To overcome the premature problem of the traditional flower pollination algorithm, a novel niche flower pollination algorithm is proposed by combining the niche strategy with flower pollination algorithm. It is designed for the parameter inversion of the space fractional order diffusion equation, so as to provide some theoretical basis for the pollutants source identification and air pollution prevention. Firstly, twenty multimodal functions were selected to verify the performance of the flower pollination algorithm and its improved algorithm. Then, we carried out direct simulation with implicit finite difference scheme. Based on the forward simulation results, the flower pollination algorithm and the improved algorithm were applied to invert the source term and the diffusion coefficients of the space fractional differential equation. The sensitivity analysis of the proposed algorithms regarding initial interval, perspectives of population and transition probability has also been completed. Furthermore, the anti-noise properties of the proposed algorithms were discussed. The numerical results demonstrate that the improved flower pollination algorithm has achieved a higher precision and accuracy.