1764

IEEE TRANSACTIONS ON POWER DELIVERY, VOL. 26, NO. 3, JULY 2011

An Adaptive Feedforward Compensation for Stability Enhancement in Droop-Controlled Inverter-Based Microgrids

Mohammad B. Delghavi, Student Member, IEEE, and Amirnaser Yazdani, Senior Member, IEEE
AbstractThis paper proposes an adaptive feedforward compensation that alters the dynamic coupling between a distributedresource unit and the host microgrid, so that the robustness of the system stability to droop coefcients and network dynamic uncertainties is enhanced. The proposed feedforward strategy preserves the steady-state effect that the conventional droop mechanism exhibits and, therefore, does not compromise the steady-state power sharing regime of the microgrid or the voltage/frequency regulation. The feedforward compensation is adaptive as it is modied periodically according to the system steady-state operating point which, in turn, is estimated through an online recursive leastsquare estimation technique. This paper presents a discrete-time mathematical model and analytical framework for the proposed feedforward compensation. The effectiveness of the proposed control is demonstrated through time-domain simulation studies, in the PSCAD/EMTDC software environment, conducted on a detailed switched model of a sample two-unit microgrid. Index TermsAdaptive control, current control, distributed generation (DG), distributed resource (DR), droop, dynamics, feedforward, microgrid, model, power sharing.

I. INTRODUCTION HE microgrid concept [1] involves the coexistence of multiple distributed energy resource (DR) units within a prespecied part of an electrical network. One critical control task within the microgrid framework is the regulation of the network frequency and voltage magnitude, in the islanded (offgrid) mode of operation. Fundamentally, this objective is fullled through a shared contribution of real and reactive power by a multitude of dispatchable embedded DR units. The most widely adopted technique to ensure power sharing and coordinated voltage/frequency regulation is to droop the frequency and magnitude of the point-of-coupling voltage of each DR unit, versus the real and reactive powers that the DR unit delivers to the network [2][4]. The popularity of the method can be attributed to its ease of implementation, based merely on local voltage and current information, in addition to the facts that it

Manuscript received August 16, 2010; revised January 16, 2011; accepted February 20, 2011. Date of publication April 07, 2011; date of current version June 24, 2011. This work was supported in part by the Natural Sciences and Engineering Research Council (NSERC) of Canada. Paper no. TPWRD-006172010. The authors are with the University of Western Ontario, London, ON N6A 5B9 Canada (e-mail: mdelghav@uwo.ca; ayazdan2@uwo.ca). 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/TPWRD.2011.2119497

enables decentralized control of multiple DR units, readily accommodates the grid-connected mode of operation, and enables plug-and-play operation of the DR units. The prime issue with respect to the conventional droop-based control is that, it is, in essence, a steady-state measure that is taken to prevent the DR units from competing against each other, for individually imposing the network frequency and voltage; any such competition would inevitably result in a network collapse. Consequently, the transient performance and stability of the droop-based decentralized control highly depend on the droop coefcients, and on dynamic properties of the network, DR units, and embedded loads [5]. Even in terms of steady-state performance, the droop technique is effective the most for highly inductive networks, such as high-voltage transmission networks, but performs rather poorly when adopted for distribution networks [6]. These dependencies, combined with the fact that the droop coefcients are commonly determined based merely on steady-state criteria [2][4], give rise to the likelihood of inaccurate power sharings, unsatisfactory transient performances, or even instabilities, in the islanded mode of operation, as recently noticed by researchers [5], [7][9]. References [5] and [10] report two studies on dynamic characteristics of islanded microgrids that embed droop-controlled electronically coupled DR units. In [5], the sensitivity of the overall system eigenvalues to droop coefcients is shown, whereas [10] places the emphasis on the controller parameters. Reference [7] proposes a method for improving the transient power sharing among the DR units, in which the magnitude and frequency of each DR unit voltage are also drooped versus the derivatives/integrals of the real and reactive output powers. The method proposed in [7] is further advanced in [8] by making the droop coefcients dependant on the operating point, to mitigate the eigenvalue migration phenomenon and to improve the damping of the critical eigenmodes. Based on a slightly different approach proposed in [9], the droop coefcients of a DR unit are made functions of the DR unit real and reactive output powers. In all three references [7][9], it is assumed that a corresponding tie reactor interfaces each DR unit with a common ac bus, which, in turn, is connected to a lumped load. Thus, the common bus voltage is taken as the reference voltage, and the real and reactive powers that a DR unit exchanges with the network depend only on the magnitude and phase angle of the DR unit terminal voltage, relative to those of the common bus voltage; however, the load is considered an independent power sink and, therefore, exhibits no dynamic interactions with the DR units. Although simple, this model,

