Abstract: The high penetration of renewable energy has diversified power system participants, making the fair allocation of system surplus increasingly critical. Focusing on the fair allocation of total social surplus in electricity markets, this paper constructs a cooperative game characteristic function reflecting multi-period system surplus, and proposes a Shapley value-based fair settlement method for electricity markets as a benchmark for evaluating surplus allocation fairness. To overcome the exponential computational complexity that arises as the number of participants grows, the sampling-weighted regression mechanism of KernelSHAP is introduced for efficient approximation of Shapley values. Simulation results on an IEEE 30-bus system and a real-world 182-bus system in eastern China demonstrate that KernelSHAP can accurately estimate Shapley values at practical system scales. The proposed method provides a feasible path for fair allocation of system surplus in electricity markets and, using Shapley value allocation as a fairness benchmark, quantitatively reveals the extent to which the prevailing locational marginal price (LMP) settlement mechanism deviates from fair allocation.