Sie sind auf Seite 1von 8

IEEE TRANSACTIONS ON POWER SYSTEMS, VOL. 25, NO.

1, FEBRUARY 2010

489

Scheduling of Droop Coefcients for Frequency and Voltage Regulation in Isolated Microgrids
Guzmn Daz, Member, IEEE, Cristina Gonzlez-Morn, Member, IEEE, Javier Gmez-Aleixandre, Member, IEEE, and Alberto Diez, Member, IEEE

AbstractThis paper details a procedure based on bifurcation theory to evaluate the impact that droops and primary reserve scheduling have on the microgrid stability. The methodology is based on nding the worst primary reserve sharethat is, the share closest to instabilitythat can be found after rescheduling the droops of selected generating units that support frequency (and voltage) regulation. The solutionwhich consists of a measure of the distance to instability in a given directionis found in a multi-parameter space endowed with coordinates corresponding to the droop coefcients. Two stages are proposed to achieve the solution. First, an investigation of the distance to bifurcation is computed in a one-dimensional parameter space in a dened search direction. Then the direction of this search is updated by calculating the normal vector at the found bifurcation point. The procedure is iteratively repeated until the closest bifurcation is found. The proposed approach is analyzed in a 69-bus and 11-generation unit isolated microgrid. It is shown through the analysis of some scenarios how the distances and normal vectors provide valuable insight on the correct scheduling from the stability point of view, giving advice on how the primary reserve should be more reliably scheduled. Index TermsBifurcation, dispersed storage and generation, frequency control, stability.

Fig. 1. Speed-droop (or governor) characteristic.

I. INTRODUCTION

ENERATING units (GUs) tted with speed-droop (or governor) characteristic provide primary frequency control to the synchronous zone where they are operating. Fig. 1, where the control system of a steam turbine governor is represented [1, p. 22], explains this issue. It is observed that a is followed by a valve change in frequency . If a droop coefcient is dened as opening , and a linear relationship between valve position and turbine power output is assumed, it follows that (1) This relationship is graphically represented in Fig. 1 for two GUs, using for convenience a back-to-back representation. ), After an increase in the total load demand ( the droop characteristic reestablishes the demand/generation balance at a lower frequency, using part of the GU reserves by
Manuscript received November 14, 2008; revised May 05, 2009. First published October 23, 2009; current version published January 20, 2010. This work was supported in part by the Spanish Ministry of Science under Grant DPI200762577. Paper TPWRS-00905-2008. The authors are with the University of Oviedo, Viesques 33204, Spain (e-mail: {guzman@uniovi.es; gonzalezmorcristina@uniovi.es; jgomez@uniovi.es; alberto@isa.uniovi.es). Digital Object Identier 10.1109/TPWRS.2009.2030425

applying the particular load division rule emanating from (1): , then the larger the corresponding share (i.e., the smaller the in Fig. 1). In this way, drooping governor characteristics allow for stable load division among synchronous GUs operating in parallel, preventing the inherent control conicts associated to isochronous governors [2, Section 11.1]. The drooping system is nowadays fully established in large power systems, where the system operator purchases several grades of spinning reserves to balance the system and to provide reliability. Spinning reserve is an ancillary service and as such is cleared in specialized markets. However, it is about the drooping system in microgridswith poorly regulated support for spinning reserve marketsthat we are concerned with in this paper. Microgrids are quite a novel concept, consisting of small power systems owning the capability of performing isolated from the main grid. They are mostly based on loads fed through LV or MV distribution radial systems, and their power is supplied by medium-low rated GUs, with probably inconsistent producer proles. Nevertheless, thanks to the drooping procedure, the isolated operation of a microgrid does not actually represent a control problem, even though many of the GUs in a microgrid are expected to be of the inverter-interfaced type (PV, microturbines, etc.). Indeed, Chandorkar et al. demonstrated that drooping characteristics can be successfully applied for controlling paralleled inverters in isolated ac systems [3], mimicking the performance of generator-turbine-governor units. It follows that any GU in the microgrid could ideally permode, despite of its power source (hereafter, form in control denotes the control system in which the droop methodology is applied). However, it is important to realize that the mode (see Fig. 1) can be provided compulsory reserve in only if the designated GUs have been deloaded by the required amount, or alternatively, if they are equipped with some energy

