Learning fuzzy control by evolutionary and advantage reinforcements

In this paper, evolutionary and dynamic programming based reinforcement learning techniques are combined to form an unsupervised learning scheme for designing autonomous optimal fuzzy logic control systems. A messy genetic algorithm, and an advantage learning scheme are first compared as reinforcement learning paradigms. The messy genetic algorithm enables flexible coding of a fuzzy structure for global optimisation, resulting in a coarsely optimised feedforward type neurofuzzy structure. Local pruning and fine tuning of the neurofuzzy system is then achieved effectively by advantage learning by directly interacting with the environment without the use of a supervisor. The methodology is illustrated and tested in detail through application to two nonlinear control systems.