Voltage Sensor Fault Detection in Li-ion Battery Energy Storage Systems

Abstract

Safe and optimal operation of battery energy storage systems requires correct measurement of voltage, current, and temperature. Therefore, fast and correct detection of sensor faults is of great importance. In this paper, model-based and non-model-based voltage sensor fault detection methods are developed for a comprehensive comparison. The residual is generated from the difference of measured voltage and estimated voltage. In the model-based method, the voltage is estimated using an extended Kalman filter (EKF). In the non-model-based method, the voltage is predicted using a recurrent neural network (RNN) with long short-term memory (LSTM). For both methods, a scalar generalized likelihood ratio (GLR) detector is developed to detect changes in the sequence of residual signal data and compared with a systematically computed threshold. The parameters threshold (h) and window-size (M) used in the GLR detector, are computed based on the probability of false alarm (P f ) and probability of correct detection (P d ). The GLR detector demonstrates the ability to effectively detect the voltage sensor fault with a maximum delay of 500 ms for the model-based residual and 200 ms for the non-model-based method.