You are on page 1of 12

This article has been accepted for publication in a future issue of this journal, but has not been

fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TSG.2018.2871413, IEEE
Transactions on Smart Grid
1

Transient Optimization of Parallel Connected


Inverters in Islanded AC Microgrids
Anushka M. Dissanayake, Student Member, IEEE and Nishantha C. Ekneligoda, Member, IEEE

 control [2],[10]. Droop controller mimics the behavior of the


Abstract--Along with the rapid growth in green energy synchronous generator and it is the most widely used
utilization, microgrids are becoming an interesting part of the autonomous, decentralized MG control method.
future power systems. This paper presents a transient trajectory Transients in MGs occur as a result of various disturbances
optimization of parallel connected inverter system in an islanded
in the system. Transient improvements of MGs and parallel
microgrid. Optimal state and control transient trajectories are
generated which drive each individual inverter from given initial inverter systems are reported in the literature [11]-[17].
conditions to their respective desired steady state equilibrium. Among them, major attention has been posed to improve the
This trajectory optimization is performed in decentralized transient behavior of the droop based MGs with distributed
manner and they are the open loop, local optimal control signals generators (DGs). Subsequent work done in [13]-[17] depict
of the inverters. The Pontryagin’s minimum principle is the optimum control of MGs operated with DGs. Optimized
employed to obtain the optimal state and control signal
sensitivity analysis based decentralized control method for
trajectories. Then, a stability analysis of the system with the local
optimal control signals is done by a Lyapunov approach. distribution network is proposed in [13]. Voltage regulation,
Dynamic model of the system and the optimum trajectories are minimization of the active power losses and the reactive
obtained in the d-q reference frame. Two example microgrid power exchange with the distributed generation units are
systems one with three inverters and the other one with five achieved by employing an artificial intelligent based
inverters were used to demonstrate the effectiveness and the optimization technique. Optimum dynamic response of
performance of the proposed approach which can be easily
parallel inverters is achieved in [14] by setting the optimum
extended to any type of microgrid such as AC DC, hybrid
microgrids in both islanded or grid connected mode. droop gains obtained by the differential evolution global
search technique. This method uses the complete system
Index Terms-- d-q reference frame, islanded microgrids, matrix of the whole system to find the optimum droop gains.
optimal control, parallel inverters, Pontryagin’s minimum Transient optimization of MGs subjected to a mode change is
principle, transient optimization. addressed in [15]-[17]. Based on the small signal model of the
droop controlled MG, optimum droop and PI controller gain
I. INTRODUCTION calculation methodology is proposed in [15]. Here, a genetic

W ITH the development in technological and conceptual


arenas in power engineering, microgrids (MGs) have
become an exciting part in modern electric power systems [1]-
algorithm based optimization technique is employed to
minimize the deviation between the instantaneous power and
the nominal output power of the inverter during the switching
[3]. Distributed energy resources (DERs) [4], distributed between the grid-connected and the islanded modes. In [16]
control [5] and high penetration of renewable energy [6] are and [17], particle swarm optimization technique is employed
some attractive aspects which can be recognized in these to tune the PI controller gains and the droop gains of the droop
MGs. Since renewable energy sources as well as small scale based MG. Objective functions in [16] are proposed to
power generating units dominate in these systems, low minimize the error in the measured power and to enhance the
spinning reserve, inertia and damping can be observed in damping characteristics in each mode of operation. On the
contrast to the large scale power systems [7]. Therefore, novel other hand, in [17], performance index comprises of the active
modeling tools and control techniques are required to analyze and reactive power errors and the voltage and frequency
these future microgrids. Further, decentralized control of the deviations from their nominal values. However, these methods
MGs has more attractive features than the centralized control have limited focus on the use of decentralized optimal control
[8],[9]. In decentralized control the active components in such strategies which minimize local objective functions. In
control systems use locally available measurements to take contrast, this paper proposes a decentralized optimal control
individual decisions, which make the power system operations approach to find the optimal behavior of the control and states
fast and economical. Several decentralized MG control of DGs in islanded microgrids.
methods have been proposed in recent few decades and the Continuous time optimal control method has been applied
most dominant primary control methodology is the droop in various engineering applications for dynamic optimization
[18],[19] where the Pontryagin’s minimum principle is the
A. M. Dissanayake is with the Oklahoma State University, Stillwater, OK basis which is inspired by the variational calculus [19]. In
74078 USA (e-mail: anushkd@okstate.edu). electrical systems it has been used in applications such as
N. C. Ekneligoda is with the Oklahoma State University, Stillwater, OK
74078 USA (e-mail: nishantha.ekneligoda@okstate.edu). hybrid electrical vehicles [20]-[22] and in power system

1949-3053 (c) 2018 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TSG.2018.2871413, IEEE
Transactions on Smart Grid
2

