摘要
井下示功图的自动识别是建造有杆抽油工况诊断专家系统的一个技术关键。由于现场难以系统地收集齐全的示功图，因而给示功图识别器的构造带来较大的困难。本文基于自适应谐振理论，提出一种基于竞争学习和自稳机制的自组织神经网络示功图识别模型。该模型较之于前馈示功图网络模型，解决了以往示功图神经网络识别模型需完备训练集（各种类型及其同类型中各种形状的示功图）及学习效率非常缓慢的问题。在有杆抽油示功图基础学习上，通过无监督学习算法，神经网络还能适用于不同油田区域的示功图自动识别工作。
The identification of down-hole dynamometer card is the key to making the diagnosis expert system for pumping with rod. On the basis of adaptive resonance theory,an adaptive neural networks with competitive and self-stablized learning algorithms to enhance the system learning efficiency is presented. Compared with common feedforward networks, this neural networks can overcome some shortcomings existing during learning and training and solve the problems of the complete training series to be provided and very low learning efficiency. By the further study on the dynamometer card for pumping with rod ,the non-supervi-sion networks system can be used to identify various kinds of patterns from different oil fields.
出处
《石油学报》
EI
CAS
CSCD
北大核心
1996年第3期104-109,共6页
Acta Petrolei Sinica
关键词
机械采油
示功图
自组织系统
有杆抽油
mechanical recovery
dynamometer card
neural networks
adaptive system
