A Novel Method CNN-LSTM Ensembler Based on Black Widow and Blue Monkey Optimizer for Electricity Theft Detection

Abstract

Enhanced metering infrastructure is a key component of the electrical system, offering many advantages, including load management and demand response. However, several additional energy theft channels are introduced by the automation of the metering system. With data analysis techniques, adapting the smart grid significantly reduces energy theft loss. In this article, we proposed deep learning methods for the identification of power theft. A three-stage technique has been devised, which includes selection, extraction, and classification of features. In the selection phase, the average hybrid feature importance determines the most important features and high priority. The feature extraction technique employs the ZFNET method to remove the unwanted features. For the detection of electric fraud, Convolutional Neural Network based Long Short Term Memory (CNN-LSTM) is utilized. Meta-heuristic techniques, including BlackWidow Optimization (BWO) and Blue Monkey Optimization (BMO), are used to calculate optimized values for the hyperparameters of CNN-LSTM. The tuning of hyperparameters of the classifier helps in better training on data. After extensive simulation, our proposed methods CNN-LSTM-BMO and CNN-LSTM-BWO achieved an accuracy of 91% and 93%. Our proposed methods outperform all the existing compared schemes. The performance of our models has attained high accuracy and low error rate. Furthermore, the statistical analysis also shows the superiority of the proposed methods.