MIMO design of voltage controllers for Distributed Generators Anna Rita Di Fazio, Giuseppe Fusco, Mario Russo Department of Electrical and Information Engineering University of Cassino and Southern Lazio Cassino, Italy {a.difazio, fusco, russo}@unicas.it Abstract—A large penetration of Distributed Generation (DG) may severely impact on the voltage profiles along LV distribution feeders. The first, cheapest and simplest step that can be undertaken is to involve DG units in the voltage control of the LV distribution networks, adopting a fully-decentralized architecture and avoiding any additional communication infrastructure. The paper proposes a design methodology for voltage control of DG units: a local controller measures and regulates the voltage at the node to which each DG unit is connected by acting on the DG reactive power injections and without any need for data or measurement exchange with other controllers. The controller design is based on a structural MIMO model of the distribution system and ensures a satisfactorily regulation, while avoiding any operation conflict among DG units. Stability analysis is also presented. The results of numerical simulations in PSCAD/EMTDC environment are discussed to validate the proposed approach. Index Terms—Distributed Generation, Distribution Systems, Voltage Regulation, Reactive Power Control, MIMO model I. I NTRODUCTION The inflexible structure of existing distribution systems is inadequate to sustain a large penetration of Distributed Generation (DG). In particular, LV distribution systems often lack of monitoring and automated control systems, and are designed to deliver strictly-unidirectional power flows from the secondary substation through radial feeders to the consumers. In this frame, voltage control represents a fundamental issue when connecting a large amount of DG to LV distribution systems [1]–[3]. In fact, the connection of large amount of DG causes inverse power flows and possible over-voltages along the LV feeders. Consequently, maintaining the nodal voltages within the required limits is a challenge for the LV distribution system operators. At the same time, the large number of nodes along LV feeders and the large number of existing LV networks push for finding cheap solutions to the voltage control problem, avoiding the installation of new measurement and control equipment and of highly-performing communication infrastructures. The first, cheapest and simplest step that can be undertaken is to involve DG units in voltage control of the LV distribution networks, adopting a fully-decentralized architecture, that is adopting local DG control systems which do not exchange any data or measurements. A technique that has been proposed in literature is based on the partial curtailment of the DG active power injection in the case of over/under voltages [4]–[6]. Unfortunately, this technique impacts on the active powers which represent the value added product for the DG owners. An alternative solution is to act on the reactive powers which can be injected by some DG units. The advantages of such a solution are represented by the limited modifications required to the existing DG control systems and by the absence of a production cost related to reactive power. As a drawback, acting on reactive powers has a weaker impact on nodal voltages with respect to active powers in LV distribution systems. Within this latter approach, decentralized Volt/Var control [7]–[9] is adopted in some distribution networks. Such a control determines the DG reactive power to be injected as a function of the voltage at the DG terminals, according to a piecewise linear relationship Q(V ). The decentralized Volt/Var control presents some stability problems: the contemporary presence of multiple DG unit controllers without coordination poses problem of mutual interactions which may lead to the instability of the system [8], [10], [11]. For example, if the controller of a DG increases the reactive power injection so as to support voltage profile in response to a load increase, then the effects of its action