Publications: August 2012 Archives

Forecasting is an important analysis technique to support decisions and functionalities in many application domains. While the employed statistical models often provide a sufficient accuracy, recent developments pose new challenges to the forecasting process. Typically the available time for estimating the forecast models and providing accurate predictions is significantly decreasing. This is especially an issue in the energy domain, where forecast models often consider external influences to provide a high accuracy. As a result, these models exhibit a higher number of parameters, resulting in increased estimation efforts. Also, in the energy domain new measurements are constantly appended to the time series, requiring a continuous adaptation of the models to new developments. This typically involves a parameter re-estimation, which is often almost as expensive as the initial estimation, conflicting with the requirement for fast forecast computation. To address these challenges, we present a framework that allows a more efficient integration of external information. First, external information are handled in a separate model, because their linear and non-linear relationships are more stable and thus, they can be excluded from most forecast model adaptations. Second, we directly optimize the separate model using feature selection and dimension reduction techniques. Our evaluation shows that our approach allows an efficient integration of external information and thus, an increased forecasting accuracy, while reducing the re-estimation efforts.

The so-called smart grid is emerging in the energy domain as a solution to provide a stable, efficient and sustainable energy supply accommodating ever growing amounts of renewable energy like wind and solar in the energy production. Smart grid systems are highly distributed, manage large amounts of energy related data, and must be able to react rapidly (but intelligently) when conditions change, leading to substantial real-time business intelligence challenges. This paper discusses these challenges and presents data management solutions in the European smart grid project \Mirabel. These solutions include real-time time series forecasting, real-time aggregation of the flexibilities in energy supply and demand, managing subscriptions for forecasted and flexibility data, efficient storage of time series and flexibilities, and real-time analytical query processing spanning past and future (forecasted) data. Experimental studies show that the proposed solutions support important real-time business intelligence tasks in a smart grid system.

Time series forecasting is challenging as sophisticated forecast models are computationally expensive to build. Recent research has addressed the integration of forecasting inside a DBMS. One main benefit is that models can be created once and then repeatedly used to answer forecast queries. Often forecast queries are submitted on higher aggregation levels, e.g., forecasts of sales over all locations. To answer such a forecast query, we have two possibilities. First, we can aggregate all base time series (sales in Austria, sales in Belgium ...) and create only one model for the aggregate time series. Second, we can create models for all base time series and aggregate the base forecast values. The second possibility might lead to a higher accuracy but it is usually too expensive due to a high number of base time series. However, we actually do not need all base models to achieve a high accuracy, a sample of base models is enough. With this approach, we still achieve a better accuracy than an aggregate model, very similar to using all models, but we need less models to create and maintain in the database. We further improve this approach if new actual values of the base time series arrive at different points in time. With each new actual value we can refine the aggregate forecast and eventually converge towards the real actual value. Our experimental evaluation using several real-world data sets, shows a high accuracy of our approaches and a fast convergence towards the optimal value with increasing sample sizes and increasing number of actual values respectively.