0885-8977/$26.00 2011 IEEE

DELGHAVI AND YAZDANI: ADAPTIVE FEEDFORWARD COMPENSATION FOR STABILITY ENHANCEMENT

1765

Fig. 1. Schematic diagram of the electronically interfaced DR unit.

which implicitly assumes a quasi steady-state condition, offers an approximate account of dynamic interactions among the DR units, but obscures the existence of such interactions with the network/loads. The reason is that the DR unit terminal voltages are functions of dynamic, as well as steady-state, power ows within the overall network and, thus, depend also on the loads. Reciprocally, the loads real and reactive powers are functions of the network voltage and frequency. Consequently, the aforementioned simplied model may not guarantee system stability and/or a satisfactory performance, in a real-life scenario. On the other hand, due to the diversities in network topologies and equipment, dynamics of a real-life network are typically governed by complex dynamic models [5], [10], [11], even for relatively small networks with loads of prespecied (e.g., RL) congurations. Consequently, the incorporation of network/load dynamics into the control design process is understandably involved [12], does not promise sufcient generality, and expectedly renders the design prone to case-by-case renements and compromises the capability for plug-and-play operation; the authors of this paper consider these issues to be the main reasons behind the widespread adoption of the simplied model described earlier. The objective of the study reported in this paper has thus been to circumvent the aforementioned difculties by making the control design, to a large extent, independent of the droop coefcients and network/load dynamics. This paper proposes an adaptive feedforward compensation that alters the dynamic coupling between a DR unit and the host network, such that the system stability is desensitized to the droop coefcients and network dynamics. The proposed feedforward strategy preserves the steady-state effect that the conventional droop mechanism exhibits and therefore does not compromise the steady-state power sharing amongst the DR units or the voltage/frequency regulation. Rather, it reshapes the contribution of the load dynamics in the control process, such that the system stability is enhanced. The proposed feedforward compensation is adaptive since it is periodically modied based on the system steady-state operating point, which, in turn, is estimated through the recursive least square (RLS) technique. The effectiveness of the proposed droop strategy is demonstrated through time-domain simulation studies, in the

PSCAD/EMTDC software environment [13], conducted on a detailed switched model of a sample two-unit microgrid. II. STRUCTURE OF THE DR UNIT In this paper, it is assumed that the DR units are all dispatchable and of the electronically coupled type, and adopt the proposed adaptive feedforward strategy (not necessarily with identical parameters). Thus, hereafter, a DR unit under study is referred to as the DR unit, and the microgrid without the DR unit is referred to as the rest of the microgrid, irrespective of whether it embeds any other DR unit. Fig. 1 shows a schematic diagram of the DR unit. The power circuit of the DR unit consists of a conditioned prime energy source, a current-controlled voltage-sourced converter (VSC), lter. The per-phase resistance, inducand a three-phase tance, and capacitance of the lter are denoted by , , and , respectively. The resistance represents the ohmic loss of the lter inductor and also includes the effect of the on-state , resistance of the VSC valves. The three-phase variables , and are referred to as the DR unit terminal voltage, VSC ac-side current, and DR unit terminal current, respectively. The VSC dc side is parallelled with a dc-link capacitor and a voltage source. The latter represents the effect of either a dispatchable mover-generator-rectier set or an energy storage device (e.g., a battery bank). Fig. 1 also shows the control components of the DR unit. It , , and are sampled and digitized is noted that by corresponding sample-and-hold (S/H) and analog-to-digital converters (ADCs). The sampled variables are then provided to to frame transformation blocks. Fig. 1 furrespective ther indicates that the DR unit is controlled in a rotating frame whose axis makes an angle with respect to the stationary axis (i.e., the axis). is obtained from a phase-locked loop (PLL) which also determines , that is, the frequency of the DR unit terminal voltage. In the grid-connected mode of operation, the DR unit terminal voltage is dictated by the rest of the microgrid, and represents the power system frequency. In the islanded mode, however, the DR unit, in conjunction with the other DR units, must contribute to the regulation of the network voltage and frequency, based on the proper commands given to

1766

IEEE TRANSACTIONS ON POWER DELIVERY, VOL. 26, NO. 3, JULY 2011

Fig. 2. Block diagram of the current-control scheme of the DR unit. Fig. 3. Block diagram of the amplitude regulation scheme.

