An Ensemble Detection Model Using Multinomial Classification of Stochastic Gas Smart Meter Data to Improve Wellbeing Monitoring in Smart Cities

Nathan Shone,Casimiro Aday Curbelo Montanez,William Hurst,Dhiya Al-Jumeily

doi:10.1109/access.2020.2964885

Nathan Shone, Casimiro Aday Curbelo Montanez + Show 2 more

Open Access

https://doi.org/10.1109/access.2020.2964885

Copy DOI

Journal: IEEE Access	Publication Date: Jan 1, 2020
Citations: 34	License type: CC BY 4.0

Affiliation: Liverpool John Moores University

Abstract

Fuel poverty has a negative impact on the wellbeing of individuals within a household; affecting not only comfort levels but also increased levels of seasonal mortality. Wellbeing solutions within this sector are moving towards identifying how the needs of people in vulnerable situations can be improved or monitored by means of existing supply networks and public institutions. Therefore, the focus of this research is towards wellbeing monitoring solution, through the analysis of gas smart meter data. Gas smart meters replace the traditional analogue electro-mechanical and diaphragm-based meters that required regular reading. They have received widespread popularity over the last 10 years. This is primarily due to the fact that by using this technology, customers are able to adapt their consumption behaviours based on real-time information provided by In-Home Devices. Yet, the granular nature of the datasets generated has also meant that this technology is ideal for further scalable wellbeing monitoring applications. For example, the autonomous detection of households at risk of energy poverty is possible and of growing importance in order to face up to the impacts of fuel poverty, quality of life and wellbeing of low-income housing. However, despite their popularity (smart meters), the analysis of gas smart meter data has been neglected. In this paper, an ensemble model is proposed to achieve autonomous detection, supported by four key measures from gas usage patterns, consisting of i) a tariff detection, ii) a temporally-aware tariff detection, iii) a routine consumption detection and iv) an age-group detection. Using a cloud-based machine learning platform, the proposed approach yielded promising classification results of up to 84.1% Area Under Curve (AUC), when the Synthetic Minority Over-sampling Technique (SMOTE) was utilised.

Highlights

Fuel poverty remains a prevalent concern [1], [2]; where consumers with long-term health conditions, or individuals living on a low income, can find themselves in the position of whether to keep their homes at a comfortable temperature or pay their energy bills [3]
We propose a novel approach of using gas smart meter data to improve the wellbeing of occupants in residential properties
Smart meter data is a time-series dataset, and as such the majority of investigations focus on techniques that are appropriate for time-series data analytics

Summary

Introduction

Fuel poverty remains a prevalent concern [1], [2]; where consumers with long-term health conditions, or individuals living on a low income, can find themselves in the position of whether to keep their homes at a comfortable temperature or pay their energy bills [3]. Smart city technologies can play a key role in improving the wellbeing of such vulnerable households through use of existing digital technologies [1], [5]. This industry has witnessed important technological developments in the real-time data analytics. An example is the smart meter, a technology that provides real-time consumption information and automates the billing process for the customer and supplier. It is well-documented that the smart meter can play a key role in the reduction in energy poverty.

Methods

Results

Conclusion