allocation problem in which "available recourses"
convergence problems and extensive computer processing. (budget) have to be allocating among the "activities"
The heuristics algorithms can obtain "good" solution for real- (reinforcements and new facilities in each year in the
size problems, but global optimal solution cannot be considered planning period) in such a way that present worth
quarantined. Although AI methods have several advantages expansion costs in the considered planning horizon (objective
(they are robust, flexible, do not require "well behaved" function) are minimized while operational constraints are
objective functions, can be easily applied for multi-objective satisfied. The dynamic programming based procedure for

CIRED2005

Session No 5
CIRED 18th International Conference on Electricity Distribution Turin, 6-9 June 2005

solving the above allocation problem in distribution networks cases (paths) will be analyzed in more details.
is described in the following sections.

2.1. Determining one-stage planning scenarios 2 3


1 DPSC1

In the first step of the procedure the planning problem for the 3 2
DPSC2
network is solved for each year in the considered planning
horizon. The starting state of the network for determining
planning scenarios in each year is the network configuration 1 3
2 DPSC3
of the year t=0 (base year), as it is shown in Fig.1. The set of
obtained reinforcements and enhancement in year j are named 3 1
DPSC4
PSCtj.
1 2
DPSC5
3
2 1
DPSC6

t=0 PSCt1 PSCt2 PSCt3 PSCtn Fig. 2. Decision tree for three-year planning problem

.
Fig. 1. Determining one stage planning scenarios
Path (1,2,3) define that planning problem is first solved for
year t1. The obtained solution (obtained set of reinforcements
After determining planning scenarios for each year, three and enhancements) becomes the starting state of the network
possible cases may arise: and on the basis of that solution the planning problem for
a) the set of reinforcements and enhancement obtained for year t2 is solved. Afterwards, based on the solution for the
year t is the subset of the set of reinforcements and year t2 the planning scenario for the year t3 is obtained. In this
enhancements obtained for year (t+1), starting from the first way path (1,2,3) define one possible solution for three-year
year (t=1) and up to the last year (t=n) in the planning planning problem (marked as DPSC1 in the Fig. 2). This way
horizon (PSCt1 ⊂ PSCt2 ⊂ ....⊂ PSCtn), of solving multi-year planning problems is well-known as
b) the previous statement is fulfilled for only first m years in forward concept (forward fill-in) [1]. The next path which
the planning horizon i.e., the set of reinforcements and will be analyzed is (3,2,1). In this case the planning problem
enhancements obtained for the first m years is common for all is first solved for the year t3. The obtained set of
years that follow year m, reinforcements and enhancements in this case is marked as
c) neither of the previous statements are fulfilled. PSCt3. After that a planning procedure for year t2 is solved
In the first case the n separately obtained one-stage planning taking into account that available (feasible) set of
scenarios represents the optimal solution of multi-year reinforcements and enhancements in this year is restricted to
planning problem. In the second case optimal planning PSCt3. It leads to the solution in which obtained set of
scenario for the first m years is obtained as in the first case. In reinforcements and enhancements for year t2 (PSCt2) is a
this way the multi-year planning problem is reduced from n- subset of PSCt3 (PSCt2⊂PSCt3). The solution of the planning
year problem to (n-m)-year planning problem. In this case as problem for the year t1 is obtained in the similar way, bearing
well as in the third case the dynamic programming based in mind that available set of reinforcements and
procedure is applied for obtaining the best multi-year enhancements in this case is restricted to PSCt2. Thus, path
planning scenario. (3,2,1) define another possible solution for three-year
planning problem (marked as DSPSC6 in the fig. 2) which is
2.2. Dynamic programming procedure known as backward pull-out [1].

