|
|
You are here: Research
| Adaptive Load Forecasting |
 |
Accurate models for electric power load forecasting are essential to
the operation and planning of a utility company. Load forecasting helps
an electric utility to make important decisions including decisions on
purchasing and generating electric power, load switching, and infrastructure development.
Load forecasts are extremely important for energy suppliers, ISOs, financial institutions,
and other participants in electric energy generation, transmission, distribution, and markets.
This study is about developing a system that is able to conduct reliable load forecasts ranging
from 1 up to 7 days leed-time predictions in several areas in Switzerland. The applicable
topics of this project ranges from traditional machine learning up to statistical biophysics
and biocomputation. This project is founded by a KTI Grant and further supported by Meteo Schweiz.
|
| Adaptive Building Intelligence |
|
Intelligent learning systems have become increasingly common due to improvements in
user comfort and security provided by automated robots and self-adapting systems.
The field of autonomous building control systems at the Institute of Neuroinformatics(INI)
has gained my research interest. Novel building architectures have been gradually equipped
with an entire communication network. Now, this establishes a new world of possibilities since
internal building devices can programmed and controlled remotely. Depending on occupants' needs,
such a programmable system could improve user comfort and save more energy than a manually controlled
system. Currently, ordinary lighting controls systems require frequent maintenance such as replacement
or re-calibration of sensors and effectors and furthermore depend on stationary building structures.
Self-adapting systems can prevent these burdensome and costly tasks. However, this seemingly easy
task is complicated since structural changes must be detected and incorporated in real-time.
Here at the INI, we developed a novel adaptive building intelligence (ABI) system that is built on the
Open Services Gateway initiative (OSGi). The system incorporates regular sensors (i.e. presence, temperature,
illumination, humidity)) and effectors (i.e. lights, window blinds, wall-switches) that can be assessed through
a dedicated fieldbus network (LonWorks). The ABI system introduces a generic Multi-Agent-System (MAS) design that
integrates the most credible information currently available from several device agent controllers (DACs).
The building structure itself is a non-stationary environment where not only individual desires are constantly
changing but also the physical structure (e.g. mobile walls in multi-office environments, integration or dismounting
of devices).
However, a major obstacle for autonomously learning dynamic space behaviors and configurations is that
sensors perceive and react very differently from the human brain. For instance, we perceive fog with a different
luminance intensity than measured by sensors. Furthermore, ambient light levels change dramatically even with
small atmospheric changes such as a momentary scattering of particles over the sun. Hence, different skylight
levels can be found even under the same sunlight condition. Consequently, intelligent buildings (IBs) need to be
flexible enough to react to such environmental changes. An additional difficulty is that user instructions are
next to their sparseness not always self-consistent and thus make learning behavior tedious.
|
Peer-reviewed conference proceeding
| 2006 |
Stephan K. Nufer, Mathias Buehlmann, Tobi Delbruck and Josef. Joller,
A mixture of experts for learning lighting control, Accepted to the 1st Workshop on
Artificial Intelligence Techniques for Ambient Intelligence (AITAmI'06) co-located event of the
17th European Conference on Artificial Intelligence (ECAI06), Riva del Garda, Italy, 2006
|
Technical Reports / Diploma Thesis
| 2005 |
Stephan K. Nufer and Mathias Buehlmann. Intelligent, Learning Systems ABI Mark II. - a novel approach for learning dynamic space behaviors in a non-stationary environment with an adaptive Intelligent Building Framework (IBF). Diploma Thesis, University of Applied Sciences Rapperswil, Switzerland and Institute of Neuroinformatics, Swiss Federal Institute of Technology, Zurich, Switzerland
 Download
|
| 2005 |
Stephan K. Nufer and Mathias Buehlmann. Remote Building Control (RBC) Protocol. - a protocol that is used to transfer building data. Technical report, University of Applied Sciences Rapperswil, Switzerland and Institute of Neuroinformatics, Swiss Federal Institute of Technology, Zurich, Switzerland
Download |
| 2005 |
Stephan K. Nufer and Mathias Buehlmann. Remote Building Control (RBC) API. - an API that supports custom agent application to remotely control a building. Technical report, University of Applied Sciences Rapperswil, Switzerland and Institute of Neuroinformatics, Swiss Federal Institute of Technology, Zurich, Switzerland
Download |
| 2005 |
Stephan K. Nufer and Mathias Buehlmann. Intelligent, Learning System. - a new ABI System built on the Open Services Gateway initiative. Term project, University of Applied Sciences Rapperswil, Switzerland and Institute of Neuroinformatics, Swiss Federal Institute of Technology, Zurich, Switzerland
Download |
Posters
| 2005 |
Stephan K. Nufer and Mathias Buehlmann. Intelligent, Learning Systems ABI Mark II. - a novel approach for learning dynamic space behaviors in a non-stationary environment with an adaptive Intelligent Building Framework (IBF). Poster, University of Applied Sciences Rapperswil, Switzerland and Institute of Neuroinformatics, Swiss Federal Institute of Technology, Zurich, Switzerland
Download |
| 2005 |
Stephan K. Nufer and Mathias Buehlmann. Intelligent, Learning Systems ABI Mark II. - a new ABI System built on the Open Services Gateway initiative. Poster, University of Applied Sciences Rapperswil, Switzerland and Institute of Neuroinformatics, Swiss Federal Institute of Technology, Zurich, Switzerland
Download |
|
| Autonomous Speaker Clustering |
|
The goal of a speaker clustering system is to detect the number of speakers or discriminable speaker characteristics in a given audio recording.
Speaker clustering systems can be very useful for the following applications:
- Improvement of the acoustic models of Speech To Text (STT) systems. I.e. the word error rate of STT systems can be reduced through speaker clustering
- Information retrieval: Speaker indexing in speech databases
However, speaker clustering is known to be a very hard problem due to:
- Environmental noise
- Overlapping speech
- Demanding feature extraction due to similar voice characteristics
The main objectives of this research is to study, implement and evaluate different clustering algorithms and feature extraction
techniques that could improve i.e. speech recognition systems by predetermining one of their most vital parameters,
- the number of speakers. This diploma thesis addresses the most general case of speaker clustering, which is clustering of
single channel audio recordings. Thus, methods such as ICA that usually requires the audio data to be recorded with multiple
microphones are usually not applicable.
|
Poster
| 2006 |
Mark Spoerndli, Stephan K. Nufer, PD Ruedi Stoop, and Josef M. Joller. Unsupervised
Speaker Clustering. Poster, University of Applied Sciences Rapperswil, Switzerland and
Institute of Neuroinformatics (INI), Swiss Federal Institute of Technology, Zurich,
Switzerland, 2006.
Download |
|
|
|
|
|
