基于过程神经网络的优势渗流场识别方法研究

发布时间：2018-06-22 07:53

本文选题：过程神经网络 + 径向基　；参考：《东北石油大学》2017年硕士论文

【摘要】：经过长期的水驱开发,大量主力油田的开发区块已进入高含水期。由于注入溶液对油层的长期冲蚀,导致油水井之间产生优势渗流通道,致使注入溶液沿优势渗流通道无效循环而造成大量浪费,降低生产效率。要实现控水稳油、提高油田生产效率,就必须要采取一定的措施对优势渗流通道进行封堵,而有效识别优势渗流通道是进行措施方案设计和优化调整注采关系的关键前提。本课题通过研究油田开发过程中的动静态数据、优势渗流场的形成机理以及影响因素,在对传统判断识别方法分析的基础上,综合利用油田地质数据、开发数据、测试数据、储层开采历史数据等动静态资料,提出一种基于径向基函数过程神经网络结合进化算法的油田优势渗流场识别方法和模型,以油田开发地质参数和生产过程数据为输入,输出为注采井之间是否存在优势渗流场。利用沃尔什变换将过程输入数据进行规整,同时采用混沌遗传算法对过程神经网络模型的性能参数进行优化求解,提高模型的自适应能力和分类识别的准确率。本文研究所用数据均来源于基于数模处理的大庆油田某采油厂部分区块注水井与生产井的层位月数据,通过分析渗流通道形成影响因素,筛选确定模型训练所用的指标参数。将从油田开发实际历史资料数据中选取的指标参数进行预处理并建立训练样本集,经过训练得到的径向基过程神经网络判别模型应用到油田优势渗流场判断识别的实际问题中,取得了良好的应用结果。本课题在面向油田优势渗流场判断识别模型及算法研究的基础上,结合油田生产开发数据管理情况进行软件原型系统的设计开发,并实际部署应用。课题的研究为油田生产开发提供了一种新的优势渗流场判断识别方法,可为指导油田开发控水稳油方案的设计与封堵措施调整提供依据和参考。
[Abstract]:After long-term water-drive development, a large number of major oilfield development blocks have entered a high water cut period. Due to the long-term erosion of the injected solution to the reservoir, there is a dominant seepage channel between oil wells and water wells, which leads to the ineffective circulation of the injected solution along the dominant seepage channel, which results in a large amount of waste and reduces the production efficiency. In order to control water and stabilize oil and improve the production efficiency of oil field, some measures must be taken to seal the dominant seepage channel, and the key premise of the design of measure scheme and the optimization adjustment of injection-production relationship are to identify the dominant seepage channel effectively. By studying the dynamic and static data during oilfield development, the formation mechanism of predominance seepage field and the influencing factors, on the basis of analyzing the traditional judgment and identification method, the paper synthetically uses oilfield geological data, development data and test data. Based on the dynamic and static data of reservoir production history, a method and model of oilfield dominant seepage field identification based on radial basis function (RBF) process neural network and evolutionary algorithm is proposed. The data of oilfield development geological parameters and production process are taken as input. The output is whether there is an advantage seepage field between injection-production wells. The process input data are regularized by Walsh transform and the performance parameters of the process neural network model are optimized by using chaotic genetic algorithm to improve the adaptive ability of the model and the accuracy of classification and recognition. The data used in this paper are all derived from the monthly data of injection wells and production wells in some blocks of Daqing Oilfield based on mathematical model processing. By analyzing the factors affecting the formation of percolation channels, the parameters used for model training are selected and determined. The index parameters selected from the actual historical data of oilfield development are pretreated and the training sample set is established. The trained radial basis process neural network discriminant model is applied to the practical problem of judging and identifying the dominant seepage field in oil field. Good application results have been obtained. Based on the research of recognition model and algorithm of oilfield dominant seepage field, the software prototype system is designed and developed based on the data management of oilfield production and development, and the application of software prototype system is put into practice. The research provides a new method to judge and identify the predominance seepage field for oilfield production and development, and provides the basis and reference for guiding the design of water control and oil stabilization scheme and the adjustment of plugging measures for oilfield development.
【学位授予单位】：东北石油大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TE312;TP183

【参考文献】