Speaker
Details
Abstract
Renewable energy sources that exhibit high volatility and intermittency, such as wind or solar, are often paired with energy storage devices to improve the efficiency and reliability of the energy systems that incorporate them. To realize the full potential of the system, we must optimize the control policy to determine the best possible energy allocation decisions in an uncertain environment. Energy storage problems appear in many variations, characterized by the configuration of the system (which determines the controls) and the nature of the different types of uncertainties, such as prices, variability of wind and solar, and the behavior of the loads on the system.
In this dissertation we consider two very different storage control problems. First, we optimize the control of a single storage device connected to a wind farm with a forecasted power output and the larger power grid. The highly volatile electricity prices and actual wind power output make this a difficult problem. In another application, we optimize the operation of a large power grid with grid-level storage devices and large amounts of offshore wind. The challenges in this problem include the high dimensionality of the resource state (the storage devices) and the need to develop policies that are cost-effective, yet robust enough to avoid power shortages in undesirable wind power scenarios. Using two different algorithms developed for each problem setting, backward approximate dynamic programming for the first case and risk-directed importance sampling in stochastic dual dynamic programming with partially observable states for the second setting, in combination with improved stochastic modeling for wind forecast errors, we develop control policies that are more cost-effective and robust compared to policies developed with standard algorithms and modeling assumptions.