You are on page 1of 8

IEEE TRANSACTIONS ON POWER SYSTEMS, VOL. 27, NO.

2, MAY 2012

951

A Method for the Design of UFLS Schemes of Small Isolated Power Systems
Lukas Sigrist, Student Member, IEEE, Ignacio Egido, and Luis Rouco, Member, IEEE
AbstractThis paper presents a systematic method for the design of robust and efcient underfrequency load-shedding (UFLS) schemes. UFLS schemes play an important role in protecting the system integrity. The systematic method consists rst in selecting representative operating and contingency (OC) scenarios by means of a clustering algorithm and subsequently, in tuning UFLS scheme parameters by dint of a simulated annealing optimization algorithm. The approach is applied to a small isolated Spanish power system. The systematic method leads to a robust and efcient UFLS scheme. The resulting design is also compared to a design based on OC scenarios determined by the common practice of OC scenario selection. The possibility of rearranging UFLS stages and the inuence of minimum allowable frequency constraints is analyzed as well. Finally, an analysis of the impact of increasing converter-connected generation (CCG) is presented. Index TermsDesign methodology, frequency stability, load shedding, power system protection.

I. INTRODUCTION

REQUENCY stability of power systems is concerned with the ability of generators to supply their loads at an acceptable frequency after generator and load imbalances. Small isolated power systems are especially sensitive to real-power imbalances. It is crucial to avoid the frequency falling below a certain value, since low frequency may severely harm power plants and load-side equipment and hence, the system integrity [1]. Increasing integration of converter-connected generation (CCG) (e.g., photovoltaics or full-converter wind generators) enhances the risk of frequency instability due to the negligible primary control and inertia [2]. Underfrequency load-shedding (UFLS) schemes are a last-resort tool to protect the power system in case of a severe disturbance [3]. In small isolated power systems, UFLS schemes play an important role in protecting the system integrity. Most of the UFLS schemes implemented today are conventional static and semi-adaptive schemes [4], [5]. These schemes continuously measure frequency and optionally the rate of change of frequency (ROCOF) by means of type 81 relays and shed a predened amount of load in case frequency and/or the rate of change of frequency fall below a certain threshold. Electric utilities adopt different approaches to design UFLS schemes, which
Manuscript received May 04, 2011; revised July 21, 2011 and September 14, 2011; accepted October 28, 2011. Date of publication December 05, 2011; date of current version April 18, 2012. This work was supported by Red Elctrica de Espaa, Madrid, Spain. Paper no. TPWRS-00410-2011. The authors are with the School of Engineering ICAI, Universidad Ponticia Comillas, 28015 Madrid, Spain (e-mail: Lukas.Sigrist@iit.upcomillas.es). 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.2011.2174448

are mainly based on their experience, and they usually follow typical design criteria on the number of steps, step size, frequency thresholds, etc. [6][8]. Several methods have been reported in the literature to design conventional UFLS schemes, i.e., to tune frequency and ROCOF thresholds, intentional time delays, and step sizes. Some of these methods are based on iterative trial and error procedures [9]; others make use of a screening process among a host of candidate schemes [6], [7]. However, these methods do not necessarily guarantee an efcient performance of the scheme. Both deterministic and heuristic optimization algorithms have also been applied to the design of conventional UFLS schemes to minimize the amount of shed load [10][13]. In [12], frequency thresholds, intentional time delays, and step sizes are optimized by means of a quasi-Newton method. Deterministic optimization algorithms depend however on the initial guess of the decision variables. Furthermore, the objective function, which aims at minimizing the amount of shed load, becomes discontinuous and shows a step-like shape if step sizes are not considered as decision variables (which is usually not possible in small isolated power systems), impeding the use of gradient-based methods. Heuristic algorithms seem to be an appropriate alternative [13]. A genetic algorithm (GA) has been used to optimize the step size of a very simple single-stage UFLS scheme of a test system, maintaining constant frequency threshold and intentional time delay [13]. However, for small isolated power systems, it is difcult to realize the optimized step size by rearranging available feeder load-blocks. Irrespective of the design method applied, the adequate selection of hypothetical contingencies the system should be protected for is crucial to design robust UFLS schemes [9]. Different contingencies could be initially considered (line tripping, generation outages, sudden load increase, etc.), but only generation outages are contemplated here, since they lead to pronounced frequency deviations [7]. The common practice of contingency selection consists in determining system operation conditions corresponding to different load-demand levels (e.g., minimum and maximum) and to design the UFLS scheme taking into account the outages of the largest and the smallest generating unit for each of these system conditions [7], [14]. In [15], clustering techniques have been applied to select representative OC scenarios for subsequent design of UFLS schemes, but the design itself has not been addressed. Similarly, most design methods did not consider the selection of OC scenarios [9][13]. This paper proposes a method for the design of robust and efcient UFLS schemes. To guarantee a robust UFLS scheme,

