The extraction of seismic characteristic parameters includes data acquisition, data analysis, variable selection and data preprocessing. Only through these steps can we realize the effective learning and training of neural network.
Normalization
After analyzing the original data we collected, we have determined 20 input variables and 2 output variables on the basis of seismic characteristic parameters (including the actual thickness and the effective thickness of sandstone). Aside from the input variables and output variables, we need to normalize the data, which makes it easier to study and training the neural network. Before the normalization, we have to check whether there are outliers which must be removed.
Normalization transforms the data samples to the range of [-1, 1] or [0, 1] in general.
Assume that and represent the ith input variable and output variable respectively, represents the total number of samples, the number of input variables is , the number of output variables is . Transform the input variable to the range of [-1, 1], then:
(1)Considering the output of Sigmoid function is [0, 1], so transform the output variable to the range of [0, 1], then:
(2)