Improved convolutional neural network combined with rough set theory for data aggregation algorithm

Abstract

Data aggregation is a crucial method to relieve the energy consumption in wireless sensor networks (WSNs). However, how to perform data aggregation while preserving data fidelity and confidentiality is a challenging research task. Since many existing aggregation algorithms have large communication and computation overheads, this paper integrates rough set theory with an improved convolutional neural network, and proposes a novel information aggregation algorithm for wireless sensor network. Firstly, a feature extraction model is designed in our proposed algorithm and then trained in Sink node, where the rough set theory is adopted to effectively simplify information and cut down the tagged dimension. Once these data features from granular deep network are extracted by the cluster nodes, they will be sent to the Sink node by cluster heads, so as to reduce the quantity of data transmission and extend the network lifetime. Qualitative and quantitative simulation results show that compared with existing data aggregation algorithms, the energy consumption of our proposed granular CNN model can decrease obviously and the accuracy of the data aggregation can be effectively improved.