Adaptive Feedforward Neural Network Control With an Optimized Hidden Node Distribution

Composite adaptive radial basis function neural network (RBFNN) control with a lattice distribution of hidden nodes has three inherent demerits: 1) the approximation domain RBFNNs is difficult to be determined a priori ; 2) only partial persistence excitation (PE) condition can guaranteed; 3) in general, required number enormous. This article proposes an feedforward RBFNN controller optimized suitably address above demerits. The calculated by K-means algorithm optimally distributed along desired state trajectory. satisfies PE for periodic reference weights all will converge optimal values. proposed method considerably reduces nodes, while achieving better ability. scheme shares similar rationality that classical PID two special cases, which thus seen as enhanced For implemented digital devices, method, manipulator unknown dynamics, potentially achieves performance than model-based schemes accurate dynamics. Simulation results demonstrate effectiveness scheme. result provides deeper insight into coordination and deterministic learning theory. xmlns:xlink="http://www.w3.org/1999/xlink">Impact Statement —Adaptive learns robot when both structures parameters target are advance. Unfortunately, current controllers need large-scale approximate dynamics manipulator, cannot guaranteed converge. this not scale networks substantially alleviate computational burden but also evidently performance. examples show increases accuracy more nine times 35 compared traditional Intuitively, people usually believe dynamic model may achieve best However, model, even further 1.5 times. technology straightforward path engineers, who experts complicated system analysis methods, design robotic