Premium
A Hybrid Algorithm for the Reliability Evaluation Models of Chemical Systems
Author(s) -
Liu Wen,
Wei Kecheng,
Xu Juanjuan,
Ji Xu
Publication year - 2017
Publication title -
quality and reliability engineering international
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 0.913
H-Index - 62
eISSN - 1099-1638
pISSN - 0748-8017
DOI - 10.1002/qre.2108
Subject(s) - reliability (semiconductor) , algorithm , computer science , genetic algorithm , residual , artificial neural network , process (computing) , grey relational analysis , convergence (economics) , data mining , automation , artificial intelligence , machine learning , engineering , mathematics , mechanical engineering , power (physics) , physics , mathematical economics , quantum mechanics , economics , economic growth , operating system
Chemical processes are complex dynamic systems. With the chemical industry under pressure to introduce improvements through the greater use of automation and intelligence, the need for comprehensive reliability evaluation has become more urgent both theoretically and practically. The employment of intelligent algorithms based on factory data has been the recent research hotspots. But for complex systems with available data on a smaller scale, reliability evaluation models have suffered on such problems as a result of instability and over‐fitting, which have to be resolved. The GRA–GA–BP–MCRC hybrid algorithm was proposed. It combined the two‐step genetic algorithm (GA)–back propagation (BP) and grey relational analysis (GRA) with Markov chain residual correction (MCRC). Based on the technical characteristics and the management demands, 46 influential factors of process reliability were introduced, which covered man, machine, material, method, and environment. For model convergence to be assured, GRA and attribute reduction rule were introduced. Meanwhile, based on the correlation of the factors, the two‐step GA–BP was proposed to resolve the over‐fitting problem of artificial neural network with complex input parameters. As well, MCRC was applied to modify the GA–BP error. The resulting average relative error of the hybrid algorithm was 2.36%, while the conventional algorithm was 10.28%. Copyright © 2016 John Wiley & Sons, Ltd.