are viewed by other controllers as a voltage increase and, consequently, they will act reducing their reactive power injections. Reversely, these reductions are viewed by the former controller as a further load increase and consequently a “hunting” instability among the controllers is established. Keeping the decentralized architecture and using local reactive power control, the present paper proposes a design methodology for voltage control of DG units. The idea is to realize a secure plug and play connection of the DG units to the distribution system without any need of data exchange with other control systems but, at the same time, to improve the nodal voltages by acting on the DG reactive power injections. A local controller measures and regulates the voltage at the node to which each DG unit is connected. Its design is based on a structural MIMO model of the distribution system and ensures a satisfactorily regulation, while avoiding operation conflicts among DG units. Stability analysis is also presented. The results of numerical simulations are discussed to validate the proposed approach. 0 1 MV/LV MV distribution system L1 lmg lm1 L2 mg m1 2 l2 Lm1 n ln Lmg Vj−1 Vj Rj Pj−1, Qj−1 Xj Ln Pj , Q j PL,j QL,j P̄DG,j Q̄DG,j Figure 2. Electric equivalent circuit for the j − th branch. PV PV Figure 1. Distribution feeder with PV units. II. S YSTEM B. Network modelling MODELLING The aim is to derive a structural MIMO model of the LV distribution system suitable for decentralized voltage control design of DG units. Firstly, the DG models are derived. Then, the network model is developed based on the linearization of the classical DistFlow equations and on the enforcement of the border conditions for each feeder. Eventually, the network model is combined with the model of the DG units, yielding an accurate MIMO model that accounts for the interaction among the DG units along a feeder. Let a LV distribution system be considered with a radial feeder composed of n nodes; the feeder is supplied by a secondary substation and other feeders supplied by the same MV/LV transformer are included in L1 as equivalent active and reactive powers. Along the feeder, g DG units are connected to different nodes: the i − th DG unit is connected to the mi − th node and the ordered set of indexes {m1 , . . . , mg } denotes the nodes to which the DGs are connected, as shown in the one-line diagram in Fig. 1. Various types of DG units can be connected; in the present paper, without loss of generality, reference is made to PhotoVoltaic (PV) systems which are the most-widely spread in LV distribution systems. The aim is to model the network in terms of nodal voltage amplitude variations in response to the reactive power injections by the DGs. The model of the feeder in Fig. 1 is based on the steady-state DistFlow equations [12]. The generic j − th branch of the feeder is represented by the electric circuit shown in Fig. 2. The branch model is characterized by: three variables at the supplying node, namely the voltage amplitude Vj−1 , the in-flowing active Pj−1 and reactive Qj−1 powers; three variables at the receiving node, namely the voltage Vj , the out-flowing active Pj and reactive Qj powers; four branch parameters, namely, the resistance Rj and reactance Xj , the shunt load active PL,j and reactive QL,j powers. Two additional external variables are included, namely e DG,j powers related to the DG the active PeDG,j and reactive Q which may be connected to the j − th node. The DistFlow equations for the j − th branch are Pj = Pj−1 −Rj 2 Pj−1 + Q2j−1 e DG,j − QL,j + Q 2 Vj−1 −2 Rj Pj−1 + Xj Qj−1 Qj = Qj−1 −Xj 2 Vj2 = Vj−1 + A. PV system modelling The aim of this subsection is to model the PV systems by identifying the input command that can be used to control the output injected reactive power and the related input-output transfer function. The PV systems are static generators with power electronic interface to the distribution network. The interface inverter is typically equipped with a reactive power control loop: on the basis of a reactive power set-point, the inverter pulses are generated so as to inject the desired reactive power. Then, the DG reactive power closed-loop response can be characterized by a simple time constant τDG of few milliseconds. Consequently, the transfer function referred to the i − th DG is QDG,i (s) = 1 Ui (s) 1 + s τDG,i (1) where QDG,i is the injected reactive power, τDG,i is the time constant and Ui is the input command, that is the reactive power set-point to the inverter. 