Downhole Fracture Diagnosis and Mud Loss Control Technologies Bases on Neural Network Algorithm

Formation fractures of different widths are always met during drilling, resulting in partial mud losses or even lost return which seriously affect drilling in a safe and efficient manner. Presently, the success rate of controlling mud losses into fractures is not high, and the fractures plugged have low pressure bearing capacities. One cause to these problems is the lacking of understanding of the widths of the fractures into which mud is lost. In this paper, rock mechanics mechanisms of formation fracture generation are used as the bases to determine the 6 mechanical and engineering factors affecting the width of fractures produced downhole. Using the nonlinearity and big data analysis natures of neural network computation a model for analyzing the widths of downhole formation fractures was established, which included input layer, output layer and 3 hidden layers. Using this model, the calculation precision of diagnosing the widths of downhole fractures was improved to an average error of only 2.09% and the maximum error of 5.88%, resolving the problem of predicting fracture width through experiences which is inaccurate or through imaging logging which is expensive. The widths of fractures calculated with the model were also used to optimize the particle size distribution of plugging additives, helping improve the strength and stability of bridging inside the fractures, the pressure bearing capacity of the plugged fractures was increased to 12.8 MPa and the back-pressure bearing capacity to 4.5 MPa. In field application, the highest pressure squeezed on the plugged fractures was 10 MPa, and the plugged fractures were able to stand large flow rate circulation, indicating that the fractured formations were effectively and successfully plugged.