《基于模糊满意度的多目标优化控制问题研究与应用》由会员分享，可在线阅读，更多相关《基于模糊满意度的多目标优化控制问题研究与应用（80页珍藏版）》请在金锄头文库上搜索。
1、上海交通大学硕士学位论文基于模糊满意度的多目标优化控制问题研究与应用姓名：杨翊鹏申请学位级别：硕士专业：控制理论与控制工程指导教师：李少远2003.1.1摘 要基于模糊满意度的多目标优化 控制问题研究与应用摘 要在复杂系统控制中对系统的控制要求往往是多方面的反映 在控制目标系统性能指标和对系统状态及控制过程中的约束上 形成了带约束的多目标的优化决策与控制问题当控制目标和约束 不是确定量时就成为控制系统中的模糊多目标优化问题有时根 据实际情况决策者对各个目标有不同的优先级要求希望控制结 果对各种目标有不同的满意度这就形成了有优先级的模糊多目标 优化控制问题 现代的优化算法主要是启发式算法但这些算
2、法找寻全局最优 解是以大量的计算为代价的即 NP-hard 问题在控制领域尤其 是对实时性要求比较高的系统中这些方法都不太适用满意优化 的思想是在综合考虑系统约束和目标的情况下寻找控制系统的满 意解它不再拘泥与苛求系统的最优解而是寻找使控制效果达到 满意的解即认为这样的控制结果是可以接受的这就为新优化方 法的引入带来了很大的自由度 在研究这些优化方法特别是模糊多目标优化算法的基础上 本文提出将评价函数法和满意控制思想相结合的方法解决了视觉 机器人全局环境未知时的路径优化问题 机器人路径规划是机器人研究领域里重要的问题在不知道全 局环境信息的情况下要求机器人运动到目标点而总的约束是在 整个过程中
3、机器人不碰到任一障碍物本文的方法是将复杂的总 控制目标分成小的模糊控制目标平衡模糊目标与模糊约束使控 制结果达到满意的效果结合预测控制滚动优化的原理和滚动窗口 的路径规划思想通过评价函数法多目标优化策略进行在线规划 使最后的路径满足一定的指标要求 本文的另一个重要工作是利用变可行域的策略解决了含优先级 要求的模糊多目标优化问题摘 要多目标中的优化和均衡一直是两个矛盾的方面当有模糊目标 和约束存在和有优先级差别要求时就形成了优先级模糊多目标优 化问题如何在满足优先级要求的同时处理好多目标的优化和平衡 一直还没有很好的解决办法 借助于截集的优化思想本文提出 了变可行域的优化策略为模糊目标和约束的隶
4、属函数引入新的可 行域变量并让其一起参与问题的优化这样就可以把优先级模糊 多目标优化问题变为有约束的单目标确定型的优化问题同时通 过选择单目标中的相应参数可以宏观地在多目标的优化和均衡中 做出平衡这种方法在求解中不是使用严格的优先级要求而是允 许存在一定的自由度求解结果是为不同目标找到了不同层次的可 行域进而决定了不同层次的隶属度使得更容易求得高满意度的 解 在这之后本文把这种方法应用到含模糊目标和模糊约束的预 测控制和实际系统的控制问题中进行了一定的仿真研究取得了 很好的效果 最后在总结全文的基础上提出了论文研究过程中的若干思 考和结论对控制领域中优化理论的发展和应用做出展望并提出了 有待进
5、一步研究的几个问题关键词 多目标优化 模糊优化 满意控制 机器人路径规划 多 目标优先级 非线性约束AbstractMULTI-OBJECTIVE OPTIMIZATIONAND CONTROL BASED ON FUZZYSATISFYING APPROACHAbstractIn complex control systems, we have many control demands that may berepresented by the control objectives, system performances and the constraints of thesystem s
6、tate and control procedures. That is the problem of multi-objectiveoptimization and decision making with constraints. If the control objectives andconstraints are uncertain variables, it is the problem of fuzzy multi-objectiveoptimization in control area. In some cases, the decision maker has certai
7、n preferenceof the objectives and hopes that the result may have different level of satisfactions. Sohere comes the problem of multi-objective optimization with multi-priority.Modern optimization algorithms are mainly heuristic. But they get the overalloptimal solutions at the cost of heavy computat
8、ion. That is so called the NP-hardproblem. In control area, especially the real time systems, these algorithms have verylimited application. The satisfying optimization tries to find the satisfying solutions ofthe control systems by taking the objectives and constraints into accountsimultaneously. I
9、nstead of finding the optimal result, it tries to find a solution thatmay satisfy the control system, or equally the result is acceptable. So it makes theapplication of some algorithms possible.In researching of these optimization algorithms, especially the fuzzy multi-objective optimization method,
10、 this paper presents a new method that combines theevaluation function algorithm and the satisfying control idea, which solves theproblem of robot path planning when the overall circumstance is unknown.Path planning is a key issue in robotics, which means finding a proper movingpath for the robot fr
11、om a designated start point to a designed goal point in aAbstractworkspace with obstacles. The path should be collision-free so that the robot canbypass all obstacles safely. This method presented here divides the complex objectiveinto some simple fuzzy objectives so that we can get the satisfying r
12、esult by balancingthe objectives and constraints. Using the rolling optimization in predictive control andthe rolling window method in robot path planning, it takes online optimization by theevaluation function algorithm so the result can satisfy a certain requirement.Another important work of this
13、paper is solving the fuzzy multi-objectiveoptimization problem with multi-priority using the varying domain method.The optimization and equilibrium of multi-objective are two ambivalent factors.If the objectives and constraints are uncertain and there are different priorities amongthem, then here co
14、mes the fuzzy multi-objective optimization problem with multi-priority. How to balancing between the optimization and equilibrium while satisfyingthe request of the objectives has not been solved well. This paper here presents avarying domain optimization algorithm according to the levelmethod. In t
15、hisalgorithm, we introduce new variables for the domains of membership functions ofthe fuzzy objectives and constraints, and then take them into the procedure of theoptimization problem. In this way it is easy to change the multi-objective optimizationinto single objective optimization. Whats more,
16、by choosing the parameters in theobjective, we can balance between the optimization and equilibrium of the objectives.This method does not apply the strict constraint of priority order but allow a certaintolerance, so that the final results are some domains with different level for theobjectives with different priorities. A solution with high memberships will be easilyobtained in this way.After that, this paper makes some simulations by applying this algorithm topredictive control and a
