An enhanced wireless sensor network localization scheme for radio irregularity models using hybrid fuzzy deep extreme learning machines

Localization is one of the key challenges facing wireless sensor networks (WSNs), particularly in the absence of global positioning equipment such as GPS. However, equipping WSNs with GPS sensors entails the additional costs of hardware logic and increased power consumption, thereby lowering the lifetime of the sensor, which is normally operated on a non-rechargeable battery. Range-free-based localization schemes have shown promise compared to range-based approaches as preferred and cost-effective solutions. Typical range-free localization algorithms have a key advantage: simplicity. However, their precision must be improved, especially under varying node densities, sensing coverage conditions, and topology diversity. Thus, this work investigates the probable integration of two soft-computing techniques, namely, Fuzzy Logic (FL) and Extreme Learning Machines (ELMs), with the goal of enhancing the approximate localization precision while considering the above factors. In stark contrast to ELMs, FL methods yield high accuracy under low node density and limited coverage conditions. In addition, as a hybrid scheme, extra steps are integrated to compensate for the effects of irregular topology (i.e., noisy signal density due to obstacles). Signal and weight are normalized during the fuzzy states, while the ELM uses a deep learning concept to adjust the signal coverage, including the spring force error estimation enhancement. The performance of our hybrid scheme is evaluated via simulations that demonstrate the scheme's effectiveness compared with other soft-computing-based range-free localization schemes.