applications [23]-[25]. Optimal control of a multi-inverter bus is referred to as the point of common coupling (PCC).
system is presented in [23] which tries to minimize a Further, a distribution line is shown from node 𝑖 to node 𝑗 and
performance index, consists of the output voltage error, the at the end of the node 𝑗, a series RL (resistive and inductive)
inductor currents of all the inverters and the corresponding load is connected to the network. Once the initial conditions
reference signals. In this approach, single objective function is and the steady state operating points of an inverter system are
considered for the complete system including all the control provided, there will be multiple control trajectories which
variables and states, and the minimization is done in a drive the inverter from the initial conditions to the preferred
centralized manner. Continuous time optimal power flow final manifold. Among them, finding the optimum transient
control in cooperative network of MGs is introduced in [24] trajectory which minimizes a desired performance index is
where the objective is to minimize the power flows while important and will be the main objective of this paper. Further,
maintaining the operating points of the storage systems around the attention has been posed to the startup transient
the desired values. In addition, exchange of information optimization which considers the zero initial conditions.
among the active components in the system as well as the Moreover, the controllers are operated in decentralized
usage of forecasts of energy productions and consumptions are manner and use only the locally available measurements.
utilized to improve the control performances. Design of an
ilDs , ilQs ilDr, ilQr
optimal grid stabilizer for weak/islanded grids using a unified Lls Llr vbDj , vbQj
vbDi , vbQi Rlr
power quality conditioner is proposed in [25]. Here, a zero Node j
sum, two player game is formed between unified power Rls Node i vbdi , vbqi
RLj
quality conditioner control and grid disturbances. Game Coupling inductor Lci iLDj , iLQj
theoretic decentralized optimum decision making iodi , ioqi
LLj
methodology is presented in [7] for DC MGs. Modeling the Rci
power electronic converters as variable impedance loads, vodi , voqi
optimum transient trajectories, which drive the system from a LC Cfi
given initial conditions to the desired steady states, are filter Lfi
generated by employing the Pontryagin’s minimum principle. Inverter terminal
In this paper, a decentralized local optimal control
approach is introduced to transient path optimization of the ~
individual inverters in an islanded MG. In the proposed local =
optimal control approach, each individual inverter system tries Source
to minimize its own dynamic cost function by maneuvering
their own local control inputs. In this modeling the optimal Fig. 1. Single DG and one load connected to the main bus.
trajectories of the control inputs which minimize a
performance index are generated in the d-q reference frame Suppose there are 𝓃 number of DGs in the set 𝒩 which are
[26]. Stability of the local optimal controllers are assessed connected to the islanded MG. Consider the dynamic model of
using a Lyapunov stability approach [27]. The main the 𝑖 𝑡ℎ DG (𝑖 ∈ 𝒩) as shown in Fig. 1. Dynamic model of
contribution of this paper is the application of the Pontryagin’s such DG with current and voltage controllers in d-q domain is
minimum principle to generate the local optimal control well known and stability and controlling aspects are well
trajectories and Lyapunov stability assessment of the local established in [15],[28],[29]. Rotating reference frame (d-q
optimal controllers. domain) dynamic state equations of the 𝑖 𝑡ℎ DG coupling
The rest of the paper is organized as follows. In section II inductor current can be expressed as,
overview of the islanded MG followed by the mathematical 𝑑𝑖𝑜𝑑𝑖
modeling is presented. Introduction of the optimal control 𝐿𝑐𝑖 = 𝑣𝑜𝑑𝑖 − 𝑣𝑏𝑑𝑖 − 𝑅𝑐𝑖 𝑖𝑜𝑑𝑖 + 𝜔𝑖 𝐿𝑐𝑖 𝑖𝑜𝑞𝑖 (1)
𝑑𝑡
problem and the objective function modeling is given in
section III. The solution to the optimal control problem and 𝑑𝑖𝑜𝑞𝑖
𝐿𝑐𝑖 = 𝑣𝑜𝑞𝑖 − 𝑣𝑏𝑞𝑖 − 𝑅𝑐𝑖 𝑖𝑜𝑞𝑖 − 𝜔𝑖 𝐿𝑐𝑖 𝑖𝑜𝑑𝑖 (2)
the Lyapunov stability analysis of the local optimal controllers 𝑑𝑡
are presented in section IV and V. Simulation study is given in where, 𝑣𝑜𝑑𝑖 , 𝑖𝑜𝑑𝑖 , 𝑣𝑜𝑞𝑖 , 𝑖𝑜𝑞𝑖 are the d and q axis LC filter
section VI and the paper is concluded in section VII. output voltage and current components of the 𝑖 𝑡ℎ DG. The
term 𝜔𝑖 denotes the frequency, 𝐿𝑐𝑖 and 𝑅𝑐𝑖 are the inductance
II. SYSTEM MODEL IN THE D-Q REFERENCE FRAME and the resistance of the coupling inductor. Local d and q axis
In a parallel inverter system, input side of each inverter is nodal bus bar voltages at the 𝑖 𝑡ℎ DG PCC are given as 𝑣𝑏𝑑𝑖
connected to a micro source and the output terminal is and 𝑣𝑏𝑞𝑖 . In addition, two control inputs are defined for the
connected to the main bus through a LC filter and a coupling 𝑖 𝑡ℎ DG as depicted in (3).
inductor as shown in Fig. 1. This single inverter system with a
micro source, a LC filter and a coupling inductor is considered 𝑑𝑣𝑜𝑑𝑖 𝑑𝜔𝑖
= 𝑢 𝑣𝑜𝑑𝑖 & = 𝑢𝜔𝑖 (3)
as a distributed generator (DG). Hence the terms inverter 𝑑𝑡 𝑑𝑡
system and the DG are interchangeably used in this paper. The In order to bring the complete system to a common
point where the individual inverter is connected to the main reference frame, a single DG is selected as the common frame

1949-3053 (c) 2018 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TSG.2018.2871413, IEEE
Transactions on Smart Grid
3

and all the other individual DGs are transformed in to that assumed that the Q axis voltage (𝑣𝑜𝑞𝑖 ) is set to zero in the
common reference frame. Furthermore, network and the load controller. On the other hand, the control input of the 𝑖 𝑡ℎ
equations are represented on this common reference frame and system is defined as 𝑢𝑖 (𝑡) ∈ Ω𝑖 ⊆ ℝ𝑚 ; 𝑖 ∈ 𝒩 ; 𝑚 = 2,
transformation suggested in [29] is applied here. An angle 𝛿 is where, 𝑢𝑖 = [𝑢 𝑣𝑜𝑑𝑖 𝑢𝜔𝑖 ]𝑇 given by (3).
defined for each individual DG as follows,
III. OPTIMAL CONTROL PROBLEM AND MODELING OF THE
𝛿𝑖 = ∫(𝜔𝑖 −𝜔𝑐𝑜𝑚 ) 𝑑𝑡 (4) OBJECTIVES
Suppose the state vector of the individual dynamical system
where, 𝜔𝑐𝑜𝑚 is the angular frequency of the common
is defined as 𝒳𝑖 ∶ [𝑡0 , 𝑡𝑓 ] → ℝ𝑛𝑖 and the initial state vector is
frame. This angle 𝛿𝑖 is the angle between the common
reference frame and an individual DG reference frame. Then 𝒳𝑖 (𝑡0 ), where the initial and final times are 𝑡0 and 𝑡𝑓
the quantities in the individual DG reference frame (d-q respectively. Then the optimal control problem can be defined
currents or voltages - 𝑓𝑑𝑞,𝑖 ∈ ℝ2 ) can be transformed in to the as, finding the admissible control 𝑢𝑖 ∶ [𝑡0 , 𝑡𝑓 ] → Ω𝑖 ⊆ ℝ𝑚
quantities in the common reference frame (𝐹𝐷𝑄,𝑖 ∈ ℝ2 ) as, such that the dynamical constraints (10) are satisfied and such
that the cost functional given in (11) is minimized [18].
cos 𝛿𝑖 − sin 𝛿𝑖
𝐹𝐷𝑄,𝑖 = [𝑇𝑖 ]𝑓𝑑𝑞,𝑖 where, 𝑇𝑖 = [ ] (5) 𝒳̇𝑖 (𝑡) = 𝑓𝑖 (𝒳𝑖 (𝑡), 𝑢𝑖 (𝑡), 𝑡) ∀ 𝑡 ∈ [𝑡0 , 𝑡𝑓 ] (10)
sin 𝛿𝑖 cos 𝛿𝑖
𝑡𝑓
When connecting an individual inverter to the main system,
output states need to be transformed in to the common frame 𝐽𝑖 (𝑢𝑖 ) = 𝜃𝑖 [𝒳𝑖 (𝑡𝑓 ), 𝑡𝑓 ] + ∫ Υ𝑖 [𝒳𝑖 (𝑡), 𝑢𝑖 (𝑡), 𝑡] 𝑑𝑡 (11)
by employing the transformation matrix (5). According to (1) 𝑡0
𝑇
and (2), output states of the 𝑖 𝑡ℎ DG are [𝑖𝑜𝑑𝑞,𝑖 ] = [𝑖𝑜𝑑𝑖 𝑖𝑜𝑞𝑖 ] .
In this paper, it is assumed that the initial time 𝑡0 as well as the
Hence, the output quantities of the 𝑖 𝑡ℎ DG seen by the
𝑇
initial state 𝒳𝑖 (𝑡0 ) are specified and the free time (free 𝑡𝑓 )
common frame ([𝑖𝑜𝐷𝑄,𝑖 ] = [𝑖𝑜𝐷𝑖 𝑖𝑜𝑄𝑖 ] ) can be obtained as, problem is considered [18]. Further, we assume that, 𝑓𝑖 is
[𝑖𝑜𝐷𝑄,𝑖 ] = [𝑇𝑖 ][𝑖𝑜𝑑𝑞,𝑖 ] (6) Lipschitz continuous [27] and the functions 𝜃𝑖 and Υ𝑖 are at
least once continuously differentiable.
Upper case subscripts (𝐷, 𝑄) denote that the particular In (11) the fixed cost is denoted as 𝜃𝑖 [𝒳𝑖 (𝑡𝑓 ), 𝑡𝑓 ] ∈
quantity is given with respect to the common reference frame. ℝ which is a constant during a transient and it is a function of
On the other hand, in (1) and (2), inputs to the 𝑖 𝑡ℎ DG are the d states at the final manifold. On the other hand,
and q-axis bus bar nodal voltages which are defined on the Υ𝑖 [𝒳𝑖 (𝑡), 𝑢𝑖 (𝑡), 𝑡] ∈ ℝ denotes the variable or, the transient
𝑇
common reference frame ([𝑣𝑏𝐷𝑄,𝑖 ] = [𝑣𝑏𝐷𝑖 𝑣𝑏𝑄𝑖 ] ). These cost. This particular objective function considers the variable
quantities need to be converted back to the individual DG cost from the initial time 𝑡0 to the final time 𝑡𝑓 . Since this
reference frame by employing the inverse transformation of paper mainly focuses on transient optimization, 𝜃𝑖 [𝒳𝑖 (𝑡𝑓 ), 𝑡𝑓 ]
(5), is considered to be zero.
In parallel connected inverter system, individual DGs’
[𝑣𝑏𝑑𝑞,𝑖 ] = [𝑇𝑖 ]−1 [𝑣𝑏𝐷𝑄,𝑖 ] (7)
transient objective function consists of several basic sub
In order to well define the bus bar nodal voltages objectives which are, regulation of the direct and quadrature
𝑣𝑏𝐷 and 𝑣𝑏𝑄 , a sufficiently large virtual resistor (𝑟) is axis currents (𝑖𝑜𝑑𝑖 , 𝑖𝑜𝑞𝑖 ), regulation of the angle separation
introduced [29]. Based on this virtual resistance, and referred between the individual reference frame and the common
to the node 𝑖 in the Fig. 1, d and q axis bus bar voltages reference frame (𝛿𝑖 ), regulation of the d-axis voltage
defined on the common reference frame are derived as, magnitude (𝑣𝑜𝑑𝑖 ), regulation of the frequency (𝜔𝑖 ) and
minimization of the control effort (𝑢𝑖 ). As stated earlier,
𝑣𝑏𝐷𝑖 = 𝑟(𝑖𝑜𝐷𝑖 − 𝑖𝑙𝐷𝑟 − 𝑖𝑙𝐷𝑠 ) (8)
reference frame of one DG is taken as the common frame and
𝑣𝑏𝑄𝑖 = 𝑟(𝑖𝑜𝑄𝑖 − 𝑖𝑙𝑄𝑟 − 𝑖𝑙𝑄𝑠 ) (9) all the other DGs are represented with respect to that common
frame. Therefore, the angle regulation term is not included in
Dynamic equations of the loads and the network are similar the objective function of the common DG. Quadratic type cost
to (1) and (2) [29]. In case of a motor load such as an function (objective function) of the form (12) is considered.
induction motor, the corresponding RL equivalent model in d-
q reference frame can be utilized [30]. Since 𝓃 individual Υ𝑖 [𝒳𝑖 , 𝑢𝑖 , 𝑡] = 𝑄𝑖2 + 𝑢𝑖𝑇 𝑅𝑖 𝑢𝑖 (12)
inverter systems are in the MG, this procedure is repeated Here, 𝑅𝑖 ∈ ℝ2𝑥2 is the diagonal, positive definite [27] control
from 𝑖 = 1 to 𝓃. DG 1 is selected as the common DG and the
input weight matrix (gain matrix) of the 𝑖 𝑡ℎ system and 𝑄𝑖2 ∈
set of DGs without the common DG is represented as 𝒩 − ⊂
ℝ is the quadratic cost term of the 𝑖 𝑡ℎ local inverter system
𝒩. Local state of the 𝑖 𝑡ℎ DG is 𝒳𝑖 (𝑡) = defined as in (13). Suppose the steady state desired values of
[𝑖𝑜𝑑𝑖 𝑖𝑜𝑞𝑖 𝛿𝑖 𝑣𝑜𝑑𝑖 𝜔𝑖 ]𝑇 ∈ ℝ𝑛𝑖 ; 𝑖 ∈ 𝒩 , where, 𝑛𝑖 = 5 ̅𝑖 ; 𝑖 ∈ 𝒩 − . Then the
the 𝑖 𝑡ℎ local inverter system is given by 𝒳
for all the inverter systems except the common DG. For the
quadratic, positive definite state cost is,
common DG, angle state is not defined hence 𝑛𝑖 = 4. It is
̅𝑖 )𝑇 𝑃𝑖 (𝒳𝑖 − 𝒳
𝑄𝑖2 = (𝒳𝑖 − 𝒳 ̅𝑖 ) (13)

