An optimized unsupervised learning model for optical network unit placement in FiWi networks

Abstract

A key challenge in developing fiber–wireless (FiWi) access networks is determining the optimal positions of optical network units (ONUs) to maximize network performance and minimize development costs. This paper introduces backtracking search-based weighted fuzzy C-means (BWFCM), an optimized unsupervised method for ONU placement in FiWi. The proposed method integrates the local refinement strength of weighted fuzzy C-means (WFCM) with the global exploration capability of the power mutation-based backtracking search algorithm (PBSA). It is enhanced by an adaptive weighting mechanism that dynamically balances load imbalance and average communication distance without manual parameter configuration. The objective is to minimize the average Euclidean distance between ONUs and their primary wireless users while ensuring fair load distribution. Extensive simulations on standard test cases demonstrate BWFCM’s superiority over state-of-the-art ONU placement algorithms, including chaotic local search-based Levy flight distribution (CLSLFD), Harris-Hawks optimization (HHO), marine predators algorithm (MPA), and arithmetic optimization combined with fuzzy c-means (AOFCM). BWFCM achieves competitive load balance across all test cases and reduces the total combined cost compared to its counterparts, with the minimum improvement (0.24%) over MPA and the maximum (12.17%) over HHO. Scalability and complexity analyses confirm BWFCM as a tractable, competitive, and parameter-free solution for ONU placement in next-generation broadband access networks.