An optimized informer model design for electric vehicle SOC prediction

Area of Science:

• Electrical Engineering
• Artificial Intelligence
• Automotive Systems

Background:

• Accurate State of Charge (SOC) prediction is crucial for electric vehicle (EV) status assessment and performance management.
• While the Informer model shows promise for SOC prediction, its practical application is limited by existing performance gaps.

Purpose of the Study:

• To propose a novel optimized Informer model for enhanced SOC prediction in electric vehicles.
• To improve the accuracy and speed of SOC prediction by integrating a health assessment algorithm.

Main Methods:

• Developed a health assessment algorithm using historical EV data to generate a health matrix.
• Integrated the health matrix into the Informer model's Encoder and Decoder modules to refine its architecture.
• Utilized the health matrix to optimize prediction logic, mitigating truncation errors.

Main Results:

• Optimizing the Informer's Encoder-Decoder modules resulted in a ~15% increase in prediction accuracy and a ~100% increase in prediction speed.
• Further optimization of the prediction logic to reduce truncation error led to an additional ~20% improvement in SOC prediction accuracy.
• The optimized Informer model demonstrated superior performance across four diverse datasets.

Conclusions:

• The proposed health-assessment-based optimized Informer model significantly enhances SOC prediction accuracy and speed for electric vehicles.
• This approach effectively addresses practical application limitations of the standard Informer model, paving the way for more reliable EV battery management systems.