IMPLEMENTATION OF FACTORIAL HIDDEN MARKOV MODEL (FHMM) FOR INTERNET OF THINGS (IOT) BASED ENERGY DISAGGREGATION

• Main Author: Allfazira, Annisa
• Format: Final Project
• Language: Indonesia
• Online Access: https://digilib.itb.ac.id/gdl/view/50359
• Institution: Institut Teknologi Bandung

Energy is an important element in human life and in economic growth. Every year, world energy demand keeps increasing while the supply is decreasing, thus it is necessary to do energy management by involving user. In this research, a household energy consumption monitoring system for user is built with energy disaggregation method, to know the active and inactive status of each device and its energy consumption. In the process of sending data, Internet of Things (IoT) technology is used hence data can be sent, processed, and displayed on the user interface in real time. Two methods are used to perform energy disaggregation, which are combinatorial optimization and Factorial Hidden Markov Model (FHMM). Combinatorial optimization methods are further divided into three, without preprocessing, with median filter, and with median filter and moving average. The methods used are validated on the load of five lamps. The system built for validation consists of an energy meter for measuring the load power, a server for processing the data, and a user interface for visualizing the data. After validation, the method with best accuracy, which is FHMM, is implemented on AMPDs dataset containing power data of appliances in a household. The implementation of the method is done on five, six, and seven appliances. Afterward, simulation of real time monitoring is done using data of five appliances. The results of validation show that the combinatorial optimization method without preprocessing results in an accuracy of 97,73%, with median filter 98,86%, with median filter and moving average 99,97%, while the FHMM method results in a accuracy of 100%. Meanwhile, the implementation on five appliances results in an accuracy of 88,06%, six appliances 83,59%, and seven appliances 50,75%. Furthermore, the system built is able to do real time monitoring every 1 minute.