A Search History-Driven Offspring Generation Method for the Real-Coded Genetic Algorithm

Area of Science:

• Evolutionary Computation
• Artificial Intelligence
• Optimization Algorithms

Background:

• Genetic algorithms generate offspring iteratively, creating valuable search history.
• Real-coded genetic algorithms (RCGA) can benefit from improved offspring generation strategies.

Purpose of the Study:

• To propose and evaluate a novel crossover model, Search History Crossover (SHX), for RCGA.
• To enhance RCGA performance by exploiting cached search history in an online manner.
• To develop a data-driven method for offspring selection that requires no additional fitness evaluations.

Main Methods:

• Survivor individuals from past generations are archived to form the search history.
• The search history is clustered, and each cluster is assigned a score.
• A crossover model (SHX) is introduced, driven by this scored search history for offspring selection.

Main Results:

• SHX significantly enhances the performance of RCGA on 15 benchmark functions.
• Improvements were observed in both solution accuracy and convergence speed.
• The additional runtime introduced by SHX is negligible compared to the overall processing time.

Conclusions:

• The proposed SHX model effectively leverages search history to improve RCGA performance.
• SHX is particularly beneficial for optimization tasks with limited budgets or expensive fitness evaluations.
• SHX offers a computationally efficient method for boosting evolutionary algorithm performance.