基于改进AdaBoost的油井动液面自适应集成建模

发布时间：2018-01-30 16:41

本文关键词： 集成学习算法自适应建模误差率动液面泛化能力　出处：《电子测量与仪器学报》2017年08期 　论文类型：期刊论文

【摘要】：针对单一模型软测量算法在动液面预测应用过程中存在泛化能力弱、自适应性差等问题,提出利用Ada Boost集成学习的思想,突出预测误差在建模样本权重及弱学习机权重中的作用,使之更加适合回归模型预测。针对油井工况动态多变导致软测量模型随生产进行逐渐失效的问题,提出利用油田生产过程中定期巡检的少量动液面数据评估原有集成模型,利用相似度原理在保留原有模型信息的基础上增加新信息,并在此基础上根据新样本权重更新弱学习机模型,集成为强学习机模型以动态适应新的油田生产工况。通过对油田生产现场实际数据验证结果表明,该方法对油田生产波动的自适应能力强,能够提高动液面软测量模型的泛化能力及预测精度。
[Abstract]:Aiming at the problems of weak generalization ability and poor adaptive ability in the application of single model soft sensor algorithm in dynamic level prediction, the idea of integrated learning using Ada Boost is proposed. Stress the role of prediction error in the weight of modeling samples and the weight of weak learning machine, make it more suitable for regression model prediction. In view of the problem that the dynamic change of oil well conditions leads to the gradual failure of soft sensor model with production. This paper proposes to evaluate the existing integrated model by using a small number of moving liquid level data which are regularly inspected in the production process of oil fields, and to add new information on the basis of retaining the information of the existing model by using the similarity principle. On this basis, the weak learning machine model is updated according to the weight of the new sample, and integrated into a strong learning machine model to dynamically adapt to the new oilfield production conditions. This method has strong adaptive ability to oil field production fluctuation, and can improve the generalization ability and prediction accuracy of the dynamic liquid level soft sensor model.
【作者单位】：沈阳工业大学电气工程学院;渤海大学工学院;
【基金】：辽宁省博士科研启动基金(201601163) 国家自然科学基金(61403040)资助项目
【分类号】：TE319
【正文快照】： 0引言油田生产过程中地下环境复杂,传统传感器无法得到有效应用,以至于动液面数据无法实时准确测量,导致油田生产过程中措施调整滞后,油井长期处在非最佳工作状态,工程人员利用井下静态参数、示功图数据、井口压力等可测参数结合机理过程计算动液面,但过程中无法确定的参数太

【相似文献】