This week, you will learn how to derive the steps of the Gaussian sequential probabilistic inference solution, which is the basis for all Kalman-filtering style state estimators. While this content is highly theoretical, it is important to have a solid foundational understanding of these topics in practice, since real applications often violate some of the assumptions that are made in the derivation, and we must understand the implication this has on the process. By the end of the week, you will know how to derive the linear Kalman filter.

The steps of a Kalman filter may appear abstract and mysterious. This week, you will learn different ways to think about and visualize the operation of the linear Kalman filter to give better intuition regarding how it operates. You will also learn how to implement a linear Kalman filter in Octave code, and how to evaluate outputs from the Kalman filter.

A linear Kalman filter can be used to estimate the internal state of a linear system. But, battery cells are nonlinear systems. This week, you will learn how to approximate the steps of the Gaussian sequential probabilistic inference solution for nonlinear systems, resulting in the "extended Kalman filter" (EKF). You will learn how to implement the EKF in Octave code, and how to use the EKF to estimate battery-cell SOC.

The EKF is the best known and most widely used nonlinear Kalman filter. But, it has some fundamental limitations that limit its performance for "very nonlinear" systems. This week, you will learn how to derive the sigma-point Kalman filter (sometimes called an "unscented Kalman filter") from the Gaussian sequential probabilistic inference steps. You will also learn how to implement this filter in Octave code and how to use it to estimate battery cell SOC.

Kalman filtering requires that noises have zero mean. What do we do if the current-sensor has a dc bias error, as is often the case? How can we implement Kalman-filter type SOC estimators in a computationally efficient way for a battery pack comprising many cells? This week you will learn how to compensate for current-sensor bias error and how to implement the bar-delta method for computational efficiency. You will also learn about desktop validation as an approach for initial testing and tuning of BMS algorithms.

You have already learned that Kalman filters must be "tuned" by adjusting their process-noise, sensor-noise, and initial state-estimate covariance matrices in order to give acceptable performance over a wide range of operating scenarios. This final course module will give you some experience hand-tuning both an EKF and SPKF for SOC estimation.