Machine Learning Algorithms Predicting Demand Response

Machine Learning Algorithms Predicting
Demand Response Potential Utilising a
Synthetic Repository
Despoina Christantoni, Dimitrios-Stavros Kapetanakis, Donal P. Finn
University College of Dublin, Ireland
Despoina Christantoni is funded under Programme for Research in Third Level Institutions and co-funded under the European Regional Development Fund (ERDF).
Investing in Your Future
Context
Demand Response
Reference: Sonja van Renssen, Nature Climate Change
Testbed building
• 11,100 m2 floor area
• Key Features:
 Offices, Retail
 Fitness Centre
 50 m Swimming Pool
 Cinema / Theatre
 Debating Chamber
 Meeting Rooms
UCD Sports Centre
Testbed building
Building Energy Simulation Model

BEMS data archived at 15 minute intervals:



Electricity and gas consumption
Zonal parameters
Electricity: MBE: -1.6% & CVRMSE: 10.5%
EnergyPlus model
(Reference: D. Christantoni, S. Oxizidis, D. Flynn, D. P. Finn, Calibration of a commercial building energy simulation model
for demand response analysis, in: Proceedings of BS2015: 14th Conference of International Building Performance
Simulation Association, 2015, pp. 2865–2872.)
DR Strategies
 Scheduled values changed when a DR signal received
 Strategies tested :
 Chilled water temperature adjustment
 Fans (VAV, CAV & on/off)
 Zone air temperature set-point adjustment
DR Strategies
 Energy Management System
 Sensors
 Actuators
 Parametrics
Synthetic Repository
 Demand response potential in 15 minutes intervals
 1 & 2 hours duration events
 Various commencement time
Train and test the machine learning algorithms
Machine Learning Algorithms

Artificial neural networks and support vector machines
 Focus on: predicting the DR potential with real time weather data
Predictive Models Software
IBM SPSS Modeler
Input variables and target
variable are selected
Inserts the dataset
to the stream
Predictive model
Model Builder
Describes the
developed model
Information about
model performance
Reports the outcome
of prediction
Progress Summary
 Synthetic Repository from Testbed Building: Completed
 Development of predictive models: Undergoing
 Work to be completed: By the end of August
Thank you for your attention!
Despoina Christantoni: [email protected]
Dimitrios-Stavros Kapetanakis: [email protected]