DA2MODE: Dynamic Archive with Adaptive Multi-Operator Differential Evolution for numerical optimization

Abstract

This paper presents Dynamic Archive with Adaptive Multi-Operator Differential Evolution (DA2MODE), a new algorithm that aims to boost the performance of meta-heuristic and evolutionary methods in numerical optimization. DA2MODE introduces a Progressive Adaptive Selector with Exponential Smoothing (PASES), which dynamically updates the selection probabilities of both mutation and crossover operators. Unlike prior approaches that emphasize only mutation operators or rely on short-term success within the current generation, PASES adapts based on cumulative operator performance over time, thus favoring the best-performing operators more reliably. DA2MODE employs an Adaptive Non-Elite Archive Update (ANEAU) mechanism that injects a controlled fraction of non-elite solutions into the archive. ANEAU promotes early exploration, which is gradually reduced to strengthen exploitation. Additionally, the control parameters (crossover probability and mutation factor) are automatically tuned in DA2MODE, allowing full adaptivity of both operator selection and parameter control. Extensive experiments on the CEC2017/2018, CEC2020-2022, and 1000-dimensional CEC2013 benchmarks, along with four real-world engineering design problems, confirm that DA2MODE consistently outperforms 33 competitive algorithms, including CEC winners and recent advanced DE variants. It achieves top performance across all statistical tests, demonstrating superior convergence speed and final accuracy. These results establish DA2MODE as a robust, scalable, and reliable algorithm for solving complex numerical optimization problems. The source code of the DA2MODE algorithm is publicly available at: URL https://github.com/MohamedRedaMu/DA2MODE-Algorithm and URL https://uk.mathworks.com/matlabcentral/fileexchange/182019-da2mode-algorithm.