In order to visualize dynamic programming procedure, the The rest of the paths are very similar regarding to the way of
solution of the above-mentioned allocation problem is obtaining multi-year planning scenarios. Because of that only
presented in the form of decision tree. For the sake of clarity, one of them, path (2,3,1), is thoroughly analyzed. In this path
an example of a three year distribution network planning the planning problem is firs solved for the year t2.
problem that is considered. The decision tree for this case is Afterwards, based on that solution, the solution of the
shown in the Fig. 2. planning problem for the year t3 is obtained. Finally, bearing
in mind that everything which is done in the year t1 must also
Sequential solution of planning problem for each year is be in operation in the year t2, the planning problem for year t1
depicted with empty circle and numbered with Roman is solved by restricting available set of reinforcements and
numbers in the Fig. 2. Each path in the decision tree defines enhancements to the PSCt2.
one possible solution for the multi-year planning problem. Maximum number of paths in the decision tree, i.e. number
These solutions are obtained by different sequence of solving of different multi-year planning scenarios, for overall
planning problems for all years in the planning period. In network is n!, where n is the number of one-stage problems
order to clarify the proposed procedure three characteristic (number of considered time intervals (years)). However,

CIRED2005

Session No 5
CIRED 18th International Conference on Electricity Distribution Turin, 6-9 June 2005

some paths will lead to the same multi-year planning Table 1. – Multi-year planning scenarios and costs
scenarios. Thus, the number of different dynamic planning Dynamic planning Present worth expansion
scenarios could be much smaller than n!. scenario (fixed) costs (million $)
DPSC1≡ DPSC2 0.8480
One-stage planning problem for the whole network in each DPSC3≡DPSC4 0.6905
path is solved by inclusion of the local network concept [15]. DPSC5≡DPSC6 0.7226
According to this concept the whole network is divided into a
number of local networks (subproblems).In this way The best three-year planning scenario that leads to the
dimensions of the original one-stage problem are significantly minimal present worth expansion costs is DPSC3. The
reduced, enabling application of time-consuming mixed obtained planning scenario for the first year is shown in the
integer model for solving planning problems for each local Fig. 4, for the second year in the Fig. 5 and for the third year
network (subproblem). Thus, the optimal network design for in the Fig. 6. In these figures, obtained reinforcements and
one-stage problem is obtained by sequential solving of a new constructions in each year are presented with bold lines,
number of defined subproblems, i.e. the dynamic so the total number of reinforcements and new constructions
programming concept is also applied in this case as it is in the three-year period is the “ sum” of all bold lines.
presented in details in [8].

Ranking of obtained multi-year planning scenarios as well as


obtained one-stage planning scenarios and selection of the
best one is done by applying minimum expansion costs
criterion, i.e. planning scenario with minimum expansion
costs is selected as the best expansion design.

3. APPLICATION

Proposed procedure is applied on solving three-year planning


problem for medium voltage network shown in the Fig. 3.
The network consists of 4 feeders, 17 existing branches (solid
lines), 6 possibly new branches (dashed lines) and 1 possibly
new tie (dash-dot-dot-dash line). Thermal capacities of all
branches are shown in p.u. with bold numbers. The lengths of
branches that are longer than 1 km are shown in the Fig. 3.
All other branches are 1 km long. It is assumed that there is a
switch in each branch. Normally opened switches are marked
with "X" in the middle of branch where they are located. The Fig. 3. Test network in the base year
15 existing demand nodes are depicted with empty circles
while 2 future demand nodes are shown as full circles.
Demand node XVII will appear in second year while demand
node XVI will appear in third year. The original loads in the
base year for all demand nodes are presented in p.u. Based
value for thermal capacities and demand loads is 10 MVA.
The load growth rate is 3% by year for all demand nodes.
Cost of building new lines as well as reinforcing existing
lines is 0.145 million$/km while discount rate is assumed to
be 9%.

In the first step, the one-stage planning problem for each year
in the planning horizon is obtained. After analyzing the
obtained set of reinforcements and enhancements in each
year, it is found that (PSCt1 ⊄ PSCt2 ⊄ PSCt3). Decision tree
for this case is the same as the one shown in the Fig. 2.In this
case only three different multi-year planning scenarios are
obtained. These scenarios and corresponding present worth
expansion costs are given in the Table 1.
Fig. 4. Optimal planning scenario in the first year

CIRED2005

Session No 5
CIRED 18th International Conference on Electricity Distribution Turin, 6-9 June 2005

planning of real-life distribution networks and can


considerably improve efficiency and reduce expansion costs
in the modern (deregulated) distribution systems.

