Deep learning prediction and optimization of biomass-powered organic Rankine cycle

Increasing demand for sustainable energy is driving significant interest in biomass-powered ORC systems, especially for small-scale, off-grid power generation. However, due to the nonlinear coupling between the operating parameters, it remains difficult to predict many different performance indicators, or rank performance to identify the best system configurations. This study seeks to address these issues by developing a deep neural networks (DNN) model with thermodynamic simulation, multi-objective optimization, and decision analysis to predict biomass organic Rankine cycle (ORC) performance, with the aim of optimizing performance. Data was generated in matrix laboratory (MATLAB) using CoolProp based on two biomass fuels, coconut shell and cornstalk, and operating ranges that included pressures of 2.6–3.5 MPa, and lower cycle temperature of 300.15–310.15 K for R245fa as working fluid. The nine decision variables included mixture strength, biomass flowrate, and component efficiencies. The DNN predicted four exergy-based outputs: net power, cycle efficiency, exergy efficiency of heat transfer fluid–organic Rankine cycle (HTF–ORC) circuits, and the overall exergy efficiency of the biomass plant. Pareto-based optimization, was implemented using particle swarm optimization (PSO), which produced non-dominated solutions from which the optimal decision variables were identified based on technique for order preference by similarity to ideal solution (TOPSIS) method. The results indicated coconut shell performance was greater than that for cornstalk, with optimal net power ranging from 1350–1460 kW, cycle efficiency values around 18.4%, HTF–ORC exergy efficiency values between 26–27%, and plant exergy efficiency between 6.5–6.9%. The most efficient points correlated with the highest operating temperatures and pressures. The mixture strength and biomass mass flowrate showed proportional increases to power targets, reaffirming the validity of the thermodynamic analysis and optimization. The DNN-PSO-TOPSIS approach appropriately captures the multi-parametric interactions that govern ORC performance, providing a powerful and scalable framework for the design of efficient biomass-to-power systems for renewable energy uses.