I. Introduction
DUE to global warming gas emissions reduction and different environmental issues, distributed generation (DG) resources such as photovoltaic arrays (PVs), wind, fuel cells and biomass are being rapidly used in the electric power utility [1]. Multiple DGs are connected with different local loads such as constant impedance loads (CILs) and constant power loads (CPLs) at the distribution level to perform a microgrid [2]–[4]. Microgrid can operate either in island mode or in grid-connected mode [2]. In the grid-connected mode, a synchronization unit is needed to synchronize the phase of the load voltage to a grid voltage, while in the autonomous mode; it is used to generate an angle with the desired frequency [3]. Additionally, the DG inverter is controlled to provide the load with the predefined voltage and frequency values [4]. Mostly power, current and voltage controllers are implemented to control the output power, current and voltage of the DGs inverters’ respectively. Droop control scheme is usually used for sharing the power between the DG units in the autonomous microgrid [4], [5]. Depending on the droop operation, the magnitude and the frequency of the output voltage are varied. With any constant load perturbation, the microgrid stability will be affected by the low inertia of the inverter-based DG units then generate frequency deviations [5]. Centralized and decentralized control schemes are used to improve and enhance the dynamic performance of the microgrids [6] –[9]. In the decentralized schemes, new DG units can be integrated without changing the controller settings continuously; however, this type of controller cannot manage operations with high levels of coordination [6], [7]. On the other hand, the system optimization can be done using the centralized control schemes; but the desirable plug-and-play feature cannot be revealed [8]. The advantages and disadvantages of both schemes were summarized in [9]. The microgrid stability is significantly affected by the dynamics of the loads [4]. The effect of passive load dynamics was reported in [10]. It was reported that the power sharing controller parameters and load demand affect dominantly the low frequency modes while the inner voltage and current controller parameters, filter components as well as load dynamics have more effects on the damped medium and high frequency modes [12]. Therefore, the controller parameters and power sharing parameters should be adapted to enhance the dynamic performance of the autonomous microgrid especially when CILs are included [4]. Previous studies show that fixed-gain PI controllers cannot easily acclimate to load changes and disturbances even with parameters variation especially in large microgrids therefore, continuous tuning process is required to adjust the controller gains to overcome these problems [13]. With significant drawbacks such as falling to obtain the optimal settings and time-consuming, different trial and error approaches has been reported in the literature [6]– [13]. Recently, in addition to classical approaches, computational intelligence algorithms such artificial neural networks, fuzzy logic and particle swarm optimization (PSO) have been applied to solve many power system problems with remarkable success [14]. Although, most of these algorithms increase the control system complexity, researchers have used these algorithms to improve the transient performance of the microgrid. As one of the promising optimization technique, PSO has been widely implemented since it has many advantages such as robustness, simplicity, computational efficiency and enhancing the global and local exploration abilities [15], [16]. It is worth mentioning that PSO is used as an efficient tool for optimization that gives a balance between local and global search techniques.