You are on page 1of 12

1922

EEE Transactions on Power Delivery, Vol. 6, No. 4, M o b e r 1991


FEEDER RECONFIGURATION FOR LOSS REDUCTION: AN APPLICATION OF D I S T R I B U T I O N AUTOMATION

T.P. Wagner, A.Y. Chikhani, Senior Member D-t en of Electrical and Com$mter Enqineerinq RGyal Military College of C n d aa; Kingston, Ontario, Canada
KEYKXDS: Distribution Automation, Feeder ReconfigurMinimization, Linear Programing, ation, Loss Heuristics.

R. Hackam, Fellow Dept. of Electrical Engineering University of Windsor Windsor, Ontario, Canada

ABslRAcT

configurations would provide the m s t accurate and reliable solution, this is unrealistic as it takes a long time for a real-time application. various methods have been proposed to solve for the minimum loss configuration in acceptable t k s , each method having different advantages and disadvantages. Linear programing techniques have been used [3-51 in planning applications to minimize the costs associated with new construction. Although it was the capital or fixed costs which were considered to be of primary importance, an attempt was often r a e to rd Branch and include the minimization of 12R losses. Bound nethods have also been used to minimize losses in proposed distribution systems 161. However, when applied to distribution systems, it is difficult to determine the upper and lower bounds effectively and t k consuming for real-time the method often &oms applications. Heuristics techniques have been proposed [6-111 for finding the optdl, or near-opthl solution with fast computation t k s . In a method developed by Merlin and Back[6] and later modified by ShilmoharraMdi and Hong [ 7 ] a low loss configuration is determined by applying an optin-al load flow analysis to the system with all switches closed. The system is returned to a radial configuration by opening the branches with the lowest current, the o p t d l load flow pattern being determined after each switch is opened. Civanlar et a1 [ 1 2 ] derived a formula for estimating the loss reduction which would result from carrying out a e i c u l a r switching option. A switching option is defined here as the closing of one open tie-switch and the opening of one of the closed sectionalizing n switches i the loop which is forred, the net result being the transfer of a block of load from one feeder to another. In the present work, a col-nparison of the m r e promising methods of real-time reconfiguration for loss reduction is conducted. Linear Programing methods using a Stepping Stone algorithm [13] is, for the first t k , applied to the feeder reconfiguration problem. TWO Heuristic methods are also compared. The first Heuristic method is that of Merlin and Back [6] as inplenented by S h i m h m d i and Hong [7]. The second Heuristic Ethod, presented here, employs the Heuristic rules developed by Givanlar et a1 [ 1 2 ] . TWO subroutines are used in this method to provide an esthte of the loss reduction which would result from carrying out a particular switching option. The first subroutine uses the loss reduction formula developed by Civanlar et a1 [12]. The second subroutine mkes use of a uniformly distributed load model [13] which l a d l most distribution systems m r e accurately. roes The different reconfiguration techniques are applied to simulations of two different distribution systems. The first system is a three feeder distribution system used by Civanlar et a1 to illustrate their loss reduction formula, and the second system consists of a shulation of the m r e complex Kingston Public utilities Comnission (P.U.C.) 44 kv, 150 MW distribution system.

One of the features provided by Distribution Automation, which can result in substantial savings for the utility, is feeder reconfiguration for loss reduction. This work provides a comparison of various methods applied to feeder reconfiguration for loss minimization. A new linear p o r n rg&g method using Transportation techniques and a new Heuristic Search method are proposed for camparison with a previously developed Heuristic technique which was based on an Optimal Load Flow analysis. The methods are compared on simulations of both a small feeder distribution system, and a larger system based on a model of the Kingston Public Utility Coxmission 44 kV distribution system. This study indicates that Linear Programing, in the f o m of transportation algorithms, is not suitable for real-time application to feeder reconfiguration whilst Heuristic approaches, although not opthl, can provide substantial savings if properly formulated and are suitable for real-time iqlemntation.

l3mRmKrIoN
It was estimated that from 5% to 13% of the total power system generation is wasted in the form of 12R losses at the distribution level [ l ] . Recent advances in distribution automation technology have made it possible to reduce these losses by applying loss minimization techniques on a real-time basis. A project by the Pennsylvania Power and Light Company [ 2 ] simulated the application of loss reduction techniques to a portion of their system for the period of one year. The results showed an estimated reduction in losses of 14.6% over the one year period on a system with a peak of 230 I . N Distribution systems supply power to three basic load types, residential, comrcial, and industrial. The load profiles for each of these load types is different, causing the distribution feeders to become m r e heavily loaded at certain times of the day and less heavily loaded at other times; each feeder varying in a different m e r depending on the This shifting of the characteristics of its loads. system loads can be used to minimize, or at least reduce the system losses by reconfiguring the system from tine to t h e in order to redistribute the load currents more efficiently. Much of recent research on Distribution Automation has focused on the minimum-loss configuration problem. While an exhaustive search of all possible
91 WM 101-6 PWRD A paper recommended and approved by the IEEE Transmission and Distribution Committee of the IEEE Power Engineering Society for presentation at the IEEE/PES 1991 Winter Meeting, New York, New York, February 3 , 1991. Manuscript submitted July 30, 1990; made available for printing November 30, 1990.

LINEAR

Linear programing mthods which are used by distribution system planners have their p r w goal

018-9464/91$01.00 0 1991 IEEE

1923

to minimize the capital costs involved in constructing new systems or expanding existing ones. Often, attempts were m d e to include the cost of 12R losses in their calculations. It is therefore worthwhile to investigate these nethods to determine their suitability in solving the feeder reconfiguration for loss minimization. Linear programing is a mathematical optimization technique developed by George Dantzig in 1947. The problem is formulated as the optimization of an objective function, subject to a set of constraints. The distribution reconfiguration system is best expressed as a transportation problem. Transportation systems are expressed as:

D1

D2

D3

D4

D5 D6

D7

D8

D9

a,

s1

s2
s 3
bj

Fig. 1
. .

