Buildings are typically equipped with smart meters to measure electricity demand at regular intervals. Smart meter data for a single building have many uses, such as forecasting and assessing overall building performance. However, when data are available from multiple buildings, there are additional applications that are rarely explored. For instance, we can explore how different building characteristics influence energy demand. If each building is treated as a random effect and building characteristics are handled as fixed effects, a mixed effects model can be used to estimate how characteristics affect energy usage. In this paper we demonstrate that producing one day ahead demand predictions for 123 commercial office buildings using mixed models can improve forecasting accuracy. We experiment with random intercept, random intercept and slope, and nonlinear mixed models. The predictive performance of the mixed effects models are tested against naive, linear and nonlinear benchmark models fitted to each building separately. This research justifies using mixed models to improve forecasting accuracy and to quantify changes in energy consumption under different building configuration scenarios.

The electricity industry is collecting large volumes of data from various sources. From regional grid demand to individual sensor readings in buildings, there is a wide range of disaggregated data sources requiring new techniques for forecasting and inference. A better understanding of how electricity is being used by consumers has the potential to increase energy efficiency and improve grid planning and management. This thesis presents several novel approaches to understanding these varied data sources. Contributions include advances in hierarchical probabilistic load forecasting; inference and forecasting using smart meter data and building characteristics; and exploratory analysis of building management system data.

Understanding the impact of building characteristics on electricity demand is important for policy and management decision making. Certain building characteristics and equipment may increase or decrease electricity consumption. Due to different operating practices, these impacts on electricity consumption may vary both across the day and across seasons. Quantifying the magnitude and statistical significance of these impacts will help managers and policy makers make better informed decisions. Here we present a mixed effects model to assess the importance of several variables on building electricity consumption. We use smart meter and building attribute data for 129 commercial office buildings. Our building attribute data includes information on installed equipment and meter characteristics of each building. To account for uncertainty in both variable significance and model selection we follow a multimodel inference approach. Demand impact profiles that show the expected change in electricity demand when a characteristic is absent or present are produced for each season. A discussion of the commercial office building characteristics we use and their impact on the daily profile of electricity demand is presented. Our approach has the advantage of only requiring building level demand and characteristic data. No equipment level sub-metering is required. Furthermore, our approach can also be used to quantify changes in electricity consumption caused by other factors that do not directly draw electricity from the grid, such as management decisions or occupant behaviour. We conclude with a discussion of applications for our methodology and future research directions. (C) 2020 Elsevier B.V. All rights reserved.

When forecasting time series in a hierarchical configuration, it is necessary to ensure that the forecasts reconcile at all levels. The 2017 Global Energy Forecasting Competition (GEFCom2017) focused on addressing this topic. Quantile forecasts for eight zones and two aggregated zones in New England were required for every hour of a future month. This paper presents a new methodology for forecasting quantiles in a hierarchy which outperforms a commonly-used benchmark model. A simulation-based approach was used to generate demand forecasts. Adjustments were made to each of the demand simulations to ensure that all zonal forecasts reconciled appropriately, and a weighted reconciliation approach was implemented to ensure that the bottom-level zonal forecasts summed correctly to the aggregated zonal forecasts. We show that reconciling in this manner improves the forecast accuracy. A discussion of the results and modelling performances is presented, and brief reviews of hierarchical time series forecasting and gradient boosting are also included.

Sparse capture-recapture data from open populations are difficult to analyze using currently available frequentist statistical methods. However, in closed capture-recapture experiments, the Chao sparse estimator (Chao, 1989, Biometrics 45, 427-438) may be used to estimate population sizes when there are few recaptures. Here, we extend the Chao (1989) closed population size estimator to the open population setting by using linear regression and extrapolation techniques. We conduct a small simulation study and apply the models to several sparse capture-recapture data sets.