[Paper Review] Electric Vehicle Driver Clustering using Statistical Model and Machine Learning

Yingqi Xiong, Bin Wang+2 more|arXiv (Cornell University)|Feb 12, 2018

Electric Vehicles and Infrastructure10 references3 citations

TL;DR

This paper proposes a hybrid approach combining statistical modeling and machine learning to cluster electric vehicle (EV) drivers based on real-world charging data from UCLA and Santa Monica. By applying unsupervised clustering and a multilayer perceptron to historical records, the model enables accurate day-ahead prediction of EV parking duration and load, achieving strong performance in cross-validation for smart charging scheduling and load forecasting.

ABSTRACT

Electric Vehicle (EV) is playing a significant role in the distribution energy management systems since the power consumption level of the EVs is much higher than the other regular home appliances. The randomness of the EV driver behaviors make the optimal charging or discharging scheduling even more difficult due to the uncertain charging session parameters. To minimize the impact of behavioral uncertainties, it is critical to develop effective methods to predict EV load for smart EV energy management. Using the EV smart charging infrastructures on UCLA campus and city of Santa Monica as testbeds, we have collected real-world datasets of EV charging behaviors, based on which we proposed an EV user modeling technique which combines statistical analysis and machine learning approaches. Specifically, unsupervised clustering algorithm, and multilayer perceptron are applied to historical charging record to make the day-ahead EV parking and load prediction. Experimental results with cross-validation show that our model can achieve good performance for charging control scheduling and online EV load forecasting.

Motivation & Objective

• To address the challenge of unpredictable EV charging behaviors that complicate energy management in distribution systems.
• To develop a user modeling technique that captures driver behavior patterns using real-world EV charging data.
• To enable accurate day-ahead prediction of EV parking duration and load for optimal scheduling.
• To integrate statistical analysis with machine learning for improved forecasting performance.
• To validate the model on real datasets from UCLA and Santa Monica smart charging infrastructures.

Proposed method

• Utilized real-world EV charging datasets collected from smart charging infrastructures at UCLA and the city of Santa Monica.
• Applied unsupervised clustering algorithms to group EV drivers based on behavioral patterns such as charging start time, duration, and energy consumption.
• Employed a multilayer perceptron (MLP) neural network to model and predict future charging behavior from clustered driver profiles.
• Combined statistical analysis of historical charging records with machine learning to enhance prediction accuracy.
• Used cross-validation to evaluate model performance on day-ahead parking and load forecasting tasks.
• Integrated clustering results as input features into the MLP for improved forecasting of EV load profiles.

Experimental results

Research questions

• RQ1 Can clustering EV drivers based on charging behavior improve the accuracy of day-ahead load forecasting?
• RQ2 How effective is the integration of statistical clustering with a multilayer perceptron for EV load prediction?
• RQ3 To what extent does the proposed model reduce uncertainty in EV charging scheduling?
• RQ4 How does the model perform across different real-world datasets from diverse urban environments?
• RQ5 What is the impact of driver behavior clustering on the reliability of smart charging control systems?

Key findings

• The proposed model achieved strong performance in day-ahead EV parking and load prediction using cross-validation.
• The integration of statistical clustering with a multilayer perceptron significantly improved forecasting accuracy compared to baseline methods.
• The model effectively captured behavioral patterns in real-world EV charging data from diverse urban settings.
• The clustering approach reduced uncertainty in charging session parameters, supporting more reliable scheduling.
• The model demonstrated robustness and generalizability across datasets from UCLA and Santa Monica.
• The results support the use of hybrid statistical and machine learning techniques for scalable EV user modeling in smart grid applications.