Deep Learning-Based Consumer Profiling from Smart Meter Data

Smart meters with their automated communication system allows to collect fine-grained real time energy consumption data. By analyzing the smart meter data, the utility and customer can arrange the energy consumption in a significant way. This paper proposes a method to analyze salient characteristics of the demand side using smart meter data obtained from Ceylon Electricity Board (CEB). The proposed method is based on a deep learning approach. Customer classification is done using supervised learning algorithms and further clustering process is done using unsupervised learning algorithms. The results are also mapped with a survey carried out by CEB for validation.

I. Introduction

Most countries in the world have an increasing demand for electricity. With that, traditional electricity meters are now replaced with smart meters for more accuracy, efficiency, features, security and visibility in electricity measurements [1]. Therefore, smart meters have been deployed globally to automatically communicate information about their real time electricity consumption to consumers and service providers. The Advanced Metering Infrastructure (AMI) is designed using smart meters, together with the communication network and data management system and it enables bi-directional information flow [2].