0885-8950/$26.00 2011 IEEE

952

IEEE TRANSACTIONS ON POWER SYSTEMS, VOL. 27, NO. 2, MAY 2012

representative OC scenarios are selected by means of clustering techniques. Efciency of the UFLS scheme can be achieved by applying an optimization algorithm such as simulated annealing (SA) to adjust the UFLS scheme parameters by minimizing the amount of shed load. SA is a heuristic method suited to solve complex optimization problems [16], [17], and it has been successfully applied to problems related to power-system stability, operation, and planning [18][20]. The proposed method makes use of a simplied powersystem model, taking into account power output limitations. This model allows simulating and evaluating the behavior of a small isolated power system following a disturbance. In [10] and [12], linear power-system models have been supposed, neglecting power output limitations, and intentional time delays have been modeled as simple delays, omitting their frequency-dependent characteristics. Finally and unlike other design methods, step sizes are a priori not considered as decision variables and only frequency and ROCOF thresholds as well as intentional time delays are optimized. In very small power systems, it is difcult to realize a desired step size by rearranging available feeder load-blocks. However, the possibility exists to rearrange UFLS stages, corresponding to a somehow rough step-size optimization. Section II describes the simplied power system model the time-domain simulations are based on; in Section III, the systematic method for the design of robust and efcient UFLS scheme is presented; in Section IV, the developed method is applied to the design of the UFLS scheme of a small isolated Spanish power system. A sensitivity analysis with respect to design conditions and an analysis of the impact of increasing CCG on the design of UFLS schemes are presented as well. Section V concludes the paper.

Fig. 1. System-frequency dynamics model of the power system.

and are also modeled. The output limitations inclusion of CCG can be easily realized if inertia and parameters of the second-order generating unit model were known. Nonlinearities introduced by both power output limitations and the step-like behavior of the UFLS scheme in Fig. 1 require simulating the response of the power system to a disturbance. and shed load These responses in terms of frequency are subsequently used within the systematic method for the design of UFLS schemes. III. METHOD FOR THE DESIGN OF UFLS SCHEMES In this section, the systematic method for the design of robust and efcient UFLS schemes is presented, covering both selection of OC scenarios and optimal tuning of UFLS scheme parameters. This method comprises two tasks: the selection of representative OC scenarios and the application of the SA optimization algorithm. A. Selection of Operating and Contingency Scenarios It is crucial to select adequate OC scenarios to design robust UFLS schemes. Not only N-1 outages have to be considered, but also multiple generation outages could occur. OC scenarios are then dened as the outage of a single generating unit or as the simultaneous outage of multiple generating units for a particular system operating condition (SOC). The clustering-based method proposed in [15] is used here to identify representative OC scenarios, i.e., to nd those scenarios which represent all other possible scenarios best. Clustering is carried out with regard to system responses in terms of to all possible OC scenarios. Note also that the frequency UFLS scheme does not intervene during the OC scenario selection process. The quality of the OC scenario selection can be evaluated by statistical measurements such as principal component analysis (PCA), etc. [22]. PCA is mathematically dened as an orthogonal linear transformation that transforms the data to a new coordinate system such that the greatest variance by any projection of the data comes to lie on the rst coordinate (called the rst principal component), the second greatest variance on the second coordinate, and so on.