Transportation Tableau

Subject to the Constraints:

i=l j=1 nrn

E Xij I ai

(2)

previous change. It is only necessary to recalculate the element from which the circled element was moved. In order to increase the accuracy of the cost factors Cij, the non-linear 12R function can be linearized in a piecewise rranner. The linearization of the objective function can be done in a stepwise or piecewise linear manner as illustrated in Fig. 2. The piecewise linear method, selected for use in this work, is accomplished by assigning a cost factor equal

For the feeder reconfiguration case the pxameters are defined as follows: ai bj m u n t of power being supplied by source i amount of power demanded by load point j x . = power supplied from source i to load point j cf2 = cost factor associated with delivering Xij 2 = losses incurred in delivering the total load n = number of feeder sources m = n m h r of load points
= =

One transportation method which is suitable for the distribution system is the stepping stone A typical tableau for solving algorithm [13]. transportation problems using the stepping stone The tableau, algorithm is illustrated in Fig. 1. organized in this m e r , is useful in pre-processing the information used in the transportation algorithm process. Notice that there are two types of entries in the tableau. The circled entries represent basis A basic solution is vectors in the basic solution. any feasible combination of the basis vectors which completely satisfies the constraints of the problem. The values of the circled entries represent existing loads which are fed from the feeder sources according to the row in which they are located. The tableau therefore provides information on not only the m u n t of the load currents being demanded, but also on the present configuration. The second type of entry is the uncircled entry. These entries are calculated by applying the stepping stone algorithm. Any positive, non-zero element indicate that the system is not opti.mil,and that losses can be reduced by eliminating that element from the basic solution. In practical tern, this m u n t s to transferring the load represented by that element, to another feeder.
Some features of radial systems siniplify the solution of transportation problems. Meshed networks, i.e., networks where each load m y be fed from m r e than one source, can have m r e than one circled element in each column. In order to r a v e a positive uncircled value it is no longer a siple rratter o f w i n g one element. The method requires a further calculation to determine the best way to r a v e the positive elemnt. Radial networks do not require these extra calculations. It is also much sinpler to recalculate the uncircled element values for radial networks because fewer elements were involved in the

POWER FLOW Fig. 2 Piwewise Linear Approximation of Lasses

to the slope of the straight line sections approximating the quadratic curve, over each range of load currents. The degree of accuracy may be enhanced sinply by increasing the number of load ranges, however, computational effort will also be increased.

The method developed by Merlin and Back [6], and later modified by Shirmhammdi et a1 [ 7 ] , is primarily based on Heuristics or rules-of-thumb, with an additional subroutine added to ensure that no arrent overloads or law voltage conditions will occur due to any proposed switching. The method has an advantage over s c w other techniques in that it is independent of the initial configuration. This means that it is m r e likely to find the global optimum and not converge to some local minimum. The basic method is described in the flow diagram illustrated in Fig. 3 . The first step in the algorithm is to read in the system data. This includes static data such as feeder line impedances as well as dynamic information such as system configuration and voltage, current, and power values. The radial distribution system is then modified to a meshed system by assuming that all open tie switches have been closed. Using

1924
AND SWlTCH DATA

CLOSE Au OPEN S W C H E S CREATlNC A MESHED NE'ITWRK

C'LOSE THE LMT SWlTCH OPEND OFN THE SWlTCHCARlWNG THE NEXT WMsT

CU

AS DETERMINEDBY THE 0mM.U LOADMWANMYSIS

The Heuristic search method is used to find the optimal or near optimal configuration for a distribution feeder. It employs two different subroutines to calculate system loss reductions. Generally the method works by considering all possible switching options carrying out the option which provides the greatest loss reduction and then repeating the procedure until no further loss reductions are possible. An additional subroutine is included to ensure that no constraint violations would occur due to carrying out the proposed switching. It is a s s d that no violations exist before the loss minimization algorithm is applied and that if such a condition did exist, other routines such as load balancing or CrAR control would be used to rectify the problem. If a configuration is investigated and found to cause under-voltage or over-voltage conditions, then that option is disregarded. A flow diagram of the mthcd is shown in Fig. 4.

YEJ

Fig. 3

Optimal Load Flow Method

the current system data, an ac load flow is p e r f o m to determine the actual load currents. This step is required to represent the loads as constant power loads, such that load currents will vary with changing voltages. The load currents calculated in the ac load flow are then used in the optimal load flow to determine individual line currents. The optimal load flow is similar to the ac load flow only with line section reactances set to zero. This procedure gives the o p t b l load flow pattern, in terms of minimal I2R losses for the given loads and system configuration. The o p t h l load flow results provide the optimal currents, in terms of minimal losses, for the mshed system. In order to find the radial system which will supply all of the loads with minimal losses, some of the switches must be opened. In order to disturb the system the least mount, the switch in the section with the lowest current is opened (assuming that the switch f o m part of a loop feed). Once this switch has been opened, the load flow is repeated, and again the section with the lowest current is opened. This process is repeated until the system is in a radial configuration

Fig. 4

Heuristic Search Method

An additional subroutine is used to ensure that no violations of voltage or current constraints would occur due to the resulting configuration. After each load flow analysis, all voltages and currents are checked against system limitations. If a violation has occurred, the last switch to be opened is closed and the section with the next h e s t current is then opened.

In order to reduce the number of options which must be considered, two Heuristics have been incorporated into the program. It has been shown [12] that in order to reduce the losses on a distributed system, it is necessary to transfer loads fromthe feeder on the low voltage side of the open switch to the feeder on the high voltage side of the switch. Therefore, when considering the open tie-switch, it is necessary to transfer loads from the side with the greatest voltage drop from the substation to the switch, over to the side with the lower voltage drop from the source to the switch. This is done by closing a normally-open tie-switch and opening one of the closed sectionalizing switches in the loop feed which has been formd. The second rule states that a reduction in losses is only possible if there is a substantial voltage drop across the open switch. By employing these two Heuristic rules, it is possible to reduce the number of options which must be considered. The mthod begins by considering each open switch as an option. The Heuristic rules are then applied. If the voltage drop across the open tie-switch is negligible, then that option is disregarded, as are all of the options which branch fromthat option. If a large voltage drop does exist, then the switch is considered closed and all the possible options which branched fromthis decision are then tested. All the possible switches which could be used to break the parallel circuit are considered in turn. It is only necessary to consider opening switches on the side of

It is assumed that no voltage or current violations exist on the system prior to application of the reconfiguration process. If such violations do exist, other programs such as load shedding or voltage correction would be mployed to correct them.

