Role of optimisation method on kinetic inverse modelling of biomass pyrolysis at the microscale. (15th February 2020)

• Record Type: Journal Article
• Authors: Purnomo, Dwi M.J.; Richter, Franz; Bonner, Matthew; Vaidyanathan, Ravi; Rein, Guillermo
• Is Part Of: Fuel. Volume 262(2020)
• Publication Date: 2020-02-15
• Subjects: Biomass -- Pyrolysis -- Kinetics -- Optimisation; Fuel -- Periodicals; Coal -- Periodicals; Coal; Fuel; Periodicals; 662.6
• DOI: 10.1016/j.fuel.2019.116251
• Languages: English
• ISSNs: 0016-2361

Abstract

Highlights: Optimisation methods for inverse modelling of biomass pyrolysis are compared. Investigated cellulose and wood for different algorithms and objective functions. All combinations performed equally for cellulose. Large differences in time and accuracy were found for wood. Best combination was Shuffle Complex algorithm with mean square error function. Abstract: Understanding biomass pyrolysis is important for biofuel production and fire safety. Inverse modelling is an increasingly used technique to find values for the kinetic parameters that control pyrolysis. The quality of this inverse modelling depends on, in order of importance, the quality of the experimental data, the kinetic model, and the optimisation method used. Unlike the two former components, the optimisation method chosen, i.e. the combination of algorithm and objective function, is rarely discussed in the literature. This work compares the accuracy and efficiency of five commonly used advanced algorithms (Genetic Algorithm, AMALGAM, Shuffled Complex Evolution, Cuckoo Search, and Multi-Start Nonlinear Program) and a simple algorithm (Random Search) to find the kinetic parameters for cellulose and wood pyrolysis at the microscale via thermogravimetric measurements in the literature. These algorithms are combined with seven objective functions comprising concentrated and dispersed functions. The results show that for cellulose (simple chemistry) the use of an advanced optimisation algorithm is unnecessary, since a simple algorithm achieves similarly high accuracy with higher efficiency (40% to 350% faster). However, for wood (complex chemistry) a combination of an advanced algorithm and a concentrated function greatly improves accuracy. Among the 25 possible combinations we investigated for wood, Shuffled Complex Evolution with mean square error objective function performed best with 0.91% error in mass loss rate and 0.88 × 10 13 CPU time. These findings can guide the selection of the optimal optimisation method to use in inverse modelling of kinetic parameters, improving accuracy and efficiency.