Title: Research on well production prediction based on improved extreme learning machine
Authors: Wenbo Na; Quanmin Zhu; Zhiwei Su; Qingfeng Jiang
Addresses: College of Electric and Mechanical Engineering, China Jiliang University, Hangzhou, China ' Department of Engineering Design and Mathematics, University of the West of England, Bristol, UK ' College of Electric and Mechanical Engineering, China Jiliang University, Hangzhou, China ' College of Electric and Mechanical Engineering, China Jiliang University, Hangzhou, China
Abstract: To overcome traditional extreme learning machine (ELM) disadvantages in over-fitting problem and the local minimum, consequently to improve the prediction accuracy in production of oil well, this paper proposes an improved extreme learning machine (RWELM) prediction model so that the structural risk minimisation principle is integrated into the model and the common activation function is substituted by wavelet function. So we get the improved extreme learning machine algorithm (RWELM) and give the parameter selection method. This paper takes dynamic data from an oil well production of LunNan (China) oilfield as the research background to demonstrate the new approach in forecasting production of oil well. The experimental results show that the forecasting model is better than ELM, BP (back propagation) network with time delay sequence, LM-BP neural networks in both generalisation performance and predictive accuracy.
Keywords: modelling; structural risk; wavelet function; generalisation performance; oil well production; well production prediction; extreme learning machine; ELM; oil wells; production forecasting; China; oilfields.
DOI: 10.1504/IJMIC.2015.069945
International Journal of Modelling, Identification and Control, 2015 Vol.23 No.3, pp.238 - 247
Published online: 16 Jun 2015 *
Full-text access for editors Full-text access for subscribers Purchase this article Comment on this article