An improved gene expression programming algorithm for function mining of map-reduce job execution in catenary...

Area of Science:

• Computational Intelligence
• Machine Learning
• Data Mining

Background:

• Gene Expression Programming (GEP) is a key algorithm in function mining.
• Accurate function models are crucial for optimizing configuration parameters-execution efficiency (CP-EE) in map-reduce jobs, particularly for high-speed railway catenary monitoring systems.

Purpose of the Study:

• To develop a novel algorithm, Multi-Strategy Gene Expression Programming (MS-GEP), for more accurate function modeling in CP-EE.
• To enhance GEP's ability to escape premature convergence and local optima.

Main Methods:

• The MS-GEP algorithm incorporates an adaptive mutation rate based on evolutionary generations, population diversity, and fitness values.
• A manual intervention strategy identifies and addresses local optima based on evolutionary stagnation.
• Population diversity is managed through random individual replacement and ancestral population tracing to alter evolutionary direction.

Main Results:

• MS-GEP demonstrated superior solution quality and increased population diversity compared to traditional GEP algorithms on benchmark function mining tasks.
• The algorithm achieved higher accuracy in modeling the CP-EE of high-speed railway catenary monitoring systems.

Conclusions:

• MS-GEP offers significant improvements over standard GEP for function mining tasks.
• The proposed strategies effectively prevent local optima and enhance overall model accuracy, showing promise for complex system optimization.