【作者】 刘华蓥；

【导师】 许少华；

【作者基本信息】 东北石油大学 ， 石油工程计算技术， 2012， 博士

【摘要】 石油工程中的很多问题都可以抽象为优化问题。面对这些复杂优化问题，传统优化方法往往无能为力，于是智能优化方法成为解决复杂优化问题的有效方法，而粒子群优化算法是得到广泛关注和应用的一种智能优化方法。粒子群优化算法计算简单、控制参数少、易于实现、具有较强的鲁棒性，非常适合于求解复杂优化问题，但它有易陷入局部最优、收敛精度不高等缺点。因此，本文对其在求解无约束单目标、有约束单目标和有约束多目标等优化问题时的性能进行了研究与改进，并将改进后的算法应用于几个典型的石油工程优化问题，取得了令人满意的效果。1．本文提出一种基于混沌变异的动态量子粒子群优化算法。该算法根据群体进化因子动态划分子群，当种群进化速度减慢时，对由适应值较差的粒子组成的子群采用混沌变异，并对全局最优位置加一小扰动，以保持种群的多样性，提高算法的全局搜索能力。对典型高维复杂函数的测试结果表明，该算法不易陷入局部极值，收敛速度快，优化效果明显优于混沌优化算法和量子粒子群优化算法，体现出良好的全局优化性能。将该方法与罚函数法相结合，应用于油田注水系统运行调度优化，取得了较好的效果。2．目前最常用的约束条件处理方法是惩罚函数法，但确定适当的罚因子是很困难的，常常需要多次实验来不断调整。本文提出一种基于双适应值的量子粒子群优化算法。该算法将目标函数和约束条件分离，从而赋予每个粒子双适应值，并根据这两个适应值来决定粒子优劣，同时提出保持不可行解比例的自适应策略。数值实验证明该算法在求解精度和稳定性上明显优于采用罚函数的量子粒子群优化算法和其他几种算法。将该方法应用于油田注水管网布局优化设计，取得了较好的优化效果。3．本文提出基于空间划分树的多目标粒子群优化算法。该算法把外部集所对应的目标空间划分为多个单元格，使用空间划分树来索引非空单元，降低了算法的时间复杂度。优先选择拥挤距离密度比最大的粒子作为全局极值，使全局极值的选取更加准确，从而使非劣解集的多样性有了进一步提高。数值实验验证了该方法的有效性。将该方法应用于油品调和优化，取得了较好的优化效果。4．将基于双适应值的量子粒子群优化算法分别应用于分层开发指标动态劈分预测和管道保温优化设计，都取得了很好的优化效果；将基于空间划分树的多目标粒子群优化算法分别应用于配注方案优化和管道保温优化，也都取得了令人满意的结果。更多还原

【Abstract】 Many problems in petroleum engineering can be abstracted to optimization problems.The traditional optimization methods are powerless in dealing with these complexoptimization problems. Intelligent optimization methods have become the effective methodsto solve complex optimization problems. Particle Swarm Optimization (PSO) is an intelligentoptimization method which is concerned and used widely.PSO has simple calculation, less control parameters, easy realization and strongrobustness. PSO is very suitable for solving complex optimization problems. It has theshortcomings of being easy to fall into local optimum and low convergence precision.Therefore its performance is studied and improved in this paper as it is used to solve theproblems of unconstrained single objective optimization, constrained single objectiveoptimization and constrained multi-objective optimization. The improved algorithms areapplied to several typical petroleum engineering optimization problems and satisfactoryresults have been achieved.1. Dynamic Quantum-behaved Particle Swarm Optimization Based on Chaos (CDQPSO)is proposed in this paper. According to population evolution factor, a particle swarm will bedivided dynamically into two subgroups. When the evolution of the population slows down,chaotic mutation will be used to update the particles in the subgroup which is composed ofparticles having worse fitness values, and a small perturbation will be given to the globaloptimal particle to keep population diversity and improve the global searching ability. Thetest results of typical complex high dimension functions indicate that CDQPSO is not easy tofall into local extremum and its convergence speed is high. Its optimization effect is betterthan that of CO and QPSO. It shows good global optimization performance. Combining withpenalty function, better effect is achieved as it is applied to operation optimization of oilfieldwater injection system.2. Currently penalty function is most commonly used to handle the constraints. It isdifficult to determine appropriate penalty factor. It needs to be adjusted through manyexperiments. In this paper, Quantum-behaved Particle Swarm Optimization with DoubleFitness (DFQPSO) is proposed for constrained optimization. Double fitness values aredefined for every particle by separating objective function and the constraints. Whether theparticle is better or not will be decided by its two fitness values. An adaptive strategy is usedto keep a proper proportion of infeasible particles. Numerical experimental results show that DFQPSO is better on precision and convergence than QPSO using a penalty function and afew other algorithms. The effect is good when it is applied to layout optimization design ofoilfield water injection pipe network.3. Multi-objective Particle Swarm Optimization Based on Spatial Partition Tree(SPTMOPSO) is proposed in this paper. The target space, corresponding to archive set, isdivided into many cell-grids. Nonempty cell-grids are indexed by spatial partition tree. As aresult, the time complexity of the algorithm is cut down. The particle, whose density ratio ofcrowding distance is the largest, has priority to be selected as the global extremum. Globalextremum selection is more accurate. Pareto optimal set has better diversity. Numericalexperimental results show that SPTMOPSO is effective. The effect is good when it is appliedto oil blending optimization.4. The effects are good when DFQPSO is applied respectively to dynamic divisionprediction of development indexes and pipeline insulation optimization. The results aresatisfactory when SPTMOPSO is applied respectively to injection allocation schemeoptimization and pipeline insulation optimization.更多还原