2204

IEEE TRANSACTIONS ON POWER SYSTEMS, VOL. 22, NO. 4, NOVEMBER 2007

Effect of Load Models in Distributed Generation Planning

Devender Singh, R. K. Misra, and Deependra Singh
AbstractThe effect of load models on distributed generation (DG) planning in distribution system is investigated in this work. It is shown that load models can signicantly affect the DG planning. Normally a constant power (real and reactive) load model is assumed in most of the studies. Such assumptions may lead to inconsistent and misleading results about deferral values, loss reduction, payback period, and other subsequent calculations. It has been demonstrated that DG planning based on such assumptions would not be effective after implementation. It is shown that load models can signicantly affect the optimal location and sizing of DG resources in distribution systems. A comparative study of real and reactive power loss, real and reactive power intake at the main substation and MVA support provided by installing DG resources for different type of loads models has been performed. Index TermsDistributed generation, distribution system, load models.

System real and reactive power losses , Real and reactive power losses with DG MVA ow in line Voltage of th node WDG WODG System with DG. System without DG. I. INTRODUCTION OWER deregulation and restructuring have created increased interest in distributed generation (DG), which is expected to play an increasingly important role in the electric power system infrastructure planning and market operations. Distributed generation is power source that can be connected to a distribution network by a distribution company (DISCO) at any node or by the customer at the customer side of the meter. Distributed resources are strategically located and operated in the system to defer or eliminate system upgrades, improve voltage prole, reduce system losses, reinforce grid, and to improve system reliability and efciency. Recent studies have predicted that by year 2010, distributed generation will account for up to 25% of all new generation [1]. In the last few years there has been signicant contribution to research in the eld of DG resource planning. Normally, DGs are integrated in the existing distribution system and the planning studies have to be performed for optimal location and size of DGs to yield maximum benets. Y. G. Hegazy et al. [1] evaluated the effect of DGs controlled by the customer on system power capacity to satisfy the total system load and predicted the average amount of unsupplied demand for a given year. The authors used Monte Carlo approach to model the operating histories of the installed distributed generators. El-Khattam et al. [2] proposed a method of solving distributed generation planning problem (location and size) in different utility scenarios as an optimization problem. The objective function was based on supply-demand chain which aimed to minimize the investment and operating costs of local candidate DGs, payments toward purchasing the required extra power by the DISCO, payments toward loss compensation services, as well as the investment cost of other chosen new facilities for different market scenarios. Wang et al. [3] proposed an analytical method to determine the best location of candidate DGs for minimum loss conguration. A rigorous analysis for uniformly distributed, centrally distributed, and increasingly distributed loads has been carried out for constant and time varying loads on a feeder. The proposed approach was non-iterative unlike power

NOMENCLATURE Voltage exponents of real and reactive loads. Const., Res. Ind., Com. Mix. Constant and residential load models. Industrial and commercial load models. Mixed load model. MVA capacity of line - . MVA intake at bus 1. Total system MVA intake by DISCO. NV, NL Number of voltage and line limit violations. Real and reactive load at bus at nominal voltage. Total system real and reactive power demands. Total size of DGs. Real and reactive power injections at bus . Real and reactive power intake at bus 1.
Manuscript received March 14, 2007; revised May 24, 2007. Paper no. TPWRS-00173-2007. Devender Singh and R. K. Misra are with the Electrical Engineering Department, Institute of Technology, Banaras Hindu University, Varanasi, UP, India (e-mail: dsingh@bhu.ac.in; rkmisra@bhu.ac.in). Deependra Singh is Lecturer in Electrical Engineering Department, Kamla Nehru Institute of Technology, Sultanpur, UP, India. Color versions of one or more of the gures in this paper are available online at http://ieeexplore.ieee.org. Digital Object Identier 10.1109/TPWRS.2007.907582