the amplitude and frequency setpoints of the DR unit terminal voltage; the setpoints are, in turn, determined by a droop-based power sharing mechanism, which enables decentralized control of the network voltage and frequency, and ensures proper power sharing among the DR units. III. BASIC CONTROL The control schemes of the DR unit of Fig. 1 are extensively discussed in [14], but are also briey reviewed here for ease of reference. In the following formulations, the - and -axis are denoted by components of a three-phase signal and , respectively. A. Current-Control Scheme The function of the current-control scheme is to regulate the - and -axis components of the VSC ac-side current , by means of the pulsewidth modulation (PWM) switching strategy. This is primarily to ensure that the DR unit is protected against network faults, but also enables the regulation of the amplitude . and frequency of the DR unit terminal voltage Fig. 2 shows a block diagram of the current-control scheme and illustrates that two respective compensators process the error signals and and, based on the method described and . These in [14], deliver the signals two last signals are then delayed by one sampling period and and for the PWM scheme generate the signals of the VSC; the delays are essential in a microprocessor-based implementation due to the fact that the control signals, which are calculated based on the feedbacks at the th sampling instant, are practically computed during the time interval between th sampling instants and, consequently, the th and th sampling instant. cannot be released earlier than the Finally, the modulating signals , , and are , , and , and determine the calculated from switching instants of the VSC valves. The goal is achieved based on the symmetrical regular sampled PWM technique [15], which renders the VSC switching frequency equal to the control system sampling frequency. As explained in [14], ensures that and track their proper design of corresponding reference commands in two sampling periods, that is

Fig. 4. Block diagrams of the d- and q -axis closed loops equivalent to the amplitude regulation scheme of Fig. 3.

B. Amplitude Regulation Scheme Fig. 3 shows a block diagram of the amplitude regulation scheme whose function is to regulate and (i.e., the and axis components of the DR unit terminal voltage) and . This obat their respective setpoints and . In jective is accomplished by the control of the scheme of Fig. 3, the and axis compensators process the error signals and , respectively, and generate the setand for the current-control scheme. As points will be discussed in Section III-C, is indirectly employed in the frequency regulation scheme and has zero steady, that is, , state value. Hence, the amplitude of is predominantly determined by and, as such, effectively serves as the setpoint for the amplitude of the DR unit terminal voltage. As explained in [14], the control scheme of Fig. 3 transforms the closed-loop system to the two decoupled control loops of Fig. 4 in which the effective control plant is . ( is the sampling period.) C. Frequency Regulation Scheme The objective of the frequency regulation scheme is to regulate , that is, the frequency of the DR unit terminal voltage, at its setpoint . As Fig. 1 shows, a PLL processes by and determines in such a way that the lter is forced to zero [16]. In the grid-connected mode, is imposed by the rest of the microgrid and becomes equal to the power system angular frequency. However, in the islanded mode

DELGHAVI AND YAZDANI: ADAPTIVE FEEDFORWARD COMPENSATION FOR STABILITY ENHANCEMENT

1767

The calculated values of the real- and reactive-power outputs of the DR unit are formulated as (2)
Fig. 5. Block diagram of the frequency regulation loop.

and (3) Filtering each calculated value or by a corresponding discrete-time rst-order low-pass lter yields (4) and

Fig. 6. Block diagram of a microgrid embedding a droop-controlled DR unit.

(5) where and signify the ltered versions of and , respectively, and is a constant parameter. This , where is the parameter can be determined as corner frequency of an equivalent continues-time, rst-order, should be low-pass lter. From the viewpoint of dynamics, adequately large to not signicantly alter the controller dynamic properties, but reasonably small to ensure adequate ltering. Then, as Fig. 6 indicates, the amplitude and frequency setpoints and of the DR unit terminal voltage are drooped versus as (6) (7) and are the real- and reactive-power droop coefwhere is an auxiliary signal with zero cients, respectively, and steady-state value, similar to the signal which was introand are the no-load duced in Section III-C. The setpoints frequency and amplitude of the DR unit terminal voltage, respectively. and in (4) and (5), from (2) and Substituting for (3), one deduces

of operation, is regulated by . As explained in [14],

, through the control of is of the form (1)

