Voltage, Var and Watt Control in a Distribution System with High PV Penetration Using NSGA-III Kamel Alboaouh § , Salman Mohagheghi §§ , Senior Member, IEEE, and Ronald G. Harley  , Fellow, IEEE § Electrical Engineering Department, Colorado School of Mines, Golden, CO 80401 USA. Email: alboaouh@mymail.mines.edu §§ Electrical Engineering Department, Colorado School of Mines, Golden, CO 80401 USA. Email: smohaghe@mines.edu  School of Electrical and Computer Engineering, Georgia Institute of Technology, Atlanta, GA 30332 USA Abstract— In an earlier work, the authors showed that a central coordination of PV panels as well as reactive power support devices is an effective solution to facilitate PV integration in distribution system. The problem was formulated in the form of a multi-objective function and was solved using evolutionary algorithm (EA). The objective functions considered were the system losses, operational variations of voltage regulators (VRs) and switching capacitors (SCs), as well as curtailment of the PV active power. In this work, instead of using the classical EA, the Non-dominated Sorting Genetic Algorithm III (NSGA-III) is employed in order to ensure Pareto optimality of the solution. A deterministic simulation over a one-day period is conducted using the modified version of the IEEE 123-bus test feeder. The objective is to assess the centralized control on the distribution grid considering high penetration of PVs. Index Terms—Distribution system, NSGA-III, rooftop photovoltaic, reactive power control, and voltage var and watt optimization. I. INTRODUCTION Photovoltaic generation is increasingly being used at the distribution level due to its positive impact on the environment and global warming [1] as well as being able to provide ancillary services to the main grid “upstream” [2]. However, at high penetration levels, problems may emerge that could interfere with the operation of the grid. For instance, the distribution feeders may experience instances of overvoltage or voltage flicker [3], the operation of the protective equipment may be affected due to the bi-directional flow of power [4], and more operational stress may be placed on switching components such as VRs [5]. Therefore, an optimal strategy to operate and control the distribution network in the presence of high PV penetration levels seems essential. Many researchers have proposed reactive power control strategies in order to maintain the network variables within acceptable operating limits [6], [7]. PV panels can participate in reactive power support by using the interfaced inverters [8], [9]. As such, it has been shown that instead of merely trying to maintain the system variables within acceptable limits, it is possible to enhance and/or optimize the system performance by properly utilizing the existing reactive power resources to regulate the voltage profile and reduce system losses [10-15]. This is often referred to in the literature as volt/var optimization (VVO), or simply voltage/var control (VVC). It is possible to attain VVO at the design stage by reconfiguring the distribution system and/or optimal allocation of capacitor banks [16]. However, it is certainly less expensive to optimize the system performance by utilizing the existing reactive power resources [14], such as the on-load tap changing transformer (OLTC), voltage regulators (VRs), switching capacitors (SCs), or inverter-interfaced distributed generators [17]. Since PVs can absorb or inject the available reactive power quickly, they can contribute greatly to achieving this feat because they respond significantly faster than VRs, SCs and OLTCs. The fast response property of PVs becomes more attractive at higher penetration levels because it dominates other voltage control devices [18]. The contribution of PVs in reactive power support has also been acknowledged in recent standards, most notably the latest revisions of IEEE 1547, which allows PVs to participate in voltage regulation by changing their active and reactive powers to keep the voltage within ±5% [19]. In the literature, optimal coordination among electrical devices in a distribution network has been achieved either in a centralized or a decentralized fashion. The decentralized coordination uses local measurements to find the optimum settings of the electrical devices, whereas the centralized approach relies on a communication network to allow for data collection and transmission [17], [20]. Although the decentralized approach is simpler, it allows all devices to operate autonomously in a non-coordinated manner which may not guarantee the voltages to stay within the acceptable limits [21] and may lead to operational stresses on OLTC and VRs [5]. Because of this, a centralized approach has been adopted in this work. At the distribution level, most available central control approaches either use VVO or only optimize the active power dispatch [10-13], [15], often with the goal of minimizing overall system losses. Although in [14], both active and reactive