Stress-sensitive widely exists in fractured reservoir. In this paper, a mathematical model of flow in stress-sensitive reservoir with horizontal well is established based on experimental data and with process of linearization. By using of Lord Kelvin point-source solution, Bessel function integration and Poisson superimpose formula, the dimensionless pressure response function of horizontal well in infinite stress-sensitive reservoir is obtained. And then the derivative type curve is calculated. Based on the type curve, the characteristics and influencing factors of the fluid flow through porous medium of horizontal well in stress-sensitive gas reservoir are analyzed.

BP neural network has been successfully used in the gas well productivity prediction, but as a result of neural network is sensitive to the number of input parameters, we had to ignore some factors that is less important to the gas well productivity. In addition, the existing various productivity prediction method cannot consider the influence of some important qualitative factors. This article integrated the advantages of fuzzy comprehensive evaluation and BP neural network, fuzzy comprehensive evaluation method is used to construct the BP neural network's input matrix, and BP neural network learning function is used to solve the connection weights, so as to achieve the aim of predicting gas production. This method not only can consider as many factors influence on gas well production, ut also can consider qualitative factors, so the forecast results of the new model are more realistically close to the actual production situation of reservoirs.