Abstract:
This study employs the K-means clustering algorithm, implemented in MATLAB, to analyze a real-time dataset from public electric vehicle (EV) charging stations with the goal of uncovering meaningful usage patterns. By identifying and grouping data points into four distinct clusters based on their proximity to calculated centroids, the research reveals key behavioral trends, particularly focusing on the distribution of charging events and the identification of peak usage hours associated with each cluster. This clustering approach allows for a nuanced understanding of how and when charging infrastructure is most heavily utilized. By determining the number of data points within each cluster, the study provides quantitative support for evaluating demand intensity and user behavior across different time frames. The insights derived from this analysis are instrumental in informing strategic decisions regarding the placement, scheduling, and scaling of EV charging stations. Ultimately, the research contributes to improving user convenience, reducing wait times, and promoting the efficient, sustainable growth of electric vehicle infrastructure in response to evolving demand patterns.
Aim:
This study aims to classify electric vehicle (EV) charging behavior into four distinct patterns—Convenient, Gradual, Anxious, and Urgent—using the K-means clustering algorithm applied to real-time public EV charging data. Each pattern represents a unique user charging behavior inferred from features such as start time, charging duration, energy consumption, and frequency. Convenient users typically charge during off-peak hours with longer, relaxed sessions, suggesting they plan their charging proactively. Gradual users show consistent but moderate usage patterns, indicating habitual charging aligned with daily routines. Anxious users tend to initiate charging at higher battery levels, possibly due to range anxiety, while Urgent users often plug in at very low charge levels, seeking immediate energy replenishment. By identifying and grouping these behaviors through clustering around centroids in the dataset, the analysis provides valuable insights into user profiles. These insights are critical for designing user-centric charging infrastructure, optimizing station scheduling, and supporting policy-making for more efficient and adaptive EV ecosystem development.
Objective:
This study begins by collecting comprehensive data from various EV charging stations, focusing on key session parameters such as charging time, duration, energy consumed, and frequency of use. To streamline the analysis and emphasize the most influential variables, the Pareto principle is applied, helping to identify and prioritize the critical 20% of factors that drive 80% of the charging behavior outcomes. Building on this refined dataset, the study determines the optimal number of clusters based on user charging patterns, with the objective of uncovering deeper insights into the diverse factors that influence charging decisions. The K-means clustering algorithm is then implemented using the Minkowski embedded distance metric to enhance accuracy and flexibility in pattern recognition. This approach allows for a more nuanced grouping of users and aids in calculating the proportion of consistent behaviors within each cluster. Additionally, the study analyzes the peak-hour density within each individual cluster to understand when different user types are most active. These findings offer valuable implications for demand forecasting, charging station deployment, and the development of smarter, more user-responsive EV infrastructure.
