Machine Learning Methods to Assess the Thermal Performance of a Novel Basalt-Fiber-Bundle Thermal Flow-Reversal Reactor

Abstract

Artificial intelligence network technique is proposed to predict the thermal performance of a basalt fiber bundle thermal flow-reversal reactor. A comparison of the prediction accuracy against Multiple Linear Regression (MLR) method and Computational Fluid Dynamics (CFD) simulations was done. It is found that the prediction accuracy of the artificial neural network based Back Propagation (BP) model gave the best accuracy, followed by MLR method, and both of them were better than the traditional CFD simulation model. The influence of the main input variables on the thermal efficiency of the reactor was also investigated. It was found that increasing the inlet gas concentration decreased the thermal efficiency of the oxidizer. When the full commutation cycle was extended and the inlet gas flow rate was raised, the thermal efficiency increased firstly and then it decreased. Based on the results from the BP model, the basalt fiber bundle thermal reversal-flow reactor can handle a wide range of exhaust gas concentrations, and it is suitable for the treatment of low methane concentration exhaust gas. The BP-model also helps to establish the theoretical basis for setting the operational parameters in such a way that the performance of proposed reactor can be maintained optimal.

Keywords: Machine Learning, thermal flow-reversal reactor, Basalt fiber bundle, CFD