Supervised Learning for Supervisory Control of Discrete-Event Systems

Abstract

This project integrates supervised learning techniques with Supervisory Control Theory (SCT) to enhance the control mechanisms of Discrete-Event Systems (DES) in complex and unpredictable environments. By applying Logistic Regression, LSTM networks, and Random Forest models, this project aims to develop a data-driven approach to supervisory control, leveraging MATLAB and the MatlabTCT toolkit for implementation.

Introduction

The project addresses the increasing complexity and evolving nature of modern discrete-event systems such as manufacturing lines and transportation systems, which demand a shift from conventional model-based control strategies to more adaptive, data-driven solutions.

Code Structure

• Parameter Initialization: Set parameters like mean Gaussian, sigma values, etc.
• User Interface Setup: Configure options for random number generation, machine learning algorithm selection, and label type selection.
• Data Pre-processing: Includes functions like parser_des, parser_dat, and sup_to_data for preparing the data for subsequent machine learning tasks.
• Label Generation: Functions like label_control_data, label_event_prediction, and label_marker to create appropriate labels for machine learning training.
• Feature Extraction: Functions data_for_LR and data_for_RNN prepare features for Logistic Regression and Recurrent Neural Networks, respectively.
• Machine Learning Models: Implementations of logistic regression, recurrent neural networks (LSTM), and random forest models for classification tasks.

Setup and Running

1. Environment Setup: Ensure MATLAB R2023a and MatlabTCT toolkit are installed.
2. Clone the Repository: git clone https://github.com/rafael-mannarelli/supervised_learning_DES
3. Navigate to the Project: cd path_to_project
4. Execute Main Script: Run the main MATLAB script to initiate the program.

Validation and Results

The system's performance is validated through accuracy metrics calculated against test datasets, demonstrating the efficacy of the integrated machine learning models.

Conclusion

This project provides a robust framework for exploring the potential of machine learning in enhancing the supervisory control of discrete-event systems, suitable for researchers and practitioners looking to incorporate data-driven techniques into their control strategies.

References

• W. M. Wonham and K. Cai. Supervisory Control of Discrete-Event Systems. Communications and Control Engineering. Springer, 2018.