0885-8950/$26.00 2009 IEEE

490

IEEE TRANSACTIONS ON POWER SYSTEMS, VOL. 25, NO. 1, FEBRUARY 2010

storage system. In large power systems, these practices are rewarded by the system operator in well-dened rates for spinning reserve service, agreed upon by the parties. But in microgrids, with a surely poorly developed micro-market, the agreement is not that straightforward, discouraging producers from providing frequency support. Therefore, it is highly likely that generating mode inany producer would prefer to resort to steadthat is, a mode in which the control system obeys the references given by a maximum power point tracking algorithm, so that 1) maximum energy production is achieved by regulating the output power, and 2) reserves (i.e., missed producing oppormode is ideal for a given producer, tunity) are not required. because it maximizes its incomes when primary reserve is not GUs for adequately rewarded. However, there is a need for a sustained operation of the microgrid (indeed these units are regarded as grid-forming units in [4]), hence the need for a racontrol. tional share of the However, a literature survey reveals that for a given microgrid , the assignation of an to stiffness each GU is usually based on an equitable load share in the form , where is the rated output power of the th GU (see, for instance, [5][9]). With is directly assigned to the largest this rationale, the lowest GU, implying that it has to support the largest available reserve, is given because the share of the th GU in a total demand by (2)

drooping scenarios (as far as we are aware, the analysis of microgrid dynamics has been to date restricted to small microgrids with up to three GUs [6][13]). As a nal remark, note that for simplicity, we have exclu. sively introduced the control of power and frequency However, as noticed in [2] and [3], drooping output voltage against reactive power also offers voltage regulation capabilities . Moreover, harmonic share has been proposed using similar drooping characteristics [5], [9], [12], [14]; and in some and approaches are exworks, alternative plored [11]. In any case, these droops can be straightforwardly included into the methodology proposed here. The paper is structured as follows. Section II explains the theoretical foundation and introduces an algorithm to compute the worst direction to perform changes in a given scenario. The algorithm also permits an evaluation of the available margin of changes to be obtained. Section III briey introduces the microgrid model. Section IV discusses three case analyses. Section V concludes. II. THEORETICAL APPROACH A. Preliminaries We consider a microgrid model as a set of differential and algebraic equations (DAE), with state variables , algebraic variables , and a vector of slowly varying parameters : (3) For convenience, (3) can be summarized as

It is remarkable the lack of technical and economic content in the based scheduling. Neither the capability of providing sufcient level of reserves nor the economy are considered in the droop coefcient assignation. Only a scheduling based on the rated power is proposed. In this respect, the recent work in [10] proposed a scheduling using (four) different droop coefcients based on factors such as producing costs or available reserves. The authors in [10] acknowledged the economic importance of smart sharing among GUs instead of the commonly accepted rule . It is to be noted nevertheless that their approach did not include stability constraints. This could be a problem because, as shown in [6], [11], and in the results of this paper, there is a limit beyond which an increase of the droop coefcients leads to instability. That is, economical justications might colgeneration with inlide with stability constraints. Indeed, may be regarded as generationthe optimum econite nomical solution for each individual GU, but inappropriate for stability supporting purposes. We discuss in this paper a methodology based on bifurcation theory of application to the scheduling of the droop coefcients. Our aim here is to provide insight on how the choice of the droop coefcients affects both to the stability of the microgrid and to the reserve requirements. The approach differentiates from other published analyses in that 1) a formal procedure is dened for evaluating the best and worst directions in which droop coefcients can be varied, in contrast with trial procedures discussed in some papers [6], [11]; and 2) it applies to a large microgrid, allowing to observe the effects of different realistic

where and . In small signal stability analysis, it is customary to linearize around an operating point to obtain the system Jacobian : (4) By using Schurs complement, the state matrix is obtained from the Jacobian in (4). The classical bifurcation theory studies the change of qualitative properties of the dynamical system in response to quantitative changes in . The most generally occurring bifurcations in power systems are the codimension one saddle-node (SNB), limit-induced (LIB), and Hopf (HB) bifurcations. At an SNB, a given system becomes unstable when becomes singularstable and unstable equilibrium points coalesce and disappear as varies. LIB, a type of transcritical bifurcation, is associated with limitations (hard constraints or hard limits in [15]) especially in the generation reactive power capability, and it is attributed to the instantaneous change of the system structure and dynamics when the limit is reached [16]in LIBs, stable and unstable equilibrium points coalesce and exchange stability as varies. Both SNB and LIB have been mainly associated with voltage collapse [16]. Occurring when has a pair of purely imaginary conjugate eigenvalues that correspond to the birth of a limit cycle oscillation, HB has been however

