<p>Fractured aquifers are widespread in nature, yet most previous studies on groundwater contaminant source identification focus on non-fractured, porous aquifers. This study introduces a simulation-surrogate-optimization model tailored for fractured aquifers, using a hybrid discrete fracture network/equivalent porous media (DFN/EPM) model for flow field and solute migration simulation, a back propagation neural network (BPNN) surrogate model to improve inversion efficiency, and optimization based on sparrow search algorithm (SSA) to identify the location of pollution sources, mas release rates, and key hydrogeological parameters. To evaluate the robustness and effectiveness of SSA, two widely used optimization algorithms, particle swarm optimization (PSO) and genetic algorithm (GA), were used for comparative analysis. Two representative geological scenarios, fractured and non-fractured aquifers, were constructed to assess the framework’s adaptability. The results show that fracture-related parameters, such as fracture inclination and fracture density, exhibit sensitivity values close to 20% for pollutant concentration. In fractured aquifers, the pollutant plume spreads widely and stretches unevenly along the dominant fracture directions. while the plume in non-fractured aquifers disperses more evenly with a smoother and more symmetrical shape. Although fracture structures increase the nonlinearity and uncertainty of parameter identification, the proposed framework shows stable identification performance in both types of aquifers. In fractured aquifers, the average relative error of SSA is 9.49%, which remains within an acceptable range. SSA also exhibits the fastest convergence and highest inversion accuracy, making it suitable for preliminary source inversion tasks in complex environments.</p>