2 + Q2j−1 Pj−1 − PL,j + PeDG,j 2 Vj−1 (2) 2 (Rj2 + Xj2 ) (Pj−1 + Q2j−1 ) 2 Vj−1 standing for j = 1, . . . , n with e DG,j = 0 PeDG,j = 0 Q e DG,j = QDG,k PeDG,j = PDG,k Q ∀j ∈ / (m1 , . . . , mg ) ∀j ∈ (m1 , . . . , mg ) being k : mk = j Since (2) are non linear, they are linearized around an initial point. Let the vector of nodal variables be defined as T (3) xj = Pj Qj Vj2 and let the operating condition in which the DG power injections are null be considered as initial point and indicated as 2 T x0,j = P0,j Q0,j V0,j Linearizing the DistFlow equations (2) around such an initial operating condition yields the following branch model e DG,j 0 T (4) ∆xj = Jj ∆xj−1 + PeDG,j Q 2 V0,m 1 being ∆xj = xj − x0,j = ∆Pj ∆Qj ∆Vj2 and Jj the Jacobian matrix related to (2) ∂xj Jj = ∂xj−1 x0,j−1 T (5) (6) The linear set (4), (6) can be analytically solved and its solution can be partitioned yielding the following matrix relationship [13] ∆V = Γ Q (7) T ∆V = ∆Vm2 1 ∆Vm2 2 . . . . . . ∆Vm2 g QDG,1 QDG,2 . . . . . . QDG,g ∆Vm2 1 1 1+sτDG,1 [Γ] 2 ∆Vdes,g Cg (s) + Ug ∆Vm2 g 1 1+sτDG,g Figure 3. MIMO control scheme for the design. and the transfer function Ci (s) represents the voltage con2 troller for the i − th DG unit. It is worth noting that V0,m i can be interpreted as a bounded bias which depends on the operating conditions of the feeder, and the dimension of the plant square matrix P0 (s) is equal to g, independently from the number of nodes of the distribution system. III. C ONTROL DESIGN AND STABILITY ANALYSIS The control design aims at ensuring a satisfactorily voltage regulation at nodes at which the DGs are connected while guaranteeing system stability. Without loss of generality let the case of two DGs (g = 2) connected at nodes m1 and m2 be considered. The proposed control matrix is n 1 + τzi s o C(s) = diag Ci (s) = kci 1 + τpi s i = 1, 2 where kci , τzi and τpi are design parameters with τzi < τpi . The open loop matrix F(s) = P0 (s)C(s) takes the form kc1 γ11 (1+τz s) kc2 γ12 (1+τz s) T 1 and Γ ∈ IRg×g a non singular matrix of known scalars. C. MIMO modelling for design By combining the DG and the network modelling, see (1) and (7), the MIMO model can be written in matrix form as ∆V(s) = Γ G(s) U(s) = P0 (s)U(s) U1 2 V0,m g ∆Qn = 0 C1 (s) + + ∆V02 = 0 ∆Pn = 0 Q = 2 ∆Vdes,1 + 2 Vref,g In (4), for j ∈ / (m1 , . . . , mg ) the last term is trivially null and, consequently, can be skipped. The model for the whole feeder is composed of n sets of equations of type (4) in n + 1 variables ∆xj for j = 0, . . . n. Then, to obtain a defined problem, 3 scalar variables must be fixed by forcing the border conditions of the feeder: the squared voltage amplitude V02 at the supplying station, which is assumed to be fixed as slack bus, and the active and reactive powers, Pn and Qn , flowing out of the last node which are assumed to be always null. The border conditions can be expressed in terms of variations as with 2 Vref,1 F(s) = 2 (1+τp1 s)(1+τDG1 s) (1+τp2 s)(1+τDG2 s) kc1 γ11 (1+τz1 s) (1+τp1 s)(1+τDG1 s) kc2 γ22 (1+τz2 s) (1+τp2 s)(1+τDG2 s) (11) with γiℓ (ℓ = 1, 2) elements of matrix Γ2×2 . Each transfer function Fiℓ (s) satisfies the condition (8) lim Fiℓ (s) 6= 0 s→0 being T ∆Vm2 1 (s) ∆Vm2 2 (s) . . . . . . ∆Vm2 g (s) 1 G(s) = diag GDG,i (s) = 1 + s τDG,i T U(s) = U1 (s) U2 (s) . . . . . . Ug (s) Choosing kci such that ∆V(s) = kci >> (12) it results Fii (s)|s=0 >> 1 with U(s) the input vector. From (5) specified for j = mi , the voltage output Vm2 i can be written as 2 + ∆Vm2 i (s) Vm2 i (s) = V0,m i 1 γii (9) Defining the i − th regulation error Ei (s) = ∆V 2des,i − ∆Vm2 i (s) (13) Then, the voltage