Machine Learning based Theft Detection by Processing the Smart Meter Data

Abstract

The intentional and illegal use of electricity by various means is referred to as energy theft. Several studies have been conducted using machine learning methods to detect energy theft in advanced metering infrastructure. However, there is a problem with using machine learning for energy theft detection in that it is difficult to obtain enough electricity theft data to train a machine learning model. In this paper, we propose a method for detecting electricity theft in data streams generated by smart meters that are based on anomaly pattern detection. To train the model, the proposed method requires only normal energy consumption data. Previous usage records of customers being monitored are not required for detecting energy theft. This feature makes the proposed method applicable in real-world situations. The significance of the present study relies on collecting and analyzing existing papers to find the exact energy usage records for each customer, develop an algorithm to helps reduce theft detection, implementing a machine-learning algorithm to identify the type of electricity theft behaviors and their properties, and compare between different methods used for theft detection. The significance of the project is that power consumption increases each year, the power generation and distribution industry grow, and the need for technologies to reduce power loss is increasing. Energy theft refers to the intentional and illegal usage of electricity by various means. Therefore, a smart meter is installed in a customer-filled area. Making it nearly difficult for unauthorized individuals to tamper with it. Moreover, to follow the SDG (Sustainable Development Goals) goals the 12th and 16th goals. The 12th goal states, “Responsible Consumption and Production”, and the 16th goal states “Peace, Justice, and Strong Institutions”. Also, according to vision 2030, Prince Mohammad bin Salman stated that the industry will grow, and the power will reduce. The suggested methodology is machine learning and processing the smart meter data. Experiments were carried out using real smart meter data and artificial attack data, including the standardization of daily consumption vectors, the construction of an outlier detection model on normal electricity consumption data of randomly selected customers, and the application of anomaly pattern detection on test data streams.