II. SYSTEM MODELING This section outlines the model used to represent and simulate small isolated power systems. This model is widely used for the analysis and design of UFLS schemes and is able to reect short-term frequency dynamics of small isolated power systems [21]. Fig. 1 details the power-system model used to design UFLS schemes of a small isolated power system, consisting of generating units. Each generating unit is represented by a secondorder model approximation of its turbine-governor system. In fact, frequency dynamics are dominated by rotor and turbinegovernor system dynamics. Excitation and generator transients can be neglected for being much faster than the turbine-governor dynamics. Steam turbines can be usually represented by rstorder models as in [12], but generation mix also contains gasdriven and Diesel-driven turbines, which might require higher order models. Since frequency can be considered uniform [21], can be dened. The overall rean equivalent system inertia sponse of loads can be taken into account by means of a loaddamping factor if its value is known. The gain and parameters , , , and , of each generating unit , can be deduced from more accurate models or eld tests. Since primary spinning reserve is nite, power

SIGRIST et al.: A METHOD FOR THE DESIGN OF UFLS SCHEMES OF SMALL ISOLATED POWER SYSTEMS

953

B. Tuning of UFLS Scheme Parameters Efciency of the UFLS scheme depends on the tuning of its parameters, which is proposed to be carried out in function of the selected representative OC scenarios. The problem of UFLS tuning is formulated as an optimization problem (1).

(1) The main objective of a UFLS scheme is to protect a power system against instability by curtailing a minimum amount of aims at minimizing load; therefore, the objective function the amount of shed load, whereas stability is guaranteed by im. and posing appropriate inequality constraint functions are the lower and upper bounds of the decision variables . In its most generic formulation, decision variables of the optimization problem correspond to the frequency and ROCOF thresholds, the intentional time delays, and the step sizes of underfrequency and ROCOF stages. However, for rather small power systems, it does not make sense to adjust the step size. In fact, the step size is usually dened by the utility and also depends on the feeders connected to the relays dened by a particular stage and on the priority of the associated loads; therefore, in smaller power systems, it will be difcult to nd feeder blocks which nally sum up to the desired step size. In addition, implemented step size usually differs from real step size due to feeder-load variation, feeder outages, or breaker failures. Thus, type 81 (underfrequency and ROCOF) relay parameters, , , and ) i.e., frequency and ROCOF thresholds ( and ) are a as well as intentional time delays ( priori considered here as decision variables as shown in (2). This corresponds to a more practical, industry-oriented optimization strategy. The settings of the existing UFLS scheme can be used as initial values for the decision variables: (2) The objective function complying with the main objective is (3) where is a weighting factor, the amount of shed load in pu for the th representative OC scenario, the vector of decision variables, and is the number of representative OC scenarios. Other objective function formulations also include an additional term related to frequency deviations (see, e.g., [11]), but it seems more intuitive to include restrictions on frequency is set to unity deviation as constraints. The weighting factor here as in [12], attaching equal importance to the amount of shed load in each OC scenario. Another possibility could be to make inversely proportional to the amount of lost generation in pu. This way, load shedding for smaller outages outweighs load shedding for larger outages. Constraints can be divided into two categories. The rst one contains the constraints imposed by the power system, whereas the second category covers the constraints with respect to the performance of the UFLS scheme. The constraints of the rst

category are 1) the minimum and 2) the maximum allowable freand ). In general, quency values ( the constraint of minimum allowable frequency is accompanied , during which freby a maximum time delay, quency can stay below the minimum allowable frequency. Typical values for small isolated power systems are 47.5 Hz during maximum 3 s [23]. The maximum allowable frequency varies between 51.5 Hz and 52 Hz. Equation (4) implements these two constraints for the considered OC scenarios: . . .

. . . The constraint

(4)

requires that the time frequency that is below is smaller than . The constraint requires that the maximum frequency, , is always smaller than or equal to . The constraints with respect to the performance of the UFLS scheme are 3) the UFLS scheme does not act once the frequency has passed its minimum value since frequency is returning towards its nominal value (instant of shedding), 4) the amount of , is smaller than or at most equal to the amount shed load, (amount of shed load), and 5) the UFLS of lost real power, scheme respects the priority of loads (priority). The constraint in (5) requires that the instant of last shedding, , is prior , whereas the conto the instant of minimum frequency, requires that is smaller than or at most equal to straint . The constraint on priority is implemented such that its associated constraint function is positive whenever a UFLS stage actuates without that its preceding stage has actuated: . . .

. . .

(5)

The SA optimization algorithm [17] is applied to solve the optimization problem (1). The objective function is translated into an energy function, which also takes into account the constraints by means of a penalty function. This energy function is given by (6): (6) indicates whether the constraint is active (i.e., 1) where and are a penalty constants or not (i.e., 0), whereas taking distinct values for different constraints and is the th constraint function. In fact, these two penalty constants ensure that the importance of the constraint is taken into account. is the number of constraints and is the number of considered

954

IEEE TRANSACTIONS ON POWER SYSTEMS, VOL. 27, NO. 2, MAY 2012

Fig. 2. Flow chart of the UFLS scheme tuning process based on the SA optimization algorithm.

