Smart Charging Even Smarter
Using Predictions to Increase Performance of a Cloud-Based Smart Charging Algorithm
More Info
expand_more
Abstract
Electric Vehicles (EVs) are considered a promising solution when it comes to reducing CO2 emissions in the transport sector. To accommodate for large numbers of EVs, a considerable amount of additional charging infrastructure must be installed. Simultaneous charging of large quantities of EVs leads to peak loads in the energy grid. One way to overcome this is through smart charging. With smart charging the time of charging is postponed to a less busy moment, and the charging powers of cars are mutually balanced.
In this report, a problem around a smart charging algorithm is presented. Information on session duration, required energy and EV parameters is not available. The current algorithm handles this lack of information perfectly but there might be room for improvement. Considering practical limitations, smart charging is currently implemented based on a round-robin scheduling algorithm. This is an efficient and fair way to distribute limited resources. Additional information is available in the form of predictions, which are subject to uncertainty. In this study we aim to improve the existing smart charging algorithm by using these predictions.
Based on historical charging session data, an analysis will be made of the charging behaviour of different user types. Charging infrastructure is used by a mix of user types, each type having their own characteristics in terms of connection time and energy demands. This will serve as the basis for development of a simulation environment which will be used later to assess the performance of different smart charging algorithms.
A new performance metric is introduced that describes the amount of inconvenience that users experience as a result of smart charging. In addition, a simulation-based optimization method is introduced that, using a posteriori information, minimizes total inconvenience. Next, three heuristics are proposed that approximate the global optimum using the available predictions. Simulations show that a relative improvement of 81.6% is achievable with the current accuracy level of the predictions. Moreover, these heuristics can deliver more energy from the same grid connection than the existing algorithm. This leads to more efficient use of grid connections, higher revenues for the Charge Point Operator (CPO) and higher user satisfaction.
Files
File under embargo until 18-09-2024