PRODUCTION CAPACITY PREDICTION BASED ON NEURAL NETWORK TECHNOLOGY IN AN EFFICIENT ECONOMIC AND MANAGEMENT ENVIRONMENT OF OIL FIELD

摘要

The change of single well production in oilfield is complex and versatile, and it is difficult to predict the yield with Arps curve. However, the neural network has the characteristics of strong learning ability, high adaptability, fast processing speed and high data fusion capability. Therefore, in order to improve the prediction accuracy of single well natural production, this paper chooses neural network to construct a single well natural production prediction model. The model uses BP forward neural network, and the transfer function is hidden layer tangent SIGMOID function and output layer linear function. This makes it possible to approximate any nonlinear function with finite discontinuities with arbitrary precision. The number of hidden layer neurons is taken as 8 and the first 3 order lag terms are selected as the input variables of the model. The training function selects a momentum gradient descent method that can adaptively adjust the learning rate. After many trainings, the model can meet the set requirements. The average error of the prediction results is 2.72%, indicating the reliability of the method in single well production capacity prediction.