Growing electricity needs among the vast majority of the population
seconded by a voluminous increase in electrical appliances have led to a huge surge in
electric power demands. With thediminishing unit price of electric meters and increase
of loading, it has been observed that a certain amount of electric meters generate faulty
readings after exhaustive usage. This results in erroneous meter readings thereby
affecting the billings. We propose a fault detecting learning algorithm that is trained by
early meter readings and compares the actual meter reading (AMR) with the predicted
meter reading (PMR). The decision matrix generates an alarm if |PMR-AMR|>T;
where T equals the threshold limit. T itself is decided by the learning algorithm
depending upon the meter variance. Moreover, our system also detects if there is any
power theft as such an action would result in a sudden rise in AMR. The learning
algorithm deploys six binary classifiers which reflect an accuracy of 98.24% for the
detection module and an error rate of 1.26% for the prediction module.
Keywords: Smart Meters, Energy Prediction, Fault Detection, Machine Learning