0885-8950/$25.00 2007 IEEE

SINGH et al.: EFFECT OF LOAD MODELS IN DISTRIBUTED GENERATION PLANNING

2205

ow programs. Therefore, there is no convergence problem involved, and results could be obtained quickly. However, authors have indicated that other constraints such as voltage and line limits may affect the DG placement. Ochoa et al. [4] proposed various indices to evaluate the impact of distributed generation on distribution networks. The indices to measure the impacts of DG introduction on loss reduction, voltage prole, current carrying capacity of conductors, and short circuit currents for three-phase and single-phase ground faults are proposed. Finally, a multi-objective function is derived for such planning studies. Chiradeja and Ramakumar [5] derived indices to measure the technical benets in terms of voltage prole improvement indices (VPII), line loss reduction indices (LLRI), and emission improvement. Finally the authors devised a composite index by combining the various indices. Load models are well known in stability studies [6][9]. These studies are normally aimed toward voltage or frequency dependent representation of load especially for dynamic or static stability studies of power systems. Exhaustive review of load model to be used for power ow and dynamic studies has been presented in [7]. The effect of load models in planning studies was demonstrated in optimal capacitor placement/switching by Rizy et al. [10] in a distribution system and by Arnborg et al. [11] in under-voltage load shedding studies. The authors demonstrated that consistent with the modeling of feeder load as constant power, it was expected that the switching of capacitor bank to improve the power factor would result in decreased real and reactive power injections at the substation. This decrease in real and reactive power injections is due to improved voltage prole. In fact the measured real and reactive power injections increase. In this case, contrary to the constant power load modeling, a reduction in real and reactive power injection reecting the reduced line losses is not observed at the substation. An assumption of constant power load modeling leads to general misunderstanding of reduced power injection due to reduced losses. Analysis of the experimental results [10] using voltage sensitive load model showed that while the feeder losses are reduced following the capacitor placement, the attendant improvement in voltage prole results in an increase in loads that exceeds the amount of loss reduction. In context of optimization, problem of DG placement is similar in nature to that of capacitor placement discussed above. Most of the planning methods invariably use power ow programs which normally utilize constant real and reactive power load model representation. It is observed from the literature review that load models are not included in planning the location and size, and calculating the said indices except Gozel et al. [12] who have used analytical approach considering the load model to show that the location does not change as signicantly as size of DG. However, the analysis did not take the constraints of voltage and line limits and therefore has only instructive value. In this paper, a detailed study of effects of load models in DG planning is made to asses the technical impacts and feasibility of DG planning. Extent up to which load models can affect the location and size of the DGs in such studies has also been investigated. The voltage dependent load models for residential, industrial, and commercial loads are adopted from [7]. A 38-bus radial distribution system from [13] is utilized for demonstrating

TABLE I LOAD TYPES AND EXPONENT VALUES

the results. In the practical situation, loads are not explicitly residential, industrial, and commercial; rather, load class mix may be seen by distribution system depending on the nature of area being supplied. Therefore, a load class mix of residential, industrial and commercial load has also been investigated. II. LOAD MODEL-BASED TEST CASES To quantify the effect of various load models on distributed generation planning, a 38-node distribution system is adopted (Fig. 9). The data for p.u. line impedances, load data and the line MVA limits are given in the Appendix (Table VIII). The effects of selected voltage dependent load models are investigated in different planning scenarios (test cases). Practical voltage dependent load models, i.e., residential, industrial, and commercial, given in [7] have been adopted for investigations. The load models can be mathematically expressed as (1) (2) In a constant power model, conventionally used in power ow studies, is assumed. The values of the real and reactive exponents used in the present work for industrial, residential, and commercial loads are give in Table I [7]. During investigations the comparison of constant power load model assumption with the practical load models are emphasized. While investigating effect of residential load, the 38-node system is assumed to be supplying residential consumers only (all loads are residential type). Similarly, for industrial and commercial load, it is assumed that all the loads are industrial and commercial type, respectively. In a practical situation, a load class mix may be seen; therefore, a load class mix of residential, commercial and industrial loads is also adopted. This is indicated in Table VIII along with the load data. The following test cases are developed in which DG locations and sizes for case 1 and case 2 have been chosen arbitrarily. Case 1) Single DG Case: In the given load scenario, customer decides to embed a single distributed generation resource of 0.8 p.u. on bus 14. The effect of load models on the following studies is to be made: a) number of voltage limit violation (NV); b) number of line limit violations (NL); , real p.u. demand , c) MVA and reactive p.u. demand on the main substation; d) loss reduction in real and reactive power; e) saving of MVA capacity on the main substation. Case 2) Multiple DG Case: In the given load scenario customers decide to embed three distributed generators

