IIIT Hyderabad Publications |
|||||||||
|
Non-intrusive load monitoring for low sampled aggregated dataAuthor: Ronak Aghera 20161002 Date: 2022-01-27 Report no: IIIT/TH/2022/21 Advisor:Vishal Garg AbstractEnergy consumption of the residential sector is increasing year after year with economic development, urbanization, and improvement of people’s living standards. In order to achieve more efficient energy consumption, it is crucial to provide appliance-wise energy consumption feedback to the users to save energy. Appliance wise energy consumption feedback allows users to identify faulty appliances, reduce energy demand and identify unnecessary active devices. Individual appliance energy consumption is monitored by two approaches intrusive load monitoring and non-intrusive load monitoring. Intrusive load monitoring uses smart plugs and smart strips on each device to monitor their power consumption.Non-intrusive load monitoring (NILM) estimates the individual appliance power consumption from the main meter aggregated data without using additional sensors. NILM is a cost-effective approach for giving appliance-wise energy consumption feedback. Non-intrusive load monitoring (NILM) is a blind source separation problem that requires a system to estimate the electricity usage of individual appliances from the aggregated household energy consumption. In this thesis, we will discuss about NILMmethods and algorithms on low sampled (1 min) aggregated data in detail and identify the research gaps. To addresses those research gaps, we propose a novel deep neural network-based approach for performing load disaggregation on low-frequency power data (1 min) obtained from residential households. We combine a series of one-dimensional Convolutional Neural Networks and Long Short Term Memory(1D CNN-LSTM) to extract features that can identify active appliances and retrieve their power consumption given the aggregated household power value. We took appliances which are multi-state appliances(washing machine, dishwasher, microwave, and refrigerator ) to test the algorithm. The disaggregation performance of the algorithm is measured using six metrics and compared with five stateof-the-art algorithms. To explore how well the algorithms generalize to unseen houses, the performance of the algorithms was measured in two separate scenarios: one using test data from a house not seen during training and a second scenario using test data from houses that were seen during training. Our neural net achieves better F1 scores (across all four appliances) than state-of-the-art algorithms and generalizes well to unseen houses with a lower number of trainable parameters. The algorithm is designed for low-power offline devices. Empirical calculations show that our model outperforms the state-of-theart on Reference Energy Disaggregation Dataset (REDD) and UK-dale dataset. We have collected the electricity consumption data of 11 houses in India for 19 days as part of the Residential Building Energy Demand in India (RESIDE) Project. The proposed model on RESIDE dataset achieved better F1 scores and generalised well on the unseen houses. Full thesis: pdf Centre for IT in Building Science |
||||||||
Copyright © 2009 - IIIT Hyderabad. All Rights Reserved. |