12 95 the tie-switch with the largest voltage drop from the substation to the switch, which is obviously the low voltage side of the switch. Once the number of switching options has been reduced by the application of the Heuristic rules, the rmining options are evaluated to determine which option would provide the greatest loss reduction. This option is then tested by performing a load flow analysis on the proposed configuration to ensue that no over-current or low voltage conditions would result from carrying out the switching option. If these conditions are discovered, then that switching option is disregarded and the switching option providing the next greatest loss reduction is considered. Once a feasible switching option has been selected, the switching is considered to have been carried out. The entire process is then repted until no further loss reductions are possible. The minimum loss configuration having been determined, switching is then carried out on the actual system to reflect the proposed optimal ( o r near-optimal) configuration.
T o methods are used to provide the estimate of w the reduction in losses for each possible configuration. The first one is based on a formula developed for calculating the loss reduction using the km n load currents, tie-switch voltage, and feeder resistance values [12]. The second method used to calculate the reduction in losses is based on the uniformly distributed load model [13].

the length of the section. The use of lumped load models artificially inflates the value of the calculated system losses. In addition, the o p t h l configuration may not be the s m as the configuration which would result if the system were modelled m r e accurately. The second weakness is that an esthte of the total system losses is not provided, only an esthte of the loss reduction. Although an esthte of total losses is not required for the method to work, the information is certainly desirable for budgetary reasons. In order to provide total system loss estimates using this technique, a separate subroutine must be included which reduces the speed advantage of the method. Uniformly Distributed Load Model Although the total load d m d e d from each feeder section is knwn to the Distribution Automation system, the distribution of that load over the feeder length is not. Rather than d e l the load as lumped at the feeder section end, a unifody distributed load d e l [13] is used in order to more accurately estimate system losses. Estimates of total system losses before and after the proposed switching change are used to calculate the loss reduction. Although this method requires m r e computations, it has the advantages of greater accuracy and also providing a value for the total system losses and loss reduction. The uniformly distributed load d e l is illustrated in Fig. 5.
I2

m s Reduction F o m l a The loss Reduction Formula [12] estimates the reduction in losses which would occur when a particular switching option is carried out. The formula requires voltage and current information only at the locations of the tie and sectionalizing switches. The formula is given as:

Id

DOWNSTREAM LOADS

where :
D

(4)

Fig. 5

Uniformly Distributed Load Model

is the set of buses which are disconnected from Feeder I1 and connected to Feeder I. is the tie bus of Feeder I to which loads m from Feeder 11 will be connected. is the tie bus of Feeder I that will be n connected to bus m via a tie-switch. is the Complex bus current as bus i. Ii is the series resistance of the path connectRloop ing the two substation buses of Feeder I and Feeder I1 via closure of the specified tie switch. is the component of E = Qus 1~~~correspondE , ing to bus m. Qus is the bus resistance matrix of Feeder I before the load transfer which is found using the substation bus as reference. 1~~~ is the vector of bus currents for Feeder I. is similar to Embut defined for bus n of En Feeder 11. Ret 1, *, I I are the real part, complex conjugate, and magnitude operators, respectively. The formula provides a fast method of esthating the reduction in losses, however, it has t w weaknesses. The method is based on modelling the feeder section loads as being lumped at the feeder section ends. This is an unrealistic model for m s t distribution systems as loads are usually tapped off of each feeder section at a number of locations along

In a typical distribution automation system, monitors at each switch location would provide the values of I1 and 12, from which Id can be calculated. Losses over the feeder section are then calculated by:
I

12R

3
0

[ I1 - I1x + 12x]

WX

where x is the length of the feeder section. Solving for the integral, and given that I = 11 d 12, we get an expression for feeder section losses in terms of load d m d e d at that section (Id), y d current emanating from the end of the feeder section (12).

This qression is then used to calculate the losses in each feeder branch, which can then be sunned to provide an estimate of total system losses. total cmtation time for estimating system losses will increase sht -a over the lumped load model, however, the extra effort is justified by the increased

1926

accuracy of the esthte of losses.

Tw(EEFEEDER?EsTsysIEM
The first system selected for comparing the various reconfiguration schemes is the three-feeder distribution system used by Civanlar et a1 [ U ] to illustrate their loss reduction formula. The system consists of 3 feeders, 13 nomlly closed sectionalizing switches, 3 normally open tie-switches, and 13 load pints. Feeder section impedances, system loads, and system configuration are illustrated in Fig. 6. This particular system was chosen as it provides a sinple, well defined system for which the optimal configuration is easily determined. The system is used as described, with the loads assumed to be constant as specified.

realistic system in which the loads vary with t m . In this way, the amount of savings which would result from each method can be compared. It is therefore necessary to establish load flow patterns for each of
FEEDER FEEDER
FEEDER

FEEDER

EEDER

FEEDER

33M9

FEEDER 1

LDADCEKIFR

--- OPEN SwrrcH

Fig. 7

Kingston P.U.C. 44 kV Distribution System

the three load types, residential, cmrcial, and industrial.


A study conducted by the Pennsylvania Power and Light Co. monitored loads on eight distribution feeders for a period of one year [ 2 ] . By emmining the recorded data, it has been determined that in order to &l e the load changes on a distribution system over a one year period, only 18 different daily load profiles are required.