DAZ et al.: SCHEDULING OF DROOP COEFFICIENTS

491

associated with oscillatory phenomena. We will focus on this latter type of bifurcation in this paper because of the oscillatory nature of frequency control, and because it has been shown that a reduction of the system stiffness is mainly concerned with HB [17]. This has been corroborated by our experience from investigating a considerable amount of scenariosthough a slight displacement of real eigenvalues towards the right half plane can be observed, the prevalent bifurcation is always of the HB type when the stiffness of the system is modied by acting on the droops. For a given equilibrium, the robustness of the system stability can be characterized by its parametric distance to hyperof critical solutions (bifurcations), where the surfaces behavior of the system changes qualitatively. Such hypersurfaces bound the feasibility regions, which consist of all equilibrium points that can be reached from an equilibrium by continuous variations of without losing stability along the path [15]. for which 1) For the generic HB, is the set of parameters has an HB at with having a pair of eigenvalues , and all other eigenvalues with nonzero real parts and 2) verifying the transversality condition [18, Section 3.4]. Verication of the above conditions 1) and 2) during the HB search ensures a transversal crossing of the imaginary axis; and note that condition 2) amounts to verifying that the gradient of the real part of the complex eigenvalue is nonzero at bifurcationthat is, it amounts to verifying that below.1 B. Search Algorithm It is computationally too demanding to compute the set of parameters at which the system undergoes an HB. Instead, from a given scenario in which is dened, it is more efcient to extract the closest HB point using an iterative method, as explained below. The method follows a two-step procedure that is iteratively repeated until a convergence criterion is reached. We assume that initially the system is stable for a xed parameter vector . In practice, we shall dene as the droop coefcients of the candidate GUs to be endowed with con. trol. That is, First Step (One-Dimensional Parameter Space): We rst recast the problem by parameterizing the system by the single parameter , i.e., (5) be the critical value of the parameter vector at bifurLet cation (i.e., ), and the corresponding equilibrium point. Though pertaining to and , and are to be found by progressing in a one-dimensional parameter space . This is valid because HB and SNB are codimension one bifurcations, and therefore they can be fully described by considering the quantitative change in a single parameter.
1For SNB, the hypersurfaces 6 are described by the set of parameters that make the system matrix singular at an equilibrium point (x ;  ) satisfying the transversality conditions @=@ F (x ;  ) 6= 0 and @ =@ x F (x ;  ) ( ;  ) 6 = 0[18, Section 3.4]. The transversality conditions for LIB require, however, a formulation in which the system is dened before and after the limits are reached; for brevity, we refer to [19] for details, where both SNB and LIB are thoroughly analyzed. We point out that [19] provides a relevant guidance on the interpretation of SNB versus LIB.

By convenience, we dene a ray in the form . and are here xed values, namely the vector of nominal parameters and the direction of the ray. With this denition, the HB can be found from the following set of equations [20]: (6) (7) (8) (9) The rst equation indicates that the solution must be an equilibrium point. The second and third equations are the expansion into real and imaginary parts (superscripts and ) of the expression (10) where it is required that the real part of the eigenvalue at the ]. The last (9) forces the corbifurcation be null [i.e., to not be null to nd a nontrivial responding left eigenvector solution. to start the search from a nominal , Given a direction (7)(10) can be solved for and hence nd the critical parameter , because . We dene hereafter the computed as the margin to HB for a given direction in the parameter space. Its value is important because it shows how far the system is from undergoing limit cycle oscillations. In this paper, there are two such parametric distances of interest: 1) the distance to an HB which gives a stability margin when the system stiffness is modied, but not the GU share, ; and 2) the distance to the closest HB point, . This latter gives insight on how to schedule the GUs to improve the system stability margin, yet a modication of the initial power share is probably needed. Equations (6)(9) are a direct method for computing the distance to an HB. However, some authors [21][23] and our own experience show that the system (6)(9) has a remarkable bad convergence feature, because of the difculty of nding good initial guesses. This problem is overcome if the location of is found by some indirect method instead. In this respect, we refer to the approach by Xiaoyu et al. in [24], where they proposed an iterative method to nd the HB point using the tangent information to move towards in a onedimensional parameter space. The procedure implies updating the value of at each iteration using the derivative of the real part of traced eigenvalues: (11) so that (12) This amounts to use the sensitivity of the eigenvalue with respect to the parameter as an indicator of the best search direction. This procedure was already used in [22], using at each step the right and left eigenvectors to compute the sensitivity (i.e., the denominator in the previous expression): (13)

