by AllbeSmart
Summary
The B-Model experiment uses the FIESTA platform to gather large volumes of IoT observations, which are then used to validate advanced Machine Learning algorithms for the prediction of energy consumption in office buildings and data centres. The experiment uses data from two different testbeds:
- Real DC: a data centre with very high-power consumption, where energy consumption depends mostly on computing resources’ usage, cooling needs and weather;
- Smart ICS: an office environment with personal monitoring of energy usage, where energy consumption depends mostly on human occupancy.
To feed the prediction algorithms with multidomain data, this experiment collects heterogeneous data, such as historical energy consumption, cooling temperature, outdoor weather and building’s occupancy across the RealDC and the SmartICS testbeds.
Two forecast algorithms for energy consumption have been implemented and validated in this experiment: Deep Learning LSTM (Long Short-Term Memory) univariate and multi-variate models. This experiment has shown that the multi-variate model outperforms the univariate model because is able to exploit underlay correlations between IoT data coming from heterogenous sources.
A dashboard with a set of visualization tools helps to understand these environments, the associated energy consumption and the performance of these two different predictive algorithms.
This experiment is also validating some key features of the FIESTA platform, namely:
- Testbed-agnostic access to different resources, which means that the hosting testbed is irrelevant to the resource access method;
- Unique platform entry point, which means that all the resources of FIESTA-IoT platform are only accessible through the only entry point with a validated set of credentials.
Thanks to this experiment, ALLBESMART will kick off the development of new products for smart buildings and operational intelligence derived from the analysis of data integrated across multiple and heterogeneous IoT data streams.