Applying Modern Meta-Heuristics in Intelligent Systems

1. Introduction

Engineering and business problems are increasingly impenetrable due to the new economics triggered by big data, artificial intelligence, and the Internet of things. Exact algorithms and heuristics are not sufficient to conquer such extremely large and unstructured problems. Metaheuristic algorithms emerge as prevailing methods in this context. A generic metaheuristic framework guides the course of search trajectories beyond the local optimality, thus overcoming the impairment of traditional computation methods. The application of modern metaheuristics has a large coverage, ranging from unmanned aerial and ground surface vehicles, unmanned factories, resource-constrained production, and humanoids to green logistics, renewable energy, the circular economy, technology agriculture, environmental protection, finance technology, and the entertainment industry. The aim of this Special Issue is to collect new proposals for marrying modern metaheuristics and intelligent systems. The manuscript submission to this Special Issue was closed on 31 August 2022; we received 22 submissions, of which 8 papers were published, an acceptance rate of 36%. I believe the publication of this Special Issue will position itself at the research frontier across many aspects of applied sciences.

2. Modern Metaheuristics and Its Applications

In previous decades, we saw many novel metaheuristic algorithms including the genetic algorithm (GA), particle swarm optimization (PSO), ant colony optimization (ACO), differential evolution (DE), simulated annealing (SA), adaptive memory programming (AMP), and tabu search (TS). Recently, two interesting branches of metaheuristics have absorbed researchers’ attention. Both of the two modern metaheuristics come from a marriage between two different disciplines, which can be easily recognized by their compound names: matheuristics and hyperheuristics. Matheuristics embeds mathematics into a metaheuristic framework, or vice versa. The purpose is to reach an elaborate mechanism melding together the effectiveness of mathematics and the efficiency of metaheuristics. The hyperheuristics aims to construct an automatic heuristic-selection machine, taking advantage of many lower-level heuristics which have been proposed in academia and industry.

This Special Issue presents two papers contemplating new matheuristics. The first paper, authored by P. Yin, P. Chen, Y. Wei, and R. Day, proposes a matheuristic approach embedding several memory programming strategies from AMP into the metaheuristic framework of the firefly algorithm [1]. The useful strategies include multiple guiding solutions, pattern search, multi-start, swarm rebuilding, and the objective landscape analysis. The second paper, authored by E. Cuevas, H. Becerra, H. Escobar, A. Luque-Chang, M. Pérez, H. Eid, and M. Jiménez, proposes matheuristic search schemes with the trajectory courses assisted by the second-order systems [2]. The second-order systems have different temporal responses depending on the set values of the parameters. Such temporally varying responses can be embedded into a metaheuristic to facilitate the search patterns adapted to complex landscapes. One hyperheuristics paper was chosen for this Special Issue. The paper, authored by X. Sánchez-Díaz, J. Ortiz-Bayliss, I. Amaya, J. Cruz-Duarte, S. Conant-Pablos, and H. Terashima-Marín, develops a feature-independent hyperheuristic approach for solving the knapsack problem. The proposed hyperheuristics does not rely on problem features to map the problem states into suitable existing heuristics [3]. Instead, a fixed sequence of existing heuristics is defined to improve the problem-solving performance within the hyperheuristic framework.

We also collected papers for classic applications of modern metaheuristics, namely vehicle routing and wireless networking. A paper authored by S. Nucamendi-Guillén, D. Flores-Díaz, E. Olivares-Benitez, and A. Mendoza presents a memetic algorithm for the cumulative capacitated vehicle routing problem [4]. The proposed method is a bi-objective optimization scheme that minimizes the total latency and total tardiness of the vehicle routing simultaneously. A mixed-integer program formulation is proposed for the cumulative capacitated vehicle routing problem. As compared to commercial software which optimally solves the problem with a small size, the proposed memetic algorithm can solve a larger-sized problem and produce the efficient bi-objective Pareto fronts. The paper by H. Rico-Garcia, J. Sanchez-Romero, A. Jimeno-Morenilla, and H. Migallon-Gomis proposes a parallel metaheuristic approach to reduce the vehicle traveling time in smart cities [5]. A Compute Unified Device Architecture (CUDA)-based implementation of the Teacher–Learner-Based Optimization (TLBO) metaheuristic is presented to target the shortest routing path for visiting a large number of points in a city. For applications in wireless networking, the paper by M. Chiang and W. Su presents a load balancing scheme for multithreaded applications on NUMA systems. When an imbalance occurs in the load on NUMA multiple cores, the load balancing mechanism of the kernel scheduler should migrate threads between NUMA cores. Threads to be migrated are selected considering the distribution of threads on nodes for inter-node load balancing [6]. The paper, authored by H. Zhang, J. Yang, T. Qin, Y. Fan, Z. Li, and W. Wei, develops a multi-strategy improved sparrow search algorithm (ISSA) for solving the node localization problem in heterogeneous wireless sensor networks (HWSNs). ISSA is an advanced version of the traditional SSA for improving the convergence speed and accuracy. This is accomplished by adopting the PSO’s individual best for solution guiding and a Gaussian disturbance for preventing falling at local optima. The experimental results show that the ISSA yields a smaller average localization error than that of other metaheuristics [7].

Future challenges in metaheuristic applications are addressed in the final paper. The paper, authored by C. Huang, W. Liang, P. Chen, and Y. Chan, uses a dual matrix to identify the opinion leaders and followers in order to reach a converged consensus on the web [8]. The effectiveness of the proposed method has been validated by a case study of extracting leading opinions on green energy and low carbon for helping make effective public policies on environmental protection.