Phase matching quantum key distribution is a promising scheme for remote quantum key distribution, breaking through the traditional linear key-rate bound. In practical applications, finite data size can cause significant system performance deterioration when data size is below 1010. In this paper, an improved statistical fluctuation analysis method is applied for the first time to two decoy-states phase matching quantum key distribution, offering new insights and potential solutions for improving the key generation rate and the maximum transmission distance while maintaining security. Moreover, we also compares the impacts of the proposed improved statistical fluctuation analysis method to those of the Gaussian approximation and Chernoff-Hoeffding boundary methods on system performance. The simulation results show that the proposed scheme significantly improves key generation rate and maximum transmission distance compared to the Chernoff-Hoeffding approach, and approaches the results obtained when the Gaussian approximation is employed. At the same time, the proposed scheme retains the same security level as the Chernoff-Hoeffding method, and is even more secure than the Gaussian approximation.