Fig. 3. Comparison of OC scenarios determined by the common practice and by the K-Means clustering algorithm. (a) Time domain. (b) Principal component analysis.

OC scenarios. The energy function is evaluated by time-domain simulations using the power-system model in Fig. 1. These simulations evaluate in turn the impact of the UFLS scheme dened by on the representative OC scenarios in every iteration of the SA algorithm. is varied according to a neighborhood function and a cooling schedule as described in [16]. The tuning process of a UFLS scheme using SA optimization algorithm and contemplating representative OC scenarios is shown in Fig. 2. IV. APPLICATION TO AN ISOLATED SPANISH POWER SYSTEM The systematic method for the design of UFLS schemes is applied to a small isolated Spanish power system. This power system consists of eleven generators and its peak load demand is around 35 MW. The parameters of the generating unit models and the different system operation conditions (SOC) are given in the Appendix. The SOC include CCG which covers up to 12% of total load demand and which corresponds to the average contribution of CCG in supplying load demand of the considered isolated Spanish power system. At rst, the systematic method is applied step by step to the existing UFLS scheme of the considered power system. The resulting UFLS scheme will be referred to as base case. Subsequently, a sensitivity analysis with respect to design conditions is presented. Finally, the impact of increasing CCG on the design of UFLS schemes is discussed. A. Base Case In a rst step, the representative operating and contingency scenarios are determined by means of K-Means algorithm. It has been found by applying the K-Means algorithm iteratively that four clusters are sufcient to represent all possible 164 OC scenarios. N-1 outages have been considered since they include generation loss of over 50% of the total demand. Fig. 3(a) shows the representative OC scenarios determined by the K-Means algorithm and compares them to the OC scenarios given by the common practice, i.e., the OC scenarios corresponding to the outages of the largest and smallest generating unit for the maximum and minimum load-demand level. Note that during the process of OC scenarios selection, the UFLS scheme does not intervene and that the worst OC scenarios are in both cases nearly the same. According to Fig. 3(a), the clustering-based method covers a wider range of possible system responses. In Fig. 3(b), rst and second principal components of the

TABLE I EXISTING (EXIST) AND SYSTEMATICALLY DESIGNED (SYST) UFLS SCHEME OF THE CONSIDERED POWER SYSTEM

representative scenarios are superposed to the principal components of all OC scenarios and the principal components of the common practice. For example, the rst three OC scenarios of the common practice seem to be very close, whereas the clustering-based scenarios are rather widespread, thus better covering the variance within all the possible OC scenarios. The existing UFLS scheme, which is currently used by the system operator of the contemplated Spanish isolated power system, is shown in Table I. After large disturbances and in order to enable the power system to withstand the same disturbances next time, this UFLS scheme has been revised and manually readjusted by the system utility using OC scenarios selected according to the common practice. The existing UFLS scheme will be redesigned using the proposed method for the design of robust and efcient UFLS scheme outlined in Section III. and Decision variables are the frequency thresholds , the ROCOF thresholds , and the intentional time . The UFLS scheme parameters are bounded above delays and below as follows: 1) , 2) , 3) , and . Step sizes of the underfrequency and 4) ROCOF stages have not been considered as a decision variable and therefore, the step sizes of the existing UFLS scheme have accounts for measurement and tripping delays. been used. The optimization constraints for the base case design are: the minimum and maximum allowable frequencies, the instant of

SIGRIST et al.: A METHOD FOR THE DESIGN OF UFLS SCHEMES OF SMALL ISOLATED POWER SYSTEMS

955

TABLE III COMPARISON OF DIFFERENT UFLS SCHEME DESIGNS

Fig. 4. (a) Comparison of the system responses of the existing and the systematically designed UFLS scheme. (b) Convergence characteristics of the optimization algorithm.

