Stochastic simulation of time irreversible processes and its use in hydrosystem control problems (Keynote talk)

Monte Carlo control methods, in which a control action, possibly expressed in terms of a parametric relationship, is tested by stochastic simulation and subsequently optimized by a global optimization procedure, are promising for complex hydrosystems with nonlinear dynamics. In particular, they have proved powerful for the management of large reservoir systems, where simulation and optimization are performed on a time scale of the order of a month. A basic requirement of these methods is a proper technique for stochastic generation of hydrological inputs, respecting characteristic behaviours of hydrological processes, such as seasonality, intermittence, long term persistence and roughness (fractality). However, most control problems in hydrosystems require time scales of study much finer than monthly, e.g., hourly or even finer. Examples are the control of spillway gates and of hydropower turbines. On fine time scales, another behaviour of hydrological processes becomes important and necessary to reproduce: time irreversibility. As common stochastic techniques produce time series whose properties are symmetric in time, a new stochastic simulation method is presented, which is capable of generating sequences with the required properties related to time irreversibility. The method is generic as it can reproduce any marginal distribution, covariance structure and irreversibility index, and can work both in simulation and forecast mode.