Auto Encoder and PMU Applied Electricity Theft Detection in Smart Grids

Electricity is being stolen at an alarming rate in almost every country around the world. As a result of this, the country's economy will be impacted. Power theft remains as a major challenge for power distribution companies due to the resultant monetary and energy losses. Electricity can be stolen in a variety of methods, including tampering with energy meters or tapping wires at the customer end. Due to the severity of this theft, manually investigating every incident is a time-consuming task. As a result, there is an emerging need for automatic detection of power theft. To automate the power theft detection process, this study has designed and developed a novel methodology that uses data from smart grid meters and AE-PMUs to detect the incidents of electricity theft. Here, feature selection technique is employed to discover the crucial factors in electricity theft detection method. This Preprocessing phase employs data normalization, missing value interpretation, and data cleansing. The Principal Component Analysis (PCA) method is used to identify the most relevant features before training the model. The proposed technique outperforms the existing GRU and LSTM models.