Shale reservoirs influenced by multiscale factors such as mineral composition and depositional processes, are characterized by the development of micro-to nanopores and pronounced bedding structures. These features lead to extreme fluid occurrence states and significantly increased structural complexity of fracture networks after hydraulic fracturing. In the case of degassed crude oil, particularly oils with high gas–oil ratios (GOR), complex three-phase flow behavior is commonly observed, posing significant challenges to production decline forecasting and Estimated Ultimate Recovery (EUR) evaluation. To address the limitations of conventional analytical models in accurately describing multiphase flow, as well as the high computational cost and inefficiency of numerical simulation methods for decline analysis, this study develops a three-phase decline prediction model for shale reservoirs by coupling analytical approaches with parameter optimization algorithms. First, based on the complex fracture network formed after hydraulic fracturing, a coupled physical model of “reservoir supply–fracture conductivity” is established, and the dominant flow regimes—including bilinear flow, linear flow, and boundary-dominated flow—are systematically identified. Subsequently, analytical production prediction models are derived for the oil and water phases based on rate transient analysis (RTA) diagnostic plots, while gas production is predicted using the produced gas–oil ratio. By incorporating the pressure–saturation (P–S) relationship within nanopore systems, a fully coupled solution for three-phase production is achieved. On this basis, three parameter optimization algorithms—Particle Swarm Optimization (PSO), Genetic Algorithm (GA), and Simulated Annealing (SA)—are employed for history matching. The algorithm with the fastest convergence and minimum fitting error is automatically selected, enabling efficient determination of key parameters through inverse modeling. These include fluid-related parameters, flow-related parameters, and geological parameters. Finally, future production performance is forecasted based on the optimized model, and EUR evaluation is conducted. Field application results demonstrate that the proposed model achieves a history matching accuracy exceeding 85% for oil, gas, and water production, and can reliably predict long-term production decline trends. The proposed methodology provides an efficient and robust tool for production evaluation and development optimization of hydraulically fractured wells in shale reservoirs.