Non-model based finite-horizon sampled-data optimal control via global extremum seeking

A non-model based approach to finite-horizon optimal sampled-data control of continuous-time nonlinear systems subject to input saturation constraints is proposed. The approach adapts the global optimisation method DIRECT to computing a periodically piecewise constant system input which minimises a possibly non-quadratic cost involving the system state and input. In the case where several similar plants are available, multi-agent extremum seeking control is considered for the purpose of expediting the search for an optimal control law within a tolerable margin of errors. An example based on pendulum dynamics illustrates the results.