Novel Research and Application on Swarm Optimization and Bioinspired Optimization Algorithms

1. Introduction

Biological behaviors have inspired the development of numerous evolutionary optimization algorithms. Among these, swarm intelligence algorithms (SIAs) represent a significant category. SIAs are stochastic global optimization methods that rely on the collective intelligence of swarms. Unlike traditional evolutionary algorithms, which use operators such as mutation, crossover, and replication, SIAs achieve optimization through the interplay of antagonistic and synergistic interactions among particles. In SIAs, an initial population of particles—each representing a potential solution—is generated randomly. Through a straightforward yet effective iterative mechanism, the algorithm refines random solutions, ultimately converging on a global optimal solution by following the most efficient path and minimizing associated costs. SIAs have found applications across diverse domains, including engineering, evolutionary computing, biomedicine, navigation and communication systems, environmental sciences, cryptography, real-time traffic control, and many other fields. An overview of SIAs is provided in [1].

2. Overview of Published Articles

Twelve articles are published in this Special Issue. They present novel results in the broad area of evolutionary and bioinspired optimization algorithms and their application in diverse domains, including the prediction of coconut yields, the optimization of laser-cutting paths, software clustering, and truck-drone parcel delivery, to name a few.

In Contribution 1, Alkhawaji et al. propose a robust and precise model for predicting coconut yields, leveraging the Bi-directional Long Short-Term Memory (BILSTM) deep learning classifier. This model is designed to facilitate informed decision making in coconut farming and agricultural research. To address the issue of overfitting, the model’s hyperparameters are fine-tuned using the Lévy Flight and Seagull Optimization Algorithm (LFSOA). Experimental results demonstrate that the proposed approach outperforms existing methods, achieving superior accuracy, precision, recall, F1-scores, and other statistical metrics.

A conformal antenna was designed to fit a surface shaped based on considerations other than electromagnetic requirements, such as aerodynamic or hydrodynamic factors [2]. During its manufacturing process, a high-resolution industrial camera is used to detect the precise positions of feature points, facilitating tasks like coating, mounting, welding, and automated optical inspection. The challenge of path detection for this purpose can be modeled as a constrained three-dimensional traveling salesman problem (TSP). To address this, Ding et al., in Contribution 2, introduce the concept of a historical optimal population, where the optimal individual is added to the historical optimal population each time the fitness improves. Based on this concept, they develop an improved genetic algorithm (CHOP-IGA). Experimental results demonstrate that the CHOP-IGA significantly enhances path planning efficiency for conformal antenna surface detection, achieving a 48.44% improvement in detection efficiency compared to traditional genetic algorithms.

Flocking, the coordinated and aligned movement of a group of individuals—such as birds or fish—appearing to behave as a unified super-organism, is a common phenomenon in nature [3]. Drawing inspiration from this natural behavior, Reynolds introduced the concept to computer science [4]. Building on this idea, Vicsek et al. [5] proposed modeling such groups as complex networks, where individuals are represented as nodes, and their interactions as edges are governed with simple, uniform rules. Expanding on these approaches, Zhao et al., in Contribution 3, develop two models that incorporate more realistic interaction mechanisms between individuals and the group during the flocking process. Simulation experiments demonstrate that these models significantly enhance the group’s alignment efficiency, even under noisy conditions.

In laser-cutting processes for manufacturing sheet metal products, optimizing the cutting path to minimize the total time required for cutting all components from the sheet is critical. This challenge is known as the cutting path problem [6]. Junior et al., in Contribution 4, explore an adaptive biased random-key genetic algorithm (ABRKGA) combined with a heuristic to address this issue. Experimental results demonstrate that the proposed method outperforms existing approaches, delivering superior solution quality and reduced computational time.

The Beetle Antennae Search (BAS) algorithm is a metaheuristic optimizer inspired by the foraging behavior of beetles [7]. Similarly, the Artificial Fish Swarm Algorithm (AFSA) draws inspiration from the schooling behavior of fish in nature and has seen significant advancements since its introduction in 2002, with numerous improved and hybrid models being developed [8]. In Contribution 5, Ni et al. combine elements of BAS and AFSA, introducing a mutation mechanism and a multi-step detection strategy to enhance the BAS algorithm’s optimization accuracy. Experimental results reveal that the proposed hybrid approach outperforms the original multi-direction detection BAS algorithm, particularly in solving high-dimensional problems.

