Skip to content

Latest commit

 

History

History
58 lines (31 loc) · 3.13 KB

ICPM2019_ppm.md

File metadata and controls

58 lines (31 loc) · 3.13 KB
Raw

Predictive Performance Monitoring of Material Handling Systems Using the Performance Spectrum

This document contains additional materials for paper Predictive Performance Monitoring of Material Handling Systems Using the Performance Spectrum (ICPM2019) (in press).

Event logs of the simulation model of a Baggage Handling System (BHS)

The BHS simulation event log and the computed Performance Spectrum (see the table below) are available along with other release files here. The source code is available in this branch of the PSM project.

File Description
PPM_BHS_Sim_log.zip The event log in the XES and CSV formats.
PPM_BHS_Sim_PerfSpec.zip The computed Performance Spectrum in the format of the PSM v.1.1.6
perf_spec-assembly-1.1.6.jar The PSM v.1.1.6
ppm-assembly-1.1.6.jar The binaries for running the command line scripts descibed below

Running the simulation model

Class for running Command line arguments
org.processmining.scala.sim.conveyors.experiments.PreSorterStarterCli output_directory days_to_simulate start_offset_hours duration_hours

For example, command line

java -cp ppm-assembly-1.1.6.jar org.processmining.scala.sim.conveyors.experiments.PreSorterStarterCli g:\logs 7 10 12

triggers simulation for 7 days, operating hours start at 10:00am, duration of operating hours is 12 hours.

Model training

Before training the PyTorch framework should be installed and configure for your Python environment. The Python scripts for training are located here.

Model training can be configured in the main file fnn_app.py:

Parameter Description
is_linear Use a Logistic regression model if True and Feedforward (FF) NN if False
hidden_layers Number of the hidden layers of the FF NN. Change also number of layers in the code of class PreSorterFnn
hidden_dim Size of the hidden layers
experiment_id Root folder of the training/test sets in the common folder for the datasets ..data
n_splits k for k-fold cross-validation

There are more parameters for fine-grained tuning, their meaning is clear from naming.

As the output, the scripts summarize values of MSE, MAE and R squared and can also generate the set with predicted values for the test set (in the subfolder test).

Data-Driven Inter-Case Feature Encoding

The Data-Driven Inter-Case Feature Encoding is implemented in class DdeInterCaseFeatureEncoder, package org.processmining.scala.prediction.preprocessing.ppm_experiments.intercase. The sample code is in ProductionLogExample.