.png){kind=icon}
.png){kind=icon}
Return
Online spatial-temporal prediction for dynamic constrained multiobjective evolutionary optimization
.png){kind=icon}
PRE
AI
DOI：10.1016/j.eswa.2026.132101.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
The presence of dynamics in dynamic constrained multiobjective optimization problems (DCMOPs) causes the various changes of Pareto optima. The existing methods extract partial historical knowledge to initialize the population at a new time, but neglect inherently temporal and spatial characteristics of dynamic Pareto optima, showing insufficient tracking performance. To solve this issue, a spatial-temporal prediction strategy based dynamic constrained multiobjective evolutionary algorithm is designed in this article, called STPS. Once an environmental change appears, the knowledge construction strategy converts the Pareto optima into the two-dimensional image. All historical images constitute a spatial-temporal series to train the prediction model based on convolutional neural network (CNN) and gated recurrent unit (GRU), initializing a population under the new environment. In addition, an incremental learning strategy is designed to periodically fine-tune the predictor, guaranteeing the prediction accuracy in adapting to the time-varying environments. The intensive experiments on 10 mainstream benchmarks and a real-world case verify that, compared with several state-of-the-art dynamic constrained multiobjective evolutionary algorithms, the proposed algorithm achieves prominent performance in solving DCMOPs.
Keywords:
Dynamic constrained multiobjective optimization
Spatial-temporal prediction
Convolutional neural network
Gated recurrent unit
Population initialization
Journal
IF:
7.5
Papers:
2.9W
Citations:
10.2W
View more organizations