Neuro-Optimizer: A New Artificial Intelligent Optimization Tool and Its Application for Robot Optimal Controller Design

Abstract

The main objective of this paper is to introduce a new intelligent optimization technique that uses a predictioncorrection<br />strategy supported by a recurrent neural network for finding a near optimal solution of a given<br />objective function. Recently there have been attempts for using artificial neural networks (ANNs) in optimization<br />problems and some types of ANNs such as Hopfield network and Boltzmann machine have been applied in<br />combinatorial optimization problems. However, ANNs cannot optimize continuous functions and discrete<br />problems should be mapped into the neural networks architecture. To overcome these shortages, we introduce a<br />new procedure for stochastic optimization by a recurrent artificial neural network. The introduced neurooptimizer<br />(NO) starts with an initial solution and adjusts its weights by a new heuristic and unsupervised rule to<br />compute the best solution. Therefore, in each iteration, NO generates a new solution to reach the optimal or<br />near optimal solutions. For comparison and detailed description, the introduced NO is compared to genetic<br />algorithm and particle swarm optimization methods. Then, the proposed method is used to design the optimal<br />controller parameters for a five bar linkage manipulator robot. The important characteristics of NO are:<br />convergence to optimal or near optimal solutions, escaping from local minima, less function evaluation, high<br />convergence rate and easy to implement.