Please use this identifier to cite or link to this item: http://hdl.handle.net/10266/5555
Title: Time Series Forecasting Using SRL Based Deep Learning In Smart Grid
Authors: Singh, Taranveer
Supervisor: Kumar, Neeraj
Keywords: Machine Learning;SRL;AI;Deep Learning;SVM
Issue Date: 1-Aug-2019
Abstract: Time-series forecasting is one of the most challenging tasks due to the ubiquitousness of the Time series data. Some examples include Astronomical data, weather data, Electricity usage, stock and exchange rates data collected over time. Whereas, in many applications, the availability of labeled data is quite less either due to the privacy or low rate of generation of data. As a result of this, the small amount of data leads to low performance and overfitting of the machine learning models. In order to deal with small data and low performance, we implemented a generative model based on statistical relational learning and a two-tier ensemble forecasting model to predict the result based on machine learning and deep learning. Considering the advent and future of the Internet of Things, we choose the smart grid environment to implement the proposed approach because of the availability of the large benchmark dataset UMass Smart* Dataset - 2017 release of smart homes in a locality by taking the reading of appliances and weather conditions with a sampling rate of 1 minute. Use of this dataset helps us in building a robust model that also gets better insights from the data in order to find out the relationship between the device data and the factors affecting the device data to generate the synthetic data from the existing data. The proposed scheme also shows a significant improvement over the existing load forecasting methods over short-term and long-term load forecasting models with an accuracy of 95.6%
URI: http://hdl.handle.net/10266/5555
Appears in Collections:Masters Theses@CSED

Files in This Item:
File Description SizeFormat 
Time Series Forecasting Thesis copy.pdf6.95 MBAdobe PDFThumbnail
View/Open


Items in DSpace are protected by copyright, with all rights reserved, unless otherwise indicated.