control scheme shown in Fig. 3 is derived, where the variational set-point is defined as the expression of the steady-state regulation error ess can be approximated as follows 2 2 2 ∆Vdes,i (s) = Vref,i − V0,m i ess ≈ [F(s)|s=0 ]−1 ∆V des (10) (14) 2 2 with ∆V des = (∆Vdes,1 ∆Vdes,2 )T the vector of desired set-points. From (14) using (11) one obtains 1 2 2 γ22 ∆Vdes,1 − γ12 ∆Vdes,2 kc1 γ11 1 ess ≈ γ22 − γ12 1 2 2 −∆Vdes,1 + ∆Vdes,2 kc2 (15) Given ∆V des , for a suitable choice of gains kci satisfying condition (12) it is possible to guarantee from (15) a limited steady-state regulation error in terms of variational quantities. By using (9), (10) and (13), from ess it is possible to determine the error in terms of voltage amplitudes, as shown in Sect. IV-A. The study of the closed-loop system stability can be approached using the Gershgorin theorem [14]. It affirms that if matrix I + F( ω) is a row or column dominant diagonal the number N of encirclements of vector det(I + F( ω)) around the origin for all ω ∈ IR is given by the sum of the numbers ν1 and ν2 of the encirclements of vectors representatives of the functions 1 + F11 ( ω) and 1 + F22 ( ω), respectively. Hence N = ν1 + ν2 (16) According to the Rosenbrok procedure, matrix I + F( ω) is a row dominant diagonal if |1 + F11 ( ω)| > |F12 ( ω)| (17) |1 + F22 ( ω)| > |F21 ( ω)| (18) Conditions (17)-(18) are satisfied if the Gershgorin circles, built on the Nyquist plot of F11 ( ω) and F22 ( ω), respectively, do not encompass the critical point (−1 + 0) for all ω ∈ IR. For each ω, the radius of the circle is given by the quantity at the right-hand side of (17) in the case of F11 ( ω), while in the case of F22 ( ω) by the quantity at the right-hand side of (18). From a practical point of view, it is sufficient to draw the Gershgorin circles only for a finite number of values of ω, in particular at low frequency, since the radius decreases when frequency increases. At this point, according to the Nyquist criterium, the closedloop is asymptotically stable if N = pol (19) being pol the number of positive real part roots of the open loop characteristic polynomial cpol (s). Parameters τzi and τpi are then designed so as to ensure that conditions (17), (18) and (19) are satisfied. IV. C ASE STUDY A 0.4 kV − 50 Hz distribution system is considered composed of a three-phase feeder with eight nodes and supplied by a 20/0.4 kV substation. The 20 kV supplying system is represented by its Thevenin equivalent seen from the MV/LV substation, assuming a 1000 M V A short-circuit power and an open-circuit voltage in the range 0.95 ÷ 1.05 p.u.. The MV system supplies the 20/0.4 kV 0.25 MVA transformer TABLE I. B RANCH ELECTRICAL PARAMETERS Branch from node to 1 2 3 4 5 6 7 TABLE II. Node number 1 2 3 4 5 6 7 8 node 2 3 4 5 6 7 8 R [Ω] 0.067 0.038 0.073 0.051 0.061 0.033 0.026 X [Ω] 0.016 0.009 0.017 0.009 0.009 0.005 0.004 R ATED NODAL LOADS PL [kW ] 64.0 0.58 21.0 1.5 0.0 4.4 1.5 4.4 QL [kV Ar] 32.0 0.26 9.45 0.68 0.0 2.0 0.68 2.0 with 6% short-circuit voltage. The electrical parameters of the seven lines composing the feeder and of the eight loads are reported Tables I and II, respectively. Two 20 kW − 25 kVA PV systems are connected at nodes m1 = 5 and m2 = 7. The interface inverter is equipped with active and reactive power control systems. In the remainder, firstly, the proposed design is described and the stability analysis performed and, then, the voltage controllers are validated by numerical simulations. A. Numerical design Referring to the model of the DG, the values assumed by the parameters are τDG,1 = 0.05 and τDG,1 = 0.06. Concerning Γ, it is obtained ! 