Predictive Modeling of Electric Bicycle Battery Performance: Integrating Real-Time Sensor Data and Machine Learning...

Area of Science:

• Sustainable mobility
• Machine learning applications
• Battery management systems

Background:

• Accurate battery state of charge (SOC) estimation is crucial for electric vehicles.
• Real-world data collection from instrumented bicycles offers a valuable alternative to lab experiments.
• Integrating diverse variables enhances predictive model performance.

Purpose of the Study:

• To develop data-driven models for lithium-ion battery SOC estimation using real-time sensor data from electric bicycles.
• To collect and preprocess multimodal datasets including operational, environmental, and route variables.
• To assess and compare the performance of various machine learning algorithms for SOC prediction.

Main Methods:

• Collected multimodal data from electric bicycle sensors over a 28-day period in Medellín, Colombia.
• Employed data preprocessing techniques, including Convolutional Neural Networks (CNNs) for sensor data smoothing.
• Utilized Long Short-Term Memory (LSTM), Support Vector Regression (SVR), AdaBoost, and Gradient Boost algorithms for Remaining Useful Life (RUL) prediction.

Main Results:

• Data preprocessing, particularly CNN-based smoothing, significantly improved the accuracy of SOC estimation models.
• Comparative analysis identified the most effective machine learning models for predicting battery RUL and SOC.
• The study demonstrated the feasibility of using real-world data for robust battery performance modeling.

Conclusions:

• Machine learning models, when trained on real-world data, can accurately predict electric bicycle battery SOC.
• Data preprocessing is essential for enhancing the performance and reliability of predictive models.
• This research contributes to sustainable mobility by providing advanced battery management strategies for electric two-wheelers.