Neural Network-Based SOC Estimation
Neural Network-Based SOC Estimation
Neural Network-Based SOC Estimation
Transfer Learning and Knowledge Distillation for SOC Estimation
Transfer Learning and Knowledge Distillation for SOC Estimation
Transfer Learning and Knowledge Distillation for SOC Estimation
The primary objective was to build a data-efficient, neural network-based system to estimate the State of Charge (SOC) of lithium-ion batteries in electric vehicles. The project focused on improving accuracy, generalization, and deployment viability using real-world EV battery datasets under different operational and thermal conditions. To overcome the challenges of limited labeled data and computational overhead, the system was designed using a combination of pre-trained BiLSTM models and seventeen advanced transfer learning techniques — enabling fast, accurate SOC predictions even in data-scarce conditions.
The primary objective was to build a data-efficient, neural network-based system to estimate the State of Charge (SOC) of lithium-ion batteries in electric vehicles. The project focused on improving accuracy, generalization, and deployment viability using real-world EV battery datasets under different operational and thermal conditions. To overcome the challenges of limited labeled data and computational overhead, the system was designed using a combination of pre-trained BiLSTM models and seventeen advanced transfer learning techniques — enabling fast, accurate SOC predictions even in data-scarce conditions.
The primary objective was to build a data-efficient, neural network-based system to estimate the State of Charge (SOC) of lithium-ion batteries in electric vehicles. The project focused on improving accuracy, generalization, and deployment viability using real-world EV battery datasets under different operational and thermal conditions. To overcome the challenges of limited labeled data and computational overhead, the system was designed using a combination of pre-trained BiLSTM models and seventeen advanced transfer learning techniques — enabling fast, accurate SOC predictions even in data-scarce conditions.
Client
Research lab at Polimi
Services
Model Development Transfer Learning Engine Bayesian Hyperparameter Tuning Transfer Learning Knowledge Distillation
Industries
Electric Transportation
Date
January 2024
The core system combines transfer learning with domain adaptation methods like MMD and CORAL to align features across datasets. A fine-tuned AI layer ensures robust SOC predictions across manufacturers and conditions, making it suitable for real-world deployment in BMS systems. particularly where data acquisition is limited or noisy. This AI-first approach enables faster, cheaper, and more generalizable SOC estimation, critical for large-scale EV adoption. It reduces training time from hours to minutes without compromising accuracy, providing a framework that battery manufacturers and mobility providers can readily adopt for smart, scalable energy solutions.
The core system combines transfer learning with domain adaptation methods like MMD and CORAL to align features across datasets. A fine-tuned AI layer ensures robust SOC predictions across manufacturers and conditions, making it suitable for real-world deployment in BMS systems. particularly where data acquisition is limited or noisy. This AI-first approach enables faster, cheaper, and more generalizable SOC estimation, critical for large-scale EV adoption. It reduces training time from hours to minutes without compromising accuracy, providing a framework that battery manufacturers and mobility providers can readily adopt for smart, scalable energy solutions.
The core system combines transfer learning with domain adaptation methods like MMD and CORAL to align features across datasets. A fine-tuned AI layer ensures robust SOC predictions across manufacturers and conditions, making it suitable for real-world deployment in BMS systems. particularly where data acquisition is limited or noisy. This AI-first approach enables faster, cheaper, and more generalizable SOC estimation, critical for large-scale EV adoption. It reduces training time from hours to minutes without compromising accuracy, providing a framework that battery manufacturers and mobility providers can readily adopt for smart, scalable energy solutions.
Feedback from academic reviewers and industry collaborators highlighted the system’s practical viability, adaptability across battery types, and technical rigor , positioning this work as a benchmark in the evolving landscape of data-efficient energy intelligence for electric vehicles.
Feedback from academic reviewers and industry collaborators highlighted the system’s practical viability, adaptability across battery types, and technical rigor , positioning this work as a benchmark in the evolving landscape of data-efficient energy intelligence for electric vehicles.
Feedback from academic reviewers and industry collaborators highlighted the system’s practical viability, adaptability across battery types, and technical rigor , positioning this work as a benchmark in the evolving landscape of data-efficient energy intelligence for electric vehicles.