492

IEEE TRANSACTIONS ON POWER SYSTEMS, VOL. 25, NO. 1, FEBRUARY 2010

The same procedure also appears in [25], under the label hit the boundary. Conceptually, the procedure for approaching the critical value of in [24] is the same as in [22] and [25]. The difference lies in that Xiaoyu et al. compute the derivatives of (7) and (8) and rearrange the formulation to give a compact way of nding the derivatives of the eigenvalues (and corresponding eigenvectors) at each step through a Newton search algorithm. Additionally, Xiaoyu et al. propose the use of a simple index: (14) This index ranks the eigenvalues at each th iteration.2 The lower the index, then the more important is the eigenvalue for HB; because then the smaller is its real part, and/or the larger is its sensitivity to changes in (note that ). Therefore, this index provides guidance on tracing the most repin the given resentative eigenvalues to nd the value of direction. The outlined procedure nds the margin in the one-dimensional parameter space in quite few steps. The computation of is then straightforward. Second Step (Multidimensional Parameter Space): Once is computed, we switch to the multidimensional parameter space . In [26], Dobson proposed the use of a normal vector to the bifurcation hypersurface as a means of nding the closest bifurcation point in a multidimensional parametric space. The normal direction to the manifold of HB is uniquely dened by [27] (15) is an tensor so that . A similar formulation is found in [26], [28], and [29]. According to Dobson, the closet HB point can be approached using an iterative procedure for nding a direction in the parametric space coincident with that of the normal vector to the hypersurface at the bifurcation point given by the direction . Therefore, to nd the closest HB point, the vector in (15) is employed to successively update the value of employed in the rst step. This procedure is repeated until and are roughly coincident. Except for very simple cases, the curvature of cannot be directly checked. This suggests that the solution obtained by the above procedure could be suboptimal should it happen that a guess of the ray were tangent to the hypersurfacein such a case, the distance could be different from the minimum distance. Dobson shows in [26] that the method based on normal vectors nds a closest bifurcation point when the distance does not exceed the inverse of the maximum principal curvature of at in short, when is close enough to . Dobson gives in [26] a set of formulas to compute the maximum principal curvature for SNB from the corresponding
2Note that since ` is updated in each iteration, the state matrix must also be updated. For coherency, this means that a new equilibrium point must be computed and that accordingly the system must be linearized around that new equilibrium point.

Fig. 2. Sample system employed to yield Fig. 3.

normal vector. Checking the curvature of HB surfaces, on the other hand, is a rather involved procedure. A possible and simple solution to this shortcoming, however, is to modify the initial guesses of and/or . The rationale behind this modication is to ensure a better location of and avoid as much as possible the tangential occurrences of with respect to when, paraphrasing Dobson, the surface is too concave. Remarks About SNB: We have focused on the HB. However, also an SNB may be achieved in some particular cases before a limit cycle oscillation occurs. This may be veried employing a test algorithm similar to that described above. As stated in [26], the iterative procedure dened in the second step is also valid for SNB, after replacing the expression for the normal vector by [26], [27] (16) Because the state matrix is singular at an SNB point, apcan be done by applying the alternaproaching the value of tive approach presented in [30], in which the value of is used near the bifurcation point to approximate . This approximates the sought direction , allowing for to be estimated by progressively increasing in such a direction until bifurcation. C. Example We now propose a simple analysis on a basic system formed by two GUs in a four-bus test system to exemplify and analyze the proposed approach (see Fig. 2). The GUs are connected to nodes 2 and 4 and interfaced through LCL lters with impedance values (PWM side), , and . Their rated power is 6 kWVA. Rated frequency is 50 Hz. Line impedances are and . Load impedances are and . Because only two GUs are present, the parametric space is of dimension two. Thus, is a line in a bi-dimensional repre. This allows a simple explasentation, with nation of the procedure outlined above as well as allows consequences about the use of the bifurcation information to be easily extracted for GU scheduling purposes. The iterative procedure is sketched in Fig. 3. The bifurcation boundary has been obtained from the system state space representation by using a continuation method. This has been possible in this bi-dimensional problem, but it results unfeasible in problems of higher dimension. It is stressed that the computation of is neither necessary nor convenient for more than two GUs; and that its representation in Fig. 3 is intended only for ease of explanation.

DAZ et al.: SCHEDULING OF DROOP COEFFICIENTS

493

Fig. 4. The 69-bus test system. Fig. 3. Feasibility boundary and the minimum distance `

6 and computation of the normal vector n


.

TABLE I LOCATION OF GUS FOR DIFFERENT SCENARIOS

The two steps of the previous algorithm are explained by following the indications represented in Fig. 3. A vector is initially dened by applying the conventional rule for a given initial system stiffness, assuming that GU no. 1 is rated at a higher power than the second GU. The rst ray for step 1 (one-dimensional space) is dened following the direction of . This means . Following this ray direction implies that the droops are progressively increased in proportion to each GU droopthe system stiffness is reduced, but in equal proportion to both GU. Executing step 1 nds after few iteration an HB at a distance , and hence . denotes the maximum allowable increase of the droops in a direction that does not favour any particular GU. Step 2 (bi-dimensional space) allows the normal vector to be found. Back to step 1, the direction of search is updated as . A new distance to is computed, and the corresponding normal vector is obtained from step 2. After repeating this procedure for the third time, the closest HB bifurcation point is obtained at a distance in the direction . The stopping criterion has been . The combined analysis of the values of , , , and gives valuable information about the reliability and reserves of the microgrid. Their interrelationship is summarized in Dobsons margin sensitivity [31], which is a scaled projection of the normal vector to . means that the droops are being increased following the most dangerous direction from the stability point of view. On the other side, (which means ) denotes the best stabilizing direction for the initial . Consequently, an equitable modication of the droops (given by ) is more sensitive to stability issues when is close to 1. Similar conclusions might be inferred for the amount of reserves required. For instance, complaints from GU no. 1 owner because of its larger primary reserves could be responded with a shift in towards increasing . However, care should be taken to avoid as much as possible dening a direction with a margin sensitivity too close to +1. Also note that droop scheduling

