基于BP神经网络的低渗透油田开发动态预测OACSTPCD

Reservoir development dynamic prediction is the fundamental basis for formulating and adjusting the development method. In order to overcome the accuracy of the traditional reservoir numerical simulation method that overly relies on the three-dimensional geological model and seepage mechanism, the BP neural network (ANN) method is adopted, based on the accurate reservoir description and production dynamic analysis in the study area. Twelve types of parameters including geological factors and development factors are selected as the basic dataset. Based on the analysis of parameter correlation and the selection of main control factors, the prediction model of annual cumulative fluid production and average water content of a single well in the study area was established. According to the results of the model, the liquid and oil production capacity of the study area gradually declined, and the water content showed a gradual upward trend, and the predicted cumulative liquid production of the block after two years was 13.7×104 m3, with a year-on-year decrease of 25.6％; the predicted annual cumulative oil production was 4.7×104 t, with a comprehensive decreasing rate of 31.9％; and the average water content rate of the block after two years was 58.1％, and the local wells in the area showed flooding, and the water content rate exceeded 98％. This study shows that the BP neural network has the advantages of simplicity, efficiency, and small error, which reduces the difficulty and workload of predicting the development dynamics of oilfields. It is suitable for the problem of oilfield development with a large sample dataset, and provides a new perspective for understanding the development dynamics of low-permeability oilfields.