Fig. 6
K" l SYSlxM

Three Feeder Distribution System


-T IY

-SS1m

44

DISIRIEWTIJ

The year is divided into three seasons; winter s m r and spring/fall. The mnths of the year are divided into the three seasons as follows:
- June, July, August - December, January, February

The second system used to test the three feeder reconfiguration nethods is the Kingston P.U.C. 44 kV distribution system shown in Fig. 7. This system consists of 6 feeders, 38 normally o p n tie-switches, and 26 load pints with a peak demand of 150 MW. The loads are prim&ily a d u r e of residential and comrcial with some light industrial. This system has been selected as a l l the relevant data have been given to the authors as well as it provides a system of w a g e able size with a reasonable number of opentieswitches for reconfiguration.
LQADM3DELFtXTHE44kVSYS'X34

*ring/Fa11

- March, April,
November

september/ October,

Loads tend to follow the same patterns for e a c h of these three time period. Daily load profiles within each season will differ according to whether it is a weekday or a weekend/holiday. Daily load profiles for each of these two categories tend to r a i n the same for each of these seasons.
Daily load profiles will also vary dependinq on the type of load: residential, co&rcialor industrial. The day has been divided into 96 t periods, of 15 minute intervals. system lcrads are considered to be constant over each time interval. Fifteen minute t b intervals will track load changes with reasonable accuracy, and also provide sufficient time for transients caused by switching and other transients to settle down. A practical Distribution Automation system would allow a minimum of 15 minutes between switching operations in order to obtain steady

The three-feeder test system is used primrily to ensure that each of the reconfiguration algorithms has been *lmntd Properly. The loads are assumed to be constant, therefore, Once the 0 t l Pconfiguration has teen found, no further switching is required. The 44 kV system on the other hand, is Used to compare the reconfiguration algorithm on a m r e

1927

state values for the system parameters. It can be seen that an entire year of distribution system load changes can be shlated using only 18 load profiles, considering 3 seasons, either weekend or holiday, for 3 load types. Residential and comrcial load profiles have been obtained from actual recordings on the Kingston P.U.C. 44 kV system, for both weekdays and weekends/holidays, during the winter season. Typical industrial load profiles for winter weekdays and weekends have been provided by Ontario Hydro. These records are sufficient to determine the 18 load profiles required over a one year period. to simulate the 44 kV system Industrial load profiles do not change significantly with the seasons as heating/air conditioning represents only a very smll portion of the total load. The two winter industrial load profiles, winter-weekday and winter-holiday, are therefore also used for the s m r spring/fall seasons. For the spring/fall season, residential and comrcial load demands are reduced to approximately 85% of the winter d m d . These load profiles are therefore derived by multiplying the equivalent winter load profiles by a factor of 0.75. The winter residential and comrcial load profiles have been d f i e d to produce the s m r residential and c m r c i a l load profiles. S m r residential and comrcial loads profiles differ from the corresponding winter profiles in that s m r air conditioning loads are m r e prominent during the daytime while winter heating loads are m r e prominent at night. In addition, in 1989, for the nearby city of Toronto (population, 2 million), the peak s m r demand was equal to the peak winter denand. These two facts were used to produce the summer residential and comercia1 load profiles. Some example load profiles are illustrated in Fig. 8. The load profiles as described above are reasonable and adequate and considered mrely as input data for applyingthe proposed and existing reconfiguration methods and
0

each point is therefore assigned to be:


LOAD
= (150/200)

X (Station Capcity) X (Load (time))

where LOAD(the) is the load profile infomtion, which is a function of the, and has been nomlized In this way, the peak to reach a maximum of 1.0. system load will be 150 MW, as for the actual system, and the total load will be distributed among the various load pints in proportion to their station capacities. This does not represent exactly the Kingston 44 kV system, but it does provide a reasonable system to which the reconfiguration methods can be applied. It was necessary to derive the load profiles as described above because of the lack of sufficient recorded data. A m r e accurate analysis can only be obtained by continuously monitoring and recording system load infomtion as was done for the P.P. & L. project [ 2 ] .
RESULTS

Three Feeder T e s t System The effectiveness of each of the selected reconfiguration techniques is first tested by comparing the results of each method, applied to the 3 feeders, 13 load point, 16 branch distribution system (Fig. 6 ) . The methods which have been compared include:
1. 2. 3.

Linear Programing 0pti.milLoad Flow Heuristic Search with: a) Uniformly Distributed Load Model loss calculating formula. b) Loss Reduction Esthtion formula.

0 0. 0.

The selected methods have been p r o g r d with additional check on the accuracy of the methods is done using a Gauss-Seidel Load Flow analysis. The load flow analysis is too slow for real-th application to the feeder reconfiguration problem, but it does provide an accurate estimate of the losses for ccanparing the results of the methods being tested. The methods have been applied to the three-feeder distribution system using the data provided, in order to determine the optimal (or near-optbl) configuration. The results of the comparison are illustrated in Table 1. These results illustrate some interesting properties of the various methods. Since both the Heuristic Search/Loss Reduction Formula nethod and the Gauss-Seidel Load Flow Analysis d e l the loads as lumped at the ends of the feeder sections, the

0. 0. 0. 0.
TlME (houn)

Fig. 8

Sample Load Profiles

I
compare their perfomce. The next step in simulating the distribution system involves assigning the various load profiles to the actual system loads. Each of the system load points is designated as either residential, comnercial or industrial, and power factors of 0.98, 0.90 and 0.85 are assigned to each load type respectively. Power factors are assumed to remain constant for each load type. Loads on the 44 kV system are in the form of transfomr stations, each station having its own capacity equal to the capacity of the transformers in that station. The total system capacity is 200 MW, and the maximum load demanded is 150 MW. The load at

ASSHOWN

0.0061244 0.0042626

0.0042626

0.0046907

0.00643

'* MFIHOD I IN ERROR S

OPllMALcoNFIC~

Table 1 System Losses Configurations

(in

P.u.)

For

Various

1928

esthted losses are higher than those estimated using the m r e realistic uniformly distributed load model. The results of the Heuristic Search/Loss Reduction Formula and Gauss-Seidel loss estimates are in reasonably close agreement. The losses for the O p t k l Load Flow method are calculated using the Uniformly Distributed Load mdel, as loss calculations are not inherent in the method. The Transportation Algorithm, which used a lumpedload model, produces lower estimtes than the Heuristic Search/mss Reduction Formula and GaussSeidel methods even though a piecewise linear model of the cost factors is used. It also produces results which are incorrect. The method indicates that, in order to reduce system losses, LOAD 10 should be roved from SOURCE 1 to S O J K E 3 . The other methods indicate an increase in losses for this switching option. This is confirmed by the Gauss-Seidel load flow analysis. Further examination of Linear PrcqrarrPning methods reveals a critical weakness in the theory as applied An to the distribution system reconfiguration. assumption which m s t be valid in order for linear prcqranuning methods to work is that the costs due to one load must be linearly independent of all other loads. In other words, the total system losses must be the sum of the losses incurred supplying each load independently. Since losses are calculated as 12R, it is not only important to minimize losses for each load independently, but also to optimize how the loads are ccanbined over the various possible routes.
As a result of not satisfying this assumption, linear programning methods can result in configurations which are not optimal or near-opthl. While linear programing techniques may be useful for providing distribution system planners with a reasonable starting pint for their design, the results should be viewed with caution. Linear PrOgrarraning methods, in the form used in this coinparison, are not considered suitable for the real-time reconfiguration of distribution systems. Consequently, these methods are not included in the carprison on the 44 kv distribution system.

