The Neurothermostat: Predictive Optimal Control of Residential Heating Systems - M.C. Mozer, L. Vidmar, R.H. Dodier
@article{Mozer1997,
author = {Mozer, MC and Vidmar, L},
journal = {Neural Information Processing Systems},
language = {de},
month = may,
title = {{The neurothermostat: Predictive optimal control of residential heating systems}},
url = {http://citeseerx.ist.psu.edu/viewdoc/download?doi=10.1.1.49.2620\&rep=rep1\&type=pdf},
year = {1997}
}
Key Points
- A USA whole house control system which controls the thermostat using a neural network
- The thermostat uses real time measurements from an in-situ sensor network to attempt to solve the optimal control problem in which bopth comfort and energy costs are part of the control objective
- Occupancy is predicted using a hybrid neural net / lookup table and a fixed planning horizon and is fed into another net for control decisions
- Compared favourably with 3 other conventional control methods
In order to make comfort and energy costs comparible to minimise upon (and therefore maximise energy savings and comfort) the common metric for both used was dollars.
In the deployment experiment, the planning horizon was set to 120 minutes.
The study shows that adaptive and predictive contorl systems are most cost effective than: contant temperature, occupancy triggered, and setback. However, offsetting the cost of the prediction infrastrucutre and equipment was not considered.