Benchmarking Aggregation-Disaggregation Pipelines for Smart Charging of Electric Vehicles

TL;DR

The paper addresses the challenge of integrating millions of EVs into energy system models by benchmarking end-to-end aggregation-disaggregation pipelines. It introduces a neutral, open-source benchmark and reimplements four core methods—Representative Profile (REP), Virtual Battery (VB), Flex Objects (FO), and Dependency-based FlexOffers (DFO)—across multiple driving and price-signal scenarios, comparing each against the true optimal (unaggregated) and uncontrolled baselines. Key findings show that no method dominates across all metrics: REP often yields the smallest cost gaps, FO and DFO provide lossless disaggregation with different runtime trade-offs, and VB benefits from enhancements like a piecewise SOC constraint (FPC) and grouping, though it generally trails REP in performance. The benchmark offers actionable guidance for method selection in both academic studies and real-world planning, and its open-source availability facilitates reproducibility and extension.

Abstract

As the global energy landscape shifts towards renewable energy and the electrification of the transport and heating sectors, national energy systems will include more controllable prosumers. Many future scenarios contain millions of such prosumers with individualistic behavior. This poses a problem for energy system modelers. Memory and runtime limitations often make it impossible to model each prosumer individually. In these cases, it is necessary to model the prosumers with representatives or in aggregated form. Existing literature offers various aggregation methods, each with strengths, drawbacks, and an inherent modeling error. It is difficult to evaluate which of these methods perform best. Each paper presenting a new aggregation method usually includes a performance evaluation. However, what is missing is a direct comparison on the same benchmark, preferably by a neutral third party that is not associated with any of the compared methods. This paper addresses this gap by introducing a benchmark to evaluate the end-to-end performance of multiple aggregation-disaggregation pipelines, specifically focusing on electric vehicles (EVs). Our study assesses the performance of the common representative profile (REP) approach, four different versions of the virtual battery (VB) approach, the Flex Object (FO) method, and the Dependency-based FlexOffer (DFO) method. The results show that each method has a clear use case. Depending on the price signal, additional median charging costs of 2%-50% are incurred using an aggregation method, compared to the optimal charging costs (i.e., the charging costs resulting from optimizing the EVs directly, without aggregation). The representative profile approach results in the lowest additional costs (2%-20%), while the FO and DFO methods allow for error-free disaggregation, which is advantageous in real-world use cases.

Figures (10)

• Figure 1: Aggregation of two EV charging events into a VB. First row, determine the power bounds of the VB. Second row, determine the energy bounds of the VB. Third row, determine arriving and departing energy.
• Figure 2: Optional additional piece-wise linear constraint for the VB. For illustration purposes, the right point is depicted as already determined.
• Figure 3: Visualization of a Flex Object. This figure is based on a visualization in siksnysAggregatingDisaggregatingFlexibility2015a.
• Figure 4: DFO aggregation. The first two rows are DFO that encode the flexibility of two individual charging events. The last row shows the aggregation of the two. The figure is based on a visualization in siksnysDependencybasedFlexOffersScalable2016.
• Figure 5: Examples for the price time series