Thursday 23 January 2025
As our reliance on electric vehicles grows, so does the need for accurate power consumption predictions. But predicting energy usage is a complex task, especially when considering the diverse driving conditions and limited energy storage capabilities of these vehicles.
To tackle this challenge, researchers have turned to machine learning algorithms that can learn from large datasets. However, existing models often lack discussion on collaborative learning and privacy issues among multiple clients. This is where the Multi-Head Heterogeneous Federated Learning (MHHFL) system comes in.
MHHFL is a novel approach that uses data preprocessing to pre-classify features, facilitating knowledge transfer between different head networks. Each head network independently classifies a feature set and serves as a carrier for federated learning. In the federated phase, the head network embedding mechanism embeds each head network into 2D vectors, allowing clients to share information (weights and embedded vectors) with a central source pool.
The proposed selection mechanism then selects appropriate source networks based on the embedded vectors and mixes the target and source networks as knowledge transfer in federated learning. This approach is characterized by its heterogeneity, asynchrony, efficiency, privacy, and security.
Experimental results show that MHHFL significantly outperforms benchmark systems, reducing prediction errors by 24.9% to 94.1%. Robustness evaluation was also performed to simulate communication latency, with MHHFL maintaining excellent performance and suppressing the state-of-the-art by 4.9% to 94.5%.
Ablation studies demonstrate the effectiveness of each proposed mechanism, particularly heterogeneous federated learning (head network embedding and selection mechanisms), which significantly outperforms traditional FedAvg and random transfer.
As our society continues to shift towards a more sustainable future, accurate power consumption predictions will play a crucial role in optimizing battery power and minimizing energy waste. MHHFL’s innovative approach to collaborative learning has the potential to revolutionize the field of electric vehicle energy management, enabling more efficient and environmentally friendly transportation systems.
Cite this article: “Federated Learning for Efficient Electric Vehicle Energy Management”, The Science Archive, 2025.
Electric Vehicles, Power Consumption Predictions, Machine Learning Algorithms, Federated Learning, Heterogeneous Networks, Data Preprocessing, Knowledge Transfer, Security, Sustainability, Energy Management
