PREDICTION OF OIL-GAS RESOURCES IN SONGNEN PLAIN BASED ON SOIL GEOCHEMICAL DATA AND BACK-PROPAGATION NEURAL NETWORK

Based on the massive data obtained from the multi-target regional geochemical survey in Northeast China, the back-propagation(BP) neural network is used to establish the model between soil geochemical property and spatial location of oil-gas fields, and construct the optimal prediction model of oil-gas resources. Taking both the 54 soil geochemical indexes and XY coordinate values as input layer of the model and whether the samples are inside the oil-gas fields (1 for inside, 0 for outside) as output layer, the study carries out the model training based on the data of each 500 soil samples randomly selected from inside and outside the oil-gas fields. The results show that the recognition accuracy remains at about 90% after repeated training, indicating that the model has good classification effect and can be used for prediction of oil-gas resources. The hydrocarbon-bearing probability of 11 291 soil samples from Songnen Plain is obtained by using the model, and then the prediction map of oil-gas resources is drawn. The study shows that neural network can play an important role in solving complex nonlinear geological problems.