1949-3053 (c) 2018 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TSG.2018.2871413, IEEE
Transactions on Smart Grid
4

Where, 𝑃𝑖 ∈ ℝ5𝑥5 ; 𝑖 ∈ 𝒩 − , is the diagonal, positive definite column vector. The superscript in the terms such as 𝒳𝑖∗ , 𝑢𝑖∗ , 𝜆∗𝑖
state weight matrix. For the common inverter system, 𝑃𝑖 ∈ denotes the optimal condition. The control signal 𝑢𝑖∗ which
ℝ4𝑥4 . Steady state desired state values of the local inverter satisfies the condition (19) can be explicitly obtained by
systems could be found using a power flow solution method evaluating (20) [19].
developed for MGs [31],[32].
∇𝑢𝑖 ℋ𝑖 = 0 (20)
Both the algebraic constraints and dynamical constraints in
microgrids has significant impact in the optimal control Relations from (15) to (20) generates a two point BVP set with
objective function modeling. In this modeling, algebraic the dynamic system given by (15) and (17) where the
constraints are used for solving the system steady state desired boundary conditions are given by (16) and (18). As mentioned
operating points (𝒳̅𝑖 ). The algebraic constraints used were the before, the fixed cost term 𝜃𝑖 is considered to be zero and
power flow equality constraints, voltage and frequency hence, according to (18), boundary conditions of the costates
margins inequality constraints. However, the major focus of at the final manifold become zero.
this paper is the dynamic optimization in microgrids and the The overall process is illustrated in the flowchart shown in
steady state computations are not included in detail. Fig. 2. For a multi-inverter system with 𝓃 number of DGs, this
procedure is repeated for 𝑖 = 1 to 𝓃. Optimum trajectories are
IV. SOLUTION OF THE LOCAL OPTIMAL CONTROL PROBLEM the solution of the generated BVP equation set along with the
Solution of the free time optimal control problem is network and load dynamic equations. Obtaining a closed form
obtained using the Pontryagin’s minimum principle. This analytical solution to this BVP problem is not feasible due to
principle generates a two pint boundary value problem (BVP) the higher nonlinearities associated with the equations.
[19] and using the solution of BVP, optimal control signals Therefore, corresponding numerical solution is obtained for
can be extracted. The Pontryagin’s minimum principle the two point BVP set. Number of methods have been
suggests the necessary conditions for the optimality and the developed to solve the two point BVPs such as steepest decent
control signals are in open loop form [18],[19]. based technique, variation of extremals, quasi-linearization
and gradient projection [33]. Simulation time and complexity
Definition 1 [18]: Hamiltonian function, ℋ𝑖 ∶ ℝ𝑛𝑖 × Ω𝑖 × of the BVP set increases with the number of DGs.
ℝ𝑛𝑖 × [𝑡0 , 𝑡𝑓 ] → ℝ , Nevertheless, number of parallel inverters are limited in an
islanded MGs and simulation time and complexity are not
ℋ𝑖 (𝒳𝑖 (𝑡), 𝑢𝑖 (𝑡), 𝜆𝑖 (𝑡), 𝑡) = Υ𝑖 [𝒳𝑖 (𝑡), 𝑢𝑖 (𝑡), 𝑡] + major issues. Furthermore, optimum trajectory calculation is
𝜆𝑇𝑖 (𝑡) 𝑓𝑖 (𝒳𝑖 (𝑡), 𝑢𝑖 (𝑡), 𝑡) (14) an offline process which generates family of optimum
trajectories under different system contingencies. Generated
optimum trajectories can be used as references to maneuver
Where, 𝜆𝑖 ∶ [𝑡0 , 𝑡𝑓 ] → ℝ𝑛𝑖 is the costate vector [19].
the system along the optimum path. The implementation of the
proposed algorithm requires all the system parameters for the
Theorem 1 [18]: If the control 𝑢𝑖∗ ∶ [𝑡0 , 𝑡𝑓 ] → Ω𝑖 is optimal, computation. If the system parameters are varied, then the
then the following conditions are satisfied. model need to be recomputed accordingly. Typically, adaptive
𝒳̇𝑖∗ = ∇𝜆𝑖 ℋ𝑖∗ = 𝑓𝑖 (𝒳𝑖∗ (𝑡), 𝑢𝑖∗ (𝑡), 𝑡) (15) controllers are employed for the systems with uncertain
parameters. However, the focus of this paper is to build the
𝒳𝑖∗ (𝑡0 ) = 𝒳𝑖 (𝑡0 ) (16) theoretical foundation for the dynamic optimization of DG
trajectories under given system contingencies. The
𝜆̇∗𝑖 (𝑡) = −∇𝒳𝑖 ℋ𝑖∗ = −∇𝒳𝑖 Υ𝑖 [𝒳𝑖∗ (𝑡), 𝑢𝑖∗ (𝑡), 𝑡] − development of an adaptive controller is beyond the scope of
𝜕 𝑇 this work and could be a possible future direction.
[ 𝑓𝑖 (𝒳𝑖∗ (𝑡), 𝑢𝑖∗ (𝑡), 𝑡)] 𝜆∗𝑖 (𝑡) (17)
𝜕𝒳𝑖 Start

