An efficient convolutional neural network based attack detection for smart grid in 5G-IOT

Abstract

The deployment of 5G networks and IoT devices in smart grid applications provides electricity-generated, distributed, and managed bidirectional transmission of real-time information between utility providers and consumers. However, this increased transmission and confidence in IoT devices also present novel security challenges, since they are vulnerable to malicious attacks. Ensuring robust attack detection mechanisms in 5G-IoT smart grid systems for reliable and efficient power distribution, and early accurate identification of attacks addressed. To solve these concerns, a novel technique called Target Projection Regressed Gradient Convolutional Neural Network (TPRGCNN) is introduced to improve the accuracy of attack detection during data transmission in a 5G-IoT smart grid environment. The TPRGCNN method is combined with feature selection and classification for improving secure data transmission by detecting attacks in 5G-IoT smart grid networks. In the feature selection process, TPRGCNN utilizes the Ruzicka coefficient Dichotonic projection regression method and aims to enhance the accuracy of attack detection while minimizing time complexity. Then selected significant features are fed into Jaspen’s correlative stochastic gradient convolutional neural learning classifier for attack detection. Classification indicates whether transmission is normal or an attack in the 5G-IoT smart grid network. The implementation results demonstrate that the proposed TPRGCNN method achieve a 5% of improved attack detection accuracy and 2% improvement in precision, recall, F-score while reducing time complexity and space complexity by 13% and 23% compared to conventional methods.