A methodological analysis of consumption patterns and anomaly detection in prepaid and postpaid metering systems: an N-BEATS-driven predictive modeling approach

Abstract

The introduction of the prepaid metering system has caused severe dissatisfaction among the masses, claiming that there was a noticeable spike in the bills of their monthly electricity bill. This issue was addressed in this thesis paper through a rigorous investigation, comparing the existing prepaid metering system with the previously used post-paid metering system to identify the underlying causes of this discrepancy. Related datasets were collected through the Dhaka Power Distribution Company (DPDC), mainly focusing on the Paribagh, the first area introduced with this intelligent prepaid meter system. The dataset includes 22,000 individual customers’ billing information, and then a robust survey was conducted to accumulate 1797 households’ information on appliance usage, family size, and other relevant factors. To predict the prepaid and postpaid consumption trends, we incorporated the deep learning model N-BEATS, which efficiently forecasts time series data. We also conducted several operations on our prepaid and postpaid data to better address the issue. Moreover, we deployed the isolation forest model to address the anomalies that would align with the underlying cause of the dissatisfaction of the masses. To assist in validating our survey data, we deployed several data validation methods, such as the Kolmogorov-Smirnov test, and Pearson’s correlation. Furthermore, Pearson’s correlation technique has been used to demonstrate the correlation between prepaid and postpaid appliance usage. With N-BEATS providing a crystal difference between prepaid and postpaid data, Isolation Forest also directed us to the irregularities that may lean towards the reason for the increased billing. To address customers’ growing concerns, the authority might find our findings fruitful in solving the irregularities to ensure a seamless transition to the prepaid system. This research also illustrates actionable recommendations to ensure a satisfactory customer experience with transparent energy billing.