An efficient energy supply policy and optimized self-adaptive data aggregation with deep learning in heterogeneous wireless sensor network

Heterogeneous wireless sensor networks (HWSNs) are energy-constrained networks. Data aggregation can conserve the energy of HWSN. Clustering protocols and data processing can be used at individual nodes to reduce the amount of transfers and extend the network's lifespan. Considering these advantages, the proposed research introduces an efficient energy supply and data aggregation using effective techniques. Initially, cluster head (CH) election and data transmission are done using an information entropy based-clustering algorithm (IECA). After successful data transmission, an efficient energy supply scheme is enabled between cluster members (CMs) and sink nodes. Then, data aggregation is performed in CH using Planar Flow-Based Variational Auto-Encoder-based data aggregation (PF-VAE-DA). Before performing data aggregation, the useless and redundant data is compressed using a Long-short-term-memory-based auto-encoder (LSTM-based auto-encoder). The compressed data is aggregated in CHs. Before transferring the aggregated data to the sink, efficient data stream collection is performed to equalize the data size utilizing self-adaptive adjustment of sliding window size (SASWS). Finally, the optimal path is selected to transmit the aggregated data from CH to the sink. The performance of the proposed method is evaluated for various performance metrics. The aim of the proposed study is to enhance the accuracy of sensing data by introducing a novel deep learning-based data aggregation approach. This will extract significant features from vast amounts of data and carry out data aggregation. In addition, to improve the dependability of aggregated data transfer, an effective Energy Supply Policy based on data transmission patterns is implemented. The results show that the proposed method outperforms other methods in terms of network energy consumption, packet delivery ratio (PDR), packet dropping ratio, data aggregation rate, transmission delay, and network lifetime. The proposed approach uses 50% less energy than the other methods. The model's transmission delay ranges from 0.1 to 0.4 s as the number of nodes increases. The proposed network contains 282 active nodes at the 400th round, which is much more than the existing networks.