On-Board SOH Estimation Using Quantized Adaptive Dual Extended Kalman Filter in Resource Constraint Battery Management Systems

State-of-Health (SOH) is a critical metric used to assess the remaining capacity and overall health of a battery. Accurate SOH estimation is essential for optimizing the performance and extending the lifespan of battery packs. While neural network-based algorithms offer precise predictions, their deployment on resource-constrained hardware and embedded systems poses significant challenges due to the high computational complexity and storage demands of such algorithms. In response, this study integrates quantization-a method that reduces 32-bit floating-point values to lower-bit representations-into a self-cognizant Kalman filter for SOH estimation. A quantized feed-forward neural network (FFNN) models the battery dynamics, which is then incorporated into a dual extended Kalman filter algorithm. Model parameter updates, performed every cycle, are also executed in a quantized manner, thereby reducing the number of floating-point operations. This approach employs quantization to balance computational speed and accuracy, while lowering storage and processing requirements. Various quantization levels were tested, with 6-bit quantization proving particularly effective, enhancing the practicality of this method for on-board SOH estimations in battery management systems. Experimental findings indicate the reduction of computational time by 20.576% and memory requirements by 41.25%, while maintaining satisfactory accuracy in SOH estimation.