should be done mindful that vector is a representation of the microgrid stiffness. In an extreme case, all owners would be willing to have its droop as higher as possible, ideally innite (which would be equivalent to generation). However, Fig. 3 is a practical representation of how avoids a stable solution for any direction that tends to indenitely increase the value of . III. CASE ANALYSES The analyses in this section are conducted on a 69-bus microgrid. Data about the grid parameters of this test system can be found in [32]. The node numbering is shown in Fig. 4. A thorough explanation about the structured method followed to represent in state space form this relatively large system is detailed in [33]. The ve scenarios employed in the case analyses are listed in Table I. A. Case A: Centralized Versus Peripheral Frequency Regulation In this case, we propose an analysis of the inuence of the location of GUs on the stability of the microgrid. Four scenarios are employed (Table I): scenarios 2 and 5 gather six GUs in a small area, whereas the placement of GUs in scenarios 1 and 4 follows a distributed pattern. The normal vector and the margin are found after successively repeating steps 1 and 2 of Section II. The number of required iterations ranges from 4 (scenario 1) to 6 (scenario 4). A representation of the eigenvalue traces is depicted in Fig. 5 for the rst execution of step 1 using scenario 1.

494

IEEE TRANSACTIONS ON POWER SYSTEMS, VOL. 25, NO. 1, FEBRUARY 2010

COORDINATES OF

TABLE II FOR DIFFERENT SCENARIOS

Fig. 5. Eginvalue traces for the 69-bus microgrid when an increase in the  = k k. Only the most droops is performed following the direction dominant eigenvalues out of 531 are represented. Squares: initial position ); asterisks: progression for increasing `; circle: nal value (i.e., (i.e., ` ` > ` ). An enlarged view is provided to show the Hopf bifurcation.

=0

m =

the microgrid. For the dispersed GUs scenarios 2 and 5, the plot shows a clear bad direction in which a reduction of the reserves could be applied. For instance, in scenario 2, the entry of corresponding to the GU at node 65 has a value of 0.997. From Section II, it is inferred that the worst direction to increase is . Therefore, it is not a good idea to increase the . The representadroop in node 65 following tion in Fig. 6 shows that when the GUs are dispersed (scenarios 1 and 4), the droops can be more evenly scheduled responding to primary reserve requests, because there is not a clear dominant GU from the stability point of view. B. Case B: One Former Control Mode Resorting to

Fig. 6. Polar plot comparing the coordinates of

for different scenarios.

In Fig. 6, a plot of the coordinates of is shown. Four scenarios are shown, each one being represented by a polygon, with each vertex of the polygon corresponding to a coordinate of for that reason, scenario 4 is represented by a hexagon (six vertices because is of order 6), and the other scenarios are shown as pentagons; see Table I. We nd useful this type of representation because it gives quick insight on how uniform the coordinates of are by distributing them in radii of equal angular separation (the six coordinates of a six-dimensional vector would be projected into six radii uniformly separated by 60 degrees)the more regular the polygon, the more equitable the share of incremental droops would be; conversely, when the polygon is markedly irregular, it means that there is a dominant direction. It is stressed that a direct equivalence exists between this representation and the representation of Dobsons margin sensitivity . Indeed, each vertex of a polygon represents the value of when the value of the droop of an only GU (the GU associated with such a vertex) is modied. Therefore, the normal vector gives valuable information about the stability of

