Traditionally, the market of energy in the buildings and residential sector is seen as a “necessary evil” and not a core activity for building management and maintenance.
The state of the art of energy management using ICT technology in the “residential and civil industry” is characterized by a traditional monitoring approach which could assess the energy consumption of the building but cannot manage and act the required action to improve the energy management according to a demand-side approach.
The objective of iAE4 (DIH4CPS Project) was the development and application of a cloud platform, SimonLab, to monitor and control energy-consuming assets in buildings. Indeed, the platform aims not only at monitoring consumption but also actuating the equipment. The control strategy is defined as relying on data from sensors and simulations, considering as constraints the comfort for people in the building. The solution has been applied to a real scenario, namely, one of the medical centres of Humanitas Medical Care, the Fiordaliso Medical Centre, situated within a commercial centre in the south of Milan. More specifically, sensors were installed to monitor the main variables related to the comfort of the occupants (temperature, carbon dioxide (CO2) and humidity), energy consumption, and the status of the machines for heating, cooling and air conditioning. Then, an alerting system was implemented to automatically intercept anomalies from the data.
Finally, Machine Learning (ML) models have been developed and applied to simulate the building behaviour and predict the evolution of the parameters of interest. Thus, the monitored and predicted parameters are aggregated to define the optimal control strategy.
The experiment targets concerning model accuracy and energy consumption reduction have been reached, even considering the restrictions introduced due to the pandemic that enforced higher air quality standards:
1. AHU energy consumption reduction
The average monthly electric consumption of the AHU over a year decreased compared to the 2019 consumption of 41% (target value 30%).
2. Polyvalent pump energy consumption reduction
The average monthly electric consumption of the polyvalent pump over a year decreased compared to the 2019 consumption of 13.8% (target value 15%).
3. ML model of building behaviour
The model was validated for the interior temperature on a test dataset of 15 days. The achieved mean absolute error is 0.3°C (target value <0.5°C).
4. Malfunction or discomfort notification
Alarm activation based on configurable thresholds for the following parameters:
● CO2 > 800ppm
● average humidity > 60%
● average temperature > setpoint + 2°C
● average temperature < setpoint – 2°C
Over 2 months, 97% of the alerts were detected (target value >90%).