| Summary: | 博士 === 逢甲大學 === 電機與通訊工程所 === 95 === The evolutionary strategy (ES) which is a branch of the evolutionary algorithms is proposed. Its basic features are the distinction between a parent population and an offspring population. Through normally distributed mutations and the incorporation of the self-adaptation principle for strategy parameters, they are essential feature of the ES with regard to its general applicability to complex fitness functions in order to evolve the objective variable effectively.
In order to simultaneously optimize the parameters and structure of controllers or system modules, a DNA computing algorithm is also proposed. DNA computing algorithm (DNACA) with an electron-ion interaction potential (EIIP) decoding scheme is another branch of evolutionary algorithms to deal with the former problems. DNACA involves the basic and traditional operations of evolutionary algorithms such as selection, crossover, mutation and elite to deal with precision of solutions and avoidance of the local solutions. Moreover, the DNACA also has special operations which include enzyme and virus operators to provide a highly modular, flexible, and accurate self-organizing structure environment. Some demonstrative examples of test functions are given to show the merits of the proposed methodology.
For applications, this thesis first presents the evolutionary strategy to deal with the selection of PID controller gains and parameters of the neural net-based compensator for a planar motion platform. The cross-coupling problem of the platform is effectively solved by a neural net-based decoupling compensator with a sufficient condition which can ensure closed-loop stability. Numerical studies and a real-world experiment with a watch cambered surface polishing platform verify tracking performance and applicability of the proposed design.
In order to prove the major merit of DNACA in the field of control systems, identification of a class of the transfer functions, the reduced-order robust controller and the self-organizing PID control design have been proposed. A new coding method makes each DNA chromosome string divide into two fragments, one of which is an information fragment to deal with the structure information of a constructed transfer function which applies to system identification and H-inf controller design, the other is parameter fragments to deal with the presentation of coding parameters. Selection of structure components and the tuning of a constructed transfer function are done with two fragments simultaneously. Moreover, it presents a self-organizing PID control design based on DNACA. The algorithm uses a coding method originated from the structure of biological DNA molecules mapping control gains as well as the control structure (such as P, PI, PD, and PID controller) into DNA strings. Finally, Numerical studies of the above topics are given to show the merits of the proposed approach.
|
|---|