𝜆∗𝑖 (𝑡𝑓 ) = ∇𝒳𝑖 𝜃𝑖 [𝒳𝑖∗ (𝑡𝑓 ), 𝑡𝑓 ] (18) Local Transient Objective


Function
Steady State
Desired Values

For all 𝑡 ∈ [𝑡0 , 𝑡𝑓 ], the Hamiltonian ℋ𝑖 (𝒳𝑖∗ (𝑡), 𝑢𝑖 (𝑡), 𝜆∗𝑖 (𝑡), 𝑡) Local
Dynamic Hamiltonian
has a global minimum with respect to 𝑢𝑖 ∈ Ω𝑖 at 𝑢𝑖 = 𝑢𝑖∗ (𝑡) Equations
i.e.,
Pontryagin s Minimum Principle
ℋ𝑖 (𝒳𝑖∗ (𝑡), 𝑢𝑖∗ (𝑡), 𝜆∗𝑖 (𝑡), 𝑡) ≤ ℋ𝑖 (𝒳𝑖∗ (𝑡), 𝑢𝑖 (𝑡), 𝜆∗𝑖 (𝑡), 𝑡)
Two Point BVP Set
∀ 𝑢𝑖 ∈ Ω𝑖 , ∀ 𝑡 ∈ [𝑡0 , 𝑡𝑓 ] (19)
Solve Two Point BVP Network and
Set Load Dynamics
Proof of the theorem can be found in [18] and [19]. In the
above equations and in the upcoming relationships, ∇𝒳𝑖 , ∇𝜆𝑖 Optimal Trajectories
and ∇𝑢𝑖 represent the gradients with respect to 𝒳𝑖 , 𝜆𝑖 and
𝑢𝑖 respectively. In this paper gradient is considered as a
Fig. 2. Flowchart of the optimal trajectory generation process.

1949-3053 (c) 2018 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TSG.2018.2871413, IEEE
Transactions on Smart Grid
5

Among the generated optimum trajectories, optimal written in compact from as,

trajectories of the d axis voltage (𝑣𝑜𝑑𝑖 ) and frequency (𝜔𝑖∗ ) can
be extracted and stored for the tracking purpose. These 𝒳̇𝑖 = 𝐹𝑖 (𝒳𝑖 ) + 𝐵𝑖 𝑢𝑖 + 𝑔𝑖 (𝒳
̿𝑖 ) (22)
optimal trajectories are used as the references in the tracking Where, 𝐹𝑖 (𝒳𝑖 ) ∈ ℝ𝑛𝑖 is the local state matrix, 𝐵𝑖 ∈ ℝ𝑛𝑖×𝑚 is
controller. Block diagram of the local optimal tracking the control matrix and 𝑔𝑖 (𝒳 ̿𝑖 ) ∈ ℝ𝑛𝑖 is the interconnected
controller is shown in Fig. 3. Two cascade controllers are term which is a function of states not belongs to the
implemented and shown as voltage controller and the current ̿𝑖 ). All the terms in (22) are
considered local sub system (𝒳
controller. Outer voltage controller takes the d and q axis
∗ ∗ assumed to be Lipschitz continuous [27]. The dynamical
voltage references (𝑣𝑜𝑑𝑖 , 𝑣𝑜𝑞𝑖 ) and compare them with the
system (22) is nonlinear, time invariant and affine in the input
actual d and q axis voltages (𝑣𝑜𝑑𝑖 , 𝑣𝑜𝑞𝑖 ) to generate the [34] with the equilibrium at the origin. Now consider the same
∗ ∗
references to the inner current control loop (𝑖𝑡𝑑𝑖 , 𝑖𝑡𝑞𝑖 ). These cost function (11) with 𝜃𝑖 = 0 and Υ𝑖 is given as (12).
reference current profiles are compared with the measured LC However, now the positive definite [27] term 𝑄𝑖2 is modified
filter inductor currents (𝑖𝑡𝑑𝑖 , 𝑖𝑡𝑞𝑖 ) to generate the desired as 𝑄𝑖2 = (𝒳𝑖 )𝑇 𝑃𝑖 (𝒳𝑖 ) since the equilibrium is at the origin.
∗ ∗
voltage profiles at the inverter terminal (𝑣𝑡𝑑𝑖 , 𝑣𝑡𝑞𝑖 ). Then, the Hence the modified cost function starting from 𝑡0 = 0 to the
generated d and q inverter terminal voltage references are free final time 𝑡𝑓 = ∞ takes the form,
transformed to three phase abc domain voltages utilizing the ∞
optimal frequency trajectory (𝜔𝑖∗ ). These abc domain signals
can be used as the modulation signals in the pulse width 𝐽𝑖 = ∫ 𝑄𝑖2 + 𝑢𝑖𝑇 𝑅𝑖 𝑢𝑖 𝑑𝑡 (23)
modulator (PWM) to generate the optimal switching actions to 0

the inverter. Two proportional integral controllers (PI For any set of admissible control policies [34] 𝑢𝑖 ∈ Ω𝑖 ; 𝑖 ∈
controllers) are used inside the voltage control loop to track 𝒩, the infinitesimal version of the cost function (23) which is
the d axis voltage reference and to regulate the q axis voltage the nonlinear Lyapunov equation can be written as [34],

to zero (𝑣𝑜𝑞𝑖 = 0). Moreover, in the current control loop, two
PI controllers are used to control the LC filter inductor current ̿𝑖 )) = 0
𝑄𝑖2 + 𝑢𝑖𝑇 𝑅𝑖 𝑢𝑖 + 𝜆𝑇𝑖 (𝐹𝑖 (𝒳𝑖 ) + 𝐵𝑖 𝑢𝑖 + 𝑔𝑖 (𝒳 (24)
and consequently to produce the desired inverter terminal
From (20), the optimal control polices of the 𝑖 𝑡ℎ inverter
voltage references.
system is,
v*odi ,v*oqi ω*i
Source
1
vtdi ,vtqi itdi ,itqi iodi ,ioqi 𝑢𝑖∗ = − 𝑅𝑖−1 𝐵𝑖𝑇 𝜆𝑖 (25)
= 2
Voltage Current ~
controller controller Substituting this optimal control in (24), and using algebraic
i*tdi ,i*tqi v*tdi ,v*tqi
Main bus
manipulations following relationship can be obtained.

