Decomposition-based multi-objective firefly algorithm for RFID network planning with uncertainty

Highlights

• Uncertainty is considered in RFID network planning.

• RFID network planning is modelled as a multi-objective optimization.

• A novel multi-objective firefly algorithm is proposed to solve multi-objective RFID network planning.

• Numerical experiments show the effectiveness of the proposed firefly algorithm for multi-objective RFID network planning.

Abstract

Radio frequency identification (RFID) is widely used for item identification and tracking. Due to the limited communication range between readers and tags, how to configure a RFID system in a large area is important but challenging. To configure a RFID system, most existing results are based on cost minimization through using 0/1 identification model. In practice, the system is interfered by environment and probabilistic model would be more reliable. To make sure the quality of the system, more objectives, such as interference and coverage, should be considered in addition to cost. In this paper, we propose a probabilistic-based multi-objective optimization model to address these challenges. The objectives to be optimized include number of readers, interference level and coverage of tags. A decomposition-based firefly algorithm is designed to solve this multi-objective optimization problem. Virtual force is integrated into random walk to guide readers moving in order to enhance exploitation. Numerical simulations are introduced to demonstrate and validate our proposed method. Comparing with existing methods, such as Non-dominated Sorting Genetic Algorithm-II and Multi-objective Particle Swarm Optimization approaches, our proposed method can achieve better performance in terms of quality metric and generational distance under the same computational environment. However, the spacing metric of the proposed method is slightly inferior to those compared methods.

Introduction

RFID devices transmit data in the radio frequency band which automatically perform item identification and data acquisition without direct contact between readers and tags via wireless communication. Since RFID is a key part of “Internet of Things”, there are significant developments of RFID in many industrial and commercial applications such as logistic management, supply chain management [1], construction site management as well as intelligent transportation systems [2] in recent years.

A basic RFID system has several key components, including RFID readers, tags and a computer system linked to readers. A RFID network typically consists of numerous tags distributed in a working area such as in a warehouse or a construction site [3]. Since the communication range of a RFID system is limited, it is required to deploy multiple RFID readers and antennas in a network area in order to cover all the tags. In this case, we need to determine the total number of readers and locations where they should be placed in order to cover tags efficiently. This problem refers to the RFID network planning (RNP). During the planning phase, multiple objectives and specific constraints such as the total number of readers and the coverage of tags should be taken into consideration. It has been shown in [4] that this problem is non-polynomial (NP) hard. Hence, efficiently finding a good solution for such kind of problems is challenging.

For a static RNP, one of the most important issues is coverage of tags. As readers are much more expensive than tags in practise, a smaller number of readers to be used will lead to lower cost of a RFID network. Thus, minimizing total number of readers to be used subject to coverage constraints is usually used for the RNP. There are some existing works based on this type of formulations. For example, a tool is developed to determine number and locations of RFID readers where a rectangle area is guaranteed to be covered [5]. A stochastic modelling of the RNP is proposed in [6]. The total number of readers is minimized subject to coverage constraints where the detected probability for each tag should be greater than a given threshold. This work requires full coverage of a special area. The cases that the tags are not uniformly distributed are investigated in [7], [8]. For example, the placement of readers and tags for aircraft part identification is formulated as a linear integer programming problem [7]. There positions of readers are selected from a set of candidates and each tag is ensured to be covered by at least one reader. A genetic algorithm (GA) is used to optimize the placement of readers for a medical asset tracking system. However, all the reader placement problems mentioned above are single objective optimization problems subject to coverage constraints [8].