is the gain. for which Fig. 5 shows a block diagram of the frequency regulation closed loop including the frequency regulation scheme. As Fig. processes the error 5 shows, a compensator and determines one component of ; the other component of , that is, is an auxiliary signal of zero steady-state value which plays an important role in the proposed adaptive feedforward strategy, as will be discussed is then tracked by in Section IV. The setpoint through the action of the axis amplitude regulation loop of is regulated at . It should be noted Fig. 4(b), and at 1, (and, therefore, that due to the pole of ) settles at zero in a steady state. Therefore, in order for to track with no steady-state error, is sufcient to be a pure gain . IV. PROPOSED ADAPTIVE FEEDFORWARD COMPENSATION Fig. 6 illustrates a generic block diagram of a microgrid, whether in the islanded mode or in the grid-connected mode, in which a DR unit is interfaced with the rest of the microgrid. As and are the responses of the rest Fig. 6 indicates, , , of the microgrid to the DR unit outputs (i.e., ). In turn, and are the responses of the and DR unit to the setpoints and , respectively, is an internal variable of the DR unit control (see while Section III-C). It is noted that (ltered measures of) the DR unit and real and reactive powers determine the setpoints , based on two respective droop characteristics. Fig. 6 also illustrates the proposed feedforward compensation. As the gure indicates, the proposed feedforward compensation with a signal and determines the augments signal for the frequency regulation scheme of the DR and . unit (see Section III-C), based on measures of As such, the feedforward compensation alters the dynamic coupling between the DR unit and the rest of the microgrid, by maand , based on and . nipulating

(8) (9) To enhance system stability, the dynamic coupling between the DR unit and rest of the microgrid [see Fig. 6] should be weakened as much as possible, but must exist in steady and quasi-steady states for real- and reactive-power sharing (i.e., conventional droop strategy). Thus, a small-signal model of the system is developed so that the steady-state and dynamic characteristics can be treated separately. To that end, (8) and (9) are linearized about a steady-state operating point to yield

(10)

(11)

1768

IEEE TRANSACTIONS ON POWER DELIVERY, VOL. 26, NO. 3, JULY 2011

where and the subscript denote the small-signal perturbation and the steady-state value, of a variable, respectively. It is zero, as exshould be noted that the steady-state value of plained in Section III-C and, therefore, does not appear in (10) and have zero or (11). It is also recalled that steady-state values. Dening (12) (13) (14) (15) one can rewrite (10) and (11) as

method of stability [17]. To that end, consider the following positive semidenite function, as a candidate Lyapunov function: (26) Then

(27) (note that , based on (12), and ), then and the system is stable [17]. Therefore, the transient response of the system is bound to decay to zero and, in other words, the large-signal system response is bound to settle, in a steady state. It should be emphasized that the stability does not depend on the droop coefcients or on any network properties. Expressing (22) and (23) in the matrix form (28) one can calculate and as (29) where (30) (31) One difculty that arises in implementing the proposed feedand , one forward compensation is that to determine needs knowledge about the steady-state values of the DR unit terminal voltage and current. On the other hand, in practice, the system operating point changes constantly as the loads shed back and forth, the power ow within the microgrid changes, etc. Consequently, the operating point needs to be estimated, for example, by means of an identication algorithm. This is explained next. V. RECURSIVE LEAST SQUARE IDENTIFICATION SCHEME Online estimation of system parameters, that is, online system identication, plays a crucial role in most adaptive control schemes [18]. For the adaptive feedforward compensation of Section IV, the parameters to be estimated are the steady-state values of the DR unit terminal voltage and DR unit terminal current (see (13)(15)). It should be pointed out that these parameters do not bear a rigorous physical meaning; rather, they are the byproducts of the mathematical formulation (i.e., the linearization process). In this paper, the RLS identication method with an exponential forgetting algorithm [18] has been employed for estimating the required parameters; the estimates of parameters are updated at every sampling instant. Equations (29)(31) indicate that the proposed adaptive feedforward compensation requires the parameters , , and . Since

(16) (17) The DR unit amplitude regulation scheme (Section III-B) enand at their respective sures rapid regulation of by in (7), and linsetpoints. Thus, replacing earizing the resultant equation, one deduces (18) . In addition, as Fig. 5 shows, On the other hand, the frequency regulation loop ensures that (Section III-C), the amplitude regulation scheme ensures that (Section III-B), and the steady-state is zero (Section III-C). Hence, one nds value of (19) and Substituting for (18) and (19), one nds in (16) and (17), based on

(20) (21) Let and be determined in such a way that (22) (23) Then, (20) and (21) assume the following forms: (24) (25) Equations (24) and (25) govern the small-signal dynamics of the real and reactive powers that the DR unit delivers to the rest of the microgrid. The stability of the system described by (24) and (25) can be established by means of Lyapunovs direct