itdi ,itqi 1
̿𝑖 )) − 𝜆𝑇𝑖 𝐵𝑖 𝑅𝑖−1 𝐵𝑖𝑇 𝜆𝑖 = 0
𝑄𝑖2 + 𝜆𝑇𝑖 (𝐹𝑖 (𝒳𝑖 ) + 𝑔𝑖 (𝒳 (26)
vodi ,voqi 4
Fig. 3. Control diagram of the tracking controller. Furthermore, the optimal cost function (𝐽𝑖∗ ) of the considered
system can be derived by substituting the optimal control
V. STABILITY ANALYSIS OF THE LOCAL INVERTER SYSTEM policy (25) in (23). These optimal cost functions can be treated
Consider the 𝑖 𝑡ℎ inverter system (𝑖 𝑡ℎ subsystem) where 𝑖 ∈ as the Lyapunov functions of each subsystem [35]. Hence the
𝒩 . Dynamic equation set is given by (1)-(4) can be Lyapunov function of the 𝑖 𝑡ℎ inverter system can be derived
represented in matrix from as, (𝒳̇𝑖 = 𝑓𝑖 (𝒳𝑖 , 𝑢𝑖 , 𝑡)) as,

1
(𝑣𝑜𝑑𝑖 − 𝑣𝑏𝑑𝑖 − 𝑅𝑐𝑖 𝑖𝑜𝑑𝑖 + 𝜔𝑖 𝐿𝑐𝑖 𝑖𝑜𝑞𝑖 ) 𝑉𝑖 = 𝐽𝑖∗
𝑇
= ∫ 𝑄𝑖2 + 𝑢𝑖∗ 𝑅𝑖 𝑢𝑖∗ 𝑑𝑡 (27)
𝐿𝑐𝑖
1
(𝑣𝑜𝑞𝑖 − 𝑣𝑏𝑞𝑖 − 𝑅𝑐𝑖 𝑖𝑜𝑞𝑖 − 𝜔𝑖 𝐿𝑐𝑖 𝑖𝑜𝑑𝑖 ) 0
𝒳̇𝑖 = 𝐿𝑐𝑖
(21)
𝜔𝑖 −𝜔𝑐𝑜𝑚 Taking the time derivative of (27) along the dynamics (22),
𝑢 𝑣𝑜𝑑𝑖 and identifying the gradient of the optimal cost is the costates
[ 𝑢𝜔𝑖 ] of the system i.e. ∇𝒳𝑖 𝐽𝑖∗ = 𝜆𝑖 [19],[34], 𝑉𝑖̇ can be expressed as,
Network dynamics which are not local to the 𝑖 𝑡ℎ DG are 1
̿𝑖 ))
𝑉𝑖̇ = 𝜆𝑇𝑖 (𝐹𝑖 (𝒳𝑖 ) − 𝐵𝑖 𝑅𝑖−1 𝐵𝑖𝑇 𝜆𝑖 + 𝑔𝑖 (𝒳 (28)
hidden in the terms 𝑣𝑏𝑑𝑖 and 𝑣𝑏𝑞𝑖 . These are referred to as 2
interconnected terms of the subsystem. Hence the local
dynamics can be seen as a combination of local states, ̿𝑖 )) from (26),
Substituting for the 𝜆𝑇𝑖 (𝐹𝑖 (𝒳𝑖 ) + 𝑔𝑖 (𝒳
interconnected terms and control inputs. Since the actual
steady state points of the above subsystem are known, the 1
𝑉𝑖̇ = −𝑄𝑖2 − 𝜆𝑇𝑖 𝐵𝑖 𝑅𝑖−1 𝐵𝑖𝑇 𝜆𝑖 ≤
equilibrium of (21) can be transferred to the origin by simple 4
algebra [27]. The transformed local system dynamics are 1
− [𝑄𝑖2 + 𝜇𝑚𝑖𝑛 (𝑅𝑖−1 )‖𝐵𝑖𝑇 𝜆𝑖 ‖2 ] (29)
4

1949-3053 (c) 2018 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TSG.2018.2871413, IEEE
Transactions on Smart Grid
6

Where, 𝜇𝑚𝑖𝑛 (𝑅𝑖−1 ) is the minimum eigenvalue of 𝑅𝑖−1 . gain values were selected according to their magnitude
According to (29), time derivative of the considered Lyapunov deviations between the initial conditions and the desired final
function can be shown to be negative definite. Hence the local values. Smaller gains were assigned to the control signals
inverter system is asymptotically stable [27] under the control since control effort minimization is not the major aim and also
signals given by (25). to achieve fast responses. Diagonal matrix is shown as 𝑑𝑖𝑎𝑔.
TABLE II
VI. SIMULATIONS STUDY AND DISCUSSION STATE AND CONTROL GAINS
Example test cases were simulated in Matlab considering 𝑃𝑖 𝑅𝑖
the simple three inverter system illustrated in Fig. 4. Inverter 1 DG 1 𝑑𝑖𝑎𝑔 (10 20 0.02 0.01)
reference frame was selected as the common reference frame (1 × 10−6 )
DG 2 𝑑𝑖𝑎𝑔 (10 20 8 × 105 0.02 0.01)
and virtual resistor (𝑟) was selected as 1000 Ω. Nominal RMS
DG 3 𝑑𝑖𝑎𝑔 (10 20 1 × 105 0.02 0.01)
voltage and the system frequency were considered as 110 V
per phase and 60 Hz respectively. Steady state desired
A. Startup Transient Trajectory Optimization
operating points are given in Table I were considered when
constructing the local objective functions for the startup Transient event of driving the system from no load initial
transient trajectory optimization. The steady state values were conditions to the desired steady state manifold was considered
computed by a power flow method considering the droop in this simulation. Generated two point BVP set was solved
active and reactive power sharing. Active and reactive power using the BVP4C in matlab [36] and the results are depicted in
Fig. 5 to Fig. 9.
droop coefficients were considered as 9.4 × 10−4 𝑟𝑎𝑑 𝑠 −1 /𝑊
and 1.3 × 10−3 𝑉/𝑉𝑎𝑟 respectively for all the DGs.
RL1 = 25
RC1 = 0.03 RL Load 1
LL1 = 50 mH
LC1 = 0.35 mH 1
DG 1

R12 = 0.23

L12 = 0.32 mH

DG 2 2
RC2 = 0.03
L23 = 1.8 mH
LC2 = 0.35 mH

R23 = 0.35

DG 3
3
RC3 = 0.03 RL2 = 35
LC3 = 0.35 mH LL2 = 50 mH RL Load 2

Fig. 4. Example test system.


