Overview
Installation.
Run the System.
Reference.

Overview

Nnet is a Windows-based neural network development environment for solving the popular traveling salesman problems (TSP) and flow shop scheduling problems. The neural network paradigms used are Hopfield-Tank (HT) net and Simulated Annealing Neural Network (SANN). The system was written by Microsoft Visual Basic 3 (professional version) and Visual C++ (version 1.5). The user interface was developed using VB to enforce the user-friendliness and the algorithm was coded in VC to maintain computational efficiency.

Installation

A student version of Nnet (restricted to no more than 12 cities for TSP and 12 jobs for FSS) was stored at my course server (under directory of m528\systems\nnet). The files were compressed to reduce storage space. First, download the system into a PC or a disk. To run the system, you must first install the system to a hard drive. To simplify the installation process, a batch file "install.bat " was provided. Simply run "install.bat" from Windows would copy and uncompress all related files into "C:\nnet_opt" and sample data files into "C:\nnet_opt\data". To remove Nnet from the hard drive, simply run "uninstall.bat".

Run the System

Once the system was installed, you can run "nnet_opt.exe" from Windows. A screen which shows the selected network type, location of parameter file, and four buttons ¾ Assign Parameters, Edit Current Parameters, Run Network, and Tools ¾ will be displayed.

Lists below are the proper sequence (procedure) of using the system:

1. Select a network type, either Hopfield-Tank or Simulated Annealing.
2. Enter parameters or retrieve a parameter file for the selected network type. All parameter files have an extension of "par". There are several sample files available in the "data" subdirectory. "City10.par" is the popular 10 city TSP problem for using HT and "SANN10.par" is the corresponding file for using SANN. "F7_7.par" and "F8_8.par" are FSS files using HT and "SA7_7.par" and "SA8_8.par" are the corresponding files using SANN. Where "F7_7" means the problem contains 7 jobs to be processed by 7 machines.
3. Generate or select an initial permutation matrix. The files have an extension of "ipm". These file can be randomly generated by the computer or retrieve from an existing database. The dimension of the file should match the parameter file. We use the same notation (e.g., city10.ipm) for easy tracking.
4. Enter or retrieve a data matrix file. The files have an extension of "dsm" for TSP or "pst" for FSS. The "dsm" files can be entered directly or generated by entering the coordinate of the cities. The matrix for dsm files can be updated by clicking the value to switch from "1" to "0" or reverse.
5. Once all the required data has been entered, the "Run Network " button can be click to process the result. The results will be displayed in the screen, where several additional options ¾ print, view matrix, view histogram, and view statistics ¾ can be selected.
6. If you would like to review previous processed results, the "Tools" button can be selected for this purpose. The results files have an extension of either "cvr" for all results, or "fpm" for final permutation matrix. A tour of TSP can also be plotted by selecting the "postprocessing" module. Again we used the same notation for easy tracking.
7. Finally, the "file" menu provides some utilities for "Open" and updates related data files such as network parameters, initial permutation matrix, distance matrix (TSP), processing time matrix (FSS), etc.

Reference:

• Chang, E. and Chu, C. H., "Neural Network Approaches to Flow Shop Scheduling," Proceedings of the National Decision Sciences Conference,, November 1997.