A Surrogate Modeling Approach for the Development of Biorefineries

Abstract

Biorefining processes are still struggling with commercial flowsheeting softwares. Challenges include the lack of property data, complexity of raw materials, and the constant influx of new processes and technologies. Surrogates are gaining attention due to their ability to exploit black-box nature problems and their attractive computational simplicity. However, existing surrogate modeling methodologies are deprived of property-based parameters and deviate from engineering principles that are useful to retain in the model. This paper introduces a new iterative model-based approach for the systematic development of low granularity models for biorefining processes. The approach develops regressed models in the context of existing commercial packages. Degrees of freedom include the selection of listed components, property methods, and process models. Stages include the selection of property and process models, the optimization of the model parameters, and regression with experimental data. The framework is applied for the modeling of a real-life biorefinery. Three case studies demonstrate the potential of the suggested procedure. The first case study has fixed options for the property and process models, while the second one has them as a degree of freedom. The third case study extrapolates the resulted metamodels to different capacities and six different feedstock types (wheat and rice straw, wood, sugarcane bagasse, banana stem, and miscanthus). Metamodels produced with more degrees of freedom for selecting and modifying the process and property models gain in accuracy, flexibility, and simplicity. The suggested framework can be integrated with artificial intelligence tools to automatize the link with commercial software.