Fault Detection in Wireless Sensor Networks Using Horse Herd Algorithm and Convolutional Neural Network with Attention Layer

Main Article Content

Ridha Mohammed Alfoudi, Mohsen Nickray.

Abstract

Reliable and real-time detection of faults in Wireless Sensor Networks (WSNs) is significant for the ongoing flow of critical data despite being a strenuous task. In this article, we present a comprehensive WSN-suitable fault detection system as a solution to this challenge. The primary step involves a thorough pre-processing of data such as partitioning, data cleaning, reordering of sample windows, and normalization through the min-max technique. These steps are fundamental to preparing the dataset, while also assisting systematize the proposed solution. A Horse Optimization Algorithm (HOA) integrated with a Convolutional Neural Network is at the heart of the method. As such, the CNN is able to capture highly advanced spatial and temporal features embedded in the processed data, as its convolutional layers are proficient in pattern extraction. Further, hyperparameter optimization of CNN learning rates, batch sizes, and the total number of convolution layers is performed using HOA, to improve the CNN performance. WSNs can greatly benefit from this, with the proposed model identifying faults with 99.47% accuracy on the test dataset, and 99.63% accuracy on the training dataset. These results show the proposed method’s effectiveness and accuracy in addressing fault detection issues in WSNs towards enabling better stability and performance within the network.

Article Details

Section
Articles