This case shows the inuence that eliminating a source has on the entire microgrid stability. Alternatively, this could be the case of an unscheduled increase of a GU droop coefcient, performed by a producer to unilaterally increase energy production. In this case, the scenarios 1 and 4 of Table I are compared. The GU that ceases to provide support is located at node 1. The most obvious consequence of the scenario change is the increase in the global system droopfrom to . This stiffness reduction makes the microgrid stability more vulnerable, and increases the range of frequency uctuations following load variations. Once the support from node no. 1 disappears, the system is also more likely to lose stability, because the margin to HB is clearly reduced from to . The data in Table II, where the normal vector computed for both scenarios is listed, provide insight on how to proceed to increase the system stiffness as well as the margin to HB. In particular, preferably the GUs at node 7 and 38 should reduce their droop at the cost of increasing their available reserves. Note that in the new scenario, the inuence of GUs at nodes 16 and 22 is quite small from the stability point of view. Therefore, it seems that increasing their primary reserves is not a reasonable option. Vector provides additional interesting information about the system. Observe from the data listed in Table II that for scenario no. 4, the corresponding entry of for the GU in node 22 is negative. Scheduling decisions can be taken from this information, because this shows that the reserves in this GU can be reduced (i.e., its can be increased) and unexpectedly the stability and damping of the system will improve. Yet the improvement would be extremely low because for changes in its droop coefcient (we recall that would be the best option to improve stability and damping).

DAZ et al.: SCHEDULING OF DROOP COEFFICIENTS

495

C. Case C: Response to One Excessive Primary Reserve

Complaining About

Scenario 3, analyzed in this case, has the same GU allocations GUs were supposed to have as scenario 1. In scenario 1, all . But now in scenario 3, the GUs at , whereas the rest nodes 7 and 61 have GUs are programmed with . of . The search The space dimension is algorithm is employed to nd again the normal vector (Table II). The margin to HB following the direction of is , whereas the minimum margin is . In our case, the GUs with the lowest droop coefcients wish to increase them, to be asked for less primary reserves. Let us suppose that the system general stiffness is not a problem with , and that increasing its value does not jeopardize the frequency values. In such a case, even without frequency constraints, observe that attending the demands of , its coordinate GU at node 7 is not very reasonable. In is 0.927. Because of this high value, moving in a direction to increase only its would result in the margin , too close to the worst direction (i.e., sensitivity ). That is, the margin is rapidly depleted by GU at node 7. On the other hand, the demands of GU at node 61 can be readily respondedan increase of its droop . coefcient only amounts to Note that an equitable reduction of the general stiffnessthat is, a modication of the droops in the direction of is quite acceptable, because . However, in this case, the reduction of primary reserves would be shared by all the producers; and the reduction in the system stiffness could lead to reach the frequency limit for the forecasted load probably before the demands of GUs in 7 and 61 were satised. Additionally, the information embedded into the margin sensitivity provides help in analyzing the system damping for changes in the droops. This is summarized in Fig. 7. is The eigenvalue for the HB with . A participation analysis shows that highly and (parrepresentative states for this mode are ticipations 0.224 and 0.272). These states correspond to the -component of the output current of GU at node 38, and to the -component of the output voltage of GU at node 34. The evolution of shows that limit cycle oscillations occur when is applied following an increment of the droops equal to (i.e., ). If the increment the direction of in droops exceeds the computed , it is observed that the system becomes unstable. However, if the ), then the system is best direction is applied ( more damped, at the cost of increasing the general stiffness. prove the usefulness of the computed biThe plots of furcations. Should the demand of GU at node 7 be accepted ), the system would become increasingly ( more undamped. The margin would be readily exceeded leading to instability. Also observe in Fig. 7 thatas expected from the discussion in Section IIthe same ) following the direction that satises increment ( ) does not the demands of GU at node 61 ( endanger the system stability.

Fig. 7. Response to perturbation.

IV. CONCLUSION In this paper, we have advocated the use of bifurcation theory in scheduling the drooping characteristics meant for frequency (and voltage) regulation in isolated microgrids. We have discussed the use of the distance to a bifurcation hypersurface as an indicative of the proximity to instability; and we have analyzed the advantages of using Dobsons margin sensitivity to evaluate the priorities and limits in primary reserve scheduling. Both and have been computed using an iterative procedure in which the normal vector to the bifurcation hypersurface is used to avoid the hypersurface difcult computation. The algorithm repeatedly switches from one-dimensional to multidimensional parameter spaces, yet it results quite simple and fast. This has been proven through the analysis of three cases in a 69-bus microgrid supplied by 11 generating units, where ve scenarios have been dened to discuss how the methodology applies advantageously to solving and evaluating the primary reserve scheduling when stability constraints are observed. REFERENCES
[1] J. Machowski, J. Bialek, and J. R. Bumby, Power System Dynamics and Stability. Chichester, U.K.: Wiley, 1997. [2] P. Kundur, N. J. Balu, and M. G. Lauby, Power System Stability and Control, ser. The EPRI Power System Engineering Series. New York: McGraw-Hill, 1994. [3] M. C. Chandorkar, D. M. Divan, and R. Adapa, Control of parallel connected inverters in standalone ac supply-systems, IEEE Trans. Ind. Appl., vol. 22, no. 4, pp. 136143, Jul. 2007. [4] N. L. Soultanis, S. A. Papathanasiou, and N. D. Hatziargyriou, A stability algorithm for the dynamic analysis of inverter dominated unbalanced lv microgrids, IEEE Trans. Power Syst., vol. 22, no. 1, pp. 294304, Feb. 2007.