0.0289 0.0294 Γ= (20) 0.0289 0.0334 Matrix Γ is calculated assuming the following initial operating condition: all loads equal to the 60% of their rated values, the nominal voltage for MV Thevenin equivalent supplying system and no active and reactive power injection by the DGs. Matrix Γ in (20) presents a peculiar structure: the elements lower the main diagonal belonging to the same column have all the same value. Starting from Γ the gains of the controllers which satisfy condition (12) are set equal to kc1 = 300 and kc2 = 250, respectively. Assuming Vref,1 = Vref,2 = 1.0 p.u., V0,5 = 0.945 p.u. and V0,7 = 0.937 p.u., from (10) it is obtained 2 2 ∆Vdes,1 = 0.1070 and ∆Vdes,2 = 0.1220. Then, using (15) −4 −2 T it is ess ≈ (−4e 1.51e ) . From (13) with s = 0, it is obtained ∆V52 = 0.1065 and ∆V72 = 0.1079; finally, from (9) it is V5 = 0.9998 p.u. and V7 = 0.9929 p.u.. These latter values, compared with the unitary references Vref,1 and Vref,2 , give evidence of the quite small regulation error achieved in terms of voltage amplitudes. 1 Imaginary Axis Imaginary Axis 5 0 -5 0.5 0 -0.5 -10 5 10 Real Axis 15 -1 -2 20 10 1 5 0.5 Imaginary Axis Imaginary Axis 0 0 -1 0 Real Axis 1 2 -1 0 Real Axis 1 2 0 -5 -0.5 -10 0 5 10 Real Axis 15 20 -1 -2 Figure 4. Nyquist plot of F11 ( ω) and Gershgorin circles (top-left). Nyquist plots of F22 ( ω) and Gershgorin circles (bottom-left). Zoomed view of F11 ( ω) around the critical point (top-right). Zoomed view of F22 ( ω) around the critical point (bottom-right). As concerns the parameter’s value of each controller it is set τz1 = 0.03, τp1 = 7, τz2 = 0.02 and τp2 = 6. Substituting in (11) the values of all parameters it results 8.67(1+0.03 s) 7.35(1+0.02 s) F(s) = (1+7 s)(1+0.05 s) (1+6 s)(1+0.06 s) 8.67(1+0.03 s) (1+7 s)(1+0.05 s) 8.35(1+0.02 s) (1+6 s)(1+0.06 s) The Nyquist plots of F11 ( ω) and F22 ( ω) together with four Gershgorin circles drawn at ω1 = 1e−4 (red), ω2 = 1e−3 (green), ω3 = 1e−2 (magenta) and ω4 = 1e−1 (black) are built and reported in Fig. 4. In particular the two plots at the right-hand side of Fig. 4 report a zoomed view which show that the Gershgorin circles do not encompass the critical point. Conditions (17)-(18) are satisfied and matrix I+F( ω) is then row dominant diagonal. Moreover, from the analysis of the Nyquist plots reported in Fig. 4 it is trivial to recognize that ν1 = ν2 = 0; then N = 0. At this point, since the roots of polynomial cpol (s) are {−20, −50/3, −1/6, −1/7} it results pol = 0. Then condition (19) is fulfilled and the closed-loop system is asymptotically stable. B. Numerical simulations The distribution system under study has been simulated in PSCAD/EMTDC environment, including the PV systems and the designed voltage controllers. To validate the performance and the stability of the system the transients in both unsaturated and saturated operation have been analyzed. 1) Unsaturated operation: The analyzed perturbations are small so that the DG voltage controllers respond in the linear range of reactive power variations, without the intervention of any saturation. Two types of perturbations are considered: the connection of a load at node 6, in between the two PV systems, and the step increase (from 1.0 p.u. to 1.005 p.u.) of the voltage amplitude at the MV busbar of the secondary substation. In addition two different operating conditions of the system are assumed: • Case A: all the loads are equal to half of their rated values and both the PV systems generate 20 kW of active power; • Case B: all the loads are increased of the 50% with respect to their rated values and both the PV systems generate 0.5 kW of active power. These cases are considered because they represent extreme operating conditions that are far from the ones assumed at the voltage control design state. To keep the DG reactive power injections in the linear range of variation, adequate voltage reference set-points are set, in particular, • Case A: Vref,1 = 1.023 p.u. and Vref,2 = 1.029 p.u., • Case B: Vref,1 = 0.942 p.u. and Vref,2 = 0.935 p.u.. In Figs. 5 and 6 the time evolution of the voltages at nodes 5 and 7 and of the reactive powers injected by the two PV systems are reported in the Case A, respectively, for the load insertion and for the step increase of the voltage at MV busbar. In all figures, the electrical quantities are expressed in p.u. of the PV systems rated powers. From these figures, it apparent that the response to both the small perturbations is stable and that the nodal voltages recover their value near to the set-point in few seconds, whereas the reactive power injections have a smooth variation and settle in few tens of seconds. Similar considerations can be made for the Case B from the Figs. 7 and 8, which report the time evolution of the same quantities in response to the same two same small perturbations. 