Abstract

This paper presents TBA-IUKF-DL, a rigorously formulated hybrid framework that couples an Iterated Unscented Kalman Filter (IUKF) with a dual-layer Long Short-Term Memory (LSTM) residual-correction network and the Tiger Beetle Algorithm (TBA) for energy-efficient target tracking in resource-constrained Wireless Sensor Networks (WSNs). Unlike existing LSTM-Kalman hybrids that treat neural correction as a decoupled post-processor, the proposed framework establishes a principled three-way coupling: (i) the IUKF executes iterative sigma-point updates whose convergence is formally proved via Lyapunov stability analysis; (ii) a dual-layer LSTM with 64 hidden units per layer, tanh cell activations, and dropout regularization (p = 0.2) is trained with Adam optimization (η = 0.001, MSE loss) to predict systematic residuals from unmodelled dynamics; and (iii) TBA—a three-phase swarm intelligence method modelling tiger-beetle predation—maximizes a composite information-energy utility function to identify the minimal active node subset at each step. Validation over 100 Monte Carlo trials in MATLAB R2023b on a 50-node, 100×100 m² WSN demonstrates RMSE = 0.726 ± 0.098 m, a 23.7% reduction over TBA-IUKF (p < 0.001), 31.4% energy saving, and 45.2% lifetime extension. The contribution of every component and environmental robustness is confirmed by performing an ablation study and sensitivity analysis respectively.