混沌量子粒子群碳酸盐岩流体识别模糊系统研究

Fluid identification has been the key and difficult problem in oil and gas exploration area. Especially for the complex carbonate reservoir in China, conventional fluid identification techniques always tend to be unsatisfactory. Thus, the focus on the research of fluid detection is of great importance to the carbonate reservoir exploration in our country. This project will start from numerical and physical models to study the difference between AVO responses of different fluid types. Then the basic theories of fluid identification under the condition of complex pore structure and high velocity of background rocks will be explored, which will provide the theory foudation for the fluid identification of carbonate reservoir fluid identification. Based on the theory foudation, chaos searching method, quantum particle swarm and the theory of fuzzy system will be combined together in order to propose a new colony intelligent optimization algorithm which involves the chaos quantum particle swarm fuzzy system. The mechanism and capability of optimization will be studied from both the theoretical and practical aspectes, for the sake of solving the problems of poor global searching capability and early convergence which appear in the current optimization algorithm. Then the method will be introduced to carbonate reservoir fluid identification which might help to improve the precision of fluid identificaiton. The research will not only help to deepen the knowledge of basic theory of fluid identification of carbonate reservoir, but also might help to provide a new research way for fluid identification.

流体识别一直是油气勘探领域中的重点和难点问题，尤其是针对我国西部复杂碳酸盐岩储层，传统的流体检测手段应用效果都不理想，加强流体探测技术攻关研究，对于指导我国碳酸盐岩油气勘探具有十分重要的现实意义。本项目拟从数值模拟和物理模拟出发，探索在孔隙结构复杂、围岩速度高的情况下流体识别的基础理论问题，从而为碳酸盐岩流体的准确识别提供理论依据。并将混沌搜索机制、量子粒子群和模糊系统理论有机融合，充分发挥三者各自的优势和互补性，研发一种"混沌量子粒子群模糊系统"的新型群体智能优化算法，从理论和实际两方面研究其机理与寻优性能，改善目前优化算法中存在的全局搜索能力差、早熟收敛等问题，并将其引入到碳酸盐岩流体识别当中，以期能有效解决传统流体检测手段进行碳酸盐岩流体识别时存在的问题，提高碳酸盐岩流体的识别精度。本项研究不仅将深化对碳酸盐岩流体识别基础理论认识，同时可望为碳酸盐岩流体识别提供一种新的研究思路。

根据不同流体性质在角度道集上所反映特征的差异，构建了多属性角度叠加数据体组合流体识别因子。并将量子粒子群与模糊神经网络相结合，利用量子粒子群方法来优化模糊神经网络中的连接权值和隶属函数参数，并进行一系列的改进措施，显著提高了算法的全局寻优能力。将近远角度叠加数据体组合流体识别因子作为改进模糊神经网络的输入，流体性质作为输出，同时引入“相控流体识别”的思想，利用碳酸盐岩储集相进行控制，建立了碳酸盐岩流体识别模型。通过塔中实际井区进行验证，证明该方法能够提高流体的识别精度。并且针对模糊C均值聚类(FCM)算法对初值敏感和易陷入局部收敛的缺点，将FCM算法和混沌量子粒子群(CQPSO)算法有机结合，提出了一种基于混沌量子粒子群的模糊C均值聚类(CQPSO-FCM)方法，该方法利用混沌量子粒子群算法来初始化隶属度矩阵，能够有效解决模糊Ｃ均值聚类算法对初始化敏感的问题，并具有很强的搜索全局最优解的能力，有效提高了模糊分类能力。并将改进的模糊C均值聚类算法引入到碳酸盐岩流体识别当中，建立流体性质与叠前弹性参数之间的关系，可给出待辨别样本属于各流体性质的概率分布，可对流体识别的不确定性进行评价。经证实，该方法能够充分利用多维叠前弹性参数，提高了碳酸盐岩流体识别的精度，具有较好的实际应用价值，为充分利用叠前弹性信息实现复杂碳酸盐岩储层流体识别提供了新的研究思路。