Use this URL to cite or link to this record in EThOS: http://ethos.bl.uk/OrderDetails.do?uin=uk.bl.ethos.628754
Title: Algorithms for appliance usage prediction
Author: Truong, Ngoc Cuong
ISNI:       0000 0004 5346 9486
Awarding Body: University of Southampton
Current Institution: University of Southampton
Date of Award: 2014
Availability of Full Text:
Access through EThOS:
Full text unavailable from EThOS. Please try the link below.
Access through Institution:
Abstract:
Demand-Side Management (DSM) is one of the key elements of future Smart Electricity Grids. DSM involves mechanisms to reduce or shift the consumption of electricity in an attempt to minimise peaks. By so doing it is possible to avoid using expensive peaking plants that are also highly carbon emitting. A key challenge in DSM, however, is the need to predict energy usage from specific home appliances accurately so that consumers can be notified to shift or reduce the use of high energy-consuming appliances. In some cases, such notifications may be also need to be given at very short notice. Hence, to solve the appliance usage prediction problem, in this thesis we develop novel algorithms that take into account both users' daily practices (by taking advantage of the cyclic nature of routine activities) and the inter-dependency between the usage of multiple appliances (i.e., the user's typical consumption patterns). We propose two prediction algorithms to satisfy the needs for fast prediction and high accuracy respectively: i) a rule-based approach, EGH-H, for scenarios in which notifications need to be given at short notice, to find significant patterns in the use of appliances that can capture the user's behaviour (or habits), ii) a graphical{model based approach, GM-PMA (Graphical Model for Prediction in Multiple Appliances) for scenarios that require high prediction accuracy. We demonstrate through extensive empirical evaluations on real{world data from a prominent database of home energy usage that GM-PMA outperforms existing methods by up to 41%, and the runtime of EGH-H is 100 times lower on average, than that of other benchmark algorithms, while maintaining competitive prediction accuracy. Moreover, we demonstrate the use of appliance usage prediction algorithms in the context of demand{side management by proposing an Intelligent Demand Responses (IDR) mechanism, where an agent uses Logistic Inference to learn the user's preferences, and hence provides the best personalised suggestions to the user. We use simulations to evaluate IDR on a number of user types, and show that, by using IDR, users are likely to improve their savings significantly.
Supervisor: Ramchurn, Sarvapali Sponsor: Not available
Qualification Name: Thesis (Ph.D.) Qualification Level: Doctoral
EThOS ID: uk.bl.ethos.628754  DOI: Not available
Keywords: QA75 Electronic computers. Computer science
Share: