Experimental investigation and machine learning optimization of a small-scale organic Rankine cycle

Abstract

The system behaviors of a small-scale organic Rankine cycle (ORC) system using a scroll expander under various operating conditions, as well as the machine learning optimization based on back propagation artificial neural network (BPANN), are investigated in this study. The effects of pump outlet pressure, expander inlet temperature and mass flow rate on net power output and thermal efficiency are examined experimentally. A BP-ORC neural network model is constructed, and the prediction accuracy is analyzed by the errors between the experiment and predicted data. The effects of eight operating parameters on system performance are discussed. The bi-objective optimization for maximum net output power and maximum thermal efficiency is addressed. Results indicate that as the mass flow rate increases, the net power output presents an increase trend, whereas the thermal efficiency yields a smooth tendency. The absolute error in the net power output from the BP-ORC model is between −0.05 kW and 0.05 kW, while that in the thermal efficiency is between −8 × 10-5 and 8 × 10-5. The Pareto-optimal solution of thermal efficiency is 9.32 % and the net output power is 2.42 kW.