A Monte Carlo and PSO based virtual sample generation method for enhancing the energy prediction and energy optimization on small data problem: An empirical study of petrochemical industries. (1st July 2017)

• Record Type: Journal Article
• Authors: Gong, Hong-Fei; Chen, Zhong-Sheng; Zhu, Qun-Xiong; He, Yan-Lin
• Is Part Of: Applied energy. Volume 197(2017)
• Page Start: 405
• Page End: 415
• Publication Date: 2017-07-01
• Subjects: Energy optimization -- Energy prediction -- Small data -- Virtual sample generation -- Petrochemical industries; Power (Mechanics) -- Periodicals; Energy conservation -- Periodicals; Energy conversion -- Periodicals; 621.042
• DOI: 10.1016/j.apenergy.2017.04.007
• Languages: English
• ISSNs: 0306-2619

Abstract

Highlights: A novel Monte Carlo and PSO based virtual sample generation method is proposed. Effective virtual samples are generated for solving the small data problem. Petrochemical industry empirical studies are carried out for performance validation. Simulation results show the proposed method can improve energy prediction accuracy. Guidance can be given to the production departments for improving energy efficiency. Abstract: Due to the imbalanced and uncompleted characteristics of complex petrochemical small datasets, it is a challenge to build an accurate prediction and optimization model of energy consumption of petrochemical systems. Therefore, this paper proposes a novel virtual sample generation (VSG) approach based on the Monte Carlo (MC) and Particle Swarm Optimization (PSO) algorithms to improve the accuracy of the energy efficiency analysis on small data set problems. The proposed approach utilizes the MC and PSO algorithms to generate appropriate virtual samples based on the underlying information extracted from the small datasets. An accurate prediction model is presented using the extreme machine learning (ELM) in view of the synthetic data. The performance of the proposed model is validated via an application using a purified Terephthalic acid (PTA) solvent system and an ethylene production system. The experiment results demonstrate that the accuracy of the prediction model can be improved, and guidance for the production department to improve the energy efficiency, energy savings and emission reduction is provided under the small data circumstance.