Elsevier

Knowledge-Based Systems

Volume 156, 15 September 2018, Pages 12-42
Knowledge-Based Systems

A scattering and repulsive swarm intelligence algorithm for solving global optimization problems

https://doi.org/10.1016/j.knosys.2018.05.002Get rights and content

Highlights

  • Three swarm intelligence algorithms (RFA, SRFA and ESRFA) have been proposed.

  • RFA accelerates the search process by avoiding unpromising regions.

  • SRFA scatters weak solutions to new search regions to increase exploration.

  • ESRFA conducts hovering, sharp-dive and best memory-guided search operations.

  • They outperform other classical and state-of-the-art optimization methods.

Abstract

The firefly algorithm (FA), as a metaheuristic search method, is useful for solving diverse optimization problems. However, it is challenging to use FA in tackling high dimensional optimization problems, and the random movement of FA has a high likelihood to be trapped in local optima. In this research, we propose three improved algorithms, i.e., Repulsive Firefly Algorithm (RFA), Scattering Repulsive Firefly Algorithm (SRFA), and Enhanced SRFA (ESRFA), to mitigate the premature convergence problem of the original FA model. RFA adopts a repulsive force strategy to accelerate fireflies (i.e. solutions) to move away from unpromising search regions, in order to reach global optimality in fewer iterations. SRFA employs a scattering mechanism along with the repulsive force strategy to divert weak neighbouring solutions to new search regions, in order to increase global exploration. Motivated by the survival tactics of hawk-moths, ESRFA incorporates a hovering-driven attractiveness operation, an exploration-driven evading mechanism, and a learning scheme based on the historical best experience in the neighbourhood to further enhance SRFA. Standard and CEC2014 benchmark optimization functions are used for evaluation of the proposed FA-based models. The empirical results indicate that ESRFA, SRFA and RFA significantly outperform the original FA model, a number of state-of-the-art FA variants, and other swarm-based algorithms, which include Simulated Annealing, Cuckoo Search, Particle Swarm, Bat Swarm, Dragonfly, and Ant-Lion Optimization, in diverse challenging benchmark functions.

Introduction

Nature inspired algorithms have gained popularity and been widely used for solving various global optimization problems. Among them, the firefly algorithm (FA) [1] is a popular metaheuristic search method that has been applied to undertaking diverse optimization problems in engineering, medical, and social sciences [2]. In FA, each firefly represents a solution in the search space. Its light intensity is determined by an objective function. Each firefly moves to the optimal region by following multiple optimal solutions in the neighbourhood. Overall, the FA search strategies enable the fireflies (i.e. solutions) with lower light intensities (i.e. fitness values) to move towards those with higher light intensities in the neighbourhood, in order to achieve global optimality. Although each firefly serves as a search agent to increase global exploration, the empirical results [3], [4] indicate that high dimensional optimization problems still pose great challenges to FA, and the random movement of FA has a high likelihood to be trapped in local optima.

One of the limitations of the original FA model is its pure attractiveness force that moves each firefly towards the brighter counterparts in the neighbourhood. If there is no brighter firefly in the neighbourhood, the attractiveness action stagnates, and there is no alternative mechanism to drive the search out of the local optima traps. In other words, there is no effective strategy to avoid poor solutions while moving towards the optimal ones. Moreover, by following the neighbouring brighter fireflies, there is an increased likelihood that part of the population could be clustered in the same region, therefore reducing the possibility of finding the global optimum residing elsewhere. Solving the first issue results in better performance and faster convergence, while resolving the second issue tackles the premature convergence problem. Both challenges constitute the key motivation of this research.

To deal with the abovementioned challenges, this research proposes three FA variants, i.e., Repulsive Firefly Algorithm (RFA), Scattering Repulsive Firefly Algorithm (SRFA), and Enhanced SRFA (ESRFA). Besides the conventional attractiveness movement of the FA model, RFA uses a repulsive force strategy to drive the search to reach global optimality in fewer iterations. SRFA, in addition to using the repulsive force action, incorporates a scattering mechanism to explore rarely exploited distinctive search spaces to increase search diversity and overcome premature convergence. ESRFA incorporates hawk-moths’ local hovering and sharp dive escaping movements to diversify the attractiveness and repulsive force actions of SRFA, as well as exchanges with historical best experiences in the neighbourhood to accelerate convergence. Specifically, the research novelties are five-fold. (1) The proposed repulsive force works together with the original attractive force to enable the search procedure to converge towards the global optima and, at the same time, avoid poor solutions. (2) The scattering strategy overcomes premature convergence by diverting a number of weak solutions to unexploited regions. (3) Exploitation-driven attractiveness and exploration-based evading mechanisms are used to enhance the search operations in diversifying the search process. (4) Interactions with the historical personal best experiences of other fireflies are conducted to accelerate convergence. (5) The proposed strategies cooperate with each other to overcome premature convergence, especially in solving high dimensional optimization problems. We evaluate the three proposed FA models with ten standard and CEC2014 [5] benchmark optimization functions. RFA, SRFA and ESRFA significantly outperform state-of-the-art FA variants and other evolutionary search methods, which include Particle Swarm Optimization (PSO) [6], Simulated Annealing (SA) [7], FA, Bat Swarm Optimization (BSO) [8], [9], Cuckoo Search Optimization (CSO) [10], Dragonfly Optimization (DFO) [11] and Ant-Lion Optimization (ALO) [12].