Fig. 5. Power variation at startup (a) Active power, (b) Reactive power.
TABLE I
STEADY STATE DESIRED OPERATING POINTS FOR STARTUP TRANSIENT Optimal active and reactive power variation of the DGs are
TRAJECTORY OPTIMIZATION
shown in Fig. 5. Transient time can be seen as around 0.03s.
𝑖̅𝑜𝑑𝑖 (A) 𝑖̅𝑜𝑞𝑖 (A) 𝛿𝑖̅ (rads) 𝑣̅𝑜𝑑𝑖 (V) 𝜔
̅ 𝑖 (rads/s) Steady state active power generation of all the DGs are same
DG 1 2.6124 -2.0334 _ 154.9491 376.4204 and equal to 607.2 W. In this case, similar values were chosen
DG 2 2.6071 -0.9833 -0.0001 155.2658 376.4204
for the active power droop coefficients for computing the
steady state desired operating points. Therefore, the active
DG 3 2.6113 -1.8171 -0.0044 155.0142 376.4204
power productions of all the DGs are equal for this example.
On the other hand, reactive power production of DG1, DG2
Even though, only inverter based test systems have been and DG3 are 470.4 Var, 231.5 Var and 422.6 Var. However,
considered in the simulations, same concept can be applied to in this example the reactive power productions become
MGs with directly coupled synchronous generators. In such different from one DG to another. Even though the reactive
scenario, the modeling of the MG need to be altered power droop coefficients were selected equal for all the DGs,
accordingly and the corresponding desired steady state the voltage deviations caused by the droop relationship,
influence this variation. The d and q-axis current variations are
computation methodology has to be utilized to compute the
shown in Fig. 6. The steady state d-axis currents of all the
desired equilibrium points, such as virtual synchronous
generators are 2.6 A. q-axis currents of DG1, DG2 and DG3
generator method instead of traditional droop. Selected state
can be observed as -2.023 A, -0.99 A and -1.817 A. These
and control gain matrices 𝑃𝑖 and 𝑅𝑖 are given in Table II. The values are very close to the expected steady state values given

1949-3053 (c) 2018 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TSG.2018.2871413, IEEE
Transactions on Smart Grid
7

in Table I. Corresponding variation of the 3 phase current


trajectories in the abc domain are shown in Fig. 7.

Fig. 8. Voltage and frequency variation at startup (a) d-axis voltage, (b)
frequency.
Fig. 6. Current variation at startup (a) d-axis, (b) q-axis.

Fig. 9. Angle variation at startup.

Fig. 7. Three phase current variation at startup (a) DG 1, (b) DG 2, (c) DG 3.

The variation of d-axis voltage and frequency are depicted


in Fig. 8. According to that, d axis voltages show 154.9 V,
155.3 V and 155 V at the steady state which are close to the
expected voltages. Frequencies of each generator starts from
the nominal value and converges to the steady state value of
376.4 rad/s after 0.04 s as shown in Fig. 8. (b).
Angle variation of DG2 and DG3 is illustrated in Fig. 9.
Since the weighting factor given to angle regulation is high,
corresponding transient time is small. According to Fig. 9,
DG2 angle converges to -0.0001 rads after 0.01s where DG3
angle converges to -0.0044 rads after 0.02 s.
Fig. 10. Variation of the quadratic state cost (a) proposed method, (b) method
given in [29].

1949-3053 (c) 2018 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TSG.2018.2871413, IEEE
Transactions on Smart Grid
8

The proposed methodology was compared with the


controller utilized in [29] considering the ability of total cost
minimization in same test system. The state cost is given by
(13) was computed for each control techniques and the results
are shown in Fig. 10. According to the results given by Fig.
10, under the proposed methodology, the state cost reaches
zero faster than the method [29]. The proposed method takes
around 0.015s to suppress the state cost to zero while the other
method takes around 0.045s. Therefore, it can be concluded
that the proposed methodology has better cost minimization
capability in contrast to the method in [29].
B. Time and Performance Analysis with a Larger System
The proposed methodology was applied to a slightly bigger
test system for better time and performance analysis. In this
case, the considered system has 5 DG, 3 RL loads and the
calculated steady state operating points are given in Table III.
Fig. 12. Voltage and frequency variation of 5 DG test system (a) d-axis
The considered test system is similar to the one given in Fig. 4 voltage, (b) frequency.
with the mirror image of DG 1, DG 2 and RL load 1 around
the virtual horizontal line through the DG 3 and RL load 2. TABLE IV
TIME TAKEN TO SOLVE THE BVP SET WITH THE NUMBER OF DGS AND LOADS
Results are given in Fig. 11 and Fig. 12. The d and q axis IN THE SYSTEM
current variations follows similar fashion to the active and
reactive power variations. Therefore, these current plots are Nature of the system Time taken (s)
not shown here for the sake of brevity. One DG & One Load 2.05
TABLE III Three DGs & Two Loads 8.21
STEADY STATE DESIRED OPERATING POINTS FOR 5 DG TEST SYSTEM Five DGs & Three Loads 22.07
𝑖̅𝑜𝑑𝑖 (A) 𝑖̅𝑜𝑞𝑖 (A) 𝛿𝑖̅ (rads) 𝑣̅𝑜𝑑𝑖 (V) 𝜔
̅ 𝑖 (rads/s)
The time taken to solve the BVP set in order to obtain the
DG 1 2.4177 -2.0918 - 154.9315 376.4630
optimal trajectory varies with the size of the system. An
DG 2 2.4123 -0.9552 0.0002 155.2743 376.4630
analysis has been performed to show how the solution time of
DG 3 2.4150 -1.5266 -0.0023 155.1018 376.4630 the BVP varies with the number of DGs and loads in the
DG 4 2.4109 -0.6497 -0.0001 155.3666 376.4630 system and the results are depicted in Table IV. The observed
DG 5 2.4200 -2.5947 -0.0069 154.7804 376.4630 results show that, with the less number of DGs and loads, the
time taken to solve BVP set is very small. However, when the
According to the results shown in Fig. 11 (a), active power system grows up with more number of DGs and loads, the
productions of all the DGs converge to 562.4 W after 0.03s. time taken to solve the BVP set increases. Nevertheles,
Reactive power productions of the DGs are 486 Var, 222.5 solving the BVP set is an offline process and the number of
Var, 356.1 Var, 150.7 Var and 602.3 Var respectively. The d DGs in an islanded MG are limited as stated before. Hence the
axis voltages and the frequencies of all the DGs converges to solving time of the BVP set is not a drawback for real time
the desired steady state values after 0.04s as shown in Fig. 12. control.
C. Optimal Transient Trajectories with a Load Disturbance
Optimal transient trajectories under a load disturbance were
obtained in this test case. Step change was given to the
resistive part of the load at bus 3 from 35 Ω to 25 Ω at t =
0.05s. The new desired steady state operating points after the
load disturbance are shown in the Table V. Resultant optimal
trajectories are given in Fig. 13 to Fig. 17. The d and q axis
current variations are same as the active and reactive power
variations and hence they are not shown here for the sake of
brevity. However, the corresponding variation of the 3 phase
currents in the abc domain are depicted in Fig. 15.
TABLE V
STEADY STATE DESIRED OPERATING POINTS AFTER THE LOAD CHANGE

𝑖̅𝑜𝑑𝑖 (A) 𝑖̅𝑜𝑞𝑖 (A) 𝛿𝑖̅ (rads) 𝑣̅𝑜𝑑𝑖 (V) 𝜔


̅ 𝑖 (rads/s)
DG 1 2.7798 -2.1088 _ 154.9264 376.3839
Fig. 11. Power variation of 5 DG test system (a) Active power, (b) Reactive
power. DG 2 2.7747 -1.1688 -0.0001 155.2098 376.3839
DG 3 2.7829 -2.6819 -0.0054 154.7542 376.3839

1949-3053 (c) 2018 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TSG.2018.2871413, IEEE
Transactions on Smart Grid
9

increasing their power generation to 647 W at steady state as


depicted in Fig. 13. Due to the load change at bus 3, reactive
power generation of the DG connected to bus 3 shows a
significant change as shown by Fig. 14. Its steady state value
reaches to 621.2 Var. Small variations can be observed in
DG1 and DG2 where the steady state values after the load
change are 489.4 Var and 274.1 Var. The d-axis voltage
variation is given in Fig. 16 (a). DG3 voltage is affected by the
load change since it is the nearest DG to the load subjected to
the change. Its voltage reach to 154.75 V at the steady state,
while DG1 and DG2 gain 154.9 V and 155.2 V. Small
undershoot can be observed in the DG3 frequency at the load
change as shown in Fig. 16 (b). All the frequencies converges
to the steady state value of 376.38 rad/s. The angle variation
after the load change is depicted by Fig. 17. Transient time is
Fig. 13. Active power variation subjected to a load change. very small subjected to the load change and it is less than
0.02s. change in steady state angle of DG2 is negligible, while
the DG3 angle changes to -0.0054 rads.

