Energy Efficiency Analysis Using a Novel VSG Based DEA: A Case Study of Ethylene Production Plants

Due to the complex reaction mechanisms and high coupling between variables in the Ethylene production process, it is difficult for manufacturers to configure the energy structure and achieve the optimal production status in experience. However, the traditional data envelopment analysis method (DEA) cannot perform well in distinguishing the effective and inefficient samples. Therefore, in this paper, a novel method integrating DEA with virtual sample generation (VSG-DEA) is proposed to improve the effective discrimination. The proposed DEA method applies the improved extreme learning machine (ELM) which utilized feature scaling of the hidden layer outputs to generate appropriate virtual samples from original samples. Then the mixed samples with virtual samples are evaluated using DEA. In order to validate the performance, the proposed VSG-DEA is utilized to analyze the energy efficiency of the ethylene production process. It is proved that the discrimination of the ethylene production unit is effective in the simulation experiments. Furthermore, the energy-saving potential can be obtained by analyzing the simulation results.