Model Predictive Control implementation on neural networks using denoising autoencoder

Model Predictive Control (MPC) is an effective control method for nonlinear control systems including quantized control systems; however, the optimization process requires huge computation for such cases and is therefore hard to realize. In this study, a controller design method based on a machine learning technique, in particular a neural network with denoising autoencoder (DAE), is proposed. The simulation results show that the neural controller emulates the behaviors of MPC. The mean and standard deviation of the control result are improved by applying DAE, compared to simple neural network. The proposed method requires short computation time, shorter than 1[ms], therefore it can be applied to fast mechanical control systems with nonlinear characteristics where MPC requires 100[ms] or longer.