A design of evolutionary recurrent neural-net based controllers for an inverted pendulum

In this paper, the method of acquiring a suitable strategy from imperfect observation inputs used a real-coded genetic algorithm and a recurrent Elman neural network, is proposed. The recurrent Elman neural network is suitable for learning the time series data. The weight parameters and the parameters of sigmoidal functions in the recurrent Elman neural network are optimized based on the imperfect observation inputs by using the real-coded genetic algorithm. The recurrent Elman neural network is used as the inverted pendulum stabilizing controller.