《智能控制智能控制智能控制 (46).pdf》由会员分享,可在线阅读,更多相关《智能控制智能控制智能控制 (46).pdf(12页珍藏版)》请在得力文库 - 分享文档赚钱的网站上搜索。
1、Development of Learning ControlDevelopment of Learning Control The imagination and research of learning machines began in the 1950s.The concept of the learning machine emerged at the same time as cyberneticsDevelopment of Learning Control Adaptive and self-learning methods were developed in the 1960
2、s.Learning control was first used in areas of control,pattern classification and communication of aircraft,then in the control of electric power systems and production processes.on-line trainable systemLearningSystemThe system can make various trial-and error,searching,quality-evaluating,decision-ma
3、king and self-revision of a priori knowledge to a certain extent.The response of the controlled system is evaluated by the rewards and punishments feedback so the control algorithms can be improved according to the feedback information.off-line trainable systemTwo types of Learning SystemsPattern Re
4、cognition Based on Learning ControlLearning control theory was studied on the topic of two-fold control and artificial neural networks beginning in the 1960s,(the control mechanism is based on the pattern recognition method.)K.S.Narendra et al.proposed a performance feedback-based correction method
5、in 1962.F.W.Smith presented a bang-bang control method by the use of adaptive technique of pattern recognition in 1964Pattern Recognition Based on Learning ControlAnother way of pattern recognition based learning control is to use linear reinforcement technique to learning control systemsThe third w
6、ay for studying pattern recognition based learning control is to use Bayes learning estimation that was proposed by K.S.Fu in 1965 Wee and Fu proposed the fuzzy learning control system in 1969,and Saridiset al.developed a hierarchically semantic learning method during 1977 to 1982.Repetitive Learnin
7、g Control Uneyama first proposed the repetitive control method in 1978,and used the method to control a robot manipulator.Inoue and Nakano developed the repetitive learning control from the point of view of frequency domain.Arimoto et al.further developed the Uneyamas preliminary research result(so-
8、called revised algorithm)and proposed a learning control method in time-domain,the iterative learning control.Iterative Learning Control Kawamura et al.studied the inverse system,limited reality,sensibility and optimal regulation of the iterative self-learning control Furuta et al.proposed an optima
9、l iterative learning control for multi-variable systems in 1986 Gu and Loh studied a multi-step iterative learning control method that improves the system robustness effectively in 1987Ddifficult selection and extraction of featuresAslow speed of convergenceCcomplex selection of classificatorBlarge
10、amounts of memoryThe disadvantage of early parameter learning control:Connectionist Learning Method Connectionist learning method has transported a new power to learning control since early 1980s.The representative works in Connectionist learning NN-based learning control with reinforcement learning
11、NN-based iterative learning controlNN-based self-learning controlRule-based learning control,etc.Connectionist Learning Method The successful application examples involve controls for inverted pendulum,robot manipulator and robot manipulator as well as flight aircraft.Controls for inverted pendulumrobot manipulator robot manipulator flight aircraftIn addition to the above,do you know the other methods of learning control?What are the things you encounter in life that may be the use of learning control methods?Summary