Hybrid evolutionary algorithm for identification and control of time varying system

The problem considered is that of identibing and control of an unknown rapidly time varying system from input-output data. In this paper a hybrid evolutionary algorithm is described which attempts to model the rapidly time varying parameters. The question of stabiliw is handled to show the converges condition of the proposed mtthod. Results show that this approach is capable of high accuracy for test problems. In recent years, there has been a rapid development of online process control technique. The use of process computers in the control and optimization of dynamic systems of many industrial application is increasing. This has attracted the attention of many researches toward online identification and control schemes. Much work has been done in the area of identifying time-invariant system. [l] An important application of on-line schemes is when the system parameters are time varying where is absolutely necessary to track parameters variation in real time. Some examples of such application are robotic systems, aerospace systems , chemical reactor system and others [2]. System modeling entails constructing a model which behaves similarly to a system whose structure is unknown based on observed data from systems. However , most of the identification methods , such as those based on least mean squares or maximum likelihood estimates, are search techniques based on gradient descent. It is well known that such approaches often fail to find the optimum solution if the parameters of the system are rapidly time varying [3] , the error function is also constructed to be differentiable. In recent years the capability of trained neural networks for approximating arbitrary input-output mapping can find an important application in devising procedures for the identification of unknown dynamical plants in order to control them. It is found in [4] that applying a neural network based controller could result in drastic over parameterization in the number of coefficient estimation made. Evolutionary algorithm differ from traditional methods. They are not fundamentally limited by restrictive assumptions about the search space , such as assumptions concerning continuity , existence of derivatives , and other matters [5-111. Therefore , Evolutionary algorithms are finding increasing applications in the area of system identification. but one of the significant draw back of these algorithms is the time computation in which limit their applications in real time system , for this reason this paper proposes a hybrid evolutionary algorithm has the ability to overcome the problem of identifying …