Applying incremental best estimate directed search to optimize fuzzy logic controllers for a ball-and-beam system

Incremental best estimate directed search (IBEDS) is a computational controller optimization algorithm developed by the authors. It is a very fast and effective off-line controller parameter search method. Using the least mean square (LMS) learning algorithm with the training set, another controller with randomly initialized parameters is trained in an iterative procedure. In each iteration, the trained controller is evaluated with cell state space based global and local performance measures. The training set is then updated based on the evaluation with best kept policy. In this way, the training set is optimized incrementally, and the controller trained by the training set is also optimized progressively. IBEDS has been found to have faster convergence speed over other computational method with an inverted pendulum as example. The paper reports the simulation results of applying IBEDS to fuzzy logic controller (FLC) optimization for a ball and beam system. It is shown that it is much easier to control a 4 dimensional ball and beam system than to control a 4 dimensional inverted pendulum system. The results also reveal that all the features of IBEDS that have been found so far on the inverted pendulum example remain the same with the ball and beam system.