Exploiting Evolutionary Computation Techniques for Service Industries
Well-controlled mechanical systems allow the development of useful and successful automated applications to benefit the service industries in Society 5.0. These systems use effective controllers to govern their dynamic behaviors, and all controllers need correct parameter tuning to maximize the syst...
Gespeichert in:
Hauptverfasser: | , , , , , |
---|---|
Format: | Buchkapitel |
Sprache: | eng |
Online-Zugang: | Volltext |
Tags: |
Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
|
Zusammenfassung: | Well-controlled mechanical systems allow the development of useful and successful automated applications to benefit the service industries in Society 5.0. These systems use effective controllers to govern their dynamic behaviors, and all controllers need correct parameter tuning to maximize the system performance and prevent undesired effects. The adjustment of controllers typically requires an accurate system model to make output predictions and evaluate performances without damaging the actual system. Due to uncertainties and unmodeled dynamics, obtaining such an accurate model is challenging, and the use of meta-heuristics from evolutionary computation or swarm intelligence is necessary. This work proposes an evolutionary machine learning framework applied to the identification and control of mechanical systems Based on the widely-studied Cross-Industry Standard Process for Data Mining (CRISP-DM). In this framework, the actual system is excited with particular control signals to generate valuable input/output information to train and test a model. Since well-established laws from classical mechanics are available to describe mechanical systems, the system model is obtained through Lagrangian analyses. Its adjustable parameters are related to the dynamic ones of the system. Different model alternatives are obtained by meta-heuristics and then evaluated to choose the most suitable. The selected model is used in the controller tuning through meta-heuristic optimization. Finally, the optimized controller parameters can be implanted in the actual system. Numerical simulations show the proposal's effectiveness for two study cases: the position regulation of a simple pendulum and the trajectory tracking with a fully-actuated inverted pendulum.
This chapter presents an Evolutionary Machine Learning Framework for Identification and Control of mechanical systems to integrate them into Society 5.0. While the system is being excited, its state variables must be measured or observed in short sampling intervals dt. The generated data is then archived in the sets of features and targets. Once the set of targets and features contains the generated system data, it is divided into two subsets, one to train the model and the other to evaluate it. Typically, 80% data is used for training and the remaining 20% for evaluation. As mentioned before, meta-heuristics are stochastic computational techniques that can find good solutions to hard optimization problems using an affor |
---|---|
DOI: | 10.1201/9781003158165-8 |