Original title: Location and manifold based dual prediction method for dynamic multiobjective optimization with stochastic changes
Guoping Li, Xicai Deng · September 2026
Many real-world computer problems require balancing multiple goals at once while conditions change unpredictably. Most existing solving methods only work well when changes follow predictable patterns. To tackle random shifts, researchers created a two-step forecasting technique called LMDP. When conditions shift unexpectedly, the algorithm tracks down the best solutions found in past environments. It finds one or two past solution sets nearest to the current one to estimate its new overall position. Next, it examines the geometric shapes of those previous solutions to estimate the exact shape of the new set. Tests on problems with random shifts showed that this dual approach adapts better and outperforms several comparative methods.
Imagine you are trying to find the best spot to set up a lemonade stand in a park, but the park layout changes randomly — sometimes a new bench appears, sometimes a tree falls. Most methods assume the changes happen on a schedule, like a bench appearing every Tuesday. But here, the changes are unpredictable. You look at past layouts that were most similar to the current one, then use the best spots from those to guess where to set up now. You also adjust the shape of your setup based on what worked in similar layouts. In this paper, the 'park' is the problem environment, the 'best spot' is the Pareto set, and the 'shape' is the manifold.
Many real-world problems, like adjusting a factory schedule when demand fluctuates unpredictably or rerouting drones in shifting wind, involve changes that are random, not fixed. Most optimization methods assume the change follows a pattern, so they break when it doesn't. This method keeps working even when the environment shifts in unpredictable ways, making it useful for systems that must adapt on the fly.
The abstract does not specify how the method performs on problems with very rapid or extreme stochastic changes, nor does it compare against methods designed for non-stochastic changes. It also does not discuss computational cost or scalability to problems with many objectives or variables. The tests are limited to DMOPs designed with stochastic changes, so generalizability to other types of randomness is not established.
Ask anything about this paper. Answers come from its abstract, and say so when the abstract doesn't cover your question.
Location and manifold based dual prediction method for dynamic multiobjective optimization with stochastic changes — Guoping Li, Xicai Deng — Scientific Reports — doi:10.1038/s41598-026-72099-5
Abstract licence: CC BY
Explanation generated by Synthesis from the abstract — it is not the original paper.