Optimising PSS using learning automata

Stochastic learning automata have been applied to a wide range of problems including image compression, relaxation labelling, real-time process control, multiprocessor scheduling and communications network routing and flow control. As with the present state of the art in neural networks there have been no shortage of potential applications buy many suggestions have been ill-conceived. The author specifies the criteria which should be satisfied to provide a worthwhile learning automata control application. They are as follows: the system must be sufficiently complex and involve large operational uncertainties so that no good dynamic model exists; the system must be amenable to decentralised control. At each location where control can be exercised the number of choices must be limited; and feedback to every decentralised controller must be provided by some random realisation of a global performance criterion. The only application which so far satisfies the above criteria and shows significant promise is the use of learning automata for the adaptive control of routing information in circuit switched and packet-switched communication networks. In the paper the author concentrates on variable structure automata, convergence characteristics, and adaptive optimisation