Integrating Energy Management of Autonomous Smart Grids in Electricity Market Operation

This study presents a market operation model integrated with energy management programs of independent smart grids using bilevel optimization. In this framework, autonomous smart grid entities, in the lower levels, operate their own networks and send decisions to the upper level market operator that clears the day ahead market based on unit commitment and second order conic AC power flow models. A single-leader multi-follower game is thus developed, in which every smart grid derives optimal schedules of its own renewable energy resources, storage devices, and responsive demands that are interconnected through a distribution grid using mixed integer linear programming. Given the mixed integer nature of the upper and lower level decisions, we develop and customize an exact reformulation-decomposition method to compute this bilevel optimization program. Through numerical experiments performed on three test systems, we demonstrate that the proposed modeling paradigm can accurately represent the physics of the transmission and distribution grids and achieves reasonable results with significant computational efficacy.

I. Introduction

With the improved control and communication capabilities of today’s power system infrastructures and diverse demand side technologies, it becomes feasible to replace the current power system with a more flexible one. From this and given the presence of intelligent technologies of distribution clients, more elaborate models of both market and smart grids in market operation are critical to reflect the actual network physics and attain accurate pictures of the system. Nevertheless, linear dispatch models still remain the mainstream formulation to solve the day-ahead and real-time markets because they are fast enough under such circumstances. A common alternative for the linear dispatch model is the AC optimal power flow (OPF) variant which accounts for alternating current’s mathematical complexities but is non-convex and needs much more time to solve using existing methods.