21

n
W
33

U
33

LI U
33

21

33
34

3 6

Table 2 Open Switches for Proposed Configurations Switch numbers are defined in Fig. 7 . other switches are considered closed. is different and that they all propose changes to the original configuration. The different results for the two Heuristic Search methods are expected as two different & l es are used for the system loads, the uniformly distributed load model, and the lumped-load del. These two load mdels are sufficiently different to result in different optimal configurations. The optimal Load Flow method also proposed a configuration which is different from the other two methods. This method should have produced the same results as the Heuristic Search/Loss Reduction Formula method which models the loads in the same way. The losses for the optbra1 Load Flow configuration are higher than the Heuristic Search results and the configurations are considerably different. A load flow analysis of the two configurations has been used to confirm that the Optimal k m d Flow results are worse, in terms of minimizing losses, than the Heuristic Search method which uses the s e model for system loads.
A further test has been conducted to check the reliability of the O p t h l Load Flow method. If the proposed configuration is truly o p t h l (or nearoptimal), opening some of the proposed switches prior to applying the method should not change the end result by a significant amount. In other words, if some of the switches which are listed in Table 2 as being open in the optimal configuration are physically r w v e d from the system, application of the optimal bad Flow mthod should propose opening the other The OptirOal Load Flow switches listed in T a b l e 2. algorithm has been applied to the system with different ccanbinations of the 13 switches pre-opened. The results, listed in Table 3 , show that different configurations result for each case. The method obviously does not produce optimal results.

Heuristic Methods The final configuration, as determined by each of the three Heuristic methods, should be optimal in tern of minimal losses. This is confirmed by the Gauss-Seidel Load Flow Analyses which was perfornk?d on each of the possible configurations in order to determine the one with the lowest losses. Both the Optimal Load Flow method and the Heuristic Search technique have been seleded for further camparison on the m r e complex 44 kV system.

K I "

Puc 44 kV DISlRIIWTION

S Y m

Analysis For Constant Loads The first step in the comparison of the reconfiguration methods on the 44 kV system is to apply the algorithms to the system with load currents fixed, in order to compare the "optimal" configurations as determined by each technique. The load currents for WINER-WEEKDAY, at 24 hrs., are used and assumed to be constant. The Heuristic Search methods and the optimal Load Flow technique are then applied to determine the uopthl" system configuration. Thirteen switches are required to be open in order to maintain the radial nature of the system and still supply all The configuration resulting f m t h e of the loads. application of the various reconfiguration methods are described in Table 2 which lists the 13 open switches for each configuration. Notice that each of the recomnded configurations
b 9
10

9
11
14

LO
12 14
19

10

11
14
19

I2

IS
I9

1s

1V

Denota pre-opcocd svitcha Note: mtch numben am defined io Figure 6

Table 3 Configurations Switches

for

Different

Pre-opened

1929 The Optimal Load Flaw method works by closing all switches, calculating the optimal load flow for minimum losses, and then opening the branch with the lowest current. Opening the branch with the least current is said to disturb the o p t h l flaw the least a u n t possible. A new load flaw is then perfonned and the process is repeated until the system is in a radial configuration. The method is theoretically sound up to the point of opening the branch with the lowest current. The optimal load flow does indeed provide the currents in the system which would minimize the losses. opening the switch with the lowest current is not, however, guaranteed to lead to an optirral solution. opening of each switch changes the current in other branches and the system is very sensitive to the order in which the switches are opened. The final configuration will only be o p t h l by coincidence and not by virtue of the method. The authors of the O p t k l Load Flow nethod 161, did not claim that the Ethod is optimal or nearopthl, only that it provides a degree of loss reduction. Their results show that, when applied to a distribution system which has not been previously optimized, the method reduces the losses. This is also the case for the Kingston P.U.C. 44 kV system, as can be seen in Table 2 . The method does not, hcwever, always provide optimal or near-optimal configurations, and therefore may not be suitable for real-time applications. Analysis for Chanqinq Loads The next step in the comparison of the feeder reconfiguration methods is to apply them to the 44 kV system under changing load conditions. The load profiles described above are used to simulate load changes over a period of one year. Each method is then applied to the system, the resulting switching is recorded, and the total losses for the year are calculated. The system losses which would occur for each method, over a one year period, are listed in Table 4 . The losses for the non-automated system are compared to the losses for each reconfiguration method to determine the savings in MWH. The dollar value of Automation system and, even i it is, the distribution f of the total branch load to each transfomr is not. Therefore, for typical distribution systems, the uniformly distributed load model provides a better W e 1 of the branch loads and a more accurate estimate of the system losses. The Kingston P.U.C. 44 kV system, however, is sorewhat unusual. It is a high voltage distribution system supplying loads through transfomr stations which are relatively few in nmker. Soine of these stations supply specific loads (comnercial or industrial) while other supply the 4.16 kV distribution system. This tvpe of distribution system is better modelled by the lumped-load mdel which m r e accurately represents the actual system. The subroutine for calculating the losses in the Uniformly Distributed Load Model variation of the two Heuristic Search techniques has been d f i e d , for comparison, by changing the losses formula for each branch from Eqn. (6) to:
\

KDSSES = 12$

(7)

