Abstract: The fracturing construction curve contains information of artificial fractures and reservoir, which is the basis for evaluating fracturing results. At present, the evaluation of fracturing results mainly relies on the theoretical and statistical methods, which have limited guidance for the improvement and optimization of fracturing technology. In order to fully tap the hidden information in the construction curve, a sample library is constructed for the image of fracturing construction curve based on the classification of open flow rate after fracturing. The convolution neural network CNN in artificial intelligence is used for training, and an evaluation model is established based on the capacity classification, Then, the interpretability study is conducted by using Grad-CAM to find out the main reference position for artificial intelligence identification, so as to guide the optimization and improvement of the fracturing technology. The research results show that the accuracy of fracturing curve classification by CNN is higher than 85%. The key to the fracturing results lies in the early and late stages of fracturing construction, mainly including the initial fracturing displacement and corresponding pressure rise rate, pump stop pressure, and slug duration, and production capacity can be improved by changing fracturing construction parameters. This method enables to optimize and improve fracturing construction with targeted measures.