Chenda Liao

Agent-based and graphical modelling of building occupancy

We propose a novel stochastic agent-based model of occupancy dynamics in a building with an arbitrary number of zones and occupants. Simulation of the model yields time-series of the location of each agent (a software representation of an occupant). The model is meant to provide realistic simulation of occupancy dynamics in non-emergency situations. Comparison of the models prediction of distributions of random variables such as first arrival time of a building is provided against those estimated from measurements in commercial buildings. We also propose a lower complexity graphical model of occupancy evolution in multi-zone buildings. The graphical model captures information on mean occupancy and correlation among occupancy at various zones in the building. The agent-based model can be used in conjunction with building performance simulation tools, while the graphical model is more suitable for real-time applications, such as occupancy estimation with noisy sensor measurements.

An integrated approach to occupancy modeling and estimation in commercial buildings

The problem of real-time estimation of occupancy in a commercial building (number of people in various zones at every time instant) is relevant to a number of emerging applications, such as green buildings that achieve high energy efficiency through feedback control. Due to the high deployment cost and large errors that people counting sensors suffer from, measuring occupancy throughout the building accurately from sensors alone is not feasible. Fusing sensor data with model predictions is essential. Due to the highly uncertain nature of occupancy dynamics, modeling and estimation of occupancy is a challenging problem. This paper makes two contributions toward addressing these challenges. We develop an agent-based model to simulate the behavior of all the occupants of a building, and extract reduced-order graphical models from Monte-Carlo simulations of the agent-based model. The agent-based model is validated with sensor data for the special case of one room and one occupant. Noisy measurements from a few sensors are fused with the graphical model predictions using the classical LMV estimator to estimate room-level occupancy in the building. Simulations illustrate the effectiveness of the proposed method.

Identification of multi-zone building thermal interaction model from data

Constructing a model of thermal dynamics of a multi-zone building requires modeling heat conduction through walls as well as convection due to air-flows among the zones. Reduced order models of conduction in terms of RC-networks are well established, while currently the only way to model convection is through CFD (Computational Fluid Dynamics). This limits convection models to a single zone or a small number of zones in a building. In this paper we present a novel method of identifying a reduced order thermal model of a multi-zone building from measured space temperature data. The method consists of first identifying the underlying network structure, in particular, the paths of convective interaction among zones, which corresponds to edges of a building graph. Convective interaction among a pair of zones is modeled as a RC network, in a manner analogous to conduction models. The second step of the proposed method involves estimating the parameters of the RC network model for the convection edges. The identified convection edges, along with the associated R and C values, are used to augment a thermal dynamics model of a building that is originally constructed to model only conduction. Predictions by the augmented model and the conduction-only model are compared with space temperatures measured in a multi-zone building in the University of Florida campus. The identified model is seen to predict the temperatures more accurately than a conduction-only model.

Distributed clock skew and offset estimation from relative measurements in mobile networks with Markovian switching topology

Abstract We analyze a distributed algorithm for the estimation of scalar parameters belonging to nodes in a mobile network from noisy relative measurements. The motivation comes from the problem of clock skew and offset estimation for the purpose of time synchronization. The time variation of the network was modeled as a Markov chain. The estimates are shown to be mean square convergent under fairly weak assumptions on the Markov chain, as long as the union of the graphs is connected. Expressions for the asymptotic mean and correlation are also provided.

Time-synchronization in mobile sensor networks from difference measurements

We examine distributed time-synchronization in mobile ad-hoc and sensor networks. The problem is to estimate the skews and offsets of clocks of all the nodes with respect to an arbitrary reference clock. Pairs of nodes that can communicate with each other can obtain noisy measurements of the relative skews and offsets between them. We propose a distributed algorithm with which each node can estimate its offset/skew from these noisy relative measurements by communicating only with its neighbors. The algorithm is simple and easy to implement. We model the change in the communication network due to the moving nodes as a Markov chain whose state space is the set of graphs that can occur. Using tools from Markov Jump Linear Systems, we provide a sufficient condition for the mean square convergence of the estimation error. A conjecture on mean square convergence under weaker conditions is discussed. Monte Carlo simulations are provided that corroborate the predictions and justify the conjecture.

A novel stochastic agent-based model of building occupancy

We propose a novel stochastic agent-based model of occupancy dynamics in a building with an arbitrary number of zones and occupants. Simulation of the model yields time series of the location of each agent (occupant) over time, from which a time-series of occupancy (number of people in each zone) can be determined. The model is meant to provide realistic simulation of occupancy dynamics in non-emergency situations, which can be used for extraction of reduced order models of occupancy dynamics for estimation and control purposes. Comparison of the models prediction of mean occupancy, and distributions of random variables such as first arrival time, are provided against those estimated from measurements in a commercial building.

Experimental evaluation of occupancy-based energy-efficient climate control of VAV terminal units

Results are presented from a nearly week-long experimental evaluation of a scalable control algorithm for a commercial building HVAC system based on real-time measurements of occupancy obtained from motion detectors. The control algorithm decides air flow rate and amount of reheat for each variable air volume terminal box based on real-time measurements of occupancy and space temperature. It is a rule-based controller, so the control computations are simple. The experiments showed that the proposed controller resulted in 37% energy savings over baseline on average without sacrificing indoor climate. In contrast to prior work that reports energy savings without a careful measure of the effect on indoor climate, it is verified that the controller indeed maintains indoor climate as well as the buildings baseline controller does. This verification is performed from measurements of a host of environmental variables and analysis of before/after occupant survey results. A complete system required to retrofit existing buildings with the controller is presented, which includes a wireless sensor network and a software execution platform. Two useful observations from this work are: (i) considerable energy savings—along with compliance with ASHRAE ventilation standards—are possible with simple occupancy-based control algorithms that are easy to retrofit; and (ii) these savings are attained with binary occupancy measurements from motion detectors that do not provide occupancy-count measurements. Results also show that there is a large variation in energy savings from zone to zone and from day to day.