The outer relationship was projected to inner latent variables space with orthogonal components by the partial least squares（PLS）algorithm while simultaneously compressing data blocksSome advantages of using this approach as part of control system design include automatic decoupling and efficient loop pairingWhile the method could not erase the correlation of latent variables completely，PID controller would not be competent enough to handle the controlA methodology was proposed for controller design with optimization and control algorithm in the subspace defined by dynamic latent variable models，and its application to a pressurized breast box and a distiller was illustrated to show how this strategy works.