cs.NEAug 24, 2026

Coronavirus Optimization Algorithm: A Success-History Adaptive Evolutionary Framework with Archive-Assisted Search and Stagnation Recovery for Global Optimization

Authors: Hari Mohan Pandey

Organizations: Department of Computing & Engineering Bournemouth University UK

Read PDF
Ask AI
Cite

Copy citation

DOINot available for this paper

Abstract

This paper proposes the Coronavirus Optimization Algorithm (COA), a SARS-CoV-2-inspired success-history adaptive evolutionary optimizer for box-constrained continuous global optimization. COA does not model disease transmission; instead, it maps selected coronavirus mechanisms to explicit search operators, including elite-guided attraction, trial-vector generation, adaptive parameter variation, stagnation recovery, and population-size scheduling. The algorithm combines opposition-based initialization, current-to-pbest mutation, binomial crossover, an external archive, success-history adaptation, population reduction, and partial restart. COA is evaluated on 29 CEC 2017 benchmark functions at 10, 30, and 50 dimensions against 15 competitive optimizers. Results show that COA achieves the best overall Friedman rank across all dimensions, with particularly strong performance on composition functions. The findings demonstrate that COA is a compact, transparent, and competitive adaptive evolutionary optimizer, while also highlighting limitations on some hybrid functions and the need for further high-dimensional validation.

