Model Menu (ASCMO-DYNAMIC)

The Model menu consists of the following entries:

Configurations

Opens the "Model Configurations" window where you can define model settings individually for each output in a separate tab.

Configurations for All

Opens the "Model Configurations" window where you can define model settings for all outputs.

Train Models

You can choose to train Untrained Models or to train Models with changed data.

Configuration Manager

Opens the "Model Configuration Manager" Window where you can edit, compare, and select configurations for training.

Model Manager

Opens the "Model Manager" window, which gives you an overview off all existing models and allows you to edit them.

Automated Machine Learning

Opens the Automated Machine Learning where you find hyperparameters automatically for Recurrent Neural Networks and NARX modeling method: Cell type, no. of layers, memory size and optimizer settings.

NARX Feature Search

Opens the "NARX Feature Search" window, where you can manage and perform a NARX feature search.

Set Working Configurations as References

Sets the current configurations of the working model as configurations for the reference model.

Delete Configurations

Opens the "Delete Configurations" window where you can choose which configurations to delete. Models depending on those configurations are also deleted.

Activate the checkboxes of the configurations you want to delete.

(De)Select All activates or deactivates all checkboxes.

OK closes the window and accepts your settings.

Cancel closes the window and discards your settings.

Delete Unused Configurations

Deletes all unused configurations.

Delete all but Working Configurations

Deletes all configurations except the working configuration.

NARX Model Options

This submenu contains options for NARX Structure models.

One Step Ahead Prediction

Use these menu options to select the mechanism of the model prediction.

Multi Step Ahead Prediction

Initial State of NARX Values

Opens the "Initial State of NARX Values" window where you can determine the way the initial NARX values are determined.

See Initial NARX Values for details.

Measured vs. Predicted

Opens a scatter plot window that plots the measured data against the model prediction. The expected result is a diagonal line, which would indicate a perfect match of the prediction and the measurements. In addition, scatter plots of the inputs and the datasets are shown.

Input Relevance (RMSE)

You can choose to open the "Input Relevance" Window (ASCMO-DYNAMIC ) for all or a single output where you can define settings for the "Relevance of Inputs" Window (ASCMO-DYNAMIC) .

Show Statistics

Opens the Closed Statistics window that shows statistical information on the project.

Anomaly Detection: Receiver Operating Characteristic

Uses the Receiver Operating Characteristic (ROC) curve to evaluate the performance of classifiers for anomaly detection and to measure the method performance. It can be used to visualize the performance. The ROC curve can be used to evaluate true and false positives (Classification: True vs. False and Positive vs. Negative).

Anomaly Detection: Visualization

Opens the "Anomaly Detection Scope View" window.

You can find out if the model reconstruction worked, thus if the model is good. In case of a successful reconstruction, you can identify which signal has the anomalies.

Model Memory Impact

Opens the "Model Memory Impact" window, which provides a tool to show the impact of model memory on model evaluation and to facilitate the selection of a stable model.

Cross Validation on Training Set

Opens the Closed "Cross Validation" window, where you can enter the number of cross validations and start the procedure. See also Cross Validation on Training Set.

CCR Validation

Note
This item is only available when the Advanced Settings are enabled.

Performs a cross-correlation analysis (CCR), i.e. checks the correlation of the selected residuals.

Residuals Training Data	Opens the "Cross Correlation of Residuals" window, which shows the auto-correlation of the prediction's residuals regarding the training or test data. The desired behavior is an exponentially decreasing correlation with a correlation of 1.0 at time lag 0.
Residuals Test Data
Residuals and Inputs Training Data	Opens the "Cross Correlation of Residuals and Inputs" window, which shows the cross-correlation between the residuals and the inputs for training data or test data. The expected cross-correlation is zero, without any significant peaks.
Residuals and Inputs Test Data