Using decision tree to select forecasting algorithms in distinct electricity consumption context of an office building

The flexibility and management in the storage and control of building expertise in the energy optimization can be enhanced with the support of algorithms involved in forecasting tasks. These play an important role on obtaining anticipated and accurate consumption predictions associated to different contexts through extensive consumption patterns analysis. This paper evaluates the most viable forecasting algorithm for consumption predictions of a building in different contexts according to two alternatives: artificial neural networks and k-nearest neighbors. These algorithms use patterns of data from consumptions integrated in different contexts while retaining additional information from sensors data. The different contexts are classified on a sequence of periods that take place from five-to-five minutes. The decision criterion to evaluate which of the two forecasting algorithms is the most suitable in each five minutes periods is supported with decision trees that select the forecasting algorithms that looks to be more suitable followed by a logical answer that clarifies if the selection was the most viable option. Parameterization updates concerning the depth are studied to understand the forecasting accuracy impact. The decision trees approach has the potential to improve the accuracy of prediction as it plays a promising role in decision making.