Home | Publications | Research | Awards | Bio

Projects

Reduced Consumption Prediction

• Electric utilities use Demand Response (DR) to reduce electricity consumption during anticipated peak periods.
• We formalize the problem of reduced consumption prediction.
• We propose an ensemble model that combines diverse base models for prediction.
• We use a single ensemble for all customers to make scalable reduced consumption prediction while achieving prediction accuracy comparable to ensembles learned individually for each customer.

Prediction Using Partial Observations

• While sensors collect data at a very high speed, the data cannot always be sent back to the central nodes at the same high rate due to network limitations or privacy restrictions.
• This leads to "partial data" at the central nodes which is inadequate for fast and real-time prediction and decision-making.
• We use real-time data from only a small subset of influential sensors to do predictions for all sensors, without compromising the prediction accuracy.

Evaluation of Smart Grid Prediction Models

• Prediction models are often evaluated on the basis of "abstract metrics", which may not always be meaningful for end-use domains and applications.
• We propose a set of performance measures to compare models along the dimensions of scale independence, reliability, volatility and cost.
• We include both independent and application-dependent measures.
• We demonstarate their usefulness for three Smart Grid applications: planning, customer education and demand response.

Automated Dynamic Demand Response

• We introduce the notion of Dynamic Demand Response (D²R) as process of balancing supply and demand in real-time and adapting to dynamically changing conditions.
• We present a set of insights into very-short-term predictability of electricity consumption for a diverse large-scale dataset comprising both small residential customers and large buildings in the context of D²R.
• We also deploy a D²R system for the USC microgrid and identify key challenges associated with it.

Emotion Lexicon

• We develop a novel computational method to build an emotion lexicon using the classification system of the Roget's Thesaurus.
• We use features drawn from this lexicon in classification models to automatically categorize sentences in a text into one of Ekman's six basic emotion categories: happiness, sadness, anger, disgust, surprise, fear.

Emotion Recognition in Text

• Human annotators identify emotion category, emotion intensity and the words/phrases that indicate emotion in a corpus of text drawn from blogs.
• We define the annotation scheme and perform annotation agreement study.
• We also develop emotion classification models to identify the type of emotion in text.
• Our emotion-annotated corpus has been widely used by researchers worldwide and is available upon request. If you are interested in obtaining the corpus, please send an email to Prof. Stan Szpakowicz (szpak@eecs.uottawa.ca) and CC me (saman@usc.edu). Please include your background and a brief description of what you intend to do with the corpus.