Individual, Aggregate, and Cluster-based Aggregate Forecasting of Residential Demand

Pervasive installation of smart meters opens new possibilities for advanced analytics of electricity consumption at the level of the individual household. One of the important tasks in various Smart Grid applications, from demand-response to emergency management, is the short-term electricity load forecasting at different scales, from an individual customer to a whole portfolio of customers. In this work we perform a quantitative evaluation of different machine learning methods for short-term (1 hour ahead and 24 hour ahead) electricity load forecasting at the individual and aggregate level. We discuss the relevant features that best help to improve forecasting accuracy, as well as the effectiveness of exploiting correlations between different customers. Furthermore, for aggregate forecast, we explore different clustering techniques that can be used to segment the whole customer portfolio and show that forecasting each cluster separately and then aggregating the forecast produces better accuracy compared to the traditional approach (which forecast directly the aggregate load). We also found that the improvement provided by this strategy changes as a function of total customers.

Pervasive Data Analytics for Sustainable Energy Systems

Tri Kurniawan Wijaya

With an ever growing population, global energy demand is predicted to keep increasing. Furthermore, the integration of renewable energy sources into the electricity grid (to reduce carbon emission and humanity's dependency on fossil fuels), complicates efforts to balance supply and demand, since their generation is intermittent and unpredictable. Traditionally, it has always been the supply side that has adapted to follow energy demand, however, in order to have a sustainable energy system for the future, the demand side will have to be better managed to match the available energy supply. In the first part of this thesis, we focus on understanding customers' energy consumption behavior (demand analytics). While previously, information about customer's energy consumption could be obtained only with coarse granularity (e.g., monthly or bimonthly), nowadays, using advanced metering infrastructure (or smart meters), utility companies are able to retrieve it in near real-time. By leveraging smart meter data, we then develop a versatile customer segmentation framework, track cluster changes over time, and identify key characteristics that define a cluster. Additionally, although household-level consumption is hard to predict, it can be used to improve aggregate-level forecasting by first segmenting the households into several clusters, forecasting the energy consumption of each cluster, and then aggregating those forecasts. The improvements provided by this strategy depend not only on the number of clusters, but also on the size of the customer base. Furthermore, we develop an approach to model the uncertainty of future demand. In contrast to previous work that used computationally expensive methods, such as simulation, bootstrapping, or ensemble, we construct prediction intervals directly using the time-varying conditional mean and variance of future demand. While analytics on customer energy data are indeed essential to understanding customer behavior, they could also lead to breaches of privacy, with all the attendant risks. The first part of this thesis closes by exploring symbolic representations of smart meter data which still allow learning algorithms to be performed on top of them, thus providing a trade-off between accurate analytics and the protection of customer privacy. In the second part of this thesis, we focus on mechanisms for incentivizing changes in customers' energy usage in order to maintain (electricity) grid stability, i.e., Demand Response (DR). We complement previous work in this area (which typically targeted large, industrial customers) by studying the application of DR to residential customers. We first study the influence of DR baselines, i.e., estimates of what customers would have consumed in the absence of a DR event. While the literature to date has focused on baseline accuracy and bias, we go beyond these concepts by explaining how a baseline affects customer participation in a DR event, and how it affects both the customer and company profit. We then discuss a strategy for matching the demand side with the supply side by using a multiunit auction performed by intelligent agents on behalf of customers. The thesis closes by eliciting behavioral incentives from the crowd of customers for promoting and maintaining customer engagement in DR programs.

EPFL2015

Cluster-based Aggregate Forecasting for Residential Electricity Demand using Smart Meter Data

Karl Aberer, Samuel François Roger Joseph Humeau, Matteo Vasirani, Tri Kurniawan Wijaya

While electricity demand forecasting literature has focused on large, industrial, and national demand, this paper focuses on short-term (1 and 24 hour ahead) electricity demand forecasting for residential customers at the individual and aggregate level. Since electricity consumption behavior may vary between households, we first build a feature universe, and then apply Correlation-based Feature Selection to select features relevant to each household. Additionally, smart meter data can be used to obtain aggregate forecasts with higher accuracy using the so-called Cluster-based Aggregate Forecasting (CBAF) strategy, i.e., by first clustering the households, forecasting the clusters' energy consumption separately, and finally aggregating the forecasts. We found that the improvement provided by CBAF depends not only on the number of clusters, but also more importantly on the size of the customer base.

2015

Pervasive Data Analytics for Sustainable Energy Systems

Tri Kurniawan Wijaya

EPFL2015