Smart grid distribution prediction and control using computational intelligence

Abstract

Smart-grid systems (SGS) may comprise distributed generation, automated demand response, mega-watt scale batteries, and a variety of other utility energy resources and programs. Their physical characteristics and operating flexibility within the distribution grid introduce new challenges to solving the power economic dispatch (PED) problem. Good operation of an SGS requires efficient use of available assets over both the short and long-term. Ideally resources will be scheduled and dispatched to equal loads (demand) in an optimal way, i.e., at the lowest cost, taking into account the differing operating constraints of assets and the changing conditions of the system and its environment. An experimental intelligent controller has been developed as part of the Pacific Northwest Smart Grid Demonstration (PNWSGD) project to address SGS demand-dependent non-linear cost functions, microgrid reliability zone constraints, and dynamic availability states. The controller is embedded within an existing utility control system that provides real time, historical, and forecast data used to predict energy demand for the next 72 hours and to create a near-optimal dispatch schedule for the next 24 hours. Both demand forecasts and schedules are updated every 5 minutes. The modularity of the controller architecture allows for a variety of load forecast and dispatch optimization tools and methods to be used; the current version uses computational intelligence, specifically neural networks. Its generality and versatility provides guidance for development of intelligent controllers adaptable and scalable to a variety of SGS applications.