496

IEEE TRANSACTIONS ON POWER SYSTEMS, VOL. 25, NO. 1, FEBRUARY 2010

[5] U. Borup, F. Blaabjerg, and P. N. Enjeti, Sharing of nonlinear load in parallel-connected three-phase converters, IEEE Trans. Ind. Appl., vol. 37, no. 6, pp. 18171823, Nov./Dec. 2001. [6] N. Pogaku, M. Prodanovic, and T. C. Green, Modeling, analysis and testing of autonomous operation of an inverter-based microgrid, IEEE Trans. Power Electron., vol. 22, no. 2, pp. 613625, Mar. 2007. [7] L. Tzung-Lin and C. Po-Tai, Design of a new cooperative harmonic ltering strategy for distributed generation interface converters in an islanding network, IEEE Trans. Power Electron., vol. 22, no. 5, pp. 19191927, Sep/ 2007. [8] Y. W. Li, D. M. Vilathgamuwa, and P. C. Loh, Design, analysis, and real-time testing of a controller for multibus microgrid system, IEEE Trans. Power Electron., vol. 19, no. 5, pp. 11951204, Sep. 2004. [9] F. D. Bianchi, H. De Battista, and R. J. Mantz, Optimal gain-scheduled control of xed-speed active stall wind turbines, IET Renew. Power Gen., vol. 2, no. 4, pp. 228238, 2008. [10] C. A. Hernandez-Aramburo, T. C. Green, and N. Mugniot, Fuel consumption minimization of a microgrid, IEEE Trans. Ind. Appl., vol. 41, no. 3, pp. 673681, MayJun. 2005. [11] J. M. Guerrero, J. Matas, V. Luis Garcia de, M. Castilla, and J. Miret, Decentralized control for parallel operation of distributed generation inverters using resistive output impedance, IEEE Trans. Ind. Electron., vol. 54, no. 2, pp. 9941004, Apr. 2007. [12] C. K. Sao and P. W. Lehn, Control and power management of converter fed microgrids, IEEE Trans. Power Syst., vol. 23, no. 3, pp. 10881098, Aug. 2008. [13] F. Katiraei, M. R. Iravani, and P. W. Lehn, Small-signal dynamic model of a micro-grid including conventional and electronically interfaced distributed resources, IET Gen., Transm., Distrib., vol. 1, no. 3, pp. 369378, 2007. [14] S. J. Chiang and J. M. Chang, Parallel operation of series-connected PWM voltage regulators without control interconnection, Proc. Inst. Elect. Eng., Elect. Power Appl., vol. 148, no. 2, pp. 141147, 2001. [15] V. Venkatasubramanian, H. Schattler, and J. Zaborsky, Dynamics of large constrained nonlinear systems-a taxonomy theory [power system stability], Proc. IEEE, vol. 83, no. 11, pp. 15301561, Nov. 1995. [16] I. Dobson and L. Lu, Voltage collapse precipitated by the immediate change in stability when generator reactive power limits are encountered, IEEE Trans. Circuits Syst. I, Fundam. Theory Appl., vol. 39, no. 9, pp. 762766, Sep. 1992. [17] W. Zhu, R. R. Mohler, R. Spee, W. A. Mittelstadt, and D. Maratukulam, Hopf bifurcations in a SMIB power system with SSR, IEEE Trans. Power Syst., vol. 11, no. 3, pp. 15791584, Aug. 1996. [18] J. Guckenheimer and P. Holmes, Nonlinear Oscillations, Dynamical Systems, and Bifurcations of Vector Fields, in Applied Mathematical Sciences, 5th ed. New York: Springer, 1997. [19] R. J. Avalos, C. A. Canizares, F. Milano, and A. J. Conejo, Equivalency of continuation and optimization methods to determine saddlenode and limit-induced bifurcations in power systems, IEEE Trans. Circuits Syst. I, Reg. Papers, vol. 56, no. 1, pp. 210223, Jan. 2009. [20] Y. V. Makarov, D. Zhao Yang, and D. J. Hill, A general method for small signal stability analysis, IEEE Trans. Power Syst., vol. 13, no. 3, pp. 979985, Aug. 1998. [21] Y. Zhou, X. Wen, and A. Venkataramana, Identication and tracing of oscillatory stability margin boundaries, in Proc. IEEE Power Eng. Soc. General Meeting, 2003, vol. 4, p. 2110. [22] Y. V. Makarov, D. J. Hill, and Z. Y. Dong, Computation of bifurcation boundaries for power systems: A new -plane method, IEEE Trans. Circuits Syst. I, Fundam. Theory Appl., vol. 47, no. 4, pp. 536544, Apr. 2000. [23] Y. V. Makarov, V. A. Maslennikov, and D. J. Hill, Calculation of oscillatory stability margins in the space of power system controlled parameters, in Proc. Int. Symp. Electric Power Engineering Power Tech, Stockholm, Sweden, 1995. [24] W. Xiaoyu and V. Ajjarapu, Application of a novel eigenvalue trajectory tracing method to identify both oscillatory stability margin and damping margin, IEEE Trans. Power Syst., vol. 21, no. 2, pp. 817824, May 2006. [25] A. zcan and H. Schattler, A computational method for the calculation of the feasibility boundary and clustering in differential-algebraic systems, IEEE Trans. Circuits Syst. I, Reg. Papers, vol. 52, no. 9, pp. 19401952, Sep. 2005.

