Smart Charging of Plug-in Electric Vehicles in Distribution Systems Considering Uncertainties

Abstract

Distribution feeders and equipment are designed to serve peak loads, and in the absence of Plug-in Electric Vehicle (PEV) loads, day-ahead dispatch of feeders is typically performed by optimizing feeder controls for forecasted load profiles. However, due to climate change concerns, the market share of PEVs is expected to increase, and consequently, utilities expect an increase in demand due to these loads charging from the grid. Uncontrolled charging of PEVs may lead to new peaks in distribution feeders, which would require expensive infrastructure and equipment upgrades. Furthermore, PEV loads will represent new sources of uncertainty, temporal and spatial, which will pose a challenge for the centralized control and optimal operation of the grid. In practice, these uncertainties arise as a result of variability in factors such as the number of PEVs connected to the grid for charging, the arrival and departure times of PEVs, and the initial battery State-of-Charge (SoC). Hence, the integration of PEVs into the existing distribution system, without significant infrastructure upgrades, will be possible only through smart charging of these loads, while properly accounting for these uncertainties. The elasticity of PEVs provides a level of flexibility that can be used by utilities or Local Distribution Companies (LDC) to ensure efficient feeder operation, while providing fair and efficient charging to PEV customers.

This thesis presents a novel two-step approach for the fair charging of PEVs in a primary distribution feeder, accounting for the uncertainty associated with PEVs, considering the perspectives of both the LDC and the PEV customer. In the first step of the proposed approach, the mean daily feeder peak demand and corresponding hourly feeder control schedules, such as taps and switched capacitor setpoints, are determined hourly, while minimizing the daily peak demand, taking the existence of PEVs into account. As an alternative to the conventional Monte Carlo Simulations (MCS), a nonparametric Bootstrap technique is used in conjunction with a Genetic Algorithm (GA)-based optimization model, to account for variations in the arrival and departure times, and the initial battery SoC of PEVs, at each node. In the second step, the maximum possible power that can be given to the charging PEVs at each node, while maintaining the peak demand value and corresponding feeder dispatch schedules defined in the first step, is computed every few minutes and shared fairly among the PEVs.

The proposed technique is validated using the IEEE 13-bus test feeder as well as the distribution feeder model of an actual primary feeder in Ontario, considering significant PEV penetration levels. The potential gain in PEV charging efficiency is quantified for the proposed Bootstrap feeder control schedule with respect to the base schedule (without PEVs). The presented optimization approach is also compared with the current industry practice in Ontario, and a sensitivity-based heuristic technique, demonstrating the advantages and feasibility of the presented technique. The results show that the proposed approach could be implemented in practice due to its reasonable computational burden, and its ability to charge PEV loads better than the current industry practice, or a popular heuristic method, while satisfying feeder and peak demand constraints.