Parallel particle filter for state of charge and health estimation with a long term test

The paper presents a new approach for state estimation of batteries that is able to overcome most of the obstacles for the classical Kalman filter approach. The so called particle filter is able to use any probability density function by applying Monte Carlo sampling methods for approximating the density functions for state of charge and state of health defined by the remaining capacity. Thereby the restriction of the Kalman filter to zero mean Gaussian distributions for all states and errors is overcome. The paper proves the validity of the approach by testing lithium metal oxide/graphite batteries with different states of health by applying different current and temperature profiles. A special focus of the testing is on electric vehicles and photovoltaic applications. For electric vehicles state of health determination achieves a correctness of 1 % or better and is a bit worse for photovoltaic applications with 3.75 % or better for ageing state between 100 % and 80 % of initial capacity. During long term testing the algorithm is validated with a decreasing state of health over time due to accelerated ageing. The state of charge estimation is always better than 1 % in long term testing and the state of health is correctly tracked over time.