DELGHAVI AND YAZDANI: ADAPTIVE FEEDFORWARD COMPENSATION FOR STABILITY ENHANCEMENT

1769

(39) is the vector of estimated where is the error vector, is known as Kalman parameters, gain, is referred to as the error covariance, and is known as the forgetting factor and determines the speed of adaptation. The estimation algorithm is executed at every sampling period, through the following steps: a new set of data is acquired and the prediction error is computed, based on (36), using the old estimated parameters; Kalman gain is updated based on (37); is calculated based on estimated parameters vector (38); is updated for the next sampling period, based on (39). Once the parameters , , and , are estimated, and are calculated, based on (30) and (31), and and are generated based on (29). The small-signal perturbations of the variables involved (see Fig. 7) are extracted by high-pass ltering of the variables to eliminate their steady-state components. Each high-pass lter has the following transfer function: (40) is the lter pole. where The process is illustrated by the block diagram of Fig. 7. VI. TEST CASES AND RESULTS
Fig. 8. Schematic diagram of the test microgrid.

Fig. 7. Block diagram of the scheme for the calculation of feedforward coefcients based on estimated parameters.

These parameters can be estimated if the RLS algorithm is applied to (16). Let us rewrite (16) as

(32) Let us dene the vector of parameters (33) and the regression vector (34) Then, (32) can be rewritten as (35) The estimated parameters of the plant, denoted by calculated in the th sampling period, as [18] , are then

(36) (37) (38)

To demonstrate the effectiveness of the proposed control strategy, a detailed switched model of a two-unit test microgrid has been simulated in the PSCAD/EMTDC software environment. The test microgrid is based on a typical 12.47-kV North American distribution network [19], of which a simplied schematic diagram is illustrated in Fig. 8. As Fig. 8 shows, the microgrid is divided into two subnetworks: Subnetwork 1 and Subnetwork 2, each embedding threephase (residential and industrial) loads as well as smaller singlephase networks (not shown in Fig. 8). The two subnetworks are interfaced, respectively, with Bus 1 and Bus 12 of the upstream network, through the corresponding switches S3 and S4. The subnetworks can also be interconnected by the switch S5. In turn, Bus 1 and Bus 12 are energized through the corresponding transformers Tr3 and Tr4, from a 115-kV transmission system. When S5 is open, the two subnetworks demand the real and reactive powers of (3.1 MW, 2.4 MVAr) and (1.2 MW, 0.8 MVAr), respectively. As shown in Fig. 8, the part of the system that is outside the microgrid boundaries is referred to as the grid. Fig. 8 also shows that the microgrid embeds a 4-MVA DR unit: DR1, and a 2-MVA DR unit: DR2. The two DR units are connected to Bus 2 and Bus 13, respectively. Each DR unit is interfaced with the host bus through a corresponding isolation transformer and a disconnect switch (S1 and S2, respectively, for DR1 and DR2); the transformers have a solidly grounded wye winding conguration at their low-voltage sides. The control algorithm of each DR unit (which includes the PWM, signal

