IMPLEMENTATION OF K-NEAREST NEIGHBORS ALGORITHM FOR ENERGY DISAGGREGATION BASED INTERNET OF THINGS

• Main Author: Inayah, Inayatul
• Format: Theses
• Language: Indonesia
• Online Access: https://digilib.itb.ac.id/gdl/view/50117
• Institution: Institut Teknologi Bandung

The increase in world energy consumption is not equal to the amount of energy available. Meanwhile, the availability of fossil energy cannot be renewed and will be depleted by continuous exploration, so it is necessary to take energy management actions that involve users. In this research, a monitoring system for household electrical energy consumption has been built with the energy disaggregation method based on the Internet of Things (IoT). The purpose of this study is to monitor loads and determine the electrical energy consumption of household appliances. The model validity test was carried out on five incandescent lamps connected to the kWh meter PZEM-004T and the WiFi module ESP8266. The kWh meter will measure the power of five lamp sets at a time and will be read by the reader module. After that, the power data will be sent to the server address with the MQTT protocol, then sent to Node-RED to be stored in the database. Then the data is taken from the database to be processed using a disaggregation algorithm. The machine learning model built is the K-Nearest Neighbors (KNN) Algorithm with power and time input parameters and lamp state output parameters, with the Minkowski metric and the number of closest neighbors (k) is 9. The model successfully detects and predicts lamp activity with an accuracy of 100%. Then the model was developed to disaggregate household appliances contained in the AMPds dataset. In the AMPds dataset, data preprocessing is carried out to determine the state of each device that becomes the model's output parameter. Data preprocessing is performed by clustering using the K-Means algorithm. The data classified into 2 clusters (state), namely the On state and the Off state. After that, load disaggregation was carried out on 5, 6, and 7 devices, with an accuracy of 96.20%, 91.55%, and 83.66%. Furthermore, energy consumption analysis for 30 days is carried out to determine the cost structure of the equipment used.