Fig. 14. Reactive power variation subjected to a load change.

Fig. 16. Voltage and frequency variation subjected to load change (a) d-axis
voltage, (b) frequency.

Fig. 15. Three phase current variation subjected to a load change (a) DG 1, (b)
DG 2, (c) DG 3.

Due to the resistive load change, active and reactive power


demand increases in the system. New active power
Fig. 17. Angle variation subjected to load change.
requirement is compensated by all the DGs by equally

1949-3053 (c) 2018 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TSG.2018.2871413, IEEE
Transactions on Smart Grid
10

D. Generator Disconnection Since, the power generation of DG3 was disconnected after
Behavior of the optimal trajectories subjected to a generator 0.05s, the total active and reactive power demands shared by
disconnection is explored in this test case. At t = 0.05s, DG3 the other two DGs as illustrated in Fig. 18. Steady state active
was disconnected from the system. The new steady state power productions of both DGs increase to 899.7 W. Reactive
desired operating points after the generator disconnection are powers of DG1 and DG2 increase to 593 Var and 530.7 Var.
given in the Table VI. The results are shown in Fig. 18 and The d-axis voltage and frequency variations after the generator
Fig. 19. disconnection is shown in Fig. 19. Here both the d-axis
voltages and frequencies decrease because of the high power
TABLE VI
STEADY STATE DESIRED OPERATING POINTS AFTER THE GENERATOR
productions compared to the previous case. Steady state values
DISCONNECTION of the d-axis voltages are observed as 154.8 V and 154.9 V
while the frequencies of both the DG are 376.2 rads/s. The
𝑖̅𝑜𝑑𝑖 (A) 𝑖̅𝑜𝑞𝑖 (A) 𝛿𝑖̅ (rads) 𝑣̅𝑜𝑑𝑖 (V) 𝜔
̅ 𝑖 (rads/s) angle variation is similar to the previous case and the DG2
DG 1 3.8702 -2.5385 _ 154.7972 376.1464 angle converges to -0.0004 rads/s.
DG 2 3.8684 -2.2997 -0.0004 154.8690 376.1464
E. Sensitivity to state gains
In order to see the sensitivity of the optimal trajectories to
the gain values, different two point BVP sets were obtained
for different state gain matrices. For the sake of brevity, only
the sensitivity analysis of DG 1 d-axis current is given here.
However, similar variations can be observed in the other state
gains as well. The d-axis current gain of the DG 1 was
changed considering the steady state value of 2.278 A, while
keeping the other values constant and its trajectory variation is
shown in Fig. 20. Since this analysis was carried out to
observe the optimal trajectory variation of DG 1 d axis
current, trajectory variations of the other states are not given
here.

Fig. 18. Power variation subjected to generator disconnection (a) Active


power, (b) Reactive power.

Fig. 20. Variation of the DG 1 d axis optimal current trajectory with different
state gains.

According to the results, a significant improvement can be


observed in the optimal trajectory with large gains. With
larger gains, low rise time and high slew rate can be obtained.
According to the results, it can be inferred that the percentage
increment in the slew rate is significantly increased with the
large gains. Percentage overshoots are reduced with large
gains and the reduction can be seen from Fig. 20. Settling time
is also significantly reduced with high gains. Higher gain
implies high priority of optimizing the corresponding state or
control signal. Consequently, it achieves better transient
response than the other signals in the system.

VII. CONCLUSION
In this paper, a decentralized local optimal control frame
Fig. 19. Voltage and frequency variation subjected to generator disconnection work is proposed to obtain the optimal transient response of
(a) d-axis voltage, (b) frequency.

1949-3053 (c) 2018 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TSG.2018.2871413, IEEE
Transactions on Smart Grid
11