1770

IEEE TRANSACTIONS ON POWER DELIVERY, VOL. 26, NO. 3, JULY 2011

transformation and conditioning, phase-angle extraction, current control, amplitude and frequency regulation, power calculation and droop, adaptive feedforward compensation, and online parameter estimation schemes) is implemented by a Fortran code that is linked to the PSCAD/EMTDC model of the DR unit. Parameters of the loads and transmission lines can be found in [19]. Other system parameters are given in Appendix A. The case studies demonstrate the DR unit performances in response to the connection of their corresponding host subnetworks to the grid, despite (intentional) inaccurate synchronization, and in response to the interconnection of the two subnetworks, when both of them are isolated from the grid. More important, two study cases are dedicated to highlighting the fact that the proposed feedforward compensation makes the dynamic performance and stability of the DR units insensitive to the droop coefcients and load/network dynamic properties. For each case study, the response under the proposed control is compared with its counterpart under the conventional droopbased control; the latter control strategy is invoked by setting and to zero. In the graphs the auxiliary signals to follow, the real powers, reactive powers, and currents are expressed in megawatts, megavoltamperes-reactive, and kiloamperes, respectively. A. Case 1: Transition to the Grid-Connected Mode Following Inaccurate Synchronization In this case, the switches S3, S4, and S5 are initially open (but, S1 and S2 are closed), the DR units are disabled, and, therefore, 0, the Subnetwork 1 and Subnetwork 2 are de-energized. At (see DR units are turned on, and their respective values of (7), Section IV) are ramped up from zero to 0.52 kV, in about 35 ms; hence, the (islanded) subnetworks get energized by their respective DR units. Subsequently, S3 and S4 are closed, at 0.75 s, and the subnetworks are connected to the grid, while, right before the closure of S3 and S4, the voltages of Bus 2 and Bus 13 have phase displacements of about 24 relative to the voltages of Bus 1 and Bus 12, respectively; the phase shifts are introduced intentionally to subject the DR units to a rather 2.5 s, S3 and S4 are opened again severe disturbance. At 2.5 s and isolate the subnetworks from the grid. Thus, from onwards, the two subnetworks and DR units operate under the same conditions that prevailed from 0 to 0.75 s. Figs. 9 and 10 illustrate the system response to the aforementioned sequence of events. Fig. 9 illustrates the waveforms of the real- and reactivepower outputs of the DR units, in addition to those of the - and -axis components of the terminal current of DR1, under the conventional droop-based control (hereafter, the conventional control). Similarly, Fig. 10 illustrates the waveforms of the same variables, but under the proposed feedforward compensation (hereafter, the proposed control). A comparison between Figs. 9 and 10 reveals that while the responses of the DR units to the switching incidents exhibit remarkable transient excursions and ringings under the conventional control, they are remarkably smooth and damped under the proposed control, despite the severity of the disturbances.

Fig. 9. Sample responses of the DR units to the connection of their host subnetworks to the grid, under the conventional control (Case 1).

Fig. 10. Sample responses of the DR units to the connection of their host subnetworks to the grid, under the proposed control (Case 1).

Fig. 11. Estimated parameters of DR1 under the proposed control (Case 1).

Fig. 11 illustrates the waveforms of the estimated parameters of DR1 (for instance), and indicates that subsequent to each disturbance incident, they smoothly converge to their steady-state

DELGHAVI AND YAZDANI: ADAPTIVE FEEDFORWARD COMPENSATION FOR STABILITY ENHANCEMENT

1771

Fig. 12. Sample responses of the DR units to the interconnection of the two subnetworks, under the conventional control (Case 2).

Fig. 13. Sample responses of the DR units to the interconnection of the two subnetworks, under the proposed control (Case 2).

values. As the gure shows, the settling values of , , and are, respectively, 0.0022, 0.0207, and 0.0153, at the end of the time intervals over which Subnetwork 1 is isolated from the grid. On the other hand, under the off-grid condition, the and are 4.45 kA and 3.29 kA, resettling values of is spectively (see Fig. 10), and the steady-state value of 0.475 kV (not shown in the graphs). Substituting for these three steady-state values in (13)(15), one calculates , , and as 0.00221, 0.02069, and 0.01529, respectively. It is noted that the calculated values are very close to the values estimated by the RLS identication scheme of DR1. It should be emphasized do not bear clear physical meanings, but are that , , and the byproducts of the mathematical formulation set forth in Section IV. B. Case 2: Network Topological Change in the Islanded Mode This case study demonstrates the responses of the DR units to a network topological change, resulting in a power-ow change, in the islanded mode of operation. In this case, the system continues from the same steady state as that under the operating conditions of Case 1, that is, Subnetwork 1 and Subnetwork 2 are isolated from the grid (switches S3 and S4 are open) and independently energized by their respective DR units. At 4.0 s, the switch S5 is closed (see Fig. 8), while, right before the switching incident, the voltages of Bus 8 and Bus 14 are phase-displaced by about 43 ; subsequent to the closure of S5, the aggregate of the loads of Subnetwork 1 and Subnetwork 2 is shared by the two DR units. Figs. 12 and 13 illustrate the system response to the disturbance under the conventional and proposed controls, respectively. As Fig. 12 indicates, subsequent to the switching incident, the variables plotted in the gure experience remarkable uctuations under the conventional control. By contrast, despite the disturbance severity, the system response is well damped under the proposed control, as Fig. 13 shows. A comparison between Figs. 12 and 13 conrms that the proposed control does not alter the steady-state power sharing regime that would exist under the conventional control.
Fig. 14. Sample responses of the DR units to stepwise increase in their droop coefcients, under the conventional control (Case 3).