2206

IEEE TRANSACTIONS ON POWER SYSTEMS, VOL. 22, NO. 4, NOVEMBER 2007

Fig. 1. Flowchart for creation of database. Fig. 2. Flowchart for selecting the minimum loss conguration from the database.

of 0.20, 0.30, and 0.20 p.u. on buses 14, 24, and 32, respectively. The effect of load models on the following studies is to be made: a) NV; b) NL; . and on the main subc) station; d) loss reduction in real and reactive power; e) saving of MVA capacity on the main substation. Case 3) In the given load scenario, DISCO decides to embed a distributed generator of optimal size at optimal location (without undergoing any system upgrades) for a) loss reduction or b) reduction of the main substation MVA intake (system upgrade deferment). The following studies are to be made. i) Optimal DG location and size planning: Effect of load models on the optimal location and size of distributed generators are studied. ii) Error in Load Model: Suppose planning for optimal size and location for cases (a) and (b) has been done assuming constant power load. Now, technical feasibility of already planned location and size is studied if loads in the system are actually non-constant type.

the owchart given in Fig. 1. The bus voltages occurring outside the range 0.951.03 p.u. are treated as voltage violation. The owchart for selecting the minimum loss conguration, for selected load type from the database is provided in Fig. 2. A similar algorithm is used for minimum MVA criterion also. Solutions with line limits and voltage violations are ltered out.

IV. SIMULATION RESULTS The summary of simulation results obtained for various test cases is presented for different types of load models in this sec, is tion. The total system MVA intake by the DISCO, dened as

(3) The differences in all the relevant quantities for the system with and without DG are indicated for different load types. Numbers of voltage and line limit violations have also been presented. Further, to compare the results, non-normalized versions of indices based on [4] are also developed. These indices are dened in the following paragraphs. Real and Reactive Power Loss Indices (ILP and ILQ): The real and reactive power loss indices are dened as

III. DATA PREPARATION Power ow solutions for the 38-node distribution system are obtained in the following fashion. DG size is considered in a practical range (0-4.00 p.u), decided by the total system demand which is 3.9093 p.u. The DG of 0.0 p.u. corresponds to system without DG whereas 4.00 p.u corresponds to a case when all the real power requirements are met by DG. It is considered that the DG is operated at unity p.f. Each bus of the system is considered for the placement of DG of given size from the range considered. Then load ow program is run for each of the cases. The complete procedure to create database is presented in

(4) (5) The lower the values, the better the benets in terms of loss reduction accrued to DG location and size. Voltage Prole Index (IVD): It is related to the maximum voltage drop which, in this case, considers the maximum drop between each node and the root node. This index could also be used to nd prohibitive locations for DG considering preestablished voltage drop limits. In this way, the lower the index,

SINGH et al.: EFFECT OF LOAD MODELS IN DISTRIBUTED GENERATION PLANNING

2207

TABLE II SUMMARY OF RESULTS OBTAINED FOR CASE 1: SINGLE DG CASE

Fig. 3. ILP and ILQ for different load models (Case 1).

Fig. 4. IVDs and dIVD for different load models (Case 1).

the better the network performance. The IVD can be dened as follows:
Fig. 5. IC for different load models (Case 1).

