Integration of Bitemporal PostgreSQL Temporal Database and Machine Learning for Detecting Backdate Correction in Daily kWh and Voltage Usage Data

Electricity monitoring systems require data storage mechanisms that not only record consumption
values but also preserve historical changes to ensure data integrity and accountability. This paper proposes the
integration of a bitememporal database architecture with machine learning models to support analysis and
prediction of daily electricity consumption data, including kWh and voltage measurements. The bitemporal design
is implemented using PostgreSQL with a system-versioned table approach based on the SQL:2011 standard,
enabling each record to be managed through valid time and transaction time dimensions. This mechanism allows
accurate retrieval of historical data states (as-was) and current conditions (as-is). Data processing is conducted
using Python, involving data cleaning, feature extraction, and normalization. Three machine learning algorithms,
namely Linear Regression, Support Vector Regression (SVR), and Random Forest, are employed to model the
relationship between voltage and energy consumption. Experimental results indicate that the bitemporal schema
effectively maintains complete historical records, while the machine learning models demonstrate reliable
performance in consumption modeling. The proposed approach enhances data transparency, improves
information quality, and provides a robust foundation for developing reliable electricity monitoring systems.