The basic method for calculating the losses r a i n s the same. Each feeder branch current flow is determined from the information made available by the Distribution Automation system. The loss formula can then be applied to determine the system losses. only the formula itself is chanqed to reflect modelling the load in a different m e r . This method can therefore be easily d f i e d to accorrnnodate a lumped-load model or a uniformly distributed load model. Applying the d f i e d method to the 44 kV system results in the sane losses, configurations and savings as the Heuristic Search/Loss Reduction Formula method. The results for the optimal Load Flow method show reduction in losses (2.3%), however, the losses are not reduced by nearly as much as the Heuristic Search/Loss Reduction F o m l a method (8.8%) which similarlymodels the loads as lumped at the feeder section ends. This result confirms that, although the Optimal Load Flaw method can reduce system losses, it is by no means optimal.
saw

U B(ylh)
SAVINGS

465.18

I
I

The Kingston P.U.C. 44 kV distribution system serves primarily residential and c m r c i a l loads, with a small amount of light industrial. Major industrial sites in the area are supplied directly from the Ontario Hydro transmission system and are not part of the Kingston system.
17230
I

12243 56.121.50

523.259.25

38,618.00

nEDuamN

*Loss

8.8 %

6.8 %

23 %

Table 4

savings m e to Loss Minimizations Methods

Feeder reconfiguration x h e s for loss reduction make use of the diversity of the load profiles within the system and reconfigure the system from time to time, thereby reducing losses. The greater the diversity in load profiles, the greater the potential for loss reduction. In order to determine the effect

these savings is then calculated using $0.05/k WH which is an average of the Ontario Hydro industrial rates for 1989. It can be seen that the savings which result from the Heuristic Searchflniformly Distributed Load Model method are m c h less than the savings which result from the Heuristic Search/loss Reduction Formula method. The actual losses are also less, again, due to the nature of the mdels.

m - 0
Wm(0vT

AUrUMAnON

6025.27

2851.95

6025.27

-=m) wrm
AVrOhlATMN
SAv(Ncs

5447.31 577.96 128,898.00 9.59 96

26M.11
221.84

5794.88 230.39

(Mwb)

For m s t distribution systems, the load transf o m r s are distributed along the length of the feeder branches. The spatial distribution of these transf o m r s is not normally known to the distribution

I ?= I
8-

111.09200 1.78 96

I
I

111.519.50

F a "

3.8 %

Table 5

Savings Due to Reconfiguration Modified Loads

1930
of different load types on the loss reduction, several of the residential and c m r c i a l loads on the 44 kV

system were changes to industrial load profiles. Table 5 shows the results of +roving the mixture of residential, comrcial, and industrial customers on the 44 kV system. Total system losses are increased scanewhat but, m r e importantly, the savings due to automatic reconfiguration have increased. The savings for the one year period are substantially greater for the mdified system and the percentage loss reduction is -roved. This simple test indicates that potential savings due to automatic feeder reconfiguration are greater for systems with greater diversity in the individual load profiles. Systems with a well balanced mixture of residential, comrcial and industrial load types stand to benefit m s t from the installation of Distribution Automation equipment for feeder reconfiguration.

method are that the objective function is required to be a linear function, which it is not, and that the costs associated with supplying each individual load must be linearly independent of each other, which they are not. The problem with the quadratic objective function be overcome by modelling the function in a piecewise linear fashion, at the expense of speed of calculation. The fact that the individual costs are not linearly independent is not so easily overcome. As a result, although these methods m y be suitable for system planning, they are not adequate for realt h applications to feeder reconfiguration.
( 1 2 R ) can

Heuristic Methcds Methods which guarantee optimal solutions, such as linear prcqranming, are either too t h s consuming or have other inherent weaknesses which make them unsuitable for the feeder reconfiguration problem. Since speed of application is such an important parameter, the best methods will provide fast solutions with optimal or near-optimal results. A mll reduction in the loss savings may be acceptable if the method is reliable and fast. Heuristic techniques, i properly structured, can provide fast f compu-tation times and the resulting configurations should be optimal or near-optirral. Recent efforts have concentrated on Heuristics as the best methods for obtaining fast solutions to the feeder reconfiguration problem, with good reliability and near-opthl results. T o forms of Heuristic w approaches have been compared in this work, the Heuristic Search strategy and the Optimal Load Flow method. The Heuristic methods have been tested on both the three feeder system and the Kingston P.U.C. 44 kV system. The 44 kV system was shulated with constant existing load-types and d f i e d load types. The results of applying the Heuristic methods to the sinple three feeder system showed that all three method produce the same final configuration. This configuration has been examined (by Gauss-Seidel load flow analysis of every possible configuration) and proven to be optimal. Further tests of the three Heuristic methods have been conducted on the m r e complex 44 kV system. The two Heuristic search methods propose the optimal configuration when applied to the system with the loads ass& to be constant. The Optimal Load Flow mthod, however, produces results which are sarewhat less than optkl, although some loss reduction is obtained over the original configuration. The O p t k l Load Flow method is often able to reduce the system losses but the final configuration is not guaranteed to be optimal or even near-optimal. When applied to the 44 kV system with time-varying loads, the Heuristic Search methods reduced system losses by as much as 8.8% over the one year test period. Again, the Optimal Load Flow method, while providing sone loss reduction, is not nearly as effective. Not only is the O p t k l Load Flow method less effective at reducing system losses, it is also the slowest method. Fpplication of the three methods has been sirdated on the Honeywell CP-6 mainframe computer. The system works on a th-sharing basis which makes a m a t e assessment of the computation t h s bpssible, however, an approxination of the relative speeds of calcuhtion is possible. The m u n t of t i n - e taken to complete the calculations for

Table 6 Comparison of Computation Thss


COST ANALYSIS A recent pilot project by Mississauga Hydro, which included automating six field switches, has been completed for approximately $200,000. The cost of the field switches ($10,000 per switch) has been minimized by automating the existing field switches by adding By colnparison, energy storage operators and RTJ's. load break recent prices for &ea-Brawn-Boveri automated switches with energy-storage units and RlU'S are approximately $8,000 - $10,000, depending on the degree of sophistication that is required.

