A full-featured cooperative coevolutionary memory-based artificial immune system for dynamic optimization
Introduction
Dynamic optimization problems (DOPs) are a class of optimization problems in which optima change with time in different environments. Since most real-world optimization problems are naturally dynamic, DOPs have attracted many researchers recently [1], [2].
In comparison with static optimization problems, DOPs face more challenges. While in static optimization problems, algorithms should track fixed optima and have sufficient time to locate global optima, a DOP solution needs to locate multiple optima, track the global optimum movement, and explore the environment simultaneously in a dynamic landscape [3]. Controlling diversity before and after a new change and using memory to increase convergence speed are other DOPs’ challenges. As effective solutions to address these challenges in DOPs, evolutionary algorithms (EAs) have been increasingly employed [4]. To handle dynamic landscapes, these algorithms have used such strategies as increasing and maintaining diversity [5], [6], multi-population methods [7], [8], memory approaches [9], self-adaptive mechanism [10], [11], and predicting change. As a large class of EAs, bio-inspired algorithms have proved to be an ideal solution to DOPs. Inspired by biological immune system, artificial immune system (AIS) is one of these bio-inspired algorithms.
Biological immune system has outstanding qualities, which can help it adapt to various changes in the body. AISs have been utilized in a wide variety of application areas, such as anomaly detection [12], data classification [13], global optimization [14], multi-objective optimization [15], robotics [16], and recommender systems [17]. However, they have rarely exploited in the area of dynamic optimization problems. Indeed, most proposed AISs are based on clonal selection theory, which is only one of immune system’s qualities. While a classic clonal selection algorithm (CSA) is good at maintaining diversity, it suffers from low convergence speed and premature convergence. Considering these classic CSA’s weaknesses, AIS has attracted less attention in the area of DOPs compared to other bio-inspired algorithms, such as PSO and DE [5].
Motivated by inherent qualities of biological immune system and cooperative coevolutionary [18], [19], this study develops an innovative AIS for DOPs. The main objective of this research is to overcome AIS’s deficiencies in DOPs using all biologic immune system’s characteristics. Immune cells have similar-shaped receptors, antibodies, each of which can combine with a specific part of an antigen, called an epitope. As any antigen has several epitopes, several antibodies are needed to defend against a particular antigen. In response to multiple antigens invading the body, antibodies are increased and clustered around infected area. This natural decomposition-based method is effective in locating and tracking multiple optima in DOPs [5]. Inspired by this strategy, in this paper, the entire n-dimensional search space is divided into n one-dimensional smaller search space, which are explored cooperatively and simultaneously. In this case, each dimension and its location are considered an antibody and an epitope, respectively. Based on cooperative coevolution, an n-dimensional vector named context vector is necessary to evaluate each population. To enhance the convergence speed of the classic CSA, each dimension, or population, is decomposed into several areas in which immune cells are randomly generated. This is based on the idea of generating uniformly random immune cells in bone marrow. How these areas help in accelerating the convergence speed is essential. It is achieved by incorporating a negative selection algorithm (NSA) into CSA, which helps increase convergence speed along with hyper-mutation. According to this procedure, areas are labeled as ‘self’ and ‘non-self’, indicating non-optimal and optimal areas. It happens gradually during the following stages. At the first stage, all areas are labeled ‘non-self’. At the second stage, an affinity-based hyper-mutation proposed in [20] is applied to antibodies within each area. At the last stage, areas failing to improve themselves are eliminated. As the search proceeds, the sub-populations begin to converge to a small number of areas, resulting in accelerating convergence speed. Many researchers [2], [5], [21] have used two kinds of memories, short-term and long-term memories, to track optima before and after a change occurs. In this paper, the short-term and long-term memory contain the current and the best previous environment’s solutions, respectively. It is based on primary and secondary responses qualities of natural immune system. Individuals are stored in the long-term memory in two ways. First, when a change occurs, the best solutions in the previous environment are moved to the long-term memory. Second, by applying a network suppression process to the short-term memory, unimproved current solutions are considered local optima and transferred to the long-term memory. Inspired by immune network theory, this process helps avoid premature convergence and trapping in local optima. After a change is detected, how to use the long-term memory and rebuild the short-term memory to track global optima is important. Motivated by [21], in this paper, the difference between the average of current fitness and the previous fitness is used to make certain whether an environment reoccurs. If the difference is less than a threshold, it means an environment reoccurs. Thus, the best solutions belonging to that environment are inserted into the short-term memory. Otherwise, the short-term memory is re-initialized.