TABLE II COMPARISON OF THE PERFORMANCE OF THE EXISTING AND THE SYSTEMATICALLY DESIGNED UFLS SCHEME

Fig. 5. Comparison of the system responses in terms of frequency. (a) Existing UFLS scheme. (b) Systematically designed UFLS scheme.

shedding, the amount of shed load, and the priority. In particular and taking into account (4), minimum allowable frequenof 48 Hz and 47 Hz with corresponding cies of 2 s and 0 s, respectively, have been imposed. is set to 52 The maximum allowable frequency Hz. Constraints imposed on minimum allowable frequency are more restrictive than those described in Section III (e.g., for maximum 3 s). This leads to a more conservative scheme and gives therefore a certain safety margin. Further, more restrictive constraints might also benet UFLS scheme performance in the case of assumed step sizes do not correspond with real step sizes due to feeder-load variation, feeder outages, or breaker failures. Table I shows the systematically designed UFLS scheme. The number of stages adjusted depends on the maximum amount of lost power generation in the contemplated OC scenarios. For example, the worst OC scenario in Fig. 3 corresponds to the outage of generator G17 of the SOC 4, implying a loss of about 50% of power generation. Thus, using the rst six stages of the UFLS scheme, the system can be stabilized. Remaining steps (7 to 10) could be used as backup steps in case of step-size variations or non-responding turbine-governor systems. Backup steps support the UFLS scheme during such events and guarantee system integrity, and minimizing the amount of shed load is not paramount anymore when tuning their settings. In comparison to and intenthe existing UFLS scheme, frequency thresholds tional time delays of the underfrequency relays have been and of the lowered, whereas frequency thresholds ROCOF relays have been increased. The latter allows a faster UFLS scheme intervention in case of severe disturbances. Fig. 4(a) and Table II compare the two UFLS schemes. Fig. 4(a) shows the system responses in terms of frequency to the four representative OC scenarios. It can be inferred that the minimum and maximum frequency deviations clearly exceed

the minimum and maximum allowable frequency deviations in case of the existing UFLS scheme. Frequency overshoot also indicates overshedding. In the case of the systematically designed UFLS scheme, frequency is conned within the allowable frequency range. Fig. 4(b) shows the systematic methods convergence characteristics (see also Section IV-B4). Table II provides some additional information on the performance of both the existing and the optimized UFLS scheme. Table II indicates whether the UFLS scheme satises the constraints on its performance (state) and in addition, provides a cause in case of constraint violation. It can be inferred that the existing UFLS scheme sheds too much load for the rst two contingencies (cause a). Furthermore, load is shed after the frequency has passed its minimum (cause b). Overshedding is also the reason for the large frequency overshoot and the resulting frequency instability since generators cannot lower their output to the new demand level. Finally, it can be seen from Table II that the systematically designed UFLS scheme sheds less amount of load than the existing UFLS scheme (Pshed) and that no constraint on its performance is violated. Thus, the existing UFLS scheme has been successfully optimized with respect to the considered representative OC scenarios. To come full circle, both the existing and the systematically designed UFLS scheme are applied to all possible 164 N-1 OC scenarios. The rst two lines in Table III compare the performance of both UFLS schemes. According to Table III, the amount of shed load has been remarkably reduced after implementing the systematically designed UFLS scheme. This is also reected in the lower number of relays which tripped. In addition, fewer errors (e.g., overshedding, low frequency, etc.) occurred thanks to the systematically designed UFLS scheme. have also been slightly Total frequency deviations reduced. Fig. 5 nally shows and compares the system responses in terms of frequency of the existing and the systematically designed UFLS scheme. It can be seen that in case of the systemati-

