An Enhanced Nested Partitions Algorithm Using Solution Value Prediction

doi:10.1109/TASE.2010.2090655

Return

An Enhanced Nested Partitions Algorithm Using Solution Value Prediction

2011-04-01

PRE

Weiwei Chen *

Liang Pi

Leyuan Shi

DOI：10.1109/TASE.2010.2090655

Original

Original request for help

Save

Abstract

En 中文

Metaheuristics are an important branch of optimization algorithms that attract lots of research and application efforts. In this paper, the research of predicting solution value for Nested Partitions (NP) is proposed, which is a newly developed metaheuristic algorithm for solving large-scale optimization problems. The lower bound embedded prediction procedures are developed to predict the future performance of NP based on the solution values obtained at early iterations. The prediction procedures are used in an enhanced NP algorithm to select a proper algorithm setting for NP at early stage, which saves a lot of computational resource for large-scale problems. The computational tests show the accuracy and effectiveness of the proposed algorithm. These prediction procedures can be also applied to some other metaheuristics. Note to Practitioners-Metaheuristics are popular choices to solve large-scale optimization problems in the real world. However, different algorithms (or an algorithm with different settings) usually have different performance. The solution quality is unknown without running each candidate to the end, which is a waste of time and resource. Hence, we develop prediction procedures to predict the future performance of a metaheuristic algorithm based on the solution sequence obtained at the early stage of the run. The prediction procedures provide statistical information for the solution value at certain future iteration that is of interest. The procedures are easy to implement, and have low computational cost.

Keywords:

Lower bound

metaheuristic algorithms

nested partitions

solution value prediction

AI Summary

Key information extracted from the uploaded paper, including a brief overview, abstract, background, key highlights, visual analysis, and future outlook.