Linear Ensemble Algorithm: A Novel Meta-Heuristic Approach for Solving Constrained Engineering Problems

This research This study presents the Linear Ensemble Algorithm (LEAL), which couples evolutionary search with local, linear regression–based surrogates and a neighbor-guided linear combination scheme for constrained engineering problems and standard benchmarks. For single-objective problems, LEAL frequently attains or closely approaches global optima on multimodal functions such as Rastrigin and Griewank, yielding ~55–85% lower mean error than GA, DE, and PSO, and it can occasionally uncover best-known minima in engineering tasks (e.g., Pressure Vessel), indicating an ability to exploit intricate design trade-offs. For multi-objective problems, LEAL generates feasible Pareto fronts but generally trails NSGA-II in convergence and efficiency, exhibiting higher GD⁺, longer runtimes, and greater memory usage (often by one to two orders of magnitude). These outcomes reflect the computational overhead of maintaining local surrogate ensembles: while LEAL can produce high-quality solutions, its average performance, runtime, and memory footprint are often inferior to lightweight baselines. Comparisons with CMA-ES, Bayesian Optimization, and SSA-NSGA-II confirm the same trade-offs. Overall, LEAL is a robust yet computationally intensive option best suited when ultimate solution quality outweighs runtime; future work will focus on improving efficiency, streamlining ensemble components, and extending applicability to large-scale and dynamic problems.