Electric Vehicle Battery State of Charge Prediction Based on Graph Convolutional Network

• Published in: International journal of automotive technology Vol. 24; no. 6; pp. 1519 - 1530
• Main Authors: Kim, Geunsu, Kang, Soohyeok, Park, Gyudo, Min, Byung-Cheol
• Format: Journal Article
• Language: English
• Published: Seoul The Korean Society of Automotive Engineers 01.12.2023; Springer Nature B.V; 한국자동차공학회
• Subjects: Artificial neural networks; Automobile industry; Automotive Engineering; Electric charge; Electric vehicles; Engineering; Nodes; Recurrent neural networks; State of charge; Statistical models; 자동차공학; Electric vehicle (EV); Graph convolutional network; Time series prediction; Neural network; State of charge (SoC)
• ISSN: 1229-9138; 1976-3832
• DOI: 10.1007/s12239-023-0122-6

The state of charge (SoC) of a vehicle battery can tend to vary depending on the driver’s driving patterns and circumstances. To accurately predict the SoC level, it is necessary to consider various circumstances. That is why traditional statistical models may not be sufficient. To address this issue, recurrent neural network (RNN) models have been proposed for time series prediction tasks due to their superior performance. In this paper, we propose a new approach using a graph convolutional network (GCN)-based model that shows better performance than RNN-based models. The GCN requires an adjacency matrix as input, which represents the relationships between variables. We set this matrix to be learnable during model training rather than predefined. We also use two different adjacency matrices: one with variables as nodes, and the other with timestamps as nodes, to enhance the interpretability of the data by considering different elements as nodes. This allows the model to interpret the data from different perspectives. The proposed GCN model was tested using real-world electric vehicle (EV) data and demonstrated improved performance compared to RNN-based baselines. In addition, the GCN model has advantage of being able to clearly express the relationships between variables in a graph, improving interpretabilty.