Figures & tables

Figure 1: Viral-load dynamics as population-size reduction in COA. The left part illustrates the conceptual decrease in viral load across early, middle, and late stages. The right part gives the optimization role: the active population size is gradually reduced from NP0NP_{0} to NPminNP_{min} as the number of function evaluations increases. This reduction controls the shift from broad exploration to focused exploitation.
Figure 2: Spike–receptor binding as elite-guided attraction in COA. The biological motif of spike–receptor binding is translated into the search movement from the current solution xix_{i} toward an elite solution xpbestx_{pbest} . The additional difference vector xr1−x^r2x_{r_{1}}-\hat{x}_{r_{2}} prevents purely greedy movement and helps maintain exploratory mobility.
Figure 3: Viral replication as mutation and binomial crossover. The figure describes how COA generates a new candidate solution. First, mutation creates a mutant vector viv_{i} . Then, binomial crossover forms the trial vector uiu_{i} by selecting some dimensions from the mutant vector and the remaining dimensions from the parent vector xix_{i} .
Figure 4: Antigenic drift as success-history adaptation of FF and CRCR . The figure shows the adaptive loop used in COA. Successful trials contribute values to the memories MFM_{F} and MCRM_{CR} , and future mutation and crossover parameters are sampled around these memories. This creates controlled exploratory variation rather than static or manually tuned parameter behaviour.
Figure 5: Immune evasion as opposition-based partial restart. The figure shows how COA replaces the weakest part of the population when stagnation occurs. Poor candidates are mapped to opposition-based alternatives inside the bounded domain, while stronger candidates and the best-so-far solution are preserved.
Figure 6: Overall flowchart of the Coronavirus Optimization Algorithm (COA). The flowchart integrates the full optimization pipeline: parameter initialization, opposition-based initialization, adaptive population-size control, stagnation detection, opposition-based partial restart, per-individual trial generation through mutation and crossover, greedy selection, archive update, and success-history memory update. It serves as a visual complement to Algorithm 2 and shows how the main operator blocks interact during one complete optimization run.
OperationTimeFrequency per generation
Population sortingO(NPlog⁡NP)O(NP\log NP)1
Mutation (vector)O(D)O(D)NPNP
Crossover (vector)O(D)O(D)NPNP
Boundary projectionO(D)O(D)NPNP
Objective evaluationCfC_{f}NPNP
Archive pruningO(NP)O(NP)≤1\leq 1
Table 3: Per-generation computational complexity of COA operations at population size NPNP .
Figure 7: Average Friedman ranking across 29 functions at D=10D=10 . Lower rank is better.
Figure 8: Signed log⁡10\log_{10} mean objective-value heatmap across 29 CEC 2017 functions at D=50D=50 . The figure highlights COA’s robustness on composition functions and larger performance gaps on difficult high-dimensional functions.
Figure 9: Number of best or tied-best mean results per algorithm at D=50D=50 . Values in parentheses indicate strict wins.
Figure 10: Signed log⁡10\log_{10} mean objective-value heatmap across 29 CEC 2017 functions at D=10D=10 . Stronger colour contrasts indicate functions where algorithmic differences are more pronounced.
Figure 11: Signed log⁡10\log_{10} mean objective-value heatmap across 29 CEC 2017 functions at D=30D=30 . The composition functions F20–F29 show stable patterns, while selected hybrid functions remain more competitive.
Figure 12: Number of best or tied-best mean results per algorithm at D=10D=10 . Values in parentheses indicate strict wins.
Figure 13: Number of best or tied-best mean results per algorithm at D=30D=30 . Values in parentheses indicate strict wins.
Figure 14: Average Friedman ranking across 29 functions at D=30D=30 . COA remains the top-ranked method, followed by AO, CMA-ES, and JADE.
Figure 15: Average Friedman ranking across 29 functions at D=50D=50 . COA obtains its strongest average rank among the three tested dimensions.
Figure 16: Convergence profiles across 29 CEC 2017 functions at D=10D=10 .
Figure 17: Convergence profiles across 29 CEC 2017 functions at D=30D=30 . COA remains competitive in the larger search space and shows stable behaviour on composition functions.
Figure 18: Convergence profiles across 29 CEC 2017 functions at D=50D=50 . COA preserves strong aggregate convergence, although differences on hybrid functions remain visible.
ParameterValueRole and rationale
DimensionsD=10,30,50D=10,30,50Tests low- and medium-dimensional behaviour.
Function evaluations50,000; 300,000; 500,000Budgets for D=10D=10 , D=30D=30 , and D=50D=50 , respectively.
Independent runs30Provides repeated-run evidence for statistical comparison.
Initial population NP0NP_{0}30Compact population retained after opposition-based initialization.
Minimum population NPminNP_{min}8Maintains a small late-stage population for exploitation.
Memory size HH60Stores successful parameter tendencies across generations.
Table 12: COA parameter values used in the experiments.
AlgorithmD=10D=10D=30D=30D=50D=50
COA1.24×10−21.24\times 10^{-2}8.91×10−28.91\times 10^{-2}1.53×10−11.53\times 10^{-1}
AO3.71×10−23.71\times 10^{-2}1.42×10−11.42\times 10^{-1}2.87×10−12.87\times 10^{-1}
JADE4.56×10−24.56\times 10^{-2}2.18×10−12.18\times 10^{-1}4.02×10−14.02\times 10^{-1}
CMA-ES2.89×10−22.89\times 10^{-2}1.75×10−11.75\times 10^{-1}3.14×10−13.14\times 10^{-1}
Table 18: Mean standard deviation across 29 functions. Lower = more consistent.
VariantAvg. rankRank loss vs. COA
COA (full)2.86—
COA noOpp{}_{\text{noOpp}}4.12+1.26
COA noArc{}_{\text{noArc}}5.83+2.97
COA noSHA{}_{\text{noSHA}}6.91+4.05
COA noRes{}_{\text{noRes}}4.75+1.89
COA noPop{}_{\text{noPop}}5.14+2.28
Table 19: Ablation study at D=30D=30 : average Friedman rank across 29 functions.
γ\gamma0.30.50.7 (default)0.91.01.5
Avg. rank4.183.422.863.153.675.23
Table 20: Sensitivity of COA average rank to γ\gamma at D=30D=30 .
ComparisonMinMedianMax∣d∣>0.8\lvert d\rvert>0.8 count
COA vs. AO−2.14-2.14−0.63-0.631.281.2812/29
COA vs. GWO−3.87-3.87−0.91-0.912.452.4518/29
Table 21: Cohen’s dd distribution for COA vs. AO and GWO at D=30D=30 . Negative favors COA.
Categoryκ×103\kappa\times 10^{3}
Unimodal8.42
Multimodal3.17
Hybrid1.84
Composition2.53
Table 22: Average convergence rate κ\kappa over first 30% of generations at D=30D=30 .
More in this topic

Evolutionary Optimization

Evolutionary optimization searches for strong solutions through population-based variation, selection, and adaptation.

