![]() |
![]() |
![]() |
||
SAAPS, Version 1.01 |
||||
Within SAAPS different models for the prediction of satellite anomalies have been developed. Each model is satellite specific in the sense that it has been optimize to predict the anomalies for one satellite anomaly data set . With this tool the user can submit a list of anomaly events and receive the model that best predicts the events in the data set.
When pushing the "Edit Events ..." button a dialogue window appears into which a list of anomaly events should be entered. Each row in the text field should contain the time of an event. Each row may contain the year, month, day, hour, minute, and second separated by spaces. Not all fields need to be present, e.g. it is sufficient to only specify the year, month, and day fields. Only the six first columns on each row will be read, any additional text will be omitted. Push "OK" when finished.
Times when there were no anomaly events must also be given. This can be done in two ways. Either one explicitly enters the no-anomaly data by pushing the "Edit Non-Events ..." button. Or one can push the "Random Non-Events ..." button which will select no-anomaly times at random over the time range given by the anomaly set, excluding days of anomalies. The user should enter the number of random events that should be used. The default is to set the number of non-anomalies equal to the number of anomalies.
The optimization algorithm can use either linear correlation or conditional probability as the performance measure. The two measures usually give the same result.
For a model to be considered useful the probability must be larger than 0.5, and preferably 0.8 or larger. A probability of 0.5 is approximately the same as having a linear correlation of 0 (zero), and probabilities less than 0.5 means a negative linear correlation.
When the "Submit ..." button is pushed the each anomaly prediction model will be run for the user submitted events. This may take some time so it is advisable to test first on a smaller set using e.g. 100 events. This should take around 30 seconds.
When the models have been run a ordered list will appear with the best model as the first element. Selecting a model will display the result.
Peter Wintoft |
![]() |
![]() |
![]() |
|