NEURON-LEVEL THROUGH SUBPOPULATION EVOLUTIONARY ALGORITHM

Computers & Systems Dept., Faculty of Engineering, Mansoura University

The work proposed in this paper concerns an effective genetic algorithm that works on

two levels, neuron-level and blue prints networks-level. The present algorithm differs from

standard evolutionary algorithms in that it evolves a population of neurons instead of

complete networks. These neurons are combined through different subpopulation to form

hidden layers of feed-forward networks that are then evaluated on a given problem. The

applicability of evolutionary reinforcement learning techniques was investigated through

the proposed algorithm. Bioreactor Process control simulation benchmark is considered,

which exhibit highly nonlinear behaviour and consequently complex regions of operation.

This simulation study include developing optimal neurocontroller for the reactor. The

neurocontroller developed was found to be robust as a result of generalization afforded by

the presented learning algorithm.