Electric Vehicle Charging Time Series Clustering

Time series analysis for ACN-Data

The keyword of my this line of projects is for data-driven design of adaptive EV charging networks (ACNs), motivated by answering the following questions:

  1. As more and more EV charging data collected at charging facilities, how can we take advantage of it to better predict, detect, optimize and control?

  2. Can we develop new artificial intelligence techniques and fundamental results by considering the EV charging problem as a rising application domain for ML?

ACN-Data, released publicly in the paper above, contains both session data (user’s ID, arrival time, departure time, requested energy, and actual energy delivered) and fine-grained charging data at seconds resolution (time series of control signals and charging currents). More advanced features of the dataset can be found in here.

To the best of our knowledge, ACN-Data is among the first several dynamic large-scale fine-grained charging datasets that are publicly available. We demonstrate the first analysis of the charging curves in ACN-Data and develop a systematic method to learn battery behavior from the data. Even though the number of charging curves is large, they can be classified into a small number of types (in our experiment, \(304\) charging curves are classified into \(6\) types). These battery types can be used to predict charging behavior and determine charging stage (either in the absorption stage or not) with good accuracy. The following figure exemplifies charging tail clustering results for different distance functions , with the number of clusters fixed as \(6\).

Charging tail clustering with different distance metrics.

For the hierarchical relationship between users and clusters, we plot a two-dimensional visualization (using t-SNE) of our clustering results with \(K=6\) clusters. Tails for different users are colored differently. The clusters’ colors are consistent with those used in the previous figure. The marginal probabilities \(p_1,\ldots,p_6\) represent the portions of charging sessions falling into the \(6\) clusters.

Groups of users with respect to clustered charing tails.

The related results can be found in the following two papers.


Sun, Chenxi, Tongxin Li, Steven H. Low, and Victor OK Li. “Classification of electric vehicle charging time series with selective clustering.” Electric Power Systems Research 189 (2020): 106695.

Lee, Zachary J., Tongxin Li, and Steven H. Low. “ACN-Data: Analysis and applications of an open EV charging dataset.” In Proceedings of the Tenth ACM International Conference on Future Energy Systems, pp. 139-149. 2019.