A new real-time perspective on non-linear model predictive control

This work presents a new formulation of continuous-time non-linear model predictive control (NMPC) in which the parameters defining the input trajectory are adapted continuously in real time. Continuous implementation of the control as the input parameterization is being optimized reduces the impact of computational delay, in particular in response to process disturbances. By eliminating the typical correspondence between the time partitions used for input parameterization and implementation, and instead parameterizing the input over arbitrary intervals of variable length, a means is provided to reduce the overall number of optimization parameters (and hence the dimension of the required gradient and Hessian calculations) without adversely affecting stability or optimality.