Improved State of Charge Estimation of a Lithium-Ion Battery Output: Application to Conventional Neural Network

The safety and reliability of battery storage systems are essential for the widespread adoption of electrified transportation and new energy generation. One of the crucial parameters for the safe management and effective control of batteries is the state of charge (SOC). In recent years, there has been a great deal of interest in machine-learning-based SOC estimation methods for lithium-ion batteries. However, a common issue with these models is that they frequently exhibit unstable estimation performances, which makes it challenging to use them in real-world scenarios. To address this problem, a framework based on convolutional neural networks (CNNs) uses measurements of the voltage, current, and temperature while the battery is charging to directly estimate SOC. The CNN is trained using randomized data. To increase accuracy, training data was enhanced with noise and error that included multiple layers and neurons. Additionally, the algorithm was examined for various temperature distributions, which would be common for many applications. With the aid of statistical indicator metrics, the proposed model’s accuracy and generalizability are demonstrated in the experiments using data gathered under various working conditions. The experiment’s findings show that the proposed model’s maximum error is less than 1.9%.