Adjoint sensitivity results for predictive control, state- and parameter-estimation with nonlinear models

The key-contributions of this paper are computationally efficient methods for sensitivity computation in continuous-discrete nonlinear systems using the adjoint approach. These results are relevant for predictive control in nonlinear systems described by systems of ordinary differential equations and with zero-order-hold parametrization of the manipulated variables as well as for state- and parameter-estimation in continuous-time systems observed at discrete-times. These classes of problems are often referred to as nonlinear model predictive control (NMPC), nonlinear moving horizon estimation (MHE), and parameter estimation in dynamic systems. The procedures for computing the sensitivities in continuous-discrete systems are developed by specializing the adjoint sensitivity result for continuous systems to continuous-discrete systems. Adjoint sensitivity computation is computational efficient for problems with many parameters and states, e.g. optimal control and state estimation of distributed systems.