A GENETIC ALGORITHM-BASED ADAPTIVE COVERAGE METHOD USING HEXAGONAL AREAS FOR WSN WITH DEEP REINFORCEMENT LEARNING

Abstract

Coverage optimization remains a central challenge in Wireless Sensor Network (WSN) design, directly shaping both monitoring quality and overall network performance. Because sensor nodes operate under strict energy budgets, prolonging network lifetime without sacrificing sensing coverage has become a pressing design objective. Static deployment strategies, however, tend to degrade in irregular terrains and dynamic operating conditions, where node failures and shifting energy levels continually alter the coverage landscape. This paper proposes a hybrid optimization framework for adaptive WSN coverage that combines a real-coded Genetic Algorithm, a machine-learning fitness surrogate, and deep reinforcement learning within a hexagonal cell-based coverage model. The hexagonal tessellation provides a geometrically efficient basis for evaluating sensing redundancy and coverage gaps, while the genetic algorithm performs global search over node placement to jointly minimize overlap, movement cost, and connectivity loss; a lightweight neural network surrogate reduces the computational burden of repeated fitness evaluation during this search. Beyond the initial deployment, a parameter-shared deep Q-network enables individual nodes to respond autonomously to coverage holes created by node failure, without requiring the global optimizer to be re-executed. A deterministic repair mechanism further guarantees full sensing coverage and network-wide connectivity to a hybrid fixed-and-mobile sink infrastructure, closing the gap between soft, learning-based optimization and hard operational requirements. Simulation results on a 160 m × 160 m field with 100 randomly deployed nodes show that the proposed framework sustains high coverage ratios under node attrition, recovers connectivity after failures without full redeployment, and scales effectively to dense sensor networks.