2) Saturated operation: The analyzed perturbations are large so that the DG voltage controllers respond leaving or entering saturation on reactive power injections. Two perturbations of the voltage amplitude at the MV busbar of the secondary substation are considered by applying firstly a step increase from 1.0 p.u. to 1.01 p.u. and then a step decrease back from 1.01 p.u. to 1.0 p.u. Concerning the operating conditions of the system, it is assumed that all the loads are equal to their rated values and both the PV systems generate 12.5 kW of active power. The voltage reference set-points of both the DG controllers are equal to 1.0 p.u. In Figs. 9 and 10 the time evolution of the voltages at nodes 5 and 7 and of the reactive powers injected by the two PV systems are reported respectively for the step increase and decrease of the voltage at MV busbar. From the time evolution of the reactive powers, it is evident that after the first perturbation both the voltage controllers leave the saturated operation, in which the reactive power injection is fixed at its maximum value, that is 0.5 p.u.. After the second perturbation the controllers leave the linear range of operation and reactive power is saturated again. It apparent that the response to both the perturbations is stable in spite of the saturation effect on reactive powers. Concerning the time evolution of voltages, it is apparent that, as expected, when the reactive power saturates, the voltages remain far from voltage reference setpoints. DG reactive power injections DG reactive power injections 0.150 0.100 0.360 Qdg1 0.050 Qdg1 0.340 0.000 0.320 -0.050 0.300 -0.100 0.280 Qdg2 -0.150 0.260 -0.200 -0.250 0.240 -0.300 0.220 sec Qdg2 90.0 95.0 100.0 105.0 110.0 115.0 120.0 125.0 130.0 135.0 140.0 90 100 110 Feeder nodal Voltages 1.0300 130 140 Feeder nodal Voltages 0.9420 Vdg2 1.0290 0.9410 1.0280 0.9400 1.0270 0.9390 1.0260 0.9380 1.0250 0.9370 1.0240 Vdg1 0.9360 Vdg1 1.0230 0.9350 1.0220 0.9340 1.0210 0.9330 1.0200 Vdg2 0.9320 98.0 99.0 100.0 101.0 102.0 103.0 104.0 105.0 106.0 107.0 108.0 Figure 5. Time evolution of the reactive powers injected by DG and of the voltages at the nodes where DG units are connected – Case A, load variation 98.0 99.0 100.0 DG reactive power injections 102.0 103.0 104.0 105.0 106.0 107.0 108.0 DG reactive power injections 0.400 0.050 0.350 0.000 101.0 Figure 7. Time evolution of the reactive powers injected by DG and of the voltages at the nodes where DG units are connected – Case B, load variation 0.100 0.300 Qdg2 -0.050 Qdg2 0.250 -0.100 0.200 Qdg1 -0.150 0.150 -0.200 sec 120 Qdg1 0.100 140 150 160 170 180 190 140 150 160 Feeder nodal Voltages 170 180 190 Feeder nodal Voltages 1.0320 0.9440 1.0310 0.9430 1.0300 1.0290 Vdg2 0.9420 Vdg1 0.9410 1.0280 0.9400 1.0270 0.9390 1.0260 0.9380 1.0250 0.9370 1.0240 1.0230 0.9360 Vdg1 0.9350 1.0220 148.0 Vdg2 0.9340 149.0 150.0 151.0 152.0 153.0 154.0 155.0 156.0 157.0 158.0 Figure 6. Time evolution of the reactive powers injected by DG and of the voltages at the nodes where DG units are connected – Case A, MV voltage variation 148.0 149.0 150.0 151.0 152.0 153.0 154.0 155.0 156.0 157.0 158.0 Figure 8. Time evolution of the reactive powers injected by DG and of the voltages at the nodes where DG units are connected – Case B, MV voltage variation V. C ONCLUSION DG reactive power injections Qdg1 A design methodology for voltage control of DG units in LV distribution systems has been developed. A local controller measures and regulates the voltage at the node to which each DG unit is connected by acting on the DG reactive power injection. According to a decentralized architecture, no data or measurement exchange with other controllers is needed. The design of the local controller is based on a structural MIMO model of