In short, this study proposes a new AIS for DOPs based on the following strategies. At first, inspired by cooperative coevolution, the n-dimensional search space is divided into n one-dimensional subpopulations to speed exploration. Second, each subpopulation is decomposed into several regions, which contain optimal and non-optimal solutions. Third, a negative selection algorithm is incorporated into CSA to speed exploitation. Fourth, a network suppression mechanism is applied to solutions to avoid trapping in a local optimum. Finally, previous solutions are stored in the memory and reused further to track the new global optimum better when a change occurs.
The rest of the paper is organized as follows. Section 2 presents an overview of the approaches that are a combination of memory-based and multi-population strategies as well as immune-based algorithms applied to DOPs. Section 3 describes the proposed CM-AIS in more detail. Experimental results of the performance of CM-AIS on the MPB compared to other algorithms are given in Section 4. Finally, Section 5 provides conclusions and suggests a few useful directions for future work.
Section snippets
Cooperative coevolutionary algorithms
Firstly proposed by Potter and De Jong [22], a cooperative coevolutionary (CC) has been embedded in GA to solve complex problems. The idea is to decompose an n-dimensional problem into n one-dimensional subproblems each of which is solved independently. An n-dimensional vector called a context vector (CV) was introduced, that each subproblem is responsible for optimizing one of the CV’s components. After that, Van den Bergh and Engelbrecht [18] used Potter’s idea in the basic PSO for
Proposed algorithm
Although IBAs effectively explore search space, they are not good at exploitation, which makes optimizing dynamic environments challenging [5]. Moreover, they tend to have low convergence speed and high computation. However, the biological immune system has features helping it adapt to changes in the human body. Thus, this paper has two purposes:
(1) Simulating all the immune system’s features to create a new artificial immune system the most similar to the natural one.
(2) Creating an AIS able
The moving peaks benchmark problem
The moving peaks benchmark (MPB) firstly introduced by Branke in [43] has been commonly used for evaluating dynamic optimization problems. In MPB, in a D-dimensional landscape, there are peaks, whose locations, heights, and widths are changed according to a certain frequency. The change of ith peak is determined as follows: where , , and are the width, height, and location of peak i at time t,
Conclusion and future work
AIS-based algorithms have been scarcely used for optimizing dynamic environments because of their low convergence speed and large computation. A new cooperative coevolutionary memory-based artificial immune system improved by a new clonal selection algorithm is proposed to overcome these problems. In the proposed method, inspired by cooperative coevolution, an n-dimensional population is divided into n one- dimensional subpopulations. Then, motivated by cells’ production in bone marrow, each
CRediT authorship contribution statement
Bahareh Etaati: Conceptualization, Software, Data curation, Visualization, Writing – original draft, Investigation. Zahra Ghorrati: Methodology, Writing – review & editing. Mohammad Mehdi Ebadzadeh: Methodology, Supervision.
Declaration of Competing Interest
The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.
References (53)
- et al.
Evolutionary dynamic optimization: A survey of the state of the art
Swarm Evol. Comput.
(2012) - et al.
A multi-population electromagnetic algorithm for dynamic optimisation problems
Appl. Soft Comput.
(2014) - et al.
A cluster-based clonal selection algorithm for optimization in dynamic environment
Swarm Evol. Comput.
(2019) - et al.
