TRUST-TAEA: New algorithm optimizes problems with 5000 variables

Large-scale multi-objective optimization (LSMOP) is notoriously difficult: high-dimensional spaces, complex variable interactions, and limited function evaluations create tension between convergence, diversity, and stability. Existing two-archive evolutionary algorithms (TAEAs) try to separate convergence and diversity, but often underutilize archive reliability and problem structure, leading to inefficient search and incomplete front coverage.

TRUST-TAEA, developed by JunYi Cui, introduces a trustworthiness metric that integrates evolutionary progress with convergence-archive maturity. This metric coordinates three key mechanisms: variable-grouping sparse search (to efficiently explore high-dim spaces), anchor-probing compensatory search (to fill gaps in the Pareto front), and archive stabilization (to prevent late-stage drift). The algorithm was evaluated on the LSMOP benchmark suite with 500–5000 decision variables and two or three objectives, plus a real-world three-objective day-ahead scheduling case for a grid-connected microgrid.

Results show TRUST-TAEA achieves superior or highly competitive performance across convergence, diversity, and stability metrics. In the microgrid case, it obtained the best IGD⁺ value and generated a feasible dispatch strategy balancing cost, emissions, and grid-power fluctuation. The work demonstrates how trustworthiness-guided search can systematically exploit archive information and problem structure to solve previously intractable large-scale optimization problems.