The research contributions are summarized, as follows:

  • Three FA variants, i.e., RFA, SRFA and ESRFA, are proposed. RFA incorporates a repulsive force strategy to enable fireflies with higher light intensities to jump out of unpromising search regions to achieve fast convergence.

  • SRFA employs the repulsive force action and a scattering mechanism to avoid local optima. The latter diverts a proportion of weak neighbouring solutions to an unexploited distinctive search space to increase search diversity. The repulsive behaviour and the scattering mechanism in SRFA work cooperatively to mitigate premature convergence of the original FA model. On one hand, when the repulsive force action stagnates, the scattering mechanism is able to extend the search to rarely explored regions to reduce the probability of premature convergence. On the other hand, when the scattering behaviour is unable to generate fitter solutions, the repulsive force action enables each firefly to conduct long jumps to move towards optimal regions to escape from local optima.

  • ESRFA incorporates exploitation and exploration coefficients to diversify the attractiveness and repulsive operations of SRFA, respectively, and interacts with neighbouring historical best experiences to accelerate convergence. It has three key properties. Firstly, the exploitation factor simulates the mid-air hovering of hawk-moths around attraction, which enables a refined random examination of a promising neighbourhood and overcomes the local optima traps, as compared with the original attractiveness operation in FA. Secondly, the exploration-driven escaping coefficient simulates a sharp dive of moths in response to predators, which enables the search to explore a wider search space while evading from the worse solutions. The newly proposed attractiveness operation increases local exploitation while the updated evading action increases global exploration. In other words, both properties enable the search process to balance well between local exploitation and global exploration. Thirdly, a distinctive attractiveness-based operation guided by the historical personal best experiences of neighbouring fireflies is conducted to accelerate convergence.

  • A comprehensive evaluation with diverse unimodal, multimodal and challenging CEC2014 optimization functions is conducted. The proposed RFA, SRFA and ESRFA models outperform the original FA model and advanced FA variants, as well as other metaheuristic search methods, significantly. They also show great robustness and superiority in dealing with complex high dimensional optimization problems.

The paper is organised as follows. A literature review on related work is presented in Section 2. The proposed RFA, SRFA and ESRFA models are introduced in detail in Section 3. Comprehensive experiments for evaluation of three proposed models are presented in Section 4. Finally, conclusions and directions for future work are given in Section 5.

Section snippets

Related work

Swarm intelligence (SI) based optimization methods have gained popularity recently [13], [14]. As a recent SI algorithm, FA is an effective metaheuristic search method on par with other existing models, in solving diverse optimization problems, which include PSO, Genetic Algorithm (GA), CSO, Artificial Bee Colony (ABC) and Ant Colony Optimization (ACO). In this section, we discuss the basic concepts of FA, different FA variants, and other recently proposed metaheuristic search methods.

Methodology

In this research, we propose three FA variants, i.e., RFA, SRFA and ESRFA, to mitigate the premature convergence problem of the original FA model. RFA employs the repulsive force strategy while SRFA uses both the repulsive force and the scattering mechanisms to increase local and global search capabilities. Motivated by both RFA and SRFA, another enhanced algorithm, ESRFA, is proposed, which embeds the hovering-based attractiveness behaviour, exploration-based repulsive action, and interaction

Evaluation

A comprehensive evaluation has been conducted to evaluate RFA, SRFA and ESRFA models. To indicate efficiency of the proposed models, several FA variants proposed in recent years, i.e. ODFA [3], SFA [34], LSFA [34], and CFA [35] and other metaheuristic search methods, i.e. PSO, SA, FA, BSO, CSO, DFO and ALO, have been implemented for comparison. All algorithms are implemented using MATLAB R2016b. Details of the experimental study are as follows.

