Topics
Papers
- 1 proposes an architecture for a smart city simulation platform. It consists of several services including transportation, emissions monitoring, energy consumption, and networking.
- 2 introduces a proof of concept framework for integrating stochastic multi-agent models with building planning and simulation. The aim is to bridge the gap between what is simulated in the planning phase and how the building is actually used.
- It combines stochastic models for occupant behavior, multi-agent reinforcement learning for more complex behaviors, and building simulation.
- The following stochastic models were used.
- Models for occupant activity (i.e., cooking or cleaning)
- Occupant state based on their current activity. (done using a Finite state machine)
- Occupant presence and location
- Occupant metabolic gains based on standard physical and personal parameters
- Window actions (when do agents open windows)
- Shading actions (do agents illuminate or dim the household)
- Lighting (do agents use indoor lights)
- To simplify things, the stochastic models were tested to determine if they had a significant impact on the overall model’s predictive power. Agents are equipped with their own stochastic models to introduce heterogeneity
- It also makes use of the Belief-Desire-Intention model for activities not covered by existing stochastic models .
- Social interaction between agents is governed by agent state. Interactions can lead to conflicts which are resolved through negotiation and voting(i.e., Game Theory). Agents are assumed to be perfectly rational.
- The quality of the simulation is augmented using Reinforcement Learning. RL is used to enable agents to learn actions and develop new stochastic behaviors, especially as an adaptation to environment changes
Links
Footnotes
-
Dong (2021) A Smart City Simulation Platform with Uncertainty ↩
-
Chapman (2017) Multi-Agent Stochastic Simulation of Occupants in Buildings ↩