(6) MVA Capacity Index (IC): As a consequence of supplying power near to loads, MVA ows may diminish in some sections of the network, thus releasing more capacity, but in other sections, they may also increase to levels beyond distribution line limits. This index gives important information about the level of MVA ows/currents through the network regarding the maximum capacity of conductors. This gives the information about need of system line upgrades. Lower values of the index indicate more amount of capacity available. Line overloads are indicated by index values above 100% Residential Load Model: A similar trend of reduction with , and exDG placement is observed for . As opposed to the case of constant load model, cept for in this case, despite the addition of a DG, there is increase in from the main substation. All the above reductions are signicantly smaller than those of constant load model. As far at the main substation is concerned, it is seen that in as case of residential load model, placement of DG raises (by 0.0567 p.u.) whereas in case of constant load model, the is decreased by 0.0419 p.u. which is almost of the same order. There is one line limit violation (line-7), in this case when DG is placed in the system as against the case of constant power load model. Industrial Load Model: The trend of reduction in , and with DG placement, as observed for constant and residential load model, is maintained. These reductions are signicantly smaller than those of constant load model and marginally larger than those of residential load is distinctly higher than models. Further, the increase in those observed for residential as well as constant load models. There are no voltage as well as line limit violations for the system with and without DG. Commercial Load Model: In this case, the reductions in are all the quantities indicated in Table II except the having their lowest values compared to all the other load models. Further, the total real power demand of the DISCO p.u.) with DG is (

(7)

A. Case 1: Single DG Case: Distributed Generation Resource of 0.8 p.u. on Bus-14 The summary of results obtained in this case for different load models is depicted in Table II. The various indices as discussed above are presented in Figs. 35. Constant Power Load Model: From Table II, it is observed that there is decrease in , and at the main substation due to placement of DG. Furthermore, and also get reduced appreciably. The overall system MVA, , of the DISCO is reduced by 0.0751. However, there are no voltage as well as line limit violations for both the situations, i.e., with and without DG.

2208

IEEE TRANSACTIONS ON POWER SYSTEMS, VOL. 22, NO. 4, NOVEMBER 2007

marginally higher than that for system without DG, i.e., 3.7987 and may in fact increase, despite the p.u. The and , taking advantage of loss reduction reduction in void. This is due to improvement in voltage prole which results in an increase in loads that exceeds the amount of loss reduction. In such a case, if the DISCO happens to pay same amount/MWh, it loses instead of gaining from DG placement. This may not be detectable if proper load model is not taken into account. There is one line limit violation (line-7) when DG is placed. Mixed Load Model: The reductions in , and are less compared to constant load model. The increase is also observed as in case of other load models exin cept for constant load model. A line limit violation (line-7) is observed in the system. The real and reactive loss indices, ILP and ILQ, are depicted in Fig. 3. It can be observed that the indices indicating the effect of DG placement are low in case of constant power load model assumption, whereas for non-constant power load model, these are higher. Hence, planning studies for evaluating the advantage of DG placement based on constant power load model may indicate lower system losses. The IVDs and the difference between IVDs (dIVD) of system with and without DG are plotted in Fig. 4 indicating the improvement in voltage prole due to DG placement. In this case also, the constant power load model assumption gives a picture of higher improvement as compared to actual system (non-constant power load model). The IC indices depicting the used line MVA capacity in the system with and without DG for different load models are shown in Fig. 5. For the system without DG, the used/available capacities for different load models are not much different. However, when DG is placed, IC values for non-constant power load model are signicantly different from that of constant power load model. In this case, in an actual system, constant power load model assumption gives false indication of high available MVA capacity, whereas the fact is otherwise.

TABLE III SUMMARY OF RESULTS OBTAINED FOR CASE 2: MULTIPLE DG CASE

B. Case 2: Multiple DG Case: Three Distributed Generators of 0.20, 0.30, and 0.20 p.u. ( p.u.) on Buses 14, 24, and 32, Respectively The summary of results obtained in this case for different load models is depicted in Table III. The differences in all relevant quantities for the system with DG and without DG are also indicated for different load types. Constant Power Load Model: From Table III, it is observed , and due that there is decrease in of the DISCO also to placement of DGs. Furthermore, MVA gets reduced. Hence, when constant load model is assumed, the and is directly translated as reduction in reduction in overall system MVA which gives erroneous indication of higher benets of DG placement. This is because the effect of voltage prole on loads is not taken into account. Residential Load Model: A similar trend of reduction in , and except in is observed with DG placement. Despite the addition of DGs, there is increase

