TUNING ENERGY PERFORMANCE SIMULATION ON BEHAVIOURAL VARIABILITY WITH INVERSE MODELLING: THE CASE OF SMART CAMPUS BUILDING

In order to establish a reliable model of occupants' interactions with indoor environment, a method accounting for stochastic factors is used. The result is no more a "single value" for the system performance, but a probability to fulfil a certain performance over time. In this way, occupant behaviour is not deterministic (e.g. the opening of windows when indoor temperature exceeds a threshold value) but coupled with a probability to perform an action.The proposed approach is built upon continuous measurements of both indoor environmental parameters and external climate conditions along with the behaviour of the building occupants (such as window opening, thermostat radiator valve, set point temperatures, occupancy sensors, etc.), performed in a sufficient number of areas and rooms representing different interaction zones in the case study. The monitoring period can range from medium (i.e. one week better if repeated in different seasons) to long-term periods (i.e. a year). The simple measurement of time series of physical quantities (such as relative humidity, temperature, pollutant concentrations, luminance, etc.) generates huge amounts of data that can be hard to "translate" directly into a behavioural model. In order to overcome these barriers, different suitable models can be defined using statistical techniques such as logistic regression and Markov chains.The various occupancy profiles are then inserted into the chosen dynamic simulation software. A monitoring and control scheme that detects users' occupancy, lighting levels, temperature, humidity, CO2 and manages HVAC system settings has been designed and installed. This system has been design to enable a faster "as-built" energy model calibration, using data from commissioning and early occupancy phases.Therefore, the actual measurement in the building will give the possibility of extracting useful information to calibrate the building energy model with a probabilistic behavioural model fitted to real metered data through inverse modelling.