发电机组汽门系统的多模型自学习控制  

Multi-model self-learning control for turbine valving control

作　　者：袁小芳[1] 王耀南[1] 孙炜[1] 吴亮红[1] 

机构地区：[1]湖南大学电气与信息工程学院,长沙410082

出　　处：《控制与决策》2007年第7期769-773,共5页Control and Decision

基　　金：国家自然科学基金项目(60375001);高校博士点基金项目(20030532004).

摘　　要：针对发电机组的非线性、大范围运行等实际问题,研究了用于汽门系统的多模型自学习控制(MMSC).首先根据各种工况下的样本数据归纳出模糊控制规则;然后由模糊聚类算法将多种工况约简为典型工况,得到相应的子模型模糊控制器(FLC).以子模型FLC输出的加权集成作为MMSC的控制输出,而加权系数取决于子模型匹配度.在子模型FLC学习优化中,由支持向量机离线逼近模糊规则曲面,再由梯度下降算法在线自学习.仿真实验验证了所设计控制器的优良性能.For the problem that turbine valving control of synchronous generator faces practical challenges as nonlinear characteristics and changing operation points, a multi-model self-learning control （MMSC） system is proposed. Fuzzy control rules for turbine valving control at variable operation points are derived from operation samples. Then a fuzzy clustering algorithm is employed to reduce variable operation points to typical points, and sub-model fuzzy logic controller （FLC） is obtained. The control output of MMSC is the sum of sub-model FLC multiplying respective weights which are decided by the matching degree of typical points using fuzzy logic. For the self-learning of submodel FLC, support vector machines are used to approximate fuzzy rules curve oft＇line firstly, then a gradient descend algorithm is used for online learning. Simulations show the capability of the proposed controller.

关 键 词：智能控制 汽门控制 模糊控制 支持向量机 自学习 

分 类 号：TP18[自动化与计算机技术—控制理论与控制工程]