the distribution system. The results of numerical simulations have shown the effectiveness of the proposed approach: the designed controllers improve the nodal voltages by ensuring a satisfactorily regulation. An important feature of the proposed design methodology is that it avoids operation conflicts among DG units and system instability, which is the main limitation of the Volt/Var controls that have previously been proposed for DG units in LV networks. Qdg2 0.500 0.480 0.460 0.440 0.420 0.400 0.380 sec 45.0 50.0 55.0 60.0 65.0 70.0 75.0 80.0 85.0 90.0 95.0 Feeder nodal Voltages Vdg1 1.0000 Vdg2 0.9990 0.9980 0.9970 0.9960 0.9950 R EFERENCES [1] P. Ferreira, P. Carvalho, L. Ferreira, and M. Ilic, “Distributed energy resources integration challenges in low-voltage networks: Voltage control limitations and risk of cascading,” IEEE Trans. on Sustainable Energy, vol. 4, no. 1, pp. 82–88, June 2013. [2] P. Chen, R. Salcedo, Q. Zhu, F. de León, D. Czarkowski, Z. Jiang, V. Spitsa, Z. Zabar, and R. Uosef, “Analysis of voltage profile problems due to the penetration of distributed generation in low-voltage secondary distribution networks,” IEEE Trans. on Power Delivery, vol. 27, no. 4, pp. 2020–2028, Oct. 2012. [3] R. Tonkoski, D. Turcotte, and T. El-Fouly, “Impact of high pv penetration on voltage profiles in residential neighborhoods,” IEEE Trans. on Sustainable Energy, vol. 3, no. 3, pp. 518–527, July 2012. [4] S. Conti, A. Greco, N. Messina, and S. Raiti, “Local voltage regulation in lv distribution networks with pv distributed generation,” in International Symposium on Power Electronics, Electrical Drives, Automation and Motion, May 2006, pp. 519–524. [5] E. Demirok, D. Sera, R. Teodorescu, P. Rodriguez, and U. Borup, “Evaluation of the voltage support strategies for the low voltage grid connected pv generators,” in Energy Conversion Congress and Exposition (ECCE), 2010 IEEE, Sep 2010, pp. 710–717. [6] S. Chalise, B. Poudel, and R. Tonkoski, “Overvoltages in lv rural feeders with high penetration of wind energy,” in IEEE Power and Energy Society General Meeting, July 2013, pp. 1–5. [7] M. Juamperez, G. Yang, and S. Kjaer, “Voltage regulation in lv grids by coordinated volt-var control strategies,” Journal of Modern Power Systems and Clean Energy, vol. 2, no. 4, pp. 319–328, Dec. 2014. [8] P. Jahangiri and D. Aliprantis, “Distributed volt/var control by pv inverters,” IEEE Trans. on Power Systems, vol. 28, no. 3, pp. 3429– 3439, Aug. 2013. [9] P. Esslinger and R. Witzmann, “Experimental study on voltage dependent reactive power control q(v) by solar inverters in low-voltage networks,” in CIRED, Jun 2013, pp. 1–4. [10] M. Kashem and G. Ledwich, “Multiple distributed generators for distribution feeder voltage support,” IEEE Trans. on Energy Conversion, vol. 20, no. 3, pp. 676–684, Sept. 2005. [11] A. R. Di Fazio, G. Fusco, and M. Russo, “Decentralized control of distributed generation for voltage profile optimization in smart feeders,” IEEE Trans. on Smart Grid, vol. 4, no. 3, pp. 1586–1596, Sept. 2013. [12] M. Baran and F. Wu, “Optimal capacitor placement on radial distribution systems,” IEEE Trans. on Power Delivery, vol. 4, no. 1, pp. 725–743, Jan. 1989. [13] A. R. Di Fazio, G. Fusco, and M. Russo, “Decentralized voltage control of distributed generation using a distribution system structural mimo model,” Control Engineering Practice, vol. 46, pp. 81–90, 2016. [14] H. Rosenbrok, Computer aided control systems design. Academic Press, 1974. 0.9940 0.9930 0.9920 0.9910 49.0 50.0 51.0 52.0 53.0 54.0 Figure 9. Time evolution of the reactive powers injected by DG and of the voltages at the nodes where DG units are connected – Case C, MV voltage step increase DG reactive power injections Qdg1 Qdg2 0.500 0.480 0.460 0.440 0.420 0.400 0.380 sec 99.0 100.0 101.0 102.0 103.0 104.0 103.0 104.0 Feeder nodal Voltages Vdg1 1.0000 Vdg2 0.9980 0.9960 0.9940 0.9920 0.9900 99.0 100.0 101.0 102.0 Figure 10. Time evolution of the reactive powers injected by DG and of the voltages at the nodes where DG units are connected – Case C, MV voltage step decrease