Multi-Gene Genetic Programming-Based Identification of a Dynamic Prediction Model of an Overhead Traveling Crane

Area of Science:

• Robotics and Control Systems
• Computational Intelligence
• System Identification

Background:

• Accurate dynamic modeling of overhead cranes is crucial for effective control.
• Traditional methods often require expert knowledge or involve complex, time-consuming nonlinear dynamics modeling.
• A need exists for automated, accurate, and efficient dynamic prediction models.

Purpose of the Study:

• To propose a novel multi-gene genetic programming (MGGP) approach for identifying overhead crane dynamic prediction models.
• To develop a method that avoids reliance on expert knowledge and compromises between accuracy and model complexity.
• To optimize both prediction accuracy (Mean Square Error) and model complexity.

Main Methods:

• Employed a multi-objective optimization framework for MGGP, minimizing Mean Square Error (MSE) and function complexity.
• Utilized a least squares approach for initial gene weight estimation.
• Applied the Levenberg-Marquardt algorithm for local optimization of k-step ahead predictors.
• Validated the approach on both simulated (Euler-Lagrange with friction) and experimental overhead crane systems.

Main Results:

• The MGGP approach successfully identified dynamic prediction models for overhead cranes.
• Models were trained and validated using diverse control inputs, rope lengths, and payload masses.
• The method demonstrated effectiveness in both simulation and real-world experimental settings.
• Achieved a balance between prediction accuracy and model simplicity without expert input.

Conclusions:

• The proposed MGGP method is effective for automated dynamic prediction model identification in overhead cranes.
• This approach offers a robust alternative to traditional modeling techniques, enhancing accuracy and reducing complexity.
• The findings highlight the potential of MGGP in complex system modeling and control applications.