Due to the complexity of a RFID network in practise, the RNP is typically a multi-objective problem. One strategy for solving multi-objective problems is to combine multiple objectives into a single objective using weighted sum method. However, proper weights are difficult to be selected to satisfy user's preferences. Population-based algorithm can be used to solve multi-objective optimization [9]. In [10], tag coverage, the number of readers and interference level are considered together. A particle swarm optimization (PSO) combined with redundant reader elimination is proposed to deploy RFID readers. In [11], tag coverage, reader collision, economic efficiency and load balance are considered together to configure a RFID network. In [12], tag coverage, signal interference and load balance of readers are integrated together for the RNP. To cope with the complexity of the formulated multi-objective optimization problem, an evolutionary computation based approach and a swarm intelligence based method are employed for solving the problem. Those algorithms, which are inspired by natural processes without need of gradient information, are easy to be implemented. Moreover, these algorithms can find near optimal solutions within reasonable time using polynomial time complexity iteration. Furthermore, these algorithms are robust to initial solutions through population search [13]. Hence, they are widely employed to solve many practical problems. In [14], a hybrid algorithm combining a PSO algorithm and a simulated annealing algorithm is developed for optimizing the locations of readers. These three objectives including coverage, signal interference and load balance, are linearly weighted together to form a comprehensive objective. However, it is difficult for users to determine the weight of each objective. In [15], a cooperative multi-objective artificial bee colony algorithm is developed to search all solutions. These solutions form Pareto front, from which the user can select a solution. However, the number of readers is not optimized simultaneously in the paper. Although several objectives including the number of readers, tag coverage, interference level, and load balance, are presented in the literature, few of them consider more than three objectives simultaneously. As a result of high reading speed of current readers, load balance has little effect on network performance. In this paper, we configure a RFID network through optimizing number of readers, signal interference, and coverage of tags simultaneously since they have significant influence on the performance and cost of a network.

It is worth noting that models of the RNP in most existing works are assumed to be deterministic. However, the signal strength received by a RFID reader suffers from multipath fading. In general, the fading effect on the received signal will increase as the distance between a reader and a tag increases. Also, received signal strength is always affected by environment. Thus, probability model should be more appropriate for a real RFID system.

As mentioned above, heuristic algorithms are the most popular methods to solve MORNP problems. Recently, a new swarm intelligence algorithm, named firefly algorithm, has been put forward to solve continuous and discrete optimization problems [16]. It has been widely applied to solve practical problems [17]. Several variants of firefly algorithms are proposed in the literature [18]. However, most works are focused on single objective firefly algorithms. Firefly algorithms are used to solve multi-objective optimization as reported in [19]. However, the multi-objectives of the problem in [19] are integrated as an objective through a linear weighted method. Thus, the firefly algorithm in the paper is still a single optimization method. In most cases, Pareto solutions of a multi-objective optimization problem are preferred to be obtained. In this paper, we propose a novel firefly algorithm to solve MORNP based on decomposition method. The contributions of our work include:

• In MORNP, three objectives, namely, total number of readers, tag coverage and interference level, are taken into consideration simultaneously. These three objectives, which are related to both the cost and performance of a RFID network, are conflicting. For instance, we can enhance tag coverage in a RFID network through increasing number of readers. However, higher cost and larger interference are inevitable with the increase of readers. Very few works in the literature have considered them simultaneously. In particular, the number of readers was seldom considered as an objective function.

• A novel firefly algorithm based on the decomposition approach is developed to solve this multi-objective optimization problem. Through decomposition, our proposed algorithm minimizes all these objective functions simultaneously in a single run. Different from existing methods for MRNP, the proposed method optimizes the number of readers with other objectives simultaneously, which is difficult to be solved due to that other parameters are in close connection with number of readers.

• In order to improve the search capability of the proposed algorithm, a virtual force method is integrated. This method guides movement of readers so as to guarantee tag coverage according to local information.

The rest of the paper is organized as follows. A multi-objective RFID network planning model is formulated in Section 2. Then, a novel multi-objective firefly algorithm (MOFA) is proposed for finding the solution of MORNP in Section 3.2. In Section 4, the implementation of MOFA is discussed. Experimental results are presented. A case study in an industry application is discussed in Section 5. Section 6 concludes the proposed work.