[26] I. Dobson, Computing a closest bifurcation instability in multidimensional parameter space, J. Nonlin. Sci., vol. 3, no. 1, pp. 307327, 1993. [27] M. Mnnigmann and W. Marquardt, Normal vectors on manifolds of critical points for parametric robustness of equilibrium solutions of ode systems, J. Nonlin. Sci., vol. 12, no. 2, pp. 85112, 2002. [28] I. Dobson, An iterative method to compute a closest saddle node or Hopf bifurcation instability in multidimensional parameter space, in Proc. IEEE Int. Symp. Circuits and Systems (ISCAS 92), 1992, vol. 5, pp. 25132516. [29] I. Dobson, F. Alvarado, and C. L. DeMarco, Sensitivity of Hopf bifurcations to power system parameters, in Proc. 31st IEEE Conf. Decision and Control, 1992, vol. 3, pp. 29282933. [30] I. Dobson, Observations on the geometry of saddle node bifurcation and voltage collapse in electrical power systems, IEEE Trans. Circuits Syst. I, Fundam. Theory Appl., vol. 39, no. 3, pp. 240243, Mar. 1992. [31] I. Dobson and L. Lu, Computing an optimum direction in control space to avoid stable node bifurcation and voltage collapse in electric power systems, IEEE Trans. Autom. Control, vol. 37, no. 10, pp. 16161620, Oct. 1992. [32] H. D. Chiang and R. Jean-Jumeau, Optimal network recongurations in distribution systems. ii. solution algorithms and numerical results, IEEE Trans. Power Del., vol. 5, no. 3, pp. 15681574, Jul. 1990. [33] G. Diaz, C. Gonzalez-Moran, J. Gomez-Aleixandre, and A. Diez, Complex-valued state matrices for simple representation of large autonomous microgrids supplied by PQ and VF generation, IEEE Trans. Power Syst., vol. 24, no. 4, pp. 17201730, Nov. 2009.

Guzmn Daz (M04) received the M.Eng. and Ph.D. degrees from the University of Oviedo, Oviedo, Spain, in 1993 and 2001, respectively, both in electrical engineering. Presently, he is an Associate Professor in the Department of Electrical Engineering at the University of Oviedo. His areas of interest include design and evaluation of induction motors, where he has worked in several research projects for ABB Automation Products, and distributed generation, where he has been working in several related research projects during the last three years. Also, he worked during some years in the detection of faults in power transformers.

Cristina Gonzlez-Morn (M08) received the M.Eng. degree in electrical engineering from the University of Oviedo, Oviedo, Spain, in 2003. She is currently pursuing the Ph.D. degree in the Department of Electrical Engineering at the University of Oviedo. She joined the Department of Electrical Engineering at the University of Oviedo in 2005 as a full-time researcher. Her areas of interest include motor designs and solar energy, this latter a eld where she was a consultant from 2003 to 2005.

Javier Gmez-Aleixandre (M94) received the M.Eng. degree from the Politechnical University of Madrid, Madrid, Spain, in 1984 and the Ph.D. degree from the University of Oviedo, Oviedo, Spain, in 1988, both in electrical engineering. Presently, he is Full Professor with the Department of Electrical Engineering of the University of Oviedo. His current interests lie in the area of optimization or asynchronous motors and microgrids, where he is been head of several research projects in the eld.

Alberto Diez (M99) received the M.S. and Ph.D. degrees from the University of Oviedo, Gijn, Spain, in 1983 and 1988, respectively. He is currently an Associate Professor in the Electrical Engineering Department at the University of Oviedo. He was nominated as a Member of the Executive Committee D2 Rolling-at products by the European Commission (19982004). His topics of interest include control systems, high-performance ac drives control, industrial automation, and supervision processes.

Das könnte Ihnen auch gefallen