5. REFERENCES

[1] H.L.Willis, 1997, Power Distribution Planning


Reference Book, Marcel Dekker Inc.,New York, USA.
[2] M.A. El-Kady, “ Computer-Aided Planning of
Distribution Substation and Primary Feeders”, IEEE
Trans. on PAS, Vol. 103, no. 6, pp. 1183-1189, 1984.
[3] T. Gonen and B.L. Foote, 1981, “ Distribution System
Planning Using Mixed Integer Programming”, IEEE
Proc-C, Vol. 133, no. 2, 70-79.
[4] O.M. Mikic, 1986, “ Mathematical Dynamic Model for
Long-term Distribution System Planning” , IEEE Trans.
on Power Systems, Vol. 1, no. 1, 384-390.
[5] T. Gonen, I. J. Ramirez-Rosado, 1986, “ Review of
Distribution System Planning Models: a model for
optimal multi-stage planning”, IEE Proc-C, Vol. 133, no.
Fig. 5. Optimal planning scenario in the second year
7, 397-408.
[6] I.J. Ramirez-Rosado and T. Gonen, 1991,
“ Pseudodynamic Planning for Expansion of Power
Distribution Systems”, IEEE Trans. on Power Systems,
Vol. 6, no. 1, 245-253.
[7] Z. N. Popovic, D. S. Popovic and M. D. Prica, 2001, “ A
Model for Development Planning of Distribution
Networks”, Proceedings of the 36th Universities Power
Engineering Conference UPEC 2001.
[8] Z. N. Popovic, D. S. Popovic, 2003, "A Novel
Decomposition Procedure for Distribution Network
Planning",in Proceedings of the 38th Universities Power
Engineering Conference UPEC 2003.
[9] K. Nara, et. al. , 1991, “ Multi-Year Expansion Planning
for Distribution Systems”, IEEE Trans. on Power
Systems, Vol. 6, no. 3, 952-958.
[10] K. Nara, et. al. , 1992, “ Distribution Systems Expansion
Planning by Multi-Stage Branch Exchange” , IEEE
Trans. on Power Systems, Vol. 7, no. 1, 208-214.
[11] S.K. Goswami, 1997, "Distribution System Planning
Using Branch Exchange Technique", IEEE Trans. on
Power Systems, Vol. 12, No. 2, 718-723.
[12] H. Kuwabura, K. Nara, 1997, "Multi-Year and Multi-
Fig. 6. Optimal planning scenario in the third year
State Distribution System Expansion Planinng by Multi-
Stage Branch Exchange", IEEE Trans. on Power
Delivery, Vol. 12, No. 1, 457-463.
4. CONCLUSION [13] V. Miranda, et. al.,1994, “ Genetic Algorithms in
Optimal Multistage Distribution Network Planning”,
IEEE Trans. on Power Systems, Vol. 9, No. 4, 1927-
In this paper is proposed an optimization procedure for multi- 1933.
year planning of large-scale distribution networks. The goal [14] I. J. Ramirez-Rosado and J. L. Bernal Agustin, 2001,
of the proposed procedure is to maximally utilize existing "Reliability and Costs Optimization for Distribution
capacities and minimize present worth expansion costs while Networks Expansion Using an Evolutionary Algorithm",
satisfying operational constraints (voltage and thermal IEEE Trans. on Power Systems, Vol. 16, No. 1, 111-118.
constraints). This goal is achieved through application of [15] D.S. Popovic, R.M. Ciric, 1999, "A Multi-Objective
dynamic programming principles, local network concept and Algorithm for Distribution Network Restoration", IEEE
mixed integer model. It is shown that proposed procedure is a Trans. on Power Delivery, Vol. 14, No. 3, 1134-1141.
powerful decision-making tool for multi-year expansion

CIRED2005

Session No 5