Conclusions

We have proposed three improved FA models, i.e., RFA, SRFA and ESRFA, to mitigate premature convergence of the original FA model. RFA attains global optimality with a fast convergence rate by utilizing a repulsive force strategy. SRFA employs both the repulsive force and scattering mechanisms to increase local and global search capabilities. ESRFA uses hawk-moth motivated exploitation and exploration mechanisms and the interaction with neighbouring historical personal best experiences to

Acknowledgement

This research is supported by European Union (EU) sponsored (Erasmus Mundus) cLINK(Centre of excellence for Learning, Innovation, Networking and Knowledge) project (EU Grant No. 2645).

References (75)

  • S. Massan et al.

    Wind turbine micrositing by using the firefly algorithm.

    Appl. Soft Comput.

    (2015)
  • A.H. Gandomi et al.

    Firefly algorithm with chaos.

    Commun. Nonlinear Sci. Numer. Simul.

    (2013)
  • M. Mitić et al.

    Chaotic fruit fly optimization algorithm

    Knowl.-Based Syst.

    (2015)
  • B. Amiri et al.

    Community detection in complex networks: multi–objective enhanced firefly algorithm

    Knowl.-Based Syst.

    (2013)
  • M. Alweshah et al.

    Hybridizing firefly algorithms with a probabilistic neural network for solving classification problems

    Appl. Soft Comput.

    (2015)
  • A. Kazem et al.

    Support vector regression with chaos-based firefly algorithm for stock market price forecasting

    Appl. Soft Comput.

    (2013)
  • C. Zhao et al.

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

    Appl. Soft Comput.

    (2017)
  • J. Dash et al.

    Design of multipurpose digital FIR double-band filter using hybrid firefly differential evolution algorithm

    Appl. Soft Comput.

    (2017)
  • M. Kaur et al.

    Network reconfiguration of unbalanced distribution networks using fuzzy-firefly algorithm

    Appl. Soft Comput.

    (2016)
  • T. Xiong et al.

    Multiple-output support vector regression with a firefly algorithm for interval-valued stock price index forecasting.

    Knowl.-Based Syst.

    (2014)
  • G. Kalantzis et al.

    Investigations of a GPU-based levy-firefly algorithm for constrained optimization of radiation therapy treatment planning

    Swarm Evolution. Comput.

    (2016)
  • O. Kisi et al.

    A survey of water level fluctuation predicting in Urmia Lake using support vector machine with firefly algorithm

    Appl. Math. Comput.

    (2015)
  • T. Kanimozhi et al.

    An integrated approach to region based image retrieval using firefly algorithm and support vector machine

    Neurocomputing

    (2015)
  • Y. Zhang et al.

    A return-cost-based binary firefly algorithm for feature selection

    Inf. Sci.

    (2017)
  • W.C. Yeh

    An improved simplified swarm optimization

    Knowl.-Based Syst.

    (2015)
  • W. Srisukkham et al.

    Intelligent leukaemia diagnosis with bare-bones PSO based feature optimization

    Appl. Soft Comput.

    (2017)
  • Y. Wan et al.

    A feature selection method based on modified binary coded ant colony optimization algorithm

    Appl. Soft Comput.

    (2016)
  • S.C. Neoh et al.

    Intelligent facial emotion recognition using a layered encoding cascade optimization model

    Appl. Soft Comput.

    (2015)
  • H. Haklı et al.

    A novel particle swarm optimization algorithm with Levy flight

    Appl. Soft Comput.

    (2014)
  • F. Kang et al.

    Damage detection based on improved particle swarm optimization using vibration data

    Appl. Soft Comput.

    (2012)
  • T.H.H. Zavaschi et al.

    Fusion of feature sets and classifiers for facial expression recognition

    Expert Syst. Appl.

    (2013)
  • H. Soyel et al.

    Application of NSGA-II to feature selection for facial expression recognition

    Comput. Electric. Eng.

    (2011)
  • B. Huang et al.

    Multi-objective feature selection by using NSGA-II for customer churn prediction in telecommunications

    Expert Syst. Appl.

    (2010)
  • Y. Zhang et al.

    Intelligent affect regression for bodily expressions using hybrid particle swarm optimization and adaptive ensembles

    Expert Syst. Appl.

    (2015)
  • S. Mirjalili

    SCA: a sine cosine algorithm for solving optimization problems

    Knowl.-Based Syst.

    (2016)
  • J. Derrac et al.

    A practical tutorial on the use of nonparametric statistical tests as a methodology for comparing evolutionary and swarm intelligence algorithms

    Swarm Evolution. Comput.

    (2011)
  • A. Fischer et al.

    Exact algorithms and heuristics for the Quadratic Traveling Salesman Problem with an application in bioinformatics

    Discrete Appl. Math.

    (2014)
  • Cited by (54)

    View all citing articles on Scopus
    View full text