parallel connected inverters in an islanded microgrid. Parallel [15] K. Yu, Q. Ai, S. Wang, J. Ni, and T. Lv, "Analysis and Optimization of
Droop Controller for Microgrid System Based on Small-Signal Dynamic
connected inverters were defined as local subsystems in the Model," IEEE Trans. Smart Grid., vol. 7, no. 2, pp. 695-705, 2016.
islanded microgrid system. The dynamic model of the system [16] M. A. Hassan and M. A. Abido, "Optimal Design of Microgrids in
and the individual distributed generator (DG) objective Autonomous and Grid-Connected Modes Using Particle Swarm
Optimization," IEEE Trans. Power Electron., vol. 26, no. 3, pp. 755-
functions were modeled in the d-q reference frame. The
769, 2011.
Pontryagin’s minimum principle was employed to obtain the [17] I. Y. Chung, W. Liu, D. A. Cartes, E. G. Collins, and S. I. Moon,
optimal transient trajectories. Stability of the local optimal "Control Methods of Inverter-Interfaced Distributed Generators in a
controllers were assessed through a Lyapunov analysis. Microgrid System," IEEE Trans. Ind. Appl., vol. 46, no. 3, pp. 1078-
1088, 2010.
Simulations were carried out to investigate the performance of [18] H. P. Geering, "Optimal control with engineering applications," Berlin
the proposed concept. Startup optimal transient trajectory Heidelberg, 2007.
generation, variation of the optimal transient trajectories with [19] A. P. Sage and C. C. White, Optimum systems control. Prentice-Hall
Englewood Cliffs, NJ, 1977.
a load disturbance and generator disconnection, and state [20] N. Kim, S. Cha, and H. Peng, "Optimal Control of Hybrid Electric
sensitivity to objective gains were simulated considering 3 DG Vehicles Based on Pontryagin's Minimum Principle," IEEE Trans.
and 5 DG test systems. Simulation results demonstrated the Control Syst. Technol., vol. 19, no. 5, pp. 1279-1287, 2011.
[21] L. Serrao, S. Onori, and G. Rizzoni, "ECMS as a realization of
applicability and the effectiveness of the proposed Pontryagin's minimum principle for HEV control," in Proc. Amer.
methodology. Further, the development of a fully autonomous Control Conf., 2009, pp. 3964-3969.
optimal feedback controller is an interesting future direction of [22] G. Rousseau, D. Sinoquet, and P. Rouchon, "Constrained optimization
of energy management for a mild-hybrid vehicle," Oil & Gas Science
this work. and Technology-Revue de l'IFP, vol. 62, no. 4, pp. 623-634, 2007.
[23] S. Xiao, W. Lik-Kin, L. Yim-Shu, and X. Dehong, "Design and analysis
VIII. REFERENCES of an optimal controller for parallel multi-inverter systems," IEEE
[1] N. Hatziargyriou, H. Asano, R. Iravani, and C. Marnay, "Microgrids," Trans. Circuits Syst. II, Exp. Briefs., vol. 53, no. 1, pp. 56-61, 2006.
IEEE Power Energy Mag., vol. 5, no. 4, pp. 78-94, 2007. [24] H. Dagdougui, A. Ouammi, and R. Sacile, "Optimal Control of a
[2] J. M. Guerrero, M. Chandorkar, T. L. Lee, and P. C. Loh, "Advanced Network of Power Microgrids Using the Pontryagin's Minimum
Control Architectures for Intelligent Microgrids-Part I: Decentralized Principle," IEEE Trans. Control Syst. Technol., vol. 22, no. 5, pp. 1942-
and Hierarchical Control," IEEE Trans. Ind. Electron., vol. 60, no. 4, pp. 1948, 2014.
1254-1262, 2013. [25] H. Nazaripouya and S. Mehraeen, "Modeling and Nonlinear Optimal
[3] A. Bidram and A. Davoudi, "Hierarchical structure of microgrids control Control of Weak/Islanded Grids Using FACTS Device in a Game
system," IEEE Trans. Smart Grid., vol. 3, no. 4, pp. 1963-1976, 2012. Theoretic Approach," IEEE Trans. Control Syst. Technol., vol. 24, no. 1,
[4] F. Katiraei, R. Iravani, N. Hatziargyriou, and A. Dimeas, "Microgrids pp. 158-171, 2016.
management," IEEE Power Energy Mag., vol. 6, no. 3, pp. 54-65, 2008. [26] A. Yazdani and R. Iravani, Voltage-sourced converters in power
[5] Z. Wang, W. Wu, and B. Zhang, "A Distributed Quasi-Newton Method systems: modeling, control, and applications. John Wiley & Sons, 2010.
for Droop-Free Primary Frequency Control in Autonomous Microgrids," [27] H. K. Khalil, Noninear Systems. Prentice-Hall, New Jersey, 1996.
IEEE Trans. Smart Grid., 2016. [28] M. Rasheduzzaman, J. A. Mueller, and J. W. Kimball, "An Accurate
[6] C. Huang, F. Li, and Z. Jin, "Maximum Power Point Tracking Strategy Small-Signal Model of Inverter- Dominated Islanded Microgrids Using
for Large-Scale Wind Generation Systems Considering Wind Turbine dq Reference Frame," IEEE J. Emerging Sel. Topics Power Electron.,
Dynamics," IEEE Trans. Ind. Electron., vol. 62, no. 4, pp. 2530-2539, vol. 2, no. 4, pp. 1070-1080, 2014.
2015. [29] N. Pogaku, M. Prodanovic, and T. C. Green, "Modeling, Analysis and
[7] N. C. Ekneligoda and W. W. Weaver, "Game-Theoretic Cold-Start Testing of Autonomous Operation of an Inverter-Based Microgrid,"
Transient Optimization in DC Microgrids," IEEE Trans. Ind. Electron., IEEE Trans. Power Electron., vol. 22, no. 2, pp. 613-625, 2007.
vol. 61, no. 12, pp. 6681-6690, 2014. [30] P. Krause, O. Wasynczuk, S. D. Sudhoff, and S. Pekarek, Analysis of
[8] T. Vandoorn, J. Guerrero, J. De Kooning, J. Vásquez, and L. electric machinery and drive systems. John Wiley & Sons, 2013.
Vandevelde, "Decentralized and centralized control of islanded [31] F. Mumtaz, M. H. Syed, M. A. Hosani, and H. H. Zeineldin, "A Novel
microgrids including reserve management," IEEE Ind. Electron. Mag., Approach to Solve Power Flow for Islanded Microgrids Using Modified
2013. Newton Raphson With Droop Control of DG," IEEE Trans. Sustain.
[9] S. Kazemlou and S. Mehraeen, "Decentralized discrete-time adaptive Energy., vol. 7, no. 2, pp. 493-503, 2016.
neural network control of interconnected DC distribution system," IEEE [32] M. M. A. Abdelaziz, H. E. Farag, E. F. El-Saadany, and Y. A. R. I.
Trans. Smart Grid., vol. 5, no. 5, pp. 2496-2507, 2014. Mohamed, "A Novel and Generalized Three-Phase Power Flow
[10] X. Wu, C. Shen, and R. Iravani, "Feasible range and optimal value of the Algorithm for Islanded Microgrids Using a Newton Trust Region
virtual impedance for droop-based control of microgrids," IEEE Trans. Method," IEEE Trans. Power Syst., vol. 28, no. 1, pp. 190-201, 2013.
Smart Grid., vol. 8, no. 3, pp. 1242-1251, 2017. [33] D. E. Kirk, Optimal control theory: an introduction. Courier
[11] J. M. Guerrero, L. G. d. Vicuna, J. Matas, M. Castilla, and J. Miret, "A Corporation, 2012.
wireless controller to enhance dynamic performance of parallel inverters [34] F. L. Lewis and D. Liu, Reinforcement learning and approximate
in distributed generation systems," IEEE Trans. Power Electron., vol. dynamic programming for feedback control. John Wiley & Sons, 2013.
19, no. 5, pp. 1205-1213, 2004. [35] A. Saberi, "On optimality of decentralized control for a class of
[12] Y. A. R. I. Mohamed and E. F. El-Saadany, "Adaptive Decentralized nonlinear interconnected systems," Automatica, vol. 24, no. 1, pp. 101-
Droop Controller to Preserve Power Sharing Stability of Paralleled 104, 1988.
Inverters in Distributed Generation Microgrids," IEEE Trans. Power [36] L. F. Shampine, J. Kierzenka, and M. W. Reichelt, "Solving boundary
Electron., vol. 23, no. 6, pp. 2806-2816, 2008. value problems for ordinary differential equations in MATLAB with
[13] V. Calderaro, G. Conio, V. Galdi, G. Massa, and A. Piccolo, "Optimal bvp4c," Tutorial notes, pp. 437-448, 2000.
Decentralized Voltage Control for Distribution Systems With Inverter-
Based Distributed Generators," IEEE Trans. Power Syst., vol. 29, no. 1, Anushka M. Dissanayake (S’18) received his B.S.
pp. 230-241, 2014. degree in electrical and electronic engineering from
[14] R. B. Godoy, J. O. P. Pinto, C. A. Canesin, E. A. A. Coelho, and A. M. the Faculty of Engineering, University of Peradeniya,
A. C. Pinto, "Differential-Evolution-Based Optimization of the Dynamic Peradeniya, Sri Lanka in 2014.
Response for Parallel Operation of Inverters With No Controller He is currently pursuing the Ph.D. degree in
Interconnection," IEEE Trans. Ind. Electron., vol. 59, no. 7, pp. 2859- electrical engineering at the School of Electrical and
2866, 2012. Computer Engineering, Oklahoma State University,
under the supervision of Dr. Nishantha C.
Ekneligoda. His research interests include optimal

1949-3053 (c) 2018 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TSG.2018.2871413, IEEE
Transactions on Smart Grid
12

control, distributed and adaptive controls in small scale power systems and
power electronics.

Nishantha C. Ekneligoda (S’08, M’12) received his


BS and MS degree in electrical engineering from the
University of Moratuwa Sri Lanka in 2003 and 2005.
He also has completed a MSME degree in
mechanical engineering at Southern Methodist
University, Dallas, TX in 2008.
Then, he obtained his PhD at Michigan
Technological University, in 2012. He is currently an
Assistant Professor in the department of Electrical
and Computer Engineering at Oklahoma State
University. His research interests include power
system controls, power electronics, non-linear, optimal and distributed
controls and renewable energy systems.

1949-3053 (c) 2018 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.

You might also like