A user selection algorithm for aggregating electric vehicle demands based on a multi‐armed bandit approach

Abstract

In systems with high penetration of renewables, demand side resources have been aggregated to facilitate system operation. However, the natural uncertainty and randomness of users' behaviour may deteriorate the demand aggregation performance, including a large mismatch from the expected aggregation target and unnecessary cost while executing aggregation. Here, the most fast‐growing demand side resource, electric vehicle is targeted, and an algorithm based on a multi‐armed bandit approach is proposed to aggregate those electric vehicle demands. In the proposed multi‐armed bandit model, each electric vehicle user's behaviour is viewed as two arms. Then, a combinatorial upper confidence bound mixed sorting algorithm, which selects the optimal set of users participating in demand aggregation, is developed. The case studies show that the proposed method can reduce the demand aggregation mismatch and eliminate the unnecessary cost. Additionally, it can be observed that the user experience is also improved.

Managing Considerable Distributed Resources for Demand Response: A Resource Selection Strategy Based on Contextual Bandit

“…Multi-armed bandit (MAB) approaches were applied to learn the behaviors of residential air conditioning [21]; heating, ventilation, and air conditioning (HVAC) [22]; renewable energy sources [23]; and energy storage [24], selecting the optimal set to participate in primary or secondary frequency regulation. Moreover, the optimal set of electric vehicles [25] and residential demand resources [26,27] were selected based on MAB-based approaches to provide demand response services to the power grid operation. However, the aforementioned approaches mainly focus on the uncertainty pertaining to the participation of a singular resource type, and the selected resources are aggregated to ensure the cumulative capacity adequately satisfies the predetermined demand.…”

“…Resource Type Service Approach [21] Residential air conditioning Primary and secondary frequency regulation Risk-averse MAB [22] Heating, ventilation, and air conditioning (HVAC) Secondary frequency regulation Risk-averse MAB [23] Renewable Energy Sources Secondary frequency regulation MAB [24] Energy storage Primary frequency regulation MAB [25] Electric vehicle Ancillary services MAB [26] Residential demand Demand response MAB [27] Residential demand Demand response Contextual MAB This paper Diesel generator, Gas turbine, Curtailable load Load balance, Demand response Contextual MAB Stochastic programming and robust optimization are two approaches typically used to address uncertainty in an operation. Stochastic programming can be further divided into chance-constrained programming (CCP) [28] and scenario-based approaches [29].…”