UCLA

The modern world has witnessed the surge of electric vehicles (EVs) driven by government policy worldwide to reduce transportation’s dependence on fossil fuels. According to (Slowik, 2019), the global EV market has grown sharply with the annual light-duty EV sales surpassing 2 million in 2018, which is about a 70% increase from 2017. The increase in EV population implies the rise in energy demand, and that introduces new challenges to the electricity sector. EV charging load demand in high penetration scenarios, which is foreseen, may lead to stability and quality issues in power grids. Generation capacity and the electricity infrastructure upgrade may be required to address those issues; however, it increases generation costs significantly. The most common EV chargers installed today deliver around 7 kW of power, which is over four times that of an average household power consumption in the US. EV charging load often shows two peaks in a day, one in the morning when people plug in the EV at the workplace and the other in the evening when people get home from work. Without proper energy management for EV charging, the vast power demand due to a large number of plugged-in EVs can stress the electric grid, degrade the electric power quality, and impact the wholesale electricity market. Although an EV battery may store energy up to 80 kWh, which requires more than 10 hours to charge at 7kW from empty, we found that most EVs need only 12 kWh per charge or 1.7 hours at 7 kW to meet daily commute requirement while they stay in the parking garage for a more extended period. This implies that EVs can have considerable time-flexibility for charging, and it is not necessary to start charging

right after plugging in, which is likely to result in the charging power add-up. A proper EV charging schedule can well allocate the charging load to prevent power peaks. Therefore, EV charging scheduling can play a significant role in mitigating the adverse effects of vast EV charging demand without upgrading the power grid capacity.

To optimize the EV charging schedule while satisfies EVs’ charging demand, each EV’s stay duration and energy need are essential parameters for the optimization. Those parameters are based on predictions to minimize human intervention. Nonetheless, the uncertainty of EV user behavior poses a challenge to the prediction accuracy. Therefore, this dissertation demonstrates an ensemble machine learning-based method to model and predict the EV loads accurately, thereby improving the performance of EV charging scheduling.

On the other hand, this smart EV-grid integration, which requires massive communication, including collecting, transmitting, and distributing real-time data within the network, makes it more susceptible to cyber-physical threats. Potential breaches could not only affect grid operation but also reduce consumers’ willingness to adopting EVs over conventional fuel-powered vehicles. This dissertation also presents the vulnerability analysis and risk assessment for a smart EV charging system to develop the countermeasures to secure the network. Also, while it is inevitable that the security has flaws, this dissertation provides a novel anomaly detection approach based on the invariant correlations of different measurements within the EV charging network.