Globally optimal scheduling of an electrochemical process via data-driven dynamic modeling and wavelet-based adaptive grid refinement

Electrochemical recovery of succinic acid is an electricity intensive process with storable feeds and products, making its flexible operation promising for fluctuating electricity prices. We perform experiments of an electrolysis cell and use these to identify a data-driven model. We apply global dynamic optimization using discrete-time Hammerstein–Wiener models to solve the nonconvex offline scheduling problem to global optimality. We detect the method’s high computational cost and propose an adaptive grid refinement algorithm for global optimization (AGRAGO), which uses a wavelet transform of the control time series and a refinement criterion based on Lagrangian multipliers. AGRAGO is used for the automatic optimal allocation of the control variables in the grid to provide a globally optimal schedule within a given time frame. We demonstrate the applicability of AGRAGO while maintaining the high computational expenses of the solution method and detect superior results to uniform grid sampling indicating economic savings of 14.1%.