Influence factor discovery

Influence factors enable you to identify the most significant linear relationships between a tag of interest and one or more potential influence factors for a given time period.

In practice, the influence factors feature can be used to identify possible contributors to an abnormal situation, or to generate simple soft sensors to explain or predict lab measurements, by a linear combination of measured tags.

How to use influence factors on active tags

The influence factor analysis can be performed on all active tags or on all indexed tags. This section explains how you can perform the influence factor analysis on the active tags. Analysis for all indexed tags is discussed in the following section.

  1. Open the tags menu .
  2. Load in all relevant tags related to the process or problem in the 'Active Tags' list.
  3. Navigate to the time period you want to analyse in the focus chart. 
  4. Open the influence factors menu .
  5. Select the 'Tag of interest' you want to analyse.
  6. Optional: click the 'Advanced settings' hyperlink (see the automatic shift identification section of this article for more information):
    • Toggle the 'Automatic shift identification' field.
    • Enter a maximum time shift in the 'Maximum shift' field.
  7. Click the 'Analyze' button. A list of 'Candidate Factors' will be shown beneath the button. See the result interpretation section of this article to learn how to interpret the influence factors results.
  8. Click the plus icon next to a result to add a 'Candidate'.

Note: A number of 'Candidates' can be added. Candidates can be removed again by clicking the remove icon .

Optional: Click the save icon to save the linear combination as a formula tag.

Note: digital tags can be included in an influence factors analysis but are translated into numeric codes. This means that state changes that influence the tag of interest will be detected but the regression itself cannot be extrapolated.

How to use influence factors on global tags

TrendMiner allows to perform an influence factor analysis on all indexed tags:

Note: the tags must be indexed prior to the influence factor analysis. Tags that are not indexed will not be considered by TrendMiner. This should be taken into account when analysing the results.

  1. Navigate to the time period you want to analyse in the focus chart. 
  2. Open the influence factors menu.
  3. Select the 'Tag of interest' you want to analyse.
  4. Check off the 'Global Search' tick box.
  5. Optional: enter a filter expression to search through tags containing a specific text only (e.g. when entering 'area1-*' in the 'Filter expression' field, only indexed tags containing 'area1-' as their prefix will be searched through.
  6. Optional: click the 'Advanced settings' hyperlink (see the automatic shift identification section of this article for more information):
    • Toggle the 'Automatic shift identification' field.
    • Enter a maximum time shift in the 'Maximum shift' field.
  7. Click the 'Analyse' button. A list of 'Candidate Factors' will be shown beneath the button.
  8. Click the plus icon  next to a result to add a 'Candidate'. Note that more than one 'Candidate' can be added. Candidates can be removed again by clicking the remove icon .
  9. Optional: click the save icon  to save the linear combination as a formula tag.

Note: digital tags can be included in influence factors but are translated into numeric codes. This means that state changes that influence the tag of interest will be detected but the regression itself cannot be extrapolated.

How to interpret influence factor results

Different candidate factors will be listed after analysis. A candidate factor explains how the tag of interest would look like (approximation) based on the values of the selected candidate tags. The list includes the name and description of the influencing tags. This list is ranked by the improvement of the approximation of the tag of interest which can be gained if a factor would be included.

Initially, no candidate factor is selected and a dashed line is visible in the focus chart. This dotted line represents the approximation of our tag of interest and is equal to the mean value of the tag of interest (in the selected time window of the focus chart) when no candidate factor is selected.

Adding a candidate factor will change the dashed line and the percentual approximation scores are updated. The linear combination between the candidate factors and the tag of interest can be obtained as a formula tag by clicking the 'Save' button.

Automatic shift identification

Within TrendMiner the influence factors feature also has the options to automatically detect lagged influences and search through all of the indexed tags (global search). These features can be toggled on or off by the user via the dedicated check mark boxes.

When automatic shift identification is selected, performing an influence factors analysis will find the optimal alignment in time of all other tags with regard to the selected tag of interest. The maximum shift parameter sets how far in the past alignments of candidate factors are checked. In the screenshots the results are shown when calculating the influences of a tag of interest (tm-demo-S1-TI22114) with and without the automatic shift identification activated.

Note: Choose your maximum shift carefully based on process knowledge. The search resolution can decrease with this parameter. The search resolution is equal to the index resolution until a maximum of 100 intervals is reached. From this point, 100 intervals are equally chosen from the period in your focus chart, up to the maximum shift you define. 

  • Without automatic time shift detection:          

  • With automatic time shift detection: 

The optimal alignment of candidate X with the tag of interest Y is a shift of 1 minute. This indicates that the candidate contains information about the value of the tag of interest, 1 minute into the future. The percent approximation score when selecting a factor will also take this shift into account.

Note: selecting automatic shift identification will reset any manual shifts conducted previously. The shifts that are indicated in the results persist into the current settings of the corresponding tags in the 'Active Tags' list.

Limitations

In order to prevent memory issues the Influence Factor functionality is restricted to;

  • 3 selected 'candidate'-factors (Although, the number of candidates you can add in Influence Factors is dependent on the resources on the system;
  • a selected time period of 2 years in the Focus Chart.
 
-- If your TrendMiner screens do not match the screens you see above, this means you probably run an older version of TrendMiner. Contact your administrator for upgrade options.