For shale oil reservoirs in the Jimsar Sag of Junggar Basin, the fracturing treatments are challenged by poor prediction accuracy and difficulty in parameter optimization. This paper presents a fracturing parameter intelligent optimization technique for shale oil reservoirs and verifies it by field application. A self-governing database capable of automatic capture, storage, calls and analysis is established. With this database, 22 geological and engineering variables are selected for correlation analysis. A separated fracturing effect prediction model is proposed, with the fracturing learning curve decomposed into two parts: (1) overall trend, which is predicted by the algorithm combining the convolutional neural network with the characteristics of local connection and parameter sharing and the gated recurrent unit that can solve the gradient disappearance; and (2) local fluctuation, which is predicted by integrating the adaptive boosting algorithm to dynamically adjust the random forest weight. A policy gradient-genetic-particle swarm algorithm is designed, which can adaptively adjust the inertia weights and learning factors in the iterative process, significantly improving the optimization ability of the optimization strategy. The fracturing effect prediction and optimization strategy are combined to realize the intelligent optimization of fracturing parameters. The field application verifies that the proposed technique significantly improves the fracturing effects of oil wells, and it has good practicability.