The integration of Electric Vehicles (EVs) into the power grid represents a transformative opportunity for energy system optimization. With the increasing penetration of renewable energy sources, grid stability has become a critical challenge for system operators and flexibility a valuable feature to address it. Vehicle-to-Grid (V2G) technology and smart charging strategies have emerged as viable solutions to enhance grid reliability by leveraging the charging flexibility of EV fleets.
However, understanding the extent of this flexibility and its potential contributions to system services requires an in-depth analysis of real-world charging data. This study aims to assess the charging flexibility potential of different EV fleets by analyzing real-world charging behaviors and clustering users based on their charging patterns.
By categorizing EV users, we identify key factors that influence their ability to participate in demand response programs, grid balancing mechanisms, and energy market optimization. The research focuses on evaluating charging flexibility across various user clusters, determining their potential for participating in flexibility services.
The key contributions of this work include:
- A detailed clustering analysis of EV charging behaviors to classify users based on flexibility potential.
- A comprehensive framework for assessing EV time- and power-related flexibility of different EV user groups.
- A case study analysis highlighting the potential of integrating EV fleets into grid services.
METHODOLOGY
A. Data requirements
The development of a framework for estimating EV charging flexibility requires a comprehensive dataset combining public and private charging events. Public data provides broad behavioral and power-level variations, while private data enriches the analysis with residential profiles, together capturing key factors like peak demand and user constraints for reliable predictions.
Critical data elements include the Transaction ID for integrity, Start/Stop Timestamps for temporal patterns, Connected Time to identify idle periods for deferred charging, and Charging Time with Total Energy to determine rescheduling flexibility. Max Power reveals peak draw to inform demand response strategies.
Data preparation involves cleaning (addressing missing values, normalizing formats) and transforming data to derive metrics like Idle Time. This prepared data is used in clustering analysis to group sessions by characteristics such as peak power and charging time, revealing behavioral patterns and flexibility potential for optimizing grid services and EV integration.
B. Clustering Mechanisms
The clustering module aggregates cleaned EV charging data from all sites to identify flexibility patterns and demand response opportunities. Charging events are grouped based on shared attributes such as plug in time, charging duration, energy consumption, and maximum power. This reveals distinct behavioral profiles, for example short duration high power or long duration low power sessions, which inform tailored strategies like load shifting for sessions with extended idle periods.
The process involves key technical steps: feature selection of flexibility relevant attributes; data normalization using StandardScaler to ensure equal weighting; and rigorous algorithm evaluation. Multiple algorithms, including K-Means, Agglomerative Clustering, Gaussian Mixture Models (GMM), and DBSCAN, are optimized and compared using metrics like inertia and silhouette scores to identify the best performing model for forming distinct clusters.
Cluster analysis then profiles each group statistically, for example by analyzing means and variances, and visually, categorizing behaviors such as commuter or overnight charging. Flexibility potential is assessed through metrics like idle time ratios and responsiveness to signals, enabling strategies such as time-shift charging, peak shaving, and dynamic power adjustments that align with grid conditions and renewable availability.
C. Flexibility assessment methods
Our methodology considers EV charging flexibility in two main dimensions: time flexibility and power-specific flexibility. Time flexibility refers to the ability to shift the charging schedule without affecting the total energy delivered, while power-specific flexibility reflects the capacity to modulate charging power during a session. These flexibility aspects are quantified using two metrics.
Graphically, we represent these metrics using FlexBars. In a FlexBars, time is plotted on the horizontal axis-the full width representing the total duration of the charging event-while the vertical axis displays power up to the maximum available rate. The resulting rectangle indicates the total possible energy delivery, which typically exceeds the actual energy charged, thereby visually emphasizing the flexibility potential. By averaging FlexBars over all charging events within a defined period, we obtain a comprehensive view of the temporal and power-specific flexibility of a site or cluster of EV charging events.
This structured approach, combining data aggregation, statistical analysis, and graphical visualization, lays the foundation for robust clustering and flexibility assessments, ultimately supporting targeted strategies for load shifting or peak shaving.
CASE STUDY
This case study analyzes a dataset of 10,000 EV charging events from EVnetNL in the Netherlands, comprising 417,141 meter readings. This data provides detailed transaction and energy metrics, enabling a thorough analysis of charging behavior and flexibility potential.
Using K-means clustering on features such as Max Power, Charging Time, Connected Time, and Total Energy, five distinct EV charging user clusters were identified. These are visualized in the scatter matrix (Figure 1), revealing behavioral differences in session length, power use, and energy consumption. The temporal distribution of charging events (Figure 2) confirms that some clusters predominantly charge during daytime (public stations) while others peak overnight (residential charging).
Figure 1 - Scatter matrix of the charging events dataset divided by 5 clusters
Figure 2 - Probability of the charging events occur at a certain hour for each clusters
Clusters differ widely in terms of idle time, which indicates flexibility for deferred charging. As shown in Figure 3, Clusters 2 and 4 exhibit the highest idle times, making them strong candidates for demand-side management. In contrast, Clusters 0 and 3 have short idle periods, reflecting optimized fast charging with limited flexibility.
Figure 3 - Boxplot regarding Idle Time of each cluster
Table 1 summarizes the cluster characteristics and corresponding flexibility potential, while Figure 4 provides an example of FlexBars visualization for Cluster 4, highlighting available temporal and power-related flexibility.
Table 1- Cluster classification summary table
Figure 4 - Flexbars for cluster 4
The results show that household and fleet users with long connection times (Clusters 2 and 4) offer the greatest potential for smart charging and load shifting. Fast-charging users (Cluster 3) and opportunistic public charging users (Cluster 0) are already optimized with limited flexibility. Structured high-power users (Cluster 1) present moderate opportunities.
Overall, the findings highlight how differentiated charging behaviors can inform targeted demand-response programs and improve renewable integration into the grid.
