Reliable Online Parameter Identification of Li-Ion Batteries in Battery Management Systems Using the Condition Number of the Error Covariance Matrix

Monitoring the state of health (SOH) for Li-ion batteries is crucial in the battery management system (BMS), for their efficient and safe use. Due to time-varying battery parameters and insufficient computation capability of the BMSs, computationally efficient online parameter identification is practically required. So, a simple equivalent circuit model (ECM) based recursive least squares (RLS) parameter identification algorithm has been widely used. However, it has long been acknowledged that this algorithm suffers from wind-up problem when the input current doesn't provide sufficient excitation. It causes numerical instability and then induces large sensitivity of identified parameter values to the noise or truncation error of sensor data, leading to large parameter identification errors. In this work, a new reliable version of ECM based RLS, called a condition number based recursive least squares (CNRLS) algorithm, is proposed to avoid large errors due to insufficient excitation by monitoring the condition number of the error covariance matrix If the condition number is greater than a certain prescribed value, currently identified parameters are considered unreliable and hence the proposed algorithm uses stored internal variables previously computed with sufficiently exciting input current, leading to small condition number of the error covariance matrix. Accordingly, the forgetting factor is also adjusted to give a larger weight to such stored internal variables in order to overcome the insufficient excitation of the input current. It is shown with a1-RC equivalent circuit model that the proposed CNRLS algorithm is more noise-tolerant and accurate than two benchmarks including the standard RLS and adaptive forgetting factor RLS (AFFRLS) in terms of mean absolute errors, with almost the same computing cost.