Smart Cities

Chapter 9Data-Driven Modeling, Control, and Tools for Smart Cities

Madhur Behl¹ and Rahul Mangharam²

¹Department of Computer Science, University of Virginia, Charlottesville, VA, USA

²Electrical and Systems Engineering, University of Pennsylvania, Philadelphia, PA, USA

Chapter Menu

Introduction

Related Work

Problem Definition

Data-Driven Demand Response

DR Synthesis with Regression Trees

The Case for Using Regression Trees for Demand Response

DR-Advisor: Toolbox Design

Case Study

Objectives

On the surface, DR may seem simple. Reduce your power when asked to and get paid. However, in practice, one of the biggest challenges with end user DR for large-scale consumers of electricity is the following: Upon receiving the notification for a DR event, what actions must the end user take in order to achieve an adequate and a sustained DR curtailment?

This is a hard question to answer because of the following reasons:

1. 1. Modeling Complexity and Heterogeneity: Unlike the automobile or the aircraft industry, each building is designed and used in a different way, and therefore, it must be uniquely modeled. Learning predictive models of building's dynamics using first principles based approaches (e.g., with EnergyPlus [9]) is very cost and time prohibitive and requires retrofitting the building with several sensors [10]; the user expertise, time, and associated sensor costs required to develop a model of a single building are very high. This is because usually a building modeling domain ...