956

IEEE TRANSACTIONS ON POWER SYSTEMS, VOL. 27, NO. 2, MAY 2012

cally designed UFLS scheme, frequency deviations are conned within the imposed values of maximum and minimum allowable frequency. By contrast, the existing UFLS scheme causes low and high frequencies and frequency instabilities since generators cannot lower their output to the new load demand level. B. Sensitivity Analysis With Respect to Design Conditions This subsection presents a sensitivity analysis of UFLS scheme designs with respect to the imposed design conditions; in particular, four cases are analyzed: 1) the use of OC scenarios selected according to the common practice, 2) the imposition of a minimum allowable frequency constraint of 47.5 Hz with of 3 s, 3) the neglect of the priority constraint, and 4) the use of a GA as in [13]. Case 3 is also compared with a complete optimization, i.e., when step sizes are considered as decision variables as well. Anew, the UFLS schemes resulting from the analyzed cases are applied to all possible OC scenarios. Table III displays and compares the four cases. 1) Common Practice of OC Scenario Selection: The principal idea of clustering-based OC scenario selection is to obtain most appropriate OC scenarios. Fig. 3 showed that clustering-selected OC scenarios cover a wider range of frequency responses and therefore, a more robust UFLS scheme could be expected. Three important characteristics with respect to the robustness of the systematic design can be deduced by comparing the base case with case 1 (Common practice) in Table III. First, the number of errors (occasions of poor performance) is remarkably reduced in the case of the base case scheme. Frequency deviations have also been reduced which is indicated by the lower sum of maximum frequency deviations (corresponding to minimum frequencies). Finally, the amount of shed load has been signicantly reduced thanks to the base case design. Thus, the systematic design given by the base case guarantees a robust and efcient UFLS scheme. : Con2) Minimum Allowable Frequency of straints imposed on minimum allowable frequency are more restrictive than the Spanish grid code [23], which leads to a more conservative scheme and gives therefore a certain safety margin. A relaxation on the minimum allowable frequency constraint by permitting the frequency to stay below 47.5 Hz for maximum 3 s could result in a lower amount of shed load since the constraint is less restrictive. By comparing the base case with case 2 (47.5 Hz-3 s) in Table III, it can be readily deduced that a relaxation on the minimum allowable frequency constraint remarkably reduces the amount of shed load as expected, at the expense of increased frequency deviations. By relaxing the constraint on allowable minimum frequency by applying the 47.5 Hz3 s constraint, one accepts very low minimum frequencies (e.g., 46 Hz or less) although power-system responses nally comply with the imposed constraint. This enhances the risk of tripping generating units by their underfrequency protection. 3) Neglect of Priority Constraint: Step sizes have not been considered as a decision variables, since for rather small power systems, it does not make sense to adjust the step size. An attractive alternative consists in rearranging the established order

of steps by neglecting the priority constraint. The omission of the constraint on the priority should intuitively lead to more efcient UFLS scheme designs since this corresponds somehow to a rough optimization of the step size. Table III shows the results of completely optimized UFLS scheme (Complete optimization) and the one obtained by omitting priority (No priority). It can be readily deduced that omitting priority reduces the amount of shed load with regard to the base case design. In addition, it is noteworthy that the efciency of the UFLS scheme designed without considering priority constraints approaches the efciency achieved by an SA-based complete optimization of the UFLS scheme. This shows that for smaller isolated power systems, where complete optimization is theoretically feasible but less practical, relaxing the constraint on priority leads to very efcient UFLS schemes with a performance similar to completely optimized UFLS scheme. 4) Use of Genetic Algorithm: SA optimization algorithm has been used so far. A GA optimization algorithm as proposed in [13] is applied using the same genetic operators and parameters. As shown in Table III, case 4 (GA) has a similar (although slightly worse) impact on all possible OC scenarios in terms of minimum frequency deviations, amount of shed load, and the number of errors as the base case. These small differences are mainly due to the fact that the objective function has a step-like shape and that for the same nal objective function value, slightly different decision variable values can be obtained. It seems that different heuristic optimization algorithms lead to similar results. Fig. 4(b) compares SA and GA convergence characteristics, and it can be seen that they are similar, although SA seems to be slightly faster in converging to the nal value. C. Analysis of the Impact of Increasing CCG The impact of CCG on the response of a simple power system in terms of frequency has been analyzed in [2]. Negligible primary control and the absence of inertial response increase frequency deviations and the initial slope of frequency. Larger amounts of shed load could be therefore expected. Two additional CCG penetration levels have been contemplated: 1) CCG covering up to 24% of total load demand and 2) CCG covering up to 34% of total load demand. Further and in order to study the impact of increasing CCG on the design process, a UFLS scheme has been designed for the CCG level of 24% using the proposed method. Table IV shows and compares the impact of increasing CCG levels on the performance of the base case and the performance of the UFLS scheme designed for the CCG level of 24% (Increased CCG). Both UFLS schemes are applied to all possible OC scenarios of the initial and the two additional levels of CCG (12%, 24%, and 34%, respectively). It can be deduced from Table IV that the total amount of shed load and the number of error increase with the increasing level of CCG. Note that most errors are due to low frequency and overshedding. It can be also seen that for CCG levels of 24% and 34%, the UFLS scheme designed for the CCG level of 24% sheds less load than the base case (designed for a CCG level of 12%) and reduces frequency deviations. Further, the number of

