Due to slow dynamics data acquisition for battery modelling is a time consuming process and sometimes increasing the measurement time is either not possible due to constraints in data acquisition hardware or it is not possible to cover the whole range of operation in one experiment, hence data from multiple experiments must be acquired. In this paper, two different methodologies to initialise the Polynomial Nonlinear State Space (PNLSS) model structure are proposed, which are based on the Local Polynomial Method (LPM) to estimate nonparametrically the Best Linear Approximation BLA) from multiple datasets.