Matrix Regressor Adaptive Observers For Battery Management Systems

Fast parameter estimation is required in several applications where unforeseen changes often occur in system characteristics. Quite often such parameter estimation must be accompanied by state estimation as well. A prime example of growing importance is battery management systems for Lithium-ion batteries. Without accurate knowledge of internal battery states and parameters we are left with a choice between safety and performance with regards to battery selection for a specific application. With the goal of simultaneous estimation of both states and parameters, and arbitrarily fast estimation of the latter, a new adaptive observer based on matrix regressors is introduced in this paper. In contrast to earlier work, this adaptive observer requires the regressors include filters of much lower order. Sufficient conditions for the exponential convergence of parameter estimates to their true values are stated and derived. Guidelines are provided for the selection of design parameters, minimizing the number of observer parameters that must be tuned to facilitate fast convergence. Two numerical simulations are included, demonstrating the performance of the proposed observer scheme as well as a comparison with existing matrix regressor based adaptive observers. Examples considered are a generic linear plant model as well as the dynamics representative of a single particle of a Lithium-ion battery cell.