Accurately estimating the state of charge (SOC) of electric bus batteries can effectively improve driving safety and mileage. Due to the strong nonlinearity and time-varying characteristics of batteries, it is difficult to estimate SOC through a simple model. Therefore, this paper proposes a deep learning method based on convolutional neural networks (CNNs) with parameters fine-tuning to estimate the SOC of electric bus batteries. This method is composed of convolutional neural networks and fully connected layers and a parameters fine-tuning method. CNNs are used to extract the features of the battery data. Fully connected layers are used to establish a mapping relationship between the features and SOCs. The parameter fine-tuning method takes the pre-trained network in a specific task as the research object, and improves or retains the learning ability of different network layers for new tasks by adjusting the learning rate factor of layers. In this paper, the weight learning rate factor of the fully connected layer network is increased and keeps other layers unchanged to achieve parameter fine-tuning, which effectively reduces the model’s demand for training data in new tasks and improves the estimation accuracy. In order to prove the effectiveness of this method, it is validated using datasets under different operating conditions. The results show that the proposed method has high accuracy and robustness, with root mean square error, mean absolute error, and R-squared reported as 0.397%, 0.091%, and 0.997, respectively. The proposed method is more suitable for the estimation of the SOC of electric bus batteries and for improving driving safety and efficient application of electric buses.
State of Charge Estimation for Electric Bus Batteries Based on Deep Learning Model with Parameters Fine-Tuning Method
2025-10-01
Article (Journal)
Electronic Resource
English
Elsevier | 2024
|A Dynamic State-of-Charge Model for Electric Vehicle Batteries
British Library Conference Proceedings | 1994
|