Distributed fiber optic temperature sensing provides significant advantages for production monitoring in complex geological environments due to its high precision, real-time capability, and long-term stability. However, its expanding application generates increasingly complex temperature datasets that challenge conventional production profile interpretation methods. To address these challenges, in this study, the researchers developed an intelligent interpretation framework to combine physically constrained forward modeling with data-driven machine learning techniques. A forward model of wellbore temperature profiles was established based on the fundamental principles of momentum conservation, energy conservation, and two-phase flow dynamics. Sensitivity analysis was used to identify key controlling factors, including production rate, geothermal gradient, reservoir thickness, crude oil heat capacity, and crude oil density, which were then used to generate representative training datasets. Three neural network architectures, including a fully connected neural network, a radial basis function network, and a back propagation network, were systematically trained and compared. The fully connected neural network demonstrated superior prediction accuracy and generalization capability, offering a robust tool for production profiling. Field validation using actual distributed fiber optic temperature-sensing monitoring data from commingled production wells confirmed the method’s practical effectiveness, with predicted production rates strongly agreeing with the measured values across multiple reservoir layers. The proposed framework provides a reliable, efficient solution for interpreting the production profiles of multilayer wells under single-phase flow conditions. This study establishes a foundational methodology that can be extended to more complex multiphase flow scenarios in future research, thereby contributing to intelligent and automated reservoir management.