The mentioned difficulties associated with the calibration process inspired the concept of inverse modelling. In this case, the experimental data become entirely integrated in the calibration process and an optimization routine is used to quantify the best set of parameters which explain the observed pyrolysis behaviour (i.e. multivariable curve fitting). The most used experimental data for model calibration have been the mass loss rate and the surface temperature [10–12]. The optimization technique used is function of the number of variables and their interactions. In the past, only the few most uncertain parameters (i.e. the kinetics parameters) were generally used as potentiometers [13]. However, sophisticated mathematical procedures have been developed to increase the number of parameters optimized simultaneously (e.g. Genetic Algorithm (GA) [10,14] or Shuffled Complex Evolution (SCE) [11]). Lautenberger and Fernandez-Pello [12] have recently investigated the influence that the choice of algorithm can have on the optimized parameters. They generated using their code GPYRO a set of synthetic data (mass loss rate and surfaces temperature) and tried with different algorithms to find back the set of input parameters. The four optimization algorithms provided results with an absolute average error between 1% and 25%. SCE was the most suitable algorithm. The use of synthetic data conveniently avoids the problem of agreement between the actual physical phenomena and any modelling assumption.
