Data-Driven Generalized Minimum Variance Control with Autoencoder based Dimensionality Reduction of Input Signals

Abstract

This paper considers data-driven type generalized minimum variance control (GMVC) for p-inputs/q-outputs (p > q) multivariable systems with static nonlinearity. In the proposed approach, an autoencoder, which can extract the feature of input data, is used. First, an encoder converts input data with p dimensions into that with q dimensions. Then, a GMV controller is designed by using the dimension-reduced input data. Finally, the nonlinearity of a plant is compensated by a decoder, which reconstructs the input data with p dimensions. The effectiveness of the presented approach is evaluated using a numerical example.