A clonal selection algorithm for dynamic multimodal function optimization
Swarm Evol. Comput.
(2019) - et al.
A novel particle swarm optimization algorithm with adaptive inertia weight
Appl. Soft Comput.
(2011) - et al.
A novel hybrid adaptive collaborative approach based on particle swarm optimization and local search for dynamic optimization problems
Appl. Soft Comput.
(2015) - et al.
Dual-archive-based particle swarm optimization for dynamic optimization
Appl. Soft Comput.
(2019) - et al.
A new hybrid approach for dynamic continuous optimization problems
Appl. Soft Comput.
(2012) - et al.
A note on the learning automata based algorithms for adaptive parameter selection in PSO
Appl. Soft Comput.
(2011) - et al.
Local feature selection based on artificial immune system for classification
Appl. Soft Comput.
(2020)
Continual learning classification method with new labeled data based on the artificial immune system
Appl. Soft Comput.
A hybrid multi-objective AIS-based algorithm applied to simulation-based optimization of material handling system
Appl. Soft Comput.
A multi-tier immuno-inspired framework for heterogeneous mobile robotic systems
Appl. Soft Comput.
Optimising latent features using artificial immune system in collaborative filtering for recommender systems
Appl. Soft Comput.
IA-AIS: An improved adaptive artificial immune system applied to complex optimization problems
Appl. Soft Comput.
History-driven particle swarm optimization in dynamic and uncertain environments
Neurocomputing
A scalable parallel cooperative coevolutionary PSO algorithm for multi-objective optimization
J. Parallel Distrib. Comput.
Distributed parallel cooperative coevolutionary multi-objective large-scale immune algorithm for deployment of wireless sensor networks
Future Gener. Comput. Syst.
Immune-based algorithms for dynamic optimization
Inform. Sci.
Innate-adaptive response and memory based artificial immune system for dynamic optimization
Int. J. Perform. Eng.
Adaptive artificial immune networks for mitigating DoS flooding attacks
Swarm Evol. Comput.
Recent advances in artificial immune systems: models and applications
Appl. Soft Comput.
Species-based particle swarm optimizer enhanced by memory for dynamic optimization
Appl. Soft Comput.
A novel multi-swarm algorithm for optimization in dynamic environments based on particle swarm optimization
Appl. Soft Comput.
A survey of evolutionary continuous dynamic optimization over two decades–part A
IEEE Trans. Evol. Comput.
An adaptive bi-flight cuckoo search with variable nests for continuous dynamic optimization problems
Appl. Intell.
Cited by (5)
Shape and sizing optimisation of space truss structures using a new cooperative coevolutionary-based algorithm
2024, Results in EngineeringA hybrid evolutionary immune algorithm for fuzzy flexible job shop scheduling problem with variable processing speeds
2023, Expert Systems with ApplicationsPhysical Internet-enabled E-grocery delivery Network:A load-dependent two-echelon vehicle routing problem with mixed vehicles
2022, International Journal of Production EconomicsCitation Excerpt :However, the degradation raised by a random iteration without guidance for the initialized clustering is still an unresolved issue (Yang et al., 2022). In recent years, one of the soft computing paradigm-artificial immune systems (AIS) has been widely used in various domain areas, such as optimization (Hatata et al., 2018), pattern recognition and classification (Li et al., 2021), analysis of data (Wang et al., 2018a), and robotics (Etaati et al., 2022). The artificial immune system is a highly distributed, parallel adaptive, evolved system with learning, anomaly detection, and optimization characteristics (Pinto and Gonçalves, 2022).
An adaptive clonal selection algorithm with multiple differential evolution strategies
2022, Information SciencesCitation Excerpt :In recent years, optimization problems have attracted great attention from researchers, and many nature-inspired computation algorithms have been proposed, such as the genetic algorithm (GA) [1], artificial immune algorithm(AIA) [2], particle swarm optimization (PSO) [3], ant colony algorithm (ACA) [4] and differential evolution (DE) [5].