ニューラルネットワークによる岩相解析

抄録

The present paper proposes the algorithm for well log data analysis by use of neural network.
The designated network is the layered network which consists of an input layer, one hidden layer and an output layer. The unit numbers of input layer and output layer are equal to well log curve to be inputted and the lithology to be outputted, respectively. The well log analysis procedure by use of neural network is divided into two stages, that is, training period and recognition period. During the training period, the initial weights which combine the units in each layer with those in the previous layer, are modified using the error back propagation algorithm so as to minimize the difference between the teacher's signal and the actual signal from output layer. The training period continues until the designated training counts are reached, or the residual between the teacher's signal and the actual signal from output layer becomes smaller than the designated value. After the completion of training period, the network is ready to analyze the new input data using the weight values.
Aa a result of studies using the synthetic and the actual data, it was clarified that the proposed algorithm analyzed the well log data with good accuracy.