Use of Artificial Neural Networks to Model Biomass Properties of Miscanthus (Miscanthus × giganteus) and Virginia Mallow (Sida hermaphrodita L.) in View of Harvest Season

Abstract

Miscanthus and Virginia Mallow are energy crops characterized by high yields, perenniality, and low agrotechnical requirements and have great potential for solid and liquid biofuel production. Later harvest dates result in lower yields but better-quality mass for combustion, while on the other hand, when biomass is used for biogas production, harvesting in the autumn gives better results due to lower lignin content and higher moisture content. The aim of this work was to determine not only the influence of the harvest date on the energetic properties but also how accurately artificial neural networks can predict the given parameters. The yield of dry matter in the first year of experimentation for this research was on average twice as high in spring compared to autumn for Miscanthus (40 t/ha to 20 t/ha) and for Virginia Mallow (11 t/ha to 8 t/ha). Miscanthus contained 52.62% carbon in the spring, which is also the highest percentage determined in this study, while Virginia Mallow contained 51.51% carbon. For both crops studied, delaying the harvest date had a positive effect on ash content, such that the ash content of Miscanthus in the spring was about 1.5%, while in the autumn it was 2.2%. Harvest date had a significant effect on the increase of lignin in both plants, while Miscanthus also showed an increase in cellulose from 47.42% in autumn to 53.5% in spring. Artificial neural networks used to predict higher and lower heating values showed good results with lower errors when values obtained from biomass elemental composition were used as input parameters than those obtained from proximity analysis.