by Instituto Politécnico de Castelo Branco
Summary
In the European Union, energy consumption in buildings represents about 40% of the total energy consumption [FIESTA-IoT Consortium, «D5.2 – Experiments Implementation, Integration and Evaluation,» Ed. Flavio Cirillo, 2017] . Accurate energy forecasting models is a key element of the building control and optimization process. However, the prediction of energy usage in buildings and modelling the nonlinear behaviour of the corresponding energy system, are complex tasks due to influential factors such as weather variables, building construction, thermal properties of the physical materials, occupants’ activities and end-users’ behaviours. To address this challenge current research work is mainly focused on machine learning techniques with single time series data, i.e. using only historical energy consumption records.
With more IoT sensors being deployed in buildings and more time series data being gathered, is important to investigate how this new data streams can improve the forecasting capabilities of buildings energy consumption. However, IoT subsystems are usually designed in a vertical logic and structured in independent and closed areas (“IoT silos”) which makes difficult access to heterogeneous sensing data to test the performance of advanced predictive models that combines heterogeneous sources of data.
In this experiment we have exploited the semantic interoperability provided by FIESTAIoT to overcome this issue by using two smart buildings environments: SMARTICS and ADREAM with hundreds of sensor nodes and associated data sets available. A dashboard with a set of visualization tools was developed to help to understand the buildings environment and associated energy consumption. Multivariate predictive models of energy consumption were validated taking advantage of FIESTA-IoT framework.
The main conclusion is that interoperability across IoT heterogeneous sensors can potentiate a better understanding on buildings energy consumption. In fact, there is a significant improvement on the accuracy of the energy consumption prediction when using multivariate time series (e.g. human activity in the building, temperature and historic power consumption). A main requirement to make this possible is an IoT architecture that allows interoperability among IoT data silos, as the one provided by FIESTA-IoT.