Pattern recognition of gases of petroleum based on RBF model

摘要

After the RBF training we obtained a recognition percentage about 66,5% for non preprocessed data. Using the preprocessing of the fractional difference, the network performance percentage was 88,7%. Considering the results obtained, we confirm that preprocessing has made a big difference and that it has interfered in the network performance. We are testing other preprocessing models such as: relative, difference, absolute final output, minimum output, and modified difference. The objective is to check which of them is the most adequate for this kind of sensor and problem.