A review of the switching operations on the simulated 44 kv system reveals that, over the one year period, only four switches are required to be operated automatically. The cost for the equipwnt required to automte these four switches is estimated to be $229,000 based on the Mississauga Hydro project costs. The total cost of the Mississauga Hydro project, $200,000 would be reduced by about $16,000 as only four switches need to be autmted, but an extra cost of $45,000 would be required to provide mnitorhg of all of the system switches. The net result is a total system cost of approximately $229,000. the estimated savings due to feeder reconfiguration for loss reduction, from Table 4, using the lumped-load model, are $23,259 per year. The total capital cost of the equiprent which is nomlly e x p z t e d to last 20 years, The annual savings are therefore worthis $229,000. while when col~lparedto the cost required to irplement the automated system.

Linear proqramrun . gt-lethcd


A linear prograrmring method in the form of a transportation problem has been developed and applied to the three-feeder distribution system in order to determine its effectiveness. Linear programning nethods have proven unsuitable for application to the feeder reconfiguration for loss reduction problem. The two difficulties with the

each of the rrethcds f o r the one year period has been recorded and is shown in Table 6. These times do not represent exact CPU times as the system is timesharing with other processes, however, they do give an indication of the relative times taken by each method. The times shown include the time taken to calculate configurations for 96 time periods, f o r each of two days, for each of three seasons. The total number of configurations to be determined is therefore 576.

1931

All of the times listed in Table 6 are acceptable for real-time reconfiguration of the Kingston 44 kV system which is a relatively small system. For large or medium size systems, the O p t k l Load Flow method may very well be too tine consuming to be practical.
KINGSION P.U.C. 44 kV SYSTEM PENNSYLVANIA POWER AND LIGHT
26

INUMBER OF FEEDERS
NUMBER OF SWITCHES PEAK LOAD ( M W TOTAL LOSSES WITHOUT RECONFIGURATIOWW)

51

161

16.927

LOSS REDUCTION WlTH


PERCENTAGE LOSS RECUCnON SAVINGS

RECONFlGURATlON(MW)
8.8 s

14.6 Q.

$23.259.00

SI 11.450.00

Table 7 Camparison of Kingston P.U.C. and p.p. and L . Systems The Heuristic Searchflnifody Distributed Load Model method, although sorewhat more t k consuming, is reasonably fast and would likely be adequate for larger systems. In addition, it has the advdnta9 of being easily d f i e d to &l e the system either with lumped or uniformly distributed loads, whichever m r e accurately describes the system. The savings i losses for the one year period, n using the lumped-load & l e and the Heuristic Search methods, are estimated to be approxbately $23,000 or 465 MWH. This compares favorably with the results of the Pennsylvania Power and Light (PPL) project [2]. A comparison is detailed in Table 7. The higher savings for the PPL system are attributed to the higher total load, higher system losses and, most importantly, to the greater number of possible switching options. Considering the relatively smill size of the Kingston 44 kV system, the savings are quite significant. The Cost Analysis described above shows that Distribution Automation is cost effective. Feeder reconfiguration for loss reduction alone provides some savings and when combined with other Distribution Automation features such as Load Management, VAR and voltage control, automatic meter reading, fault isolation and service restoration etc., the benefits of Distribution Autmtion become even more convincing.

D. Miller and J. E. Wheeler, "Distribution System Integrated Voltage and Reactive Power Control", IEEE Trans. on Power Ppparatus and Systems, Vol. PAS-101, pp.284-289, 1982. R. E. Lee and C. L. Brcoks, "A Method and its Application to Evaluate Automated Distribution Control", I= ~rans.Power Delivery, Vol. 3 , tp. 1232-1240, 1988. K.M. Hamam, A. Bracellar and K.S. Hindi, "Solution of the Transhipment Problem by Networking Modelling and its Application to Generator Scheduling and Distribution Design", Power Systems Cannutation Conference, 1975. M.J. Juricek, A. Fukutome and M.S. Chen, "Transportation Analysis of an Electric Power Distribution System", IEEE Trans. Power Apparatus & Systems, Vol. PAS-95, NO.3, A76-052-1, p. 758, 1976, D.L. Wall, G.L. Thompson and J.E.D. NorthcoteGreen, "An optimization Model for Planning Radial Distribution Networks", IEEE Trans. on Power Apparatus and Systems, Vol. PAS-98, No. 3, pp. 1061-1068, 1979. f A. Merlin, and G. Back, "Search f x a MinimmLOSS Operational spanning Tree Configuration for an U r b a n Power Distribution System",Proc. of the Fifth Power System Conference (PYX), Cambridge, pp. 1-18, 1975. 7 D. Shirmohammadi, and H.W. Hong, "Reconfiguration 1 of Electric Distribution Networks for Resistive Line Lasses Reduction", IEEE Trans. on Power Apparatus and Systems, Vol. 4, NO. 2, pp. 14921498, 1989. 81 T. Taylor, and D. L u b k m , "Implementationof Heuristic Search Strategies for Distribution Feeder Reconfiguration",IEEE PES S m r Meeting, Long Beach, California, 1989. "Network Reconfigur[91 M.E. Baran, and F.F. WU, ation i Distribution Systems for Loss Reduction n and Load Balancing", IEFE Trans. on Power Delivery, Vo1.4, pp. 1401-1407, 1989. A. I. Cohen, and M. Carson, [ 101 D.W. Roks, J. P t , a& "New Methods for Evaluating Distribution Automition and Control (DAC) System Benefits", IEEE Trans. on Power Apparatus & Systems, Vol. PAS-100, No.6, pp. 2978-2986, 1981.
[ll] G. Hadley, 'Linear P r o g r d n g r , Addison-Wesley publishing Campany, 1963. [12] S. Civanlar, J.J. Grainger, H. Yin and S.S.H. Lee, "Distribution Feeder Reconfiguration for LOSS Reduction", IEEE Trans. on Power Delivery, Vol. 3 , Pp. 1217-1223, 1988. [13] N. E . Chang, "Determination of Primry-Feeder Losses", IEEE Trans. on Power Appxatus and SyStemS, Vol. PAS-87, NO. 12, pP.1991-1994, 1968.

