Statistical Analysis
The suite of statistical modules consists of:
- Fuzzy Logic prediction
- Multi-linear regression prediction
- Neural network prediction
- Cluster analysis
- Principal component
All modules use a similar multi-well interface where a set of wells and intervals can be used to create a model and then this model can be applied to another group or wells and intervals. Discriminators can be used to limit the data use in the models.
The Fuzzy logic module divides the data up into user selected bins and uses probability theory to predict the likelihood of data being in a bin. The results are normally well controlled and quality control probability curves give the likelihood that the result is in a bin. This allows the output of a probability map, so the user can easily quality control results in wells not used in the model build stage. The module can be used to predict core facies or core permeability, for example.
The Multi-linear regression is useful for predicting core permeability from log data. It uses standard matrix algebra to solve for the fit coefficients. Normalised coefficients are also output to allow the user to see the contribution of each log to the result.
The Neural network module uses a back-propagation learning technique to train the network. The module can be used for log repair, prediction of core permeability or in a classification mode for prediction of core facies.
In log repair mode the user selects a few small training intervals and the trained network can then reproduce the whole log extraordinarily well.
The Cluster Analysis module is use to group log data into electro facies. The program uses K-Mean clustering to group the data into manageable data clusters (15-20). These clusters are then either manually or automatically (hierarchical clustering) regrouped into Geological clusters.
Plot shows a core facies prediction. The fuzziness of the prediction is shown in right track
The plot shows the result of a permeability prediction of core data.
Log plot showing a reconstructed density (track 6) from a Sonic, Gamma ray and neutron log. Only a few small training zones, track 2, were used to create the model
Multi-curve crossplots shows cluster grouping
Electro facies shown in left hand tracks. Far left track is after re-grouping original 15 clusters to 5.