from the main substation, unlike the case of constant in power load model. , and are signiThe reductions in cantly smaller than those obtained for constant load model. As at the main substation is concerned, placement far as of DG raises reactive demand (by 0.0341 p.u.) unlike the case is decreased almost of constant load model, where the by same order (by 0.0353 p.u.). Also, unlike the constant power and do not directly reect load assumption, the system and . into the Industrial Load Model: It can be observed that the trend of re, and with placement of DGs duction in is maintained in this case also. However, the reduction in demand is more than the residential load model but less than constant load model assumption. Further the increase in is almost doubled and is distinctly higher than those observed for residential as well as constant load model. Commercial Load Model: The reductions in all the quanand tities indicated in Table III except the are having their lowest values compared to all other load models. Further, total real power demand of the DISCO p.u.) with DGs is ( almost same to that of system without DG, i.e., 3.7987 p.u. has increased marginally which is not The observed in previously studied load models. Mixed Load Model: In this case, the reductions in , and are less compared to constant is observed in this case load model. The increase in also as in case of other load non-constant load models. The ILP and ILQ indices are depicted in Fig. 6. Similar to the single DG case, the indices indicating the effect of DG placement are low in case of constant power load model assumption, whereas for non-constant power load model (except for the industrial load model), these are normally higher. The load models

SINGH et al.: EFFECT OF LOAD MODELS IN DISTRIBUTED GENERATION PLANNING

2209

TABLE IV MINIMUM LOSS CONFIGURATION FOR DIFFERENT LOAD MODELS FOR CASE 3(I): OPTIMAL LOCATION AND SIZE PLANNING

Fig. 6. ILP and ILQ for different load models (Case 2).

TABLE V MINIMUM MVA CONFIGURATION FOR DIFFERENT LOAD MODELS FOR CASE 3(I): OPTIMAL LOCATION AND SIZE PLANNING

Fig. 7. IVD for different load models (Case 2).

Fig. 8. IC for different load models (Case 2).

signicantly affect the studies for evaluating the advantage of DG placement. The IVDs and the differences between the IVDs (dIVD) of the system with and without DGs, indicating the improvement in voltage prole due to DG placement, are plotted in Fig. 7. Similar to the single DG case, the constant power load model assumption gives a picture of higher improvement as compared to actual system (non-constant power load model). The IC indices depicting the used MVA capacity in the system with and without DGs for different load models are shown in Fig. 8. Compared to the system with DGs, the utilized line capacities for different load models are not much different for the system without DG. However unlike single DG case, studies based on constant power load model assumption give false indication of low available MVA capacity. C. Case 3 (i): Optimal DG Location and Size Planning and The summary of results obtained for optimal location for minimum loss (minimum ) conguration is given in Table IV and the same for minimum MVA (minimum ) is given in Table V. For different load models, the (without system upgrade) and corresponding optimal , and are also indicated in Tables IV and V. conguConstant Power Load Model: For minimum is 2.6 p.u. at bus 6. The ration (Table IV), optimal is reduced to 0.0973 p.u. as compared to that without DG

(0.1889 p.u., in Table III). The , have also decreased consistent with constant and power load model assumption. From Table V, it is seen that location of DG for minimum (to defer substation upgrades) is same (bus 6) to that of minimum loss criterion, of course, with different optimal (3.6050 p.u.). It also establishes the fact that for a constant power and are translated into reduced load model, the reduced , and therefore, the change in location may be minimal to minwhen optimality criterion is changed from minimum imum . conguraIndustrial Load Model: In this case, minimum to be 0.94 on bus 12. The tion (Table IV) requires optimal size and location are signicantly different than that for constant which can power and residential load models. The minimum be achieved by placing DG (optimally) is 0.1093 p.u. as compared to 0.0973 p.u. for constant power load model. Also the is marginally higher than that of constant power load model. conguration (Table V), optimal In case of minimum location is same to that of residential load but quite different from that of constant power load model assumption. Similarly, optimal size of DG is close to that of residential load model and is signicantly different than that of constant load model. Commercial and Mixed Load Models: For these load models, conguration (Table IV) requires optimal to minimum be 0.365 p.u. (Com.) and 0.375 p.u. (Mix.) on bus 16. is remarkably different from those It is observed that the of load models considered so far. The location is having the difference of two, four, and ten nodes when compared to that of residential, industrial, and constant load models, respectively. This location is farthest from the root node compared to the other load is signicantly higher compared to models. The minimum other load models. The and for commercial and