The Cat Swarm Optimization (CSO) algorithm, introduced by Chu et al. [9], has inspired a range of variations within the CSO family, widely utilized for solving continuous optimization problems. Building on this foundation, He et al., in Contribution 6, present a compact Cat Swarm Optimization algorithm incorporating a Small-Sample Probability Model (SSPCCSO). Like other CSO algorithms, SSPCCSO employs both a tracking mode and a searching mode to identify optimal solutions. Additionally, it introduces a novel differential operator within the searching mode and leverages the gradient descent method. Experimental evaluations demonstrate that SSPCCSO outperforms other compact evolutionary algorithms in terms of solution quality and efficiency.

Advancements in multi-core fiber technology have driven the adoption of spatial division multiplexing (SDM) within elastic optical networks (EONs). A key challenge in EONs is the routing and spectrum assignment (RSA) problem, which requires the contiguous and continuous allocation of spectrum across all links in a lightpath. The introduction of spatial dimensions further complicates this issue, transforming it into the routing, spectrum, and core allocation (RSCA) problem [10]. Xu et al., in Contribution 7, enhance existing approaches to address the RSCA problem, formulating it as an integer linear programming model. They propose a joint routing and core coding scheme that incorporates core switching and introduce a multi-objective evolutionary algorithm based on decomposition, enhanced by adaptation and multi-strategy fusion. Simulation results indicate that their proposed algorithm outperforms existing methods by achieving more diverse and dominated solutions while effectively incorporating core-switching capabilities.

The Software Module Clustering Problem (SMCP) is an optimization challenge focused on enhancing the modularity of software projects within the framework of Search-Based Software Engineering (SBSE) [11]. SBSE seeks to transition software engineering tasks from human-centric search methods to automated, machine-based approaches. This shift leverages techniques from metaheuristic search, operations research, and evolutionary computation paradigms [12]. Alshareef and Maashi, in Contribution 8, explore the multi-objective software module clustering problem, which entails assembling an optimal set of modules based on specified criteria to achieve improved modularity and functionality. Their experimental findings demonstrate that the proposed approach surpasses the performance of standalone multi-objective evolutionary algorithms.

In Contribution 9, Kusztelak et al. introduce the Genetic Algorithm with the Symmetrization Operator (GASO), a memetic variation of the traditional genetic algorithm. This approach involves the cyclic symmetrization of the population, where the parental points are symmetrized around the current leader of the population. They analyze the algorithm, provide examples of its application, and evaluate its effectiveness on 24 continuous functions within a multidimensional cube using the Black-Box Optimization Benchmarking (BBOB-2010) benchmark. The proposed algorithm enhances the exploration of the search space, especially for test functions classified in the benchmark as high-conditioning and unimodal.

Radial Basis Function (RBF) networks are a widely used type of feedforward artificial neural networks designed for function approximation tasks. These networks consist of three layers, with each performing specific functions: the input layer, the hidden layer, and the output layer [13]. In Contribution 10, Tsoulos et al. propose a two-phase hybrid approach to train RBF neural networks for classification and regression tasks. In the first phase, a range for the key parameters of the RBF network is determined, and in the second phase, a genetic algorithm is employed to find the optimal configuration. Experimental results across various classification and regression problems show that the proposed method outperforms other approaches on nearly all datasets.

In [14], Friedman and Tukey introduced the concept of projection pursuit as a technique for exploring multivariate datasets. This method aims to identify "interesting" linear projections of the data onto a line or plane [15]. Larabi-Marie-Sainte, in Contribution 11, presents four novel feature selection methods that utilize outlier detection through the projection pursuit technique. Experimental results, using nineteen real datasets tested with three classifiers (k-NN, SVM, and Random Forest), demonstrate that the proposed methods improve the accuracy of classification.

TSP-D is a variation of the traveling salesman problem, where a truck and a drone work together to deliver parcels to customers, aiming to minimize the overall delivery time. AlMuhaideb et al., in Contribution 12, suggest using metaheuristics, specifically a greedy, randomized adaptive search procedure with two local search options and a self-adaptive neighborhood selection strategy, to address this NP-hard problem. Experimental results, using a publicly available benchmark, show that the proposed approach achieves comparable performance to existing alternative algorithms in terms of tour duration.

3. Conclusions

These publications propose novel forms and uses of different swarm optimization and bioinspired optimization algorithms and thoroughly explore aspects of interest to both theory and application. As such, they contribute to expanding our current knowledge and understanding of the field and to opening up new research pathways.