Implementation of neural networks for optimal control in biotechnical exoskeleton systems

Abstract

Exoskeleton’s dynamic properties are studied on the basis of a mathematical model of an exoskeleton’s verticalisation. Received scalar quantities allow estimating quality of control system work of exoskeleton’s verticalisation. Structural solutions for neural network control of an exoskeleton are made. Genetic algorithms of neural controllers for an exoskeleton’s verticalisation control system are described: variation genetic algorithm and its input parameters; a hybrid genetic algorithm with an iterative process of increase in delays of a recurrent neural network; a hybrid genetic algorithm with the iterative process of the second criterion optimization. Implementation of the genetic algorithm synthesizes multilayered neural networks. Exoskeleton’s verticalisation control with one optimized criterion is carried out on the basis of the hybrid genetic algorithm. It is possible during control of the first criterion in allowed values. Implementation of the hybrid genetic algorithm with the iterative process of the second criterion optimization synthesizes the exoskeleton’s verticalisation control system with two optimized criteria. Experimental studies of quality indicators of exoskeleton’s verticalisation control using neural controllers are conducted. Neural controllers have their own structure and are configured according to developed hybrid genetic algorithms.