A Neural Network-Based Optimization Approach for the RNP Problem in Arbitrary Shaped Workspaces

Abstract

In order to monitor tags or tagged objects in a workspace, RFID readers need to be arranged so that the RFID reader network can cover most of the tags and simultaneously satisfy some constraints, such as minimized deployment cost, maximized interrogation efficiency, and load balancing among readers. The problem of optimizing the deployment of RFID readers is known as the RFID Network Planning (RNP) problem and is considered NP-hard. Different optimization methods have been proposed, among which nature-inspired approaches often give more impressive results. However, finding an optimal solution in a reasonable time for the multi-objective problem is always challenging. This paper presents a neural network-based approach in which the reader deployment optimization problem is formulated as an energy function and minimized by a Hopfield network. The achieved minimum energy determines the optimal deployment solution. With traditional natural-based optimization methods, discrete populations are initialized and searched in the candidate solution space, often leading to local optima. The Hopfield network-based approach minimizes an energy function of the entire network, thus overcoming the limitation. Furthermore, with the predetermined connection weights and activation threshold, the training phase is omitted, thus shortening the optimization time. Experimental results show that the optimization process converges faster, and the readers’ optimal positions are found early.