Surplus electricity production is anticipated to further increase during summer periods due to the inevitable high penetration of renewable energy sources in our energy system. Potential solutions for sustainable and efficient energy storage need to be explored in order to cope with the challenges of the energy transition. Biomass is a promising source of renewable energy that can not only be converted into products, but also into storable fuels. The multitude of products and processes for converting biomass into value-added materials and storable energy necessitates systematic design methods for the generation and analysis of alternative process configurations in order to provide decision support. Due to the high computational time of simulation-based optimization of biorefinery systems, the analysis is often limited to predefined process configurations and operating conditions. Once generated, the results are often only used for one single case study or application. In this work, we suggest a methodology that enables the decision maker to profit from results that are already generated and adapt them to case-tailored system configurations. For this purpose, a database containing the results of multi-objective superstructure optimisation regarding thermo-economic performance of candidate system configurations is used to design surrogate models that predict the thermodynamic performance of the conversion of biomass to SNG, heat and electricity. Furthermore, correlations between modelling inputs and the desired output parameters are revealed. It is shown that artificial neural networks are a suitable tool to predict the thermodynamic behaviour of the biomass conversion plant. Moreover, they allow the decision maker to profit from process data that is already available and to base their decision making off this rather than running computationally expensive simulations. The developed model is adaptable for changes in the input domain, flexible in predicting different key performance indicators and allows for displaying their correlations to different inputs.