Predicting success of energy savings interventions and industry type using smart meter and retrofit data from thousands of non-residential buildings

Abstract

This paper discusses the creation of targeting and segmentation information about non-residential buildings that are equipped with advanced metering infrastructure (AMI) meters, or smart meters. Statistics, model, and pattern-based temporal features are extracted from over 36,000 smart meters. They are then merged with a database of past energy efficiency interventions such as lighting, HVAC, and controls retrofits from 1,600 buildings. The buildings are divided into Good, Average, and Poor performing classes according to consumption from before and after the retrofits. Classification models are developed that improve the ability to predict retrofit success and standard industry class by 18.3% and 27.6% respectively over baselines. This study serves as an example of better leveraging smart meter data from non-residential buildings for utility targeted incentive programs. The methodology outlined is preliminary and further models and temporal features are to be tested. CCS CONCEPTS • Information systems → Data analytics; • General and reference → Empirical studies; • Applied computing → Multi-criterion optimization and decision-making;

State-of-the-art on research and applications of machine learning in the building life cycle

“…Table 9 summarizes the machine learning algorithms used in papers that are reviewed in this section. [132], [140] Clustering and ANN [19] Clustering [133] Random Forest [134] Linear Regression [135] Linear Regression and PCA [136] Gradient Boosting [137] Linear Regression and Clustering [138] Gradient Boosting and Clustering [139] ANN [141] Evaluate energy conservation measures ANN, SVM, K-Nearest Neighbors, and Linear Regression [143] ANN [144] Gradient Boosting and Linear Regression [145] Characterize buildings Random Forest [150] SVM [151] CNN [152] Boosted Decision Trees [153] SVM [154] 6.1 Identify retrofit potential Usually, building retrofit planning requires detailed energy audits, which are time-consuming. As the audit data accumulates, machine learning methods will be feasible to explore the underlying patterns of the data to support the generalization of the retrofit planning to large building stocks.…”

“…Temporal features of energy consumption can also be used to predict the success of the retrofit. With smart meter data of 1600 buildings that have ECMs implemented in between, temporal features such as shape and magnitude behavior of buildings can be extracted to predict the success level of the ECMs with some machine learning classifier [134].…”

Data-driven methodologies for evaluation and recommendation of energy efficiency measures in buildings. Applications in a big data environment

“…A first interesting data-driven solution to this problem was provided by Walter and Sohn, who used the data contained in the United States Building Performance Database to build a multivariate regression model able to estimate changes in energy usage intensity (EUI) due to retrofit implementations and depending on building characteristics [104]. More recently, Miller and Meggers have studied the possibility of leveraging smart meter data and building characteristics to predict potential energy savings and characterize buildings, producing very interesting results and demonstrating that there is highly valuable information that can be brought to light by analyzing big data repositories of tertiary buildings [109,180]. Finally, a significant contribution to this new research field has been provided by Xu et al, that implemented a tree-based machine learning algorithm to study a portfolio of commercial buildings located in the US, with the goal of predicting the effect of different retrofit options based on building characteristics and weather data [181].…”