Shale oil usually adopts the development method of “horizontal well + volume fracturing”. The production capacity of oil wells is affected by various factors such as geology, reservoir, and engineering, so traditional methods cannot meet the demand for production capacity prediction in actual production. To address this issue, this study selects a domestic shale oil block as the research object and constructs a Stacking integration model based on particle swarm optimization to predict the production capacity. The first layer of the model chooses Support Vector Regression (SVR), Random Forest (RF), and Extreme Gradient Boosting (XGBoost) as base learners, and uses the Particle Swarm Optimization algorithm to adjust the parameters of the base learners to improve the performance of the model on the training set and its generalization ability. The second layer of the model uses linear regression to integrate the prediction results of each base learner and output the predicted production capacity. The experimental results show that compared with the individual base learners, the Stacking model optimized by particle swarm optimization shows higher prediction accuracy and stability, and the shale oil production capacity prediction using the machine learning-based method is more scientific and accurate than the traditional method, which provides a new idea that can be used for the development of similar reservoirs.