A unified inverse modeling framework for whole-building energy interval data: Daily and hourly baseline modeling and short-term load forecasting

A considerable amount of literature on the application of inverse methods to building energy data has been published in the last three decades. These inverse models serve a variety of purpose such as baseline modeling for monitoring and verification (M&V) projects at monthly, daily, and hourly time scales; condition monitoring; fault detection and diagnosis; supervisory control; and load forecasting, to name a few. Usually, these models have distinct model structures, and separate models are developed depending on the specified need.This paper proposes a novel inverse modeling framework that attempts to unify some of the above application-specific models by allowing a single model to be incrementally enhanced. Specifically, we begin by clustering the energy interval data of a building, identifying scheduling day types and removing any outliers. This aspect of the analysis is described in the companion paper by Jalori and Reddy (2015). Subsequently, the first level is to identify models for each day type using daily average values of energy use and climatic variables; this is adequate in many M&V projects. These models are then extended to hourly time scales by including additional terms in the model that capture diurnal variations of the climatic variables and the building hourly scheduling about the daily mean value; this level is appropriate for M&V and for condition monitoring. Finally, periodic autoregressive terms are added to the model to enhance prediction accuracy for short-term load forecasting, useful for demand response programs, or for short-term supervisory control. The application of the proposed methodology is illustrated with year-long data from two different buildings, one synthetic (the Department of Energy medium-office prototype) building and an actual office building. INTRODUCTION Existing buildings account for a significant portion of the global energy consumption and present a tremendous opportunity to reduce the global energy footprint. The first step towards identifying this potential is baselining the existing building’s energy consumption before assessing the impact of any energy conservation measures (ECMs). This is conveniently done using inverse models that are simple to develop, require fewer inputs for model identification, and can capture changes in energy consumption based on changes in climatic variables and occupancy (ASHRAE 2013). This makes them easy to use and appealing to building managers for evaluating any general retrofits, implementing building energy condition monitoring routines, implementing ongoing commissioning, and performing short-term load forecasting. Historically, the field of building energy performance has evolved over the years and the trends can be broadly classified as the following: • Self-help analysis methods: The first wave of tool development included rather limited and general-purpose building energy analysis methods wherein monthly utility bills along with some on-site measured data were used to develop techniques for evaluating building energy use and informing energy-saving decisions, such as ECM identification and verification of ECMs after they have been installed (monitoring and verification [M&V]). • Customized tools and services: The next trend was the commercialization and customization of the methods and techniques developed earlier, and these were used by large energy service companies. However, only large  2015 ASHRAE. THIS PREPRINT MAY NOT BE DISTRIBUTED IN PAPER OR DIGITAL FORM IN WHOLE OR IN PART. IT IS FOR DISCUSSION PURPOSES ONLY AT THE 2015 ASHRAE ANNUAL CONFERENCE. The archival version of this paper along with comments and author responses will be published in ASHRAE Transactions, Volume 121, Part 2. ASHRAE must receive written questions or comments regarding this paper by July 20, 2015, for them to be included in Transactions. Saurabh Jalori is an energy analyst at Atelier Ten, NewYork, NY. T. Agami Reddy is the SRP professor at the Design School and the School of Sustainable Engineering and the Built Environment, Arizona State University, Tempe, AZ. PREPRINT ONLY. Authors may request permission to reprint or post on their personal or company website once the final version of the article has been published. industries or commercial enterprises could afford these services. • Big data cloud-based tools: The current trend in the industry is to leverage the recent evolution in various fields such as data mining and information technologies, along with widespread availability of smart meter data, database storage, and management systems. Data analysis techniques includes methods beyond traditional statistics and involves machine learning and artificial intelligence techniques that culminate in pre-packaged solutions that combine years of research and make these tools available to the mass market via cloud computing. This paper proposes a unified modeling methodology that combines various application-specific models to be combined together and allows a single model to be incrementally enhanced. More specifically, the model can be used for the following scenarios: • Predicting daily energy consumption of an existing building post-retrofit to evaluate the potential savings due to the application of ECMs • Predicting hourly energy consumption of an existing building for condition monitoring, ongoing commissioning, and fault detection • Short-term load forecasting for utility demand response management purposes As a preliminary step to the proposed inverse methodology, the energy interval data has to be clustered for identifying different day types, removing any outlier data points, and determining a normalized hourly operational schedule over each of the day types. The clustering methodology has been presented and discussed in a companion paper (Jalori and Reddy 2015).