中文摘要：

采用建模发现油田机采系统的潜在规律,再利用该规律优化获取机采系统的最佳决策参数,对解决机采系统效率低、能耗大等问题具有重要意义。然而,机采系统受机械、地层、人为等不确定因素的影响,难以掌握其生产参数、环境变量与系统性能之间的变化关系。为此,提出无迹粒子滤波神经网络,并用其建立机采系统的动态演化模型。该方法将无迹卡尔曼滤波作为重要性采样密度,直接通过无迹卡尔曼滤波估算状态向量（粒子）的概率密度函数,从而有效提高滤波精度以及建模精度。通过对某油田机采系统的数据样本实验,表明该方法提高了机采模型的精度,并能对动态系统突变实时跟踪,可有效指导机采系统获取最佳决策参数。

英文摘要：

The modeling is used to discover the potential rules existing in the oil field mechanical plucking system, and the rules optimization is used to acquire the optimum decision parameter, which have the important significance to solve the prob- lems of low efficiency and high energy consumption of the oil field mechanical plucking system. The mechanical plucking system is influenced by the uncertain factors such as the machinery, geological environment and artificial intervention, so it is difficult to master the change relationship among the operation parameter, environment variable and system performance. Therefore, a dy- namic evolution model for the mechanical plucking system based on unscented particle filter neural network （UPFNN） is pro- posed, which takes the unscented Kalman filtering as the important sampling density. The probability density function of the state vector （particle） is estimated with the unscented Kalman filtering to improve the filtering accuracy and modeling accuracy effectively. The data samples experiment of a certain oil field mechanical plucking system was conducted. The results show that the method has improved the accuracy of the mechanical plucking model, can track the dynamic system mutation in real time, and guide the mechanical plucking system for acquiring the optimum decision parameter.