A Voltage Sensor Fault Diagnosis Method Based on Long Short-Term Memory Neural Networks for Battery Energy Storage System

• Published in: 2021 IEEE/IAS Industrial and Commercial Power System Asia (I&CPS Asia) pp. 163 - 167
• Main Authors: Wan, Changjiang, Yu, Quanqing, Li, Jianming
• Format: Conference Proceeding
• Language: English
• Published: IEEE 18.07.2021
• Subjects: Battery sensor fault; Data collection; Deep learning; Fault diagnosis; Long Short-Term Memory; Recurrent neural networks; sensor fault diagnosis; Time series analysis; Training; Voltage
• DOI: 10.1109/ICPSAsia52756.2021.9621560

The safety of energy storage systems with lithium-ion batteries as the main energy storage component is a current research hotspot. Various battery system fault diagnosis strategies are based on the assumptions of accurate sensor data collection, and there are few studies on fault diagnosis of battery system data collection sensors, especially for voltage sensors. By using deep learning technology, a voltage sensor fault diagnosis method which can detect and classify voltage sensor fault in energy storage system is proposed in this paper. Assumption of three typical fault modes was considered, and these faults were injected into experiment data to generate dataset. The voltage sensor fault diagnosis model consists of four-layer Long Short-Term Memory (LSTM) recurrent neural network (RNN) and three dense layers. After training and testing, the ability of LSTM in processing time series on voltage sensor diagnosis is preliminarily proved, which provides a valuable reference for battery system sensor fault diagnosis.