Real-time RSS-based target localization for UWSNs using an IDE-BP neural network.

Localization is a vital task in underwater wireless sensor networks (UWSNs). Accurate location information is critical for applications such as environmental monitoring and military operations. Received signal strength (RSS)-based localization schemes are commonly employed because of their cost-effectiveness and performance in terrestrial environments. However, underwater acoustic environments are affected by not only path loss, but also absorption bias in which case these schemes perform poorly with large delays. This makes real-time underwater acoustic localization based on RSS challenging and unable to meet the localization requirements of practical applications. Thus, a real-time RSS-based target localization (RRTL) scheme for UWSNs is proposed. Since the derived closed-form expression is complex and nonconvex, an intelligent localization technique based on a backpropagation (BP) neural network is considered which employs improved differential evolution (IDE) to obtain the network weights and biases. First, improved opposition-based learning is used to obtain a robust population. Then an adaptive mutation strategy is utilized to improve global searching and convergence. Finally, optimized network parameters are obtained via IDE to train the BP neural network. Performance results are presented which demonstrate that the proposed method is superior to conventional state-of-the-art methods and is robust to absorption bias. For example, for an error cumulative distribution function (CDF) of 0.8 with <italic>N</italic> = 6, the proposed method provides an improvement of 69.7% in accuracy and 78.4% in computation time over the well-known WLS technique. [ABSTRACT FROM AUTHOR]