A hierarchical model predictive control approach for handling demand charges using battery systems.

Applications in energy systems often require to simultaneously mitigate long-term and short-term electricity costs. Demand charges, in particular, constitute an important component of the electricity bills for large consumption units such as buildings and manufacturing plants. Mitigating long-term and short-term costs poses a challenging multiscale planning problem that should make decisions at fine timescales and over long time horizons. This work presents a hierarchical model predictive control (MPC) approach to tackle this problem in the context of stationary battery systems. The goal is to determine the optimal charge-discharge policy for the battery to minimize hourly costs and a monthly demand charge. In the proposed hierarchical MPC approach, the state of charge (SOC) policy is assumed to be periodic, which allows to cast the long-term planning problem as a tractable stochastic programming problem. Here, every period (e.g., a day or week) represents an operational scenario and the targets for the periodic SOC levels and the peak cost are to be determined. The long-term planner MPC communicates the periodic SOC targets and maximum peak level to a short-term MPC controller. The short-term MPC controller determines the intra-period charge/discharge policies (at high resolution) while meeting the targets of the long-term planning. A simulation case study for a university campus is presented to demonstrate that the hierarchical MPC scheme yields optimal demand charge and charge-discharge policy under nominal (perfect forecast) conditions. Comparative studies of the proposed hierarchical MPC scheme and standard MPC schemes that use ad-hoc approaches to handle demand charges are also presented. Under imperfect forecasts, the simulations show that the hierarchical MPC scheme results in significant improvements in demand charge reduction over a standard MPC scheme that uses a discounting factor to capture long-term effects.