C. Case 3: Stepwise Increase in Droop Coefcients As has been analytically shown in [5], a sufcient increase in the droop coefcients has a destabilizing effect on the conventional control. This case study conrms the conclusion of [5], and also demonstrates the effectiveness of the proposed control in maintaining the system stability despite an increase in the droop coefcients. In this case, the system continues from the same steady state as that under the operating conditions of Case 2, that is, Subnetwork 1 and Subnetwork 2 are isolated from the grid, but the switch S5 is closed; therefore, the two DR units share the aggre6.5 s, the frequency/ gate of the two subnetwork loads. At real-power droop coefcients of DR1 and DR2 are stepped up, 2.0 to 12.0 (rad/s)/MW, and from 4.0 to 24.0 from (rad/s)/MW, respectively. Fig. 14 depicts the responses of the DR units under the conventional control and indicates that the system becomes 6.5 s. By contrast, as Fig. 15 shows, the oscillatory after system remains stable under the proposed control, despite the disturbance.

1772

IEEE TRANSACTIONS ON POWER DELIVERY, VOL. 26, NO. 3, JULY 2011

Fig. 15. Sample responses of the DR units to stepwise increase in their droop coefcients, under the proposed control (Case 3).

Fig. 17. Sample responses of the DR units to the introduction of an asynchronous machine to Subnetwork 1, under the proposed control (Case 4).

Fig. 16 shows that under the conventional control, the connection of the asynchronous machine to the subnetwork results in oscillations and instability. However, under the proposed control, the microgrid remains stable, as Fig. 17 indicates. Fig. 17 also shows that, from 6.5 s onwards, the two DR units deliver comparatively smaller (real) powers, due to the power contribution of the asynchronous machine. VII. CONCLUSION This paper proposed an adaptive feedforward compensation strategy that alters the dynamic coupling between a DR unit and the host microgrid so that the system stability is made insensitive to droop coefcients and load/network dynamic characteristics. The proposed feedforward compensation preserves the steady-state effect that the conventional droop mechanism exhibits and, therefore, does not compromise the steady-state power sharing regime and voltage/frequency regulation offered by the conventional droop-based control. The feedforward compensation is adaptive since it is periodically modied according to the system steady-state operating point which, in turn, is estimated through an online RLS estimation technique. This paper presented a discrete-time mathematical model and analytical framework for the proposed feedforward compensation. The effectiveness of the proposed control was demonstrated through time-domain simulation studies, in the PSCAD/EMTDC software environment, conducted on a detailed switched model of a sample two-unit microgrid. APPENDIX A SYSTEM PARAMETERS The system parameters are given in Table I. The transfer funcand are given by tions

Fig. 16. Sample responses of the DR units to the introduction of an asynchronous machine to Subnetwork 1, under the conventional control (Case 4).

D. Case 4: Change in Load/Network Properties As pointed out in the introduction of this paper, the performance and stability of the conventional control depend on the load/network dynamic properties. This case study supports the claim and further demonstrates that the proposed control is robust to load/network characteristic variations. In this case study, the system continues from the same steady state as that under the operating conditions of Case 2, that is, Subnetwork 1 and Subnetwork 2 are isolated from the grid, but connected to each other through S5, and the two DR units 6.5 s, share the aggregate load of the two subnetworks. At an asynchronous machine, spun by a mechanical torque of 1.0 per unit at angular speed close to 377 rad/s, gets connected to Bus 6 of Subnetwork 1; the connection is enabled by a circuit breaker and a voltage-matching transformer (none of these components are shown in Fig. 8). Parameters of the asynchronous machine are given in Appendix B. Figs. 16 and 17 illustrate the responses of the DR units, under the conventional and proposed controls, respectively.

for for

(41) (42)

DELGHAVI AND YAZDANI: ADAPTIVE FEEDFORWARD COMPENSATION FOR STABILITY ENHANCEMENT

1773

TABLE I DR UNITS CIRCUIT AND CONTROL PARAMETERS

TABLE II ASYNCHRONOUS MACHINE PARAMETERS (CASE 4)

for for

(43) (44)

