STEA-QGAMetaLoc:A Method for Indoor Positioning Based on Dynamic Network Optimization and Rapid Environmental Adaptation

Abstract Indoor positioning technology based on WiFi fingerprinting serves as a crucial measurement method for location-based services, yet its practical deployment faces significant measurement challenges, including high uncertainty in fingerprint databases due to environmental dynamics, labor-intensive offline calibration, and poor transferability of measurement models to new environments. To address these issues, this paper proposes a novel measurement framework, STEA-QGAMetaLoc, designed for dynamic network adaptation and accurate location estimation. The core of this measurement methodology involves: 1) A K-MSPCA preprocessing pipeline that fuses K-Means clustering, wavelet decomposition, and Principal Component Analysis to reduce measurement noise and redundancy in Channel State Information (CSI), thereby constructing a high-fidelity fingerprint database with quantifiable quality improvement. 2) A Spatial-Temporal-External Attention (STEA) mechanism that enhances the extraction of physical features from CSI measurements. 3) An Enhanced Quantum Genetic Algorithm (ENS-QGA) combined with Meta-Learning, which jointly optimizes the network structure and hyperparameters. This approach functions as a meta-calibration process, enabling rapid adaptation and reducing the measurement uncertainty and calibration effort when the model is transferred to new, unseen environments. Experimental validation in a local environment demonstrates that the proposed STEA-QGAMetaLoc method achieves an average positioning accuracy of 0.930 meters, corresponding to a 15.16% reduction in localization error compared to traditional Meta-Learning. This framework outperforms both classical and state-of-the-art measurement algorithms, showcasing its robustness and efficiency in addressing core instrumentation and measurement challenges in WiFi-based indoor localization.