Multi-agent stochastic simulation of occupants for building simulation

This paper introduces a new general platform for the simulation of occupants’ presence and behaviours. Called No-MASS (Nottingham Multi-Agent Stochastic Simulation platform) the platform takes a selection of well validated stochastic models to generate a synthetic population of agents, predicts their presence and, in the case of residences also their activities and inferred locations, as well as their use of windows, lights and blinds. A social interaction framework is used to emulate negotiations amongst the members of diverse populations. Furthermore, machine learning techniques allow the agents to learn dynamic behaviours that maximise energy and/ or comfort rewards. This is complemented by a belief-desire-intent framework for the representation of less sophisticated behaviours for which data is scarce. Using the Functional Mockup Interface (FMI) co-simulation standard No-MASS is coupled with EnergyPlus: EnergyPlus parses environmental parameters to No-MASS which in turns parses back the energetic consequences of agents behaviours. Simulations demonstrating the range of results that No-MASS can produce are undertaken and presented.