580 California St., Suite 400
San Francisco, CA, 94104
Figure 2 – uploaded by Chun Che Fung
Table 1 From the user’s viewpoint, it is desirable that the prediction model could generate accurate prediction on data that have not been used in the training process. The accuracy of the prediction of the properties of Well B is therefore of great interest. Two sets of predicted outputs have been generated in this case study: BPNN A that uses ail training data, and BPNN B that only uses core data that are conformed to the heuristic rules derived from the log analyst’s experience. The difference between the predicted outputs as compared to the actual core data is calculated from the Percent Similarity Coefficient expressed in (8). The results are shown in Table 1.
Related Figures (2)
