AI-Assisted Adaptive Clustering for Energy-Efficient Wireless Sensor Networks: A Mathematical Approach

Energy balancing is an important challenge in wireless sensor networks (WSNs). The challenges of WSNs are considerable in terms of maintaining a balanced energy utilization and extending the network lifetime, especially because of the need to operate the networks for extended periods of time without any human intervention and the energy constraints. We introduce A2C-WSN, a novel mathematical framework that integrates machine learning based cluster head (CH) selection and constrained optimization to realize a deterministic energy-balanced clustering. The AI-assisted model used a decision tree classification method for choosing the appropriate cluster heads using the normalized energy value. Then, the A2C-WSN framework used a greedy approach to select the location of the cluster head with a given cluster cardinality. A maximum energy-weighted utility function was also optimized to further optimize energy distribution throughout the network. The A2C-WSN was evaluated when compared with the existing WSN protocols such as LEACH and Energy Aware protocols at various network sizes, N = 50, 100, 150 and 200. The results showed that A2C-WSN has competitive half-node death (HND) performance. In particular, A2C-WSN performed at a higher HND (809 rounds) than Energy Aware (802 rounds), with only a 0.9% difference. Additionally, A2C-WSN successfully transmitted 8.5% more packets than LEACH, and this could be because of the deterministic selection of CH in each round. The ablation study showed that the optimization module helped reduce the death of the last node (LND) by 16.1% compared to uncontrolled choice of CH. Apart from that, cross validation indicated an accuracy of 99.66% ± 0.12% for the classifier. Furthermore, sensitivity analysis showed that the weights on the input features did not significantly affect the performance of the classifier as the difference in LND was less than one round with weights w1 ranging from 0.1 to 0.9. Besides, the computational time of A2C-WSN framework is moderate, which took 18.2 s for 1600 times simulation of all three protocols. The mathematical model offers a promising solution for sustainable IoT applications with predictable and balanced energy management. This work is based on previous clustering research in WSNs and also shows that hybrid approaches of combining machine learning with optimization techniques can successfully combine theoretical guarantees with practical performance enhancements.