Gradient descent observer for on-line battery parameter and state coestimation

Abstract

Transformation of energy supply towards fully renewable generation presupposes readily available distributed energy storage systems to compensate the variability of renewable sources. Battery energy storage systems (BESS) connected to the grid via power electronics are well suited to be deployed in large numbers in Smart Grid applications to provide system services like load balancing, grid stabilizing and power quality management. Control and optimization of BESS operation rely on the availability of accurate and adaptive models. In particular, dispatch optimization in energy management systems demands regularly updated, simple and reliable battery models. As time-consuming parameterization testing imposes additional costs and operating constraints on BESS, adaptive battery parameter and state coestimation provides an inexpensive on-line solution. A novel observer technique based on gradient descent optimization, that provides estimates of unknown and variable battery parameters and of the internal battery state, is proposed in this work. The coestimation is validated using experimental data from commercial Li-Ion batteries and the test results are reported.