Improved Neural Network Model Predictive Control Based on DTBO Algorithm: Applied to Manipulator Robot

This study suggests an Improved Neural Model Predictive Control approach based on the metaheuristic Driving-Training-Based-Optimization (INNMPC-DTBO) applied to control a two degree-of-freedom manipulator robot. This control strategy uses two neural network models of the Multi-Layer-Perceptron (MLP) type as a models for predicting the future angular positions of the manipulator robot, and the DTBO algorithm as an on-line solver of the improved NNMPC optimization problem which contain our contribution, which is the adding of a sequence of reduction weighting coefficients over the prediction horizon. To analyze the effectiveness of the suggested approach, a comparative study was carried out by comparing the obtained outcomes and those of distinct techniques, specifically the optimized PID by particle swarm optimization algorithm controller, model predictive control based on a neural network using teaching-learning-based optimization and driving-training-based-optimization. The obtained results demonstrate that the proposed approach offers satisfactory computing time and accuracy, which means that it can control non-linear constrained systems with fast dynamics.