Energy-efficient methane recovery from hazelnut shells via mild KOH–oxidative pretreatment and GA-assisted deep learning

Growing demand for sustainable energy has intensified interest in biogas production from lignocellulosic biomass. Although alkaline pretreatment can improve biodegradability by disrupting the lignin–carbohydrate complex, mild operating conditions and combinations with complementary reagents remain underexplored, particularly in terms of energy efficiency and digestion kinetics. Here, hazelnut shells were pretreated under mild chemical dosages using KOH, H2O2, and their combinations, and evaluated by BMP tests. Single-step pretreatments increased methane yields to 130.8 mL CH4·g−1 volatile solids (VS) (KOH) and 98.0 mL CH4·g−1 VS (H2O2). Combined approaches further improved methane recovery, with the sequential KOH→H2O2 configuration achieving the highest methane yield (179.2 mL CH4·g−1 VS followed by the simultaneous KOH + H2O2 application (162.2 mL CH4·g−1 VS). The superior methane yield of the sequential pretreatment arises from initial alkaline-driven structural opening, which enables more effective and selective oxidative modification of lignin. To support predictive monitoring, two genetic algorithm–optimised deep learning models (GA-LSTM and GA-GRU) were developed to forecast cumulative methane yield trajectories and outperformed the Modified Gompertz model (R2 up to 0.9999). Future work should prioritise pilot-scale, long-term continuous operation to verify whether the stepwise dosing advantage persists under realistic mixing and mass-transfer constraints and to assess the risk of inhibitor accumulation.