SGER: Framework for Dynamic Stochastic Optimal Power Flow (DSOPF) of the Grid

点击翻译按钮获取中文摘要

A power flow may have many operating limit violations. When such conditions occur, the operator may wish to determine if the trouble could be alleviated by appropriate corrective actions. Optimal Power Flow (OPF), developed from well documented optimization principles in Operation Research, is a most useful method to diminish/alleviate violations, and determine the best allocation of controls/resources to achieve a given objective. This cost-based optimization problem for planning or preventive/corrective control observes several constraints while optimizing the weighted objective sum leading to a multi-objective OPF. Many methods have been developed to solve this type of OPF problem, but several of them fail to solve truly dynamic and noisy systems. As such, the next generation of OPF, which can deal with dynamic problems, foresight, and noise is urgently required by the power systems community. This need has led to an innovative OPF called Dynamic Stochastic Optimal Power Flow (DSOPF). Here, Adaptive Dynamic Programming (ADP) is used as a new method to solve the DSOPF problem. The new ADP technology has enhanced capability and robustness above classical optimization in dynamically changing environments where foresight is important. This award presents an extended optimal power flow as a case for ADP where stochasticity of input data, topology, and harsh system faults are controlled using advanced Operation Research principles of an ADP computational tool.This project will focus on DSOPF to solve a large scale system that cannot be solved using conventional optimization techniques. ADP - different from existing OPF technologies - will contribute to power systems optimization by extending current work in time-scale scheduling of controls, resources, and services. It can deal with system changes and stochastic perturbation over consecutive time intervals. ADP makes the "global optimization" possible where system challenges such as unit commitment, VAR planning and security are solved within any time interval. In this proposal, three different kinds of time-dependant OPF problems are combined to be solved by the DSOPF. The ADP develops on several neural networks used in Operation Research, and a training algorithm, which compares a desired output to the actual output, and generates an error to allow the network to learn. Back-propagation will be used to get necessary derivatives of the error and train parameters and inputs of the network. Overall, this new Operation Research methodology will enable planners and decision makers to gain better insight, and better predict and protect the system under different conditions of uncertainty and stochasticity in data.