A NEW METHOD FOR GAS WELL LIQUID ACCUMULATION PREDICTING BASED ON CONVOLUTIONAL NEURAL NETWORK ENVIRONMENT

Accurate prediction of liquid accumulation in gas wells is essential for efficient and stable gas field development. At present, the commonly used methods such as droplet model, liquid film model, stability analysis method and neural network method cannot well discriminate wellbore liquid accumulation. There is a large deviation between the predictions obtained by different liquid accumulation prediction models for different types of gas reservoirs. Because the liquid accumulation process is a continuous and dynamic process, the limitations of the commonly used models are mainly reflected in that the discrimination results can only represent the transient state of the liquid accumulation. In order to better solve the problem of liquid accumulation in gas wells, in this paper, we propose a new method for liquid accumulation prediction in gas wells based on convolutional neural network (CNN). This method reversely predicts the situation of liquid accumulation downhole through preprocessing, convolution, pooling, activation function (RELU, SoftMax), regularization and other processes. Finally, we verified it with examples. The results show that the convolutional neural network can predict liquid accumulation more accurately. Moreover, this model cart detect liquid accumulation at the bottom of the well earlier than other models.