2210

IEEE TRANSACTIONS ON POWER SYSTEMS, VOL. 22, NO. 4, NOVEMBER 2007

mixed load models are signicantly higher than that of constant power load model. conguration (Table V) requires opThe minimum timal to be 2.3250 (Com.) and 2.4350 (Mix) on bus 2. This shows that the DG locations are entirely different for minimum and minimum criteria for non-constant power load criterion gives same bus lomodels. Also the minimum cation (bus 2) for non-constant power load models. This is explained as follows. Assuming constant power load model for system without DG z

TABLE VI ACTUAL LOAD MODEL SCENARIO FOR MINIMUM LOSS CONFIGURATION (DG OF 2.6 P.U. AT BUS 6) FOR CASE 3(II): ERROR IN LOAD MODEL

(8) and (9) Now assuming that DG is placed which is reducing the losses and ,

TABLE VII ACTUAL LOAD MODEL SCENARIO FOR MINIMUM MVA CONFIGURATION (DG OF 3.605 P.U. AT BUS 6) FOR CASE 3 (II): ERROR IN LOAD MODEL

(10) and (11) Since constant power load assumption makes and inand to be minsensitive to DG placement; for and are to be minimized, which is same as imum, minimum criterion. However, in case of non-constant power and also become sensitive to DG placeload models, ment (different DG placement gives different voltage proles), and . In such cases, the line limit conapart from straints become more important than the losses because values and . Thus, when of losses are negligible as compared to non-constant load models are used, the DG locations in general are near to the main substation where the line limits are maximum compared to other lines of the distribution system. maintaining the line limits, it In order to minimize the is suitable to place the DG at that node where maximum line MVA limit is found, i.e., the line from main substation to the next node, i.e., bus 2. Hence, in all the non-constant power load comes out models, the optimum location for minimum to be bus-2. Since in this study, the bus-1 was not considered to be a candidate location, the next bus, the bus-2, was selected. D. Case 3 (ii): Error in Load Model This part of the work shows that DG location and size for a system, evaluated assuming constant power load model will not reect the planned benets, when put into actual system. The optimum location and size, for constant power load model, conguration (Table IV) were bus 6 and 2.6 under minimum p.u., respectively. It is depicted in Table VI that placing a DG of given location and size in an actual load model scenario may lead to line violations. This holds for all the non-constant power load models. and are lower than those obtained during planIn fact, ning study performed assuming constant load model. This may indicate that the constant power load model gives conservative estimate as far as losses are concerned. However, more important is the fact that and from the main substation increase. This nullies the advantage of loss reduction. Further getting conservative estimate of losses for actual scenario obtained assuming constant power load model is also rendered meaningless. Compared to the case of planned scenario, in case , total system real power of non-constant load models, , and increase which is due to intake and . increased Considering the optimal solution of DG placement for min( of 3.605 p.u. at bus-6) obtained using conimum stant power load model assumption and implementing it on actual system exhibiting non-constant power load behavior, the results obtained are summarized in Table VII. In this case also, the given location and size is not suitable as far as line limits are confor non-constant load models cerned. Also the optimal is higher than that obtained using constant load model. Implementing the solution based on the constant power load model assumption on the network exhibiting non-constant power load behavior, in general, reduces the assumed benets accrued to DG placement. V. DISCUSSION In this work, only voltage dependent load models are considered for DG planning studies. Frequency dependent models have not been considered due to the following reason. System voltages are local phenomena whereas frequency is a pan-system phenomenon. This means that the bus voltage, up to an extent, can be controlled by real and reactive load and generation at a particular bus. However, the frequency does not show such a characteristic and therefore cannot be controlled locally and remains same for whole of the system. Thus, the change in frequency dependent loads is not due to presence or