[4] F. Katiraei and M. R. Iravani, Power management strategies for a microgird with multiple distributed generation units, IEEE Trans. Power Syst., vol. 21, no. 4, pp. 18211831, Nov. 2006. [5] N. Pogaku, M. Prodanovic, T. C. Green, W. L. Kling, and L. Van Der Sluis, Modeling, analysis and testing of autonomous operation of an inverter-based microgrid, IEEE Trans. Power Electron., vol. 22, no. 2, pp. 613625, Mar. 2007. [6] K. D. Brabandere, B. Bolsens, J. V. D. Keybus, A. Woyte, J. Driesen, and R. Belmans, A voltage and frequency droop control method for parallel inverters, IEEE Trans. Power Electron., vol. 22, no. 4, pp. 11071115, Jul. 2007. [7] J. M. Guerrero, L. G. de Vicuna, J. Matas, M. Castilla, and J. Miret, A wireless controller to enhance dynamic performance of parallel inverters in distributed generation systems, IEEE Trans. Power Electron., vol. 19, no. 5, pp. 12051213, Sep. 2004. [8] Y. A. I. Mohamed and E. F. El-Saadany, Adaptive decentralized droop controller to preserve power sharing stability of paralleled inverters in distributed generation microgrids, IEEE Trans. Power Electron., vol. 23, no. 6, pp. 28062816, Nov. 2008. [9] E. Rokrok and M. E. H. Golshan, Adaptive voltage droop scheme for voltage source converters in an islanded multibus microgrid, Inst. Eng. Technol. Gen., Transm. Distrib., vol. 4, no. 5, pp. 562578, 2010. [10] F. Katiraei, M. R. Iravani, and P. W. Lehn, Small-signal dynamic model of a micro-grid including conventional and electronically interfaced distributed resources, Inst. Eng. Technol. Gen., Transm. Distrib., vol. 1, no. 3, pp. 369378, 2007. [11] G. Diaz, C. Gonzales-Moran, J. Gomez-Aleixandre, and A. Diez, Complex-valued state matrices for simple representation of large autonomous microgrids suppled by P Q and V f generation, IEEE Trans. Power Syst., vol. 24, no. 4, pp. 17201730, Nov. 2009. [12] H. Karimi, E. J. Davison, and R. Iravani, Multivariable servomechanism controller for autonomous operation of a distributed generation unit: design and performance evaluation, IEEE Trans. Power Syst., vol. 25, no. 2, pp. 853865, May 2010. [13] PSCAD/EMTDC 4.2 ed. Manitoba HVDC Res. Ctr, Winnipeg, MB, Canada. [14] M. B. Delghavi and A. Yazdani, Islanded-mode control of electronically-coupled distributed resource (DR) units under unbalanced and nonlinear load conditions, IEEE Trans. Power Del., to be published. [15] S. R. Bowes, Micrroprocessor control of PWM inverters, Proc. Inst. Elect. Eng., vol. 128, no. 6, pp. 293305, 1981. [16] S. K. Chung, A phase tracking system for three phase utility interface inverters, IEEE Trans. Power Electron., vol. 15, no. 3, pp. 431438, May 2000. [17] A. Medio and M. Lines, Nonlinear Dynamics: A Primer. Cambridge, U.K.: Cambridge Univ. Press, 2001. [18] K. J. strm and B. Wittenmark, Adaptive Control, 2nd ed. Reading, MA: Addison-Wesley, 1995. [19] CIGRE Task Force C6.04.02, Benchmark systems for network integration of renewable and distributed energy resources, CIGRE Tech. Brochure 2011.

APPENDIX B ASYNCHRONOUS MACHINE PARAMETERS Asynchronous machine parameters are given in Table II. REFERENCES
[1] N. Hatziargyriou, H. Asano, R. Iravani, and C. Marnay, Microgrids, IEEE Power Energy Mag., vol. 5, no. 4, pp. 7894, Jul./Aug. 2007. [2] M. C. Chandorkar, D. M. Divan, and R. Adapa, Control of parallel connected inverters in standalone ac supply systems, IEEE Trans. Ind. Appl., vol. 29, no. 1, pp. 136143, Jan./Feb. 1993. [3] Y. Li, D. M. Vilathgamuwa, and P. C. Loh, Design, analysis, and realtime testing of a controller for multibus microgrid system, IEEE Trans. Power Electron., vol. 19, no. 5, pp. 11951204, Sep. 2004.

Mohammad B. Delghavi (S09) received the M.Sc. degree from the Iran University of Science and Technology (IUST), Tehran, Iran, in 1996 and is currently pursuing the Ph.D. degree at the University of Western Ontario (UWO), London, ON, Canada. Currently, he is a Research Assistant with UWO. His research interests include switching power converters, distributed generation, and microgrids.

Amirnaser Yazdani (M05SM09) received the Ph.D. degree in electrical engineering from University of Toronto, Toronto, ON, Canada, in 2005. Currently, he is an Assistant Professor with the University of Western Ontario, London, ON, Canada. His research interests include modeling and control of electronic power converters, renewable electric power systems, distributed generation and storage, and microgrids.