面向快速油藏历史拟合的粒子群算法研究

项目名称： 面向快速油藏历史拟合的粒子群算法研究

项目编号： No.61503150

项目类型： 青年科学基金项目

立项/批准年度： 2016

项目学科： 自动化技术、计算机技术

项目作者： 韩霄松

作者单位： 吉林大学

项目金额： 18万元

中文摘要： 本项目拟利用粒子群算法强大全局搜索能力的优势，从代理模型学习策略优化、进化策略优化以及搜索策略优化三方面针对HEB（参数高维、适应度计算高代价、适应度依赖黑盒模型输出）问题开展研究，并将研究成果应用于油藏历史拟合这一典型的HEB问题中。针对油藏历史拟合的特点，利用流形学习对群体降维，从而在低维流形上快速构建回归代理模型，利用代理模型来估计大部分个体适应度，以期克服油藏历史拟合计算代价高的困难；在算法进化和搜索策略优化研究方面，以减少适应度评价为目标，拟构建基于模式理论的新型进化操作和适应度估计策略，在算法中引入聚集和发散算子，使算法在避免早熟的同时尽快收敛。项目成果将对更深入揭示模式理论原理和有效应用该理论具有较强的指导意义，为构建和优化代理模型处理HEB问题提供新颖的思路，也为其它群智能优化方法处理HEB问题奠定理论基础，本项目的顺利实施将为油田的科学开采和可持续发展提供关键技术支持。

中文关键词： 粒子群算法；HEB问题；适应度估计；代理模型；油藏历史拟合

英文摘要： The modern engineering design optimization problems often comprise fitness functions that are based on the output of one or more simulation models, and this kind of problems is named HEB (High-dimensional, Expensive computationally, Black-box) problem. To address this, this proposal will fully take advantage of the Particle Swarm Optimization (PSO)’s powerful global searching ability, and will launch the research from following three aspects: surrogate model optimization, evolutionary and searching strategy optimization. Then, the research results will be applied to a typical HEB problem, reservoir history matching. Based on the process and feature of reservoir history matching, Manifold Learning will be used for PSO’s population dimension reduction, aimed to build regression surrogate model faster and further improve the speed of PSO on the low-dimensional manifold. In respect to evolutionary strategy, in order to reduce times of fitness evaluation, this proposal will construct novel evolutionary operators and fitness estimation strategy based on Schema Theory for PSO. Convergence operator and Dispersion operator will be introduced to speed up convergence and avoid prematurity. This proposal’s results will have the strong guiding sense to further reveal the Schema Theory and apply the theory effectively; provide new ideas to deal with HEB problem by the means of building and optimizing surrogate model; lay the theoretical foundation for solving HEB problem with other swarm intelligence algorithm. The smooth implementation of this proposal will provide key technical support for oilfield scientific exploitation and sustainable development.

英文关键词： Partical Swarm Optimization;HEB Problem;Fitness Estimation;Surrogate Model;Reservoir History Matching