Metrology for Energy forecasting on domestic installations with Renewable Energy Systems 

The implementation of a Non-Intrusive Load Monitoring algorithm that meets the majority (if not all) of the following requirements: accuracy (80-90%), no training (not significant occupant effort for the algorithm), near real time capability, scalability (run online and respond to events as they happen), and ability to work with all types of appliances (on-off, finite state, permanent consumers, or variable-power).

Maximum energy savings can be achieved by real time information in the appliance level consumption as opposed to monthly bills or weekly advice on energy usage. On the utility side, providing much finer granularity of information will help to provide more precise demand response programs. Disaggregating the load information will enable the buildings energy management systems to better execute the energy conservation strategies (e.g. scheduling).

ESR05 Natalia pursues the research started previously by Awadelrahman aiming to contribute to the scope of MEAN4SG by monitoring the energy consumption and by providing direct feedback, such as real-time appliance level consumption information, to the consumers, we can achieve significant reduction of energy wastage.

Secondments

1^st secondment

Hypertech (10.2019 – 02.2020)

Achievements

Energy conservation in residential buildings through power disaggregation has been an issue to be addressed. In this sense, providing information on individual appliances consumptions to residents can make them aware of their energy profile and thus influence them to change their consuming behaviour so that to reduce the amount of energy they consume.

• A Non-Intrusive Load Monitoring (NILM) method of disaggregation on which each appliance power demand has been modelled as a Hidden Markov Model (HMM). Based on these trained HMMs of the appliances, the total load disaggregation is modelled as a Factorial Hidden Markov Model (FHMM), and then the single most probable hidden state sequence across all appliances is inferred through the Viterbi algorithm.
• By working in the Literature review of NILM, Machine learning and Probabilistic graphical models, ESR05 did help the implementation of NILM, achieving its further development. At this stage, future research is needed to optimize it
• 2 conference papers (1 expected to be published on May 2020 / 1 in drafting mode)

Thesis title

Metrology for Energy forecasting on domestic installations with Renewable Energy Systems 

Publications

[To be updated]

Current position

Engineer at Hypertech