SIGRIST et al.: A METHOD FOR THE DESIGN OF UFLS SCHEMES OF SMALL ISOLATED POWER SYSTEMS

957

TABLE IV IMPACT OF INCREASING LEVELS OF CCG IN SOC ON UFLS SCHEME DESIGNS

TABLE VI SYSTEM OPERATION CONDITIONS (SOC)

TABLE V PARAMETERS OF THE TURBINE-GOVERNOR SYSTEM MODEL

ACKNOWLEDGMENT errors is slightly larger in the case of the base case. It seems that contemplating higher levels of CCG during the design process improves the UFLS scheme performance. However, the UFLS scheme designed for the CCG level of 24% exhibits a slightly worse performance in terms of shed load and number of errors than the base case for a CCG level of 12%. Similar to the ndings of Section IV-B1, the selection of SOC with different levels of CCG penetration and the selection of OC scenarios seem to inuence both the UFLS scheme design and its performance. Thus, a representative CCG penetration level should be used during UFLS scheme design process. V. CONCLUSIONS This paper presents a systematic method for the design of robust and efcient UFLS schemes. The proposed method consists of two tasks: rst, representative OC scenarios are selected by means of the K-Means clustering algorithm and subsequently, UFLS scheme parameters are tuned by dint of the simulated annealing optimization algorithm The approach has been successfully applied to a small isolated Spanish power system. The systematically designed UFLS scheme outperformed the existing scheme. Further, it has been shown that the systematic design is more robust than a design based on OC scenarios selected according to the common practice, emphasizing the importance of an adequate selection of OC scenarios. It has been also shown that the choice of minimum allowable frequency constraints signicantly inuences the UFLS scheme performance and that by rearranging UFLS stages, a similar performance can be obtained as by optimizing step sizes. Finally, the impact of increasing converter-connected generation on the design of UFLS schemes has been analyzed. APPENDIX Table V lists the parameters of the turbine-governor system model, and Table VI lists the system operation conditions (SOC). The authors would like to thank S. Marn, J. Ruperez, A. Rodrguez, and C. Castro for providing valuable comments and discussions throughout the collaboration. REFERENCES
[1] IEEE C37.106, Guide for Abnormal Frequency Protection for Power Generating Plant, IEEE Standards, IEEE, 2004. [2] L. Rouco et al., Impact of Wind Power Generators on the Frequency Stability of Synchronous Generators. Paris, France: Cigr Session 2008. [3] R. M. Maliszewski, R. D. Dunlop, and G. L. Wilson, Frequency actuated load shedding and restoration. Part IPhilosophy, in Proc. IEEE Summer Power Meeting and EHV Conf., Los Angeles, CA, Jul. 1217, 1970, pp. 14521459. [4] IEEE C37.117, IEEE Guide for the Application of Protective Relays Used for Abnormal Frequency Load Shedding and Restoration, IEEE Standards, IEEE, 2007. [5] UFLS Regional Reliability Standard Characteristics. Washington, DC: NERC, 2008. [6] E. J. Thalassinakis and E. N. Dialynas, A Monte-Carlo simulation method for setting the underfrequency load shedding relays and selecting the spinning reserve policy in autonomous power systems, IEEE Trans. Power Syst., vol. 19, no. 4, pp. 20442052, Nov. 2004. [7] C. Concordia, L. H. Fink, and G. Poullikas, Load shedding on an isolated system, IEEE Trans. Power Syst., vol. 10, no. 3, pp. 14671472, Aug. 1995. [8] B. Delno et al., Implementation and comparison of different under frequency load-shedding schemes, in Proc. 2001 IEEE Power Eng. Soc. Summer Meeting, Piscataway, NJ. [9] H. E. Lokay and V. Burtnyk, Application of underfrequency relays for automatic load shedding, IEEE Trans. Power App. Syst., vol. PAS-87, no. 3, pp. 720726, Mar. 1968. [10] L. Jenkins, Optimal load shedding algorithm for power system emergency control, in Proc. 1983 IEEE Int. Conf. Systems, Man and Cybernetics., Bombay and New Delhi, India. [11] Y. Halevi and D. Kottick, Optimization of load shedding system, IEEE Trans. Energy Convers., vol. 8, no. 2, pp. 207213, Jun. 1993. [12] A. D. L. Hau, A general-order system frequency response model incorporating load shedding: Analytic modeling and applications, IEEE Trans. Power Syst., vol. 21, no. 2, pp. 709717, May 2006. [13] J. A. P. Lopes et al., Optimum determination of underfrequency load shedding strategies using a genetic algorithm approach, in Proc. 32nd Annu. North American Power Symp., Waterloo, ON, Canada, 2000. [14] J. G. Thompson and B. Fox, Adaptive load shedding for isolated power systems, Proc. Inst. Elect. Eng., Gen., Transm., Distrib., vol. 141, no. 5, pp. 491496, Sep. 1994. [15] L. Sigrist et al., Representative operating and contingency scenarios for the design of UFLS schemes, IEEE Trans. Power Syst., vol. 25, no. 2, pp. 906913, May 2010.

958

IEEE TRANSACTIONS ON POWER SYSTEMS, VOL. 27, NO. 2, MAY 2012

[16] L. Ingber, Simulated annealing: Practice versus theory, J. Math. Comput. Model., pp. 2957, 1993. [17] Genetic Algorithm and Direct Search Toolbox Users Guide. Natick, MA: MathWorks, 2007. [18] J. Zhu et al., Phase balancing using simulated annealing, IEEE Trans. Power Syst., vol. 14, no. 4, pp. 15081513, Nov. 1999. [19] D. N. Simopoulos, S. D. Kavatza, and C. D. Vournas, Reliability constrained unit commitment using simulated annealing, IEEE Trans. Power Syst., vol. 21, no. 4, pp. 16991706, Nov. 2006. [20] M. A. Abido, Robust design of multimachine power system stabilizers using simulated annealing, IEEE Trans. Energy Convers., vol. 15, no. 3, pp. 297304, Sep. 2000. [21] P. M. Anderson, Power System Protection. Piscataway, NJ: IEEE Press, 1999. [22] J. Abonyi and B. Feil, Cluster Analysis for Data Mining and System Identication. Basel, Switzerland: Birkhuser, 2007. [23] Ministry of Industry Tourism and Commerce of Spain, Resolution 9613 of 26 April 2006 Establishing the Operation Procedures for Insular and Extra Peninsular Power Systems, Ofcial Bulletin of the State no 219 of 31 May 2006. [Online]. Available: http://www.ree.es. Lukas Sigrist (S08) received the M.Sc. degree in electrical and electronics engineering from cole Polytechnique Fdrale de Lausanne (EPFL), Lausanne, Switzerland, in 2007 and the Ph.D. degree from Universidad Ponticia Comillas, Madrid, Spain, in 2010. He is a researcher at Instituto de Investigacin Tecnolgica (IIT) of Universidad Ponticia Comillas, where he has been involved in a number of research projects related to power-system stability. His areas of interest are modeling, analysis, simulation, and identication of electric power systems.

Ignacio Egido was born in Arvalo (vila), Spain, in 1976. He received the M.S. and Ph.D. degrees in electrical engineering from the Universidad Ponticia Comillas, Madrid, Spain, in 2000 and 2005, respectively. He is currently an Assistant Professor in the Department of Electrical Engineering of the School of Engineering of Universidad Ponticia Comillas. He develops his research activities at the Instituto de Investigacin Tecnolgica (IIT) of the same university, where he has been involved in a number of research projects related to AGC and power system stability. His interests include control system design and power systems stability and control.

Luis Rouco (M91) received the Ingeniero Industrial and Doctor Ingeniero Industrial degrees from Universidad Politcnica de Madrid, Madrid, Spain, in 1985 and 1990, respectively. He is a Professor of electrical engineering in the School of Engineering of Universidad Ponticia Comillas, Madrid, attached to the Department of Electrical Engineering. He served as Director of the Department of Electrical Engineering from 1999 to 2005. He develops his research activities at Instituto de Investigacin Tecnologica (IIT) of the same university, where he has supervised more than 100 research and consultancy projects for Spanish and foreign companies. He has published more than 70 papers in conferences and journals. He has been a visiting researcher at Ontario Hydro, Toronto, ON, Canada; the Massachusetts Institute of Technology, Cambridge; and ABB Power Systems, Vasteras, Sweden. His areas of interest are modeling, analysis, simulation, and identication of electric power systems. Prof. Rouco is a member of Cigr, the Vice-President of the Spanish Chapter of the IEEE Power Engineering Society, and a member of the Executive Committee of the Spanish National Committee of Cigr.

You might also like