Variable Rate PEV Charging Using Machine Learning and Optimization Techniques
Limited Fossil fuels and regulations to improve air quality are creating unprecedented changes in transportation, influencing customers to purchase zero-emission vehicles (ZEVs) instead of traditional internal combustion engine vehicles. California's Governor Brown issued Executive Order B-16-2012 to lower the greenhouse gas emissions due to transportation and promote ZEVs in California, setting a goal of reaching 1.5 million ZEVs in California by 2025. ZEVs are classified as plug-in electric vehicles (PEVs), battery electric vehicles (BEVs), and fuel cell electric vehicles (FCEVs). However, PEVs and BEV's draw power from the grid, and as these vehicles become more prevalent, multiple studies have demonstrated that they will increase the electric load and overwhelm the grid.
In this study, three new protocols, Machine Learning Valley-Filling (MLVF), Selected Rate Valley-Filling (SRVF), and Rate and Interval Valley-Filling (RIVF), are presented to enhance PEV charging at the local power level while minimizing the effects of uncontrolled plug-in vehicle (PEV) charging. The goal is to create algorithms that are computationally fast, operate in real-time, scalable, satisfy customer needs, reduce transformer loss of life, and contribute to smart transportation.
SRVF and RIVF are proposed to investigate how variable rate charging can affect PEV charging profiles. SRVF evaluates many charging options and determines the best charging rate for a vehicle, while RIVF utilizes many rates to create a charging profile for a vehicle. Both algorithms can charge vehicles in sections by selecting intervals to charge in. SRVF is a simpler algorithm than RIVF, and an evaluation will be conducted to determine if RIVF’s additional advantages are necessary by quantitively showing the difference in the results. Machine Learning Valley-Filling (MLVF), a neural network and is trained on processed data that determine the best interval to begin charging a vehicle at a constant rate, given its charge needs and dwell time. MLVF is used to investigate if machine learning can be used in valley-filling applications, as it has not yet been demonstrated. The purpose of this study is to investigate if a neural network algorithm can learn to identify when to begin charging a PEV by distinguishing low and high demand sections in the forecasted baseload.
The results demonstrate that SRVF and RIVF have the ability to reduce the absolute maximum peak power reached amongst all transformers, average maximum peak power reached peak power by each transformer, and average load during charging caused by uncontrolled charging up to 51.72%, 25.14%, and 88.51%, respectively. In addition, this study indicates that a neural network algorithm can indeed identify low and high demand sections in the forecasted baseload and learn when to begin charging electric vehicles to decrease demand loads. MLVF achieves a Micro-Averaged F1-Score, a classifier's overall accuracy, of 92.84% of selecting the correct timeslot to initiate charging.