Accurate modelling of the scroll expander via a mechanism-incorporated data-driven method

Abstract

Accurate modelling of the scroll expander is essential for efficiency analysis and optimal control. In this study, we propose a mechanism-incorporated adaptive-network-based fuzzy inference system (MI+ANFIS) to establish the scroll expander model. In this method, to fully utilize the mechanism characteristics and improve the prediction performance, we firstly identify the mechanistic model parameters based on the least squares method. Then, the ANFIS is adopted to construct residual prediction model according to the residual errors from the mechanistic model. The final forecasting outputs of the MI+ANFIS model are obtained by combining the mechanistic model and the ANFIS model. Experiments on forecasting the volume flow rate and torque of the scroll expander are taken separately. To demonstrate the superiorities of the proposed MI+ANFIS, it is compared with several other popular models, including the ANFIS, the extreme learning machine (ELM), the back-propagation neural network (BPNN), and the support vector regression (SVR), while with the MI+ELM, the MI+BPNN and the MI+SVR. Experimental results indicated that the proposed MI+ANFIS exhibits higher accuracy and greater robustness due to the consideration of the mechanistic properties.