See all 293 papers

Explore similar work

CardsList
Apr 29, 2026

RCMAES: A Robust CMA-ES Variant for CEC2026 Competition

This paper proposes RCMAES, a novel variant of the Covariance Matrix Adaptation Evolution Strategy (CMA-ES) for CEC benchmark optimization. RCMAES integrates a dimension-dependent nonlinear population-size reduction strategy with an adaptive restart mechanism within a pure CMA-ES framework. RCMAES is evaluated on three benchmark suites (CEC2017, CEC2020, and CEC2022) and compared with state-of-the-art DE algorithms as well as its closely related counterpart, BIPOP-aCMAES. Experimental results show that RCMAES achieves competitive and robust performance across all benchmarks.

Khoirul Faiq Muzakka, Sören Möller, Martin Finsterbusch

Institute of Energy Materials and Devices (IMD-2), Forschungszentrum Jülich GmbH, Germany

Evolutionary OptimizationBlack-Box OptimizationEvolution StrategiesCovariance Matrix Adaptation Evolution Strategy
AskPDF
Aug 13, 2026

Large-scale Testing Global Optimization Methods with Black-box Adversarial Attacks

Existing global optimization benchmark suites are of a moderate size and are based on a small number of analytical functions that date back even to the 1970s. This causes a risk of biasing the development of global optimization methods. We argue that the tasks related to the black-box adversarial attack (BBAA) can serve as valuable global optimization benchmark in many-dimensional space. We demonstrate the efficiency of several types of evolutionary algorithms and other metaheuristics in solving example BBAA problems. Thus, we take a step towards convergence of global optimization methods to the challenges and needs that arise in the modern machine learning field.

Wojciech Zarzecki, Jarosław Arabas

Warsaw University of Technology, Institute of Computer Science

Evolutionary OptimizationBenchmark DesignMetaheuristic OptimizationGlobal Optimization
AskPDF
Mar 25, 2026

A Longitudinal Analysis of the CEC Single-Objective Competitions (2010-2024) and Implications for Variational Quantum Optimization

This paper provides a historical analysis of the IEEE CEC Single Objective Optimization competition results (2010-2024). We analyze how benchmark functions shaped winning algorithms, identifying the 2014 introduction of dense rotation matrices as a key performance filter. This design choice introduced parameter non-separability, reduced effectiveness of coordinate-dependent methods (PSO, GA), and established the dominance of Differential Evolution variants capable of preserving the rotational invariance of their difference vectors, specifically L-SHADE. Post-2020 analysis reveals a shift towards high complexity hybrid optimizers that combine different mechanisms (e.g., Eigenvector Crossover, Societal Sharing, Reinforcement Learning) to maximize ranking stability. We conclude by identifying structural similarities between these modern benchmarks and Variational Quantum Algorithm landscapes, suggesting that evolved CEC solvers possess the specific adaptive capabilities required for quantum control.

Vojtěch Novák, Tomáš Bezděk, Ivan Zelinka +2

Department of Computer Science, Faculty of Electrical Engineering and Computer Science, VSB - Technical University of Ostrava, Ostrava, Czech Republic · IT4Innovations National Supercomputing Center, VSB - Technical University of Ostrava, 708 00 Ostrava, Czech Republic · Department of Informatics and Statistics, Marine Research Institute, Klaipeda University, Lithuania +3

Evolutionary OptimizationDifferential Evolution
AskPDF