Predictive storage control for a class of power conversion systems

This paper discusses the synthesis of a predictive controller for storage problems in power conversion systems. The control algorithm is based on solving an optimization problem, to optimally schedule the power stored in a storage device so the total efficiency of the system, in which the storage device is embedded, is improved. The model employed in the controller is power-based, including losses for the main components of the system. The losses are modeled by quadratic, linear, and piecewise linear relations. In general the systems for which this approach is applicable will consist of a primary power converter that converts primary power (chemical) to secondary power (mechanical), e.g., for propulsion, but also has a power take-off for a secondary power converter that converts secondary power to tertiary power (electrical) to fulfill the needs of tertiary power users. When using a device to store tertiary power, the actuation of this device can be scheduled to minimize the consumption of primary power. To compute this schedule we formulate and solve a standard QP problem. Piecewise linear relations are handled by embedding in a larger design space. We show that this approach can be effective, because the efficiencies of the converters depend on their workloads. Taking advantage of sweet spots in the efficiency characteristics may improve the total efficiency, depending on the characteristics of the storage device. The storage device achieves these savings by decoupling the consumption of and conversion to tertiary power. It appears that a reduction of the primary power that is used to generate tertiary power is achievable in the application presented. This reduction is determined by the efficiency characteristics of converters and storage device, and by the workloads foreseen. The horizon of the predictive controller has to be large enough to detect possible sweet spots, and therefore will depend largely on the characteristic of the signals that determine the workloads. It is possible to pose the problem so the required horizon is very small.