SINGH et al.: EFFECT OF LOAD MODELS IN DISTRIBUTED GENERATION PLANNING

2211

absence of DG. Therefore, even if we model the load to be fre, quency dependent, the differences obtained in etc. cannot be attributed to DG placement. Generally, the line limits are reached due to counter line ows which are normally observed in systems with DGs. In this work, no line over-loadings due to counter ows are observed in case of constant as well as non-constant load models. The overloading of line-7 (Tables II, VI, and VII) for non-constant power load models is explained as follows. In case of non-constant power load models, when DG is placed, the voltage prole of the system changes (same is the case for constant power load model also) and the system real and reactive loads change due to their voltage dependence (unlike constant power load model). Therefore, the ows in the lines normally increase to supply the increased bus loadings. In fact, the overload of line-7 is due to increased ow (in the same direction) and not due to counter ow. It may appear that assuming low (10%15%) DG penetrations, the differences reported in this work, may not be signicant. The following investigations are carried out to clarify that it is not so. % of Let us assume that location for DG of 0.3750 p.u. ( ) is to be obtained for the following cases. the a) To Minimize MVA Requirement : The location for constant power load model is bus-16 (as in case of minimum conguration), but for mixed load, it is bus-2. The location and and the relevant quantities such as are signicantly different. b) To Minimize the System Losses: The location for constant power load model and mixed load model is found to be for constant load same, i.e., bus-16 (Table IV). However, the model comes out to be 0.1500 p.u. (reduction of 0.0389 p.u.) whereas, for the mixed load model, it is 0.1337 p.u. (reduction of 0.0325 p.u.) which is a signicant difference (around 19%). for constant load model is 0.0.0997 p.u. (showing Similarly, loss reduction of 0.0268 p.u.), whereas for the mixed load it is 0.0879 p.u. (reduction of 0.0225 p.u.) which is also signicantly are different (around 19%). Similarly, the reductions in 0.3612 and 0.3164 for constant load and mixed load, respectively. In the present work, high values of DG penetration appear due /minimum to optimization of certain parameters (minimum ). These high values suggest that planning studies show a signicantly higher possible optimal DG penetration levels when the constant power load models are assumed instead of actual load models. Though the investigations are performed on radial system, the ndings are relevant to all topologies. VI. CONCLUSION Load models signicantly affect the DG planning. It is established that DG planning based on constant power load models is not effective after implementation on actual systems. Decisions and results based on constant power load model assumptions are not technically feasible if employed on system having non-constant power load behavior. Also in the ndings, a signicant effect of load models is observed on optimal location

TABLE VIII SYSTEM AND LOAD DATA FOR 38-NODE SYSTEM

Fig. 9. The 38-node test system.

and size of DG resources in distribution system. A rise in reactive demand is observed when DG is placed in an actual system having non-constant power load model, which is contrary to the observations in case of constant power load model assumption. It is shown that though the losses get reduced, the total power intake of the system increases, which makes loss reduction redundant. It is established in this work that the quantities affected are 1) number of voltage limit violation; 2) number of line limit violations; 3) MVA, real, and reactive p.u. demand on the main