J. B. Bunch, R.

1932

Discussion

M. ETEZADI-AMOLI, EE/CS Department, University of Nevada, Reno, NV 89557-0030. The authors are commended for their work regarding feeder reconfiguration for loss reduction through distribution automation. The following comments are made regarding the described approach:
1. Based on this author's experience as a former distribution engineer, many substation transformers operate at a fixed tap which may be changed once a year depending on the load variation between summer and winter. Feeder reconfiguration as proposed may require different taps for these transformers during a regular season. Since this may not be practical, extra boundary conditions will have to be incorporated into the proposed algorithm. This in turn will reduce the number of switching options. Actual distribution feeder layout may not always accommodate allocation of load from one feeder to another due to the different ratings of the various conductors and other limitations. Since all feeders do not have the same type of conductors, representation of the actual feeders and their capabilities will further reduce the switching options. Since distribution feeders are protected using overcurrent relays and settings of these relays are dependent on load and fault currents, the proposed scheme would have to resolve the protection issues under various operating conditions. In an actual power distribution system, different feeders may be designed to operate with different reliability depending on the type of loads. Thus reconfiguration of the distribution system or the purpose of loss reduction may not be acceptable to customers with essential loads such as the casinos in the Reno area. Use of shunt capacitors for voltage regulation and power factor correction is a logical and popular option among the distribution engineers. If current compensation is used to control the switching of these capacitor banks, allocation of load from one feeder to the other may disrupt the arrangement for the proper operation of these devices. Relocation of these banks to a section of a feeder which will not be influenced by the reconfiguration may not be possible, practical, or economical. Any modification cost related to this issue must be included in the economic analysis regarding the distribution automation. Voltage regulators and line sectionalizers which may be installed on some feeders may not properly operate under the described schemes.

The above comments are not made to discourage research work in the area of distribution automation. These comments are merely stated to highlight the various 11costs8' that one has to consider when dealing with automation of an actual distribution system. Again, my congratulations to the authors for their work regarding this subject. 14anuscript received March 4 . 1991.

Dariush Shirmohammadi and Wayne Hong Pacific Gas and Electric Company The authors have compared two heuristic methods of network reconfiguration for resistive losses reduction. Both these methods have been successfully implemented in practice. The comparison seems to have been made by actually implementing the algorithms presented in the references [7] and [12] of the paper. The results presented are of importance and value to the industry.
A s emphasized by the authors, both these methods are heuristic and, hence, their implementation can profoundly impact their results. This is particularly true for the method presented in [ 7 ] --llOptimal Flow Pattern" approach. In thls method, the power flow of the distribution network must be solved accurately and repeatedly. In the absence of an efficient and accurate method for solving such power flow problem, the overall algorithm becomes inefficient and perhaps inaccurate. For example, a Gauss-Seidel approach is too inefficient and a Newton-based approach may be unstable. In both cases, increasing the mismatch tolerances to obtain results may render the power flow results useless. If, as a result of an inaccurate power flow solution the nodal current injections are inaccurately computed, switches that are selected for opening will not be the optimum or near optimum. We have found the power flow method presented in [A] to efficient and accurate for distribution networks.

2.

3.

4.

5.

Another important implementation issue is the modeling of switches and their interactions with one another in the Optimal Flow Pattern approach. For example, once a normally closed switch is opened in the process converting the meshed network to a radial one, a normally open switch must remain closed in order to keep the loads downstream from the opened switch connected. The selection of such a normally open switch is not trivial if more than one such switch exists -- which is often the case. In the implementation of the Optimal Flow

Pattern approach it is also important whether and how network ampacity constraints are taken into consideration. Including ampacity considerations impacts the final results and the efficiency of the algorithm. Implementation issues are also extremely important if numerical efficiency is to be maintained.

6.

1933

Without careful attention to implementation issues, it is expected that only small scale distribution networks can be solved without difficulty as is shown in the paper. However, as the network size becomes larger, the impact of the implementation details on the efficiency and the accuracy of the approach becomes increasingly important. Our experience with the Optimal Flow Pattern approach [7] on distribution networks of practical sizes (up to five thousand branches and 1200 switches) is very favorable. We have found the approach to be very robust and practically independent on the initial set of open switches. Obviously some dependence on the initial set of open switches should be expected since the approach only points to a near-optimum solution. We have frequently achieved loss reductions that were significantly more than 10 percent for such realistic networks. The CPU times have been a few seconds on a 6 MIP computer. Small networks, such as those studied in the paper, can be solved in a fraction of a second. In conclusion, we must emphasize that unless a heuristic approach is correctly implemented to work efficiently with large scale networks, no general conclusions could be made on its results. [A] D. Shirmohammadi, H.W. Hong, A. Semlyen, G.X. Luo, "A Compensation-Based Power . Flow technique for Weakly Meshed Distribution and Transmission Networks", IEEE Transactions on Power Systems, May 1988, pp. 753-762.
Manuscript received March 4 , 1991.

CLOSURE

T. Wagner and A . Chikhani, Electrical Engineering Department, Royal Military College of Canada, Kingston Ontario, Canada, K7K 5LO R . Hackam, Electrical Engineering Department, University of Windsor, Windsor Ontario, Canada, N9B 3 P 4 . The authors would like to thank D. Shirmohammadi and W. Hong for there thoughtful comments and advice. The discussion by M. Etezadi-Amoli is also very much appreciated and outlines some very important considerations in the application of reconfiguration schemes. It is obvious that feeder reconfiguration will only be practical if it is implemented as part of a comprehensive automation system, including a sophisticated SCADA system, which may resolve some of the concerns listed in the discussion. For example, overcurrent relay settings could be automatically adjusted to suit the present configuration. Shunt capacitor switching could also be included, not only for voltage and power factor correction, but also for loss reduction. Research is presently ongoing to examine how to best implement feeder reconfiguration in conjunction with shunt capacitor switching, in order to optimize distribution system configurations for minimal losses. As pointed out, implementation of feeder reconfiguration schemes may also require upgrading of some of the feeders to larger capacity conductors, in order to take full advantage of the loss minimization techniques. With regard to the manual operation of transformer taps, the authors agree that some examination of this restriction is required. However, since the overall effect of the feeder reconfiguration process is to more evenly distribute the feeder loads over the distribution network, it may well be that taps will not require more frequent changes. The comments regarding voltage regulators and feeder reliabilities again underline the fact that there are many considerations which need to be addressed in order to effectively implement a practical reconfiguration scheme.
Manuscript received August 27, 1991.

You might also like