油田机采过程高精度建模与生产优化应用研究

发布时间：2018-04-29 19:47

本文选题：机采过程 + 无迹粒子滤波　；参考：《西安石油大学》2016年硕士论文

【摘要】：油田是产能大户,也是耗能大户。机械采油是油田主要耗能方式,但其效率普遍不足30%,若每台采油设备节省一点能耗,则效益惊人。如何进一步提升抽油机井采油技术和管理水平成为油田普遍关心和亟待解决的关键问题。数字化油田的发展使井上井下布置了大量检测装置,记录了丰富详实的工况参数与产量、能耗数据,这意味可由数据挖掘技术,从大量生产数据中挖掘采油工艺潜在规律,并用数学模型描述;再通过智能优化技术从获取的工艺规律中寻找最佳的生产参数,以使得机采系统实时保持最佳运行状态,实现节能增效。为此,本文针对油田机采系统数据挖掘技术和生产参数智能优化技术的关键科学问题展开深入研究,以油田机采系统为研究对象,通过理论研究、仿真实验及软件开发促进油田机采系统实现自主建模、智能优化和自动决策,具体包括以下内容:(1)提出基于无迹粒子滤波神经网络(UPFNN)的油田机采系统动态演化建模方法。建立精准的机采工艺模型是实现生产优化的前提。由于机采系统受机械、地层、人为等不确定因素影响,难以准确掌握其生产参数、环境变量与系统性能之间的变化关系,为此本文提出利用无迹粒子滤波实时更新神经网络的权值和阈值,建立基于无迹粒子滤波神经网络子空间逼近的机采系统非线性动态演化模型。该方法利用无迹卡尔曼滤波对粒子进行估计,产生重要性密度,并更新粒子,以提高提高粒子滤波精度,从而改善神经网络建模精度。(2)提出基于偏好多目标优化的油田机采过程生产参数优化方法。油田为实现油藏的科学、合理化开采,通常需要根据油藏分布从全局上设计出各采油区在一段时间内的开采量(即给定生产制度)。因此,机采系统优化不能以采油量最大和用能最低为目标,而应该以采油量接近某一给定值和用能最低作为优化目标。此外,油田机采系统生产优化是在各种约束条件下求取目标函数的最优值,属于复杂的非线性优化问题,应用传统优化理论往往遇到困难。带精英策略的非支配排序遗传算法通过计算个体之间的拥挤度来回避共享参数指定问题,并采用精英策略保存父代种群的优秀个体,可实现多目标并行优化。这使得其在处理工业过程问题复杂、高维、难以解析得到的优化模型时比传统优化方法更具优势。为此,本文首先结合无迹粒子滤波神经网络建立的机采过程模型和面向生产制度的偏好函数,构建偏好多目标优化模型,然后采用带精英策略的非支配排序遗传算法求解生产参数的Pareto解集,再通过有序加权获得Pareto解集上每个方案的综合评价,最终获得最佳方案。(3)开发数字化油田抽油机群调度优化决策支撑系统。为实现理论指导实际生产,本文将上述理论研究通过C#与MATLAB混合编程方式开发出一套数字化油田抽油机群调度优化决策支撑系统,并植入油田机采生产管理平台,实现了机采系统的自主建模、智能优化和自主决策。
[Abstract]:Oil field is a large capacity and a big energy consumer. Mechanical oil production is the main energy consumption mode of oil field, but its efficiency is generally less than 30%. If each oil production equipment saves a little energy consumption, the benefit is astonishing. How to further improve the oil extraction technology and management level of pumping well becomes the key problem of common concern and urgent solution in the oilfield. A large number of detection devices are arranged in well on the well, and abundant and detailed working condition parameters and output and energy consumption data are recorded. This means that data mining technology can be used to excavate the potential law of oil production process from a large number of production data and describe it with mathematical model, and then the best production is found from the process rules obtained by intelligent optimization technology. In order to keep the optimal operating state of the mechanical production system in real time and achieve energy efficiency and increase efficiency, this paper studies the key scientific problems of the data mining technology of oil field mining system and the intelligent optimization technology of production parameters, and takes the oil field mining system as the research object, and promotes oil through theoretical research, simulation experiment and software development. The field machine mining system realizes independent modeling, intelligent optimization and automatic decision making, which includes the following contents: (1) a dynamic evolution modeling method based on the Untraced particle filter neural network (UPFNN) is proposed for the dynamic evolution of the oil field production system. The establishment of a precise process model is the prerequisite for the production optimization. With the influence of certain factors, it is difficult to accurately grasp the relation between the production parameters and the changes of the environment variables and the system performance. Therefore, this paper proposes to use the non trace particle filter to update the weights and thresholds of the neural network in real time and establish the nonlinear dynamic evolution model of the machine mining system based on the subspace approximation of the Untraced particle filter neural network. The non trace Calman filter is used to estimate the particle, generate the importance density, and update the particle to improve the precision of the particle filtering and improve the precision of the neural network modeling. (2) the optimization method of the production parameters of the oil field production process based on a lot of target optimization is proposed. The distribution of oil reservoirs is designed for a period of time (the given production system) in a period of time. Therefore, the optimization of the production system can not be aimed at the maximum oil production and the lowest energy use, but the production capacity should be close to a given value and the lowest energy use as the optimization target. The optimal value of the objective function under the constraint condition is a complex nonlinear optimization problem. It is difficult to apply the traditional optimization theory. The non dominated sorting genetic algorithm with the elite strategy avoids the shared parameter assignment problem by calculating the crowding degree among individuals, and uses the elite strategy to preserve the outstanding individuals of the parent population. Multi objective parallel optimization is realized. This makes it more advantageous than the traditional optimization method when dealing with the complicated industrial process problem, high dimension and difficult to parse. For this reason, this paper first combines the process model of the non trace particle filter neural network and the preference function facing the production system, and constructs a lot of objective optimization. The model, then using the non dominated sorting genetic algorithm with elite strategy to solve the Pareto solution set of the production parameters, and then through the ordered weighting to obtain the comprehensive evaluation of each scheme on the Pareto solution set, and finally get the best scheme. (3) developing the optimization decision support system for the scheduling optimization of the digital oilfield pumping unit. In this paper, a set of digital oilfield pumping unit scheduling optimization decision support system is developed through the mixed programming of C# and MATLAB, and the production management platform of oil field production is implanted. The autonomous modeling, intelligent optimization and independent decision of the machine production system are realized.

【学位授予单位】：西安石油大学
【学位级别】：硕士
【学位授予年份】：2016
【分类号】：TE35

【相似文献】