2212

IEEE TRANSACTIONS ON POWER SYSTEMS, VOL. 22, NO. 4, NOVEMBER 2007

substation; 4) loss reduction in real and reactive power; 5) saving of MVA capacity on the main substation; and 6) optimal location and size of DG. It was observed that most or all of the quantities are affected. APPENDIX Table VIII has the system and load data for the 38-node system. Fig. 9 shows the 38-node test system. REFERENCES [1] Y. G. Hegazy, M. M. A. Salama, and A. Y. Chikhani, Adequacy assessment of distributed generation systems using Monte Carlo simulation, IEEE Trans. Power Syst., vol. 18, no. 1, pp. 4852, Feb. 2003. [2] W. El-Khattam, Y. G. Hegazy, and M. M. A. Salama, An integrated distributed generation optimization model for distribution system planning, IEEE Trans. Power Syst., vol. 20, no. 2, pp. 11581165, May 2005. [3] C. Wang and M. H. Nehrir, Analytical approaches for optimal placement of distributed generation sources in power systems, IEEE Trans. Power Syst., vol. 19, no. 4, pp. 20682076, Nov. 2004. [4] L. F. Ochoa, A. Padilha-Feltrin, and G. P. Harrison, Evaluating distributed generation impacts with a multiobjective index, IEEE Trans. Power Del., vol. 21, no. 3, pp. 14521458, Jul. 2006. [5] P. Chiradeja and R. Ramakumar, An approach to quantify the technical benets of distributed generation, IEEE Trans. Energy Convers., vol. 19, no. 4, pp. 764773, Dec. 2004. [6] C. Concordia and S. Ihara, Load representation in power systems stability studies, IEEE Trans. Power App. Syst., vol. PAS-101, no. 4, pp. 969977, Apr. 1982. [7] IEEE Task Force on Load Representation for Dynamic Performance, Bibliography on load models for power ow and dynamic performance simulation, IEEE Trans. Power Syst., vol. 10, no. 1, pp. 523538, Feb. 1995. [8] IEEE Task Force on Load Representation for Dynamic Performance, Load representation for dynamic performance analysis, IEEE Trans. Power Syst., vol. 8, no. 2, pp. 472482, May 1993. [9] IEEE Task Force on Load Representation for Dynamic Performance, Standard load models for power ow and dynamic performance simulation, IEEE Trans. Power Syst., vol. 10, no. 3, pp. 13021313, Aug. 1995. [10] D. T. Rizy, J. S. Lawler, J. B. Patten, and W. R. Nelson, Measuring and analyzing the impact of voltage and capacitor control with high speed data acquisition, IEEE Trans. Power Del., vol. 4, no. 1, pp. 704714, Jan. 1989. [11] S. Arnborg, G. Anderson, D. J. Hill, and I. A. Hiskens, On inuence of load modelling for undervoltage load shedding studies, IEEE Trans. Power Syst., vol. 13, no. 2, pp. 395400, May 1998.

[12] T. Gozel, M. H. Hocaoglu, U. Eminoglu, and A. Balikci, Optimal placement and sizing of distributed generation on radial feeder with different static load models, in Proc. Int. Conf. Future Power Systems, Nov. 2005, pp. 16. [13] M. E. Baran and F. F. Wu, Network reconguration in distribution systems for loss reduction, IEEE Trans. Power Del., vol. 4, no. 2, pp. 14011407, Apr. 1989. Devender Singh received the B.E. degree in electrical engineering from Sardar Vallabhbhai Regional College of Engineering and Technolgy, Surat, India, in 1993, the M.E. degree in electrical engineering from Motilal Nehru Regional Engineering College, Allahabad, India, in 1999, and Ph.D. in electrical engineering from Institute of Technology (IT), Banaras Hindu University (BHU), Varanasi, India. Presently, he is Reader in the Department of Electrical Engineering, IT, BHU. His research interests are distribution generation planning, state estimation, short-term load forecasting, and AI applications in power systems.

R. K. Misra received the B.Sc. (Engg.) degree in electrical engineering and the M.Tech. degree in engineering systems from Dayalbagh Educational Institute, Agra, India, in 1995 and 1997, respectively, and the Ph.D. degree in electrical engineering from Institute of Technology (IT), Banaras Hindu University (BHU), Varanasi, India. Presently, he is Reader in the Department of Electrical Engineering, IT, BHU. His research interests are distribution generation planning, power system security, and AI applications in power systems.

Deependra Singh received the B.Tech. degree in electrical engineering from Harcourt Butler Technological Institute, Kanpur, India, in 1997 and the M.E. degree in electrical engineering from the University of Roorkee, Roorkee, India, in 1999. Currently, he is pursuing the Ph.D. degree from UP Technical University, Lucknow, India. He is a Lecturer in the Department of Electrical Engineering, Kamla Nehru Institute of Technology, Sultanpur (UP), India. His research interests are distributed generation planning and distribution system analysis.