Genetic Programming Based Identification of an Overhead Crane

Abstract

Overhead cranes carry out an important function in the transportation of loads in industry. The ability to transport a payload quickly and accurately without excessive oscillations could reduce the chance of accidents as well as increase productivity. Accurate modelling of the crane system dynamics reduces the plant-model mismatch which could improve the performance of model-based controllers. In this work the simulation model to be identified is developed using the Euler-Lagrange method with friction. A 5-step ahead predictor, as well as a 10-step ahead predictor, are obtained using multi-gene genetic programming (MGGP) using input-output data. The weights of the genes are obtained by using least squares. The results of 15 different genetic programming runs are plotted on a complexity-mean square error graph with the Pareto optimal solutions shown.

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

“…The method proposed in this paper follows the previous work [ 13 ], in which the MGGP with least square parameter estimation was applied to identify the direct k -step ahead predictors of the crane dynamics by training and validating them with data sets obtained from simulations carried out on a model derived from the Euler-Lagrange equation. The direct prediction-based method developed in [ 13 ] requires n different models to predict over the prediction horizon n .…”

“…The method proposed in this paper follows the previous work [ 13 ], in which the MGGP with least square parameter estimation was applied to identify the direct k -step ahead predictors of the crane dynamics by training and validating them with data sets obtained from simulations carried out on a model derived from the Euler-Lagrange equation. The direct prediction-based method developed in [ 13 ] requires n different models to predict over the prediction horizon n . This approach is less prone to bias compared to the iterated prediction method when there is model misspecification; however, the large number of models required reduces the interpretability of the output dynamics, which is a significant advantage of GP.…”

Soft-Computing-Based Estimation of a Static Load for an Overhead Crane

“…To the best of our knowledge, this is the first work on the crane's payload weight estimation using a data-driven model identified using multi-gene genetic programming and grammar-guided genetic programming with sparse regression. The other genetic programming-based crane modeling approaches used the MGGP for payload sway prediction [42,43]. Crane dynamics modeling and control were also studied by using a genetic algorithm (GA) in [44][45][46][47].…”