Short-term forecasting of photovoltaic (PV) output is of great significance for energy management and optimal scheduling of power grids or microgrids. A spiking neural network (SNN) based on an improved grey wolf optimization (IGWO) algorithm is constructed and applied to short-term forecasting of PV output. First, the grey relation analysis method is used to select similar days. Then, an IGWO algorithm is proposed for SNN model training, and the performance of SNN coding and forecasting is improved by introducing a nonlinear descent convergence factor and updating strategy of dynamic weight based on the regular changes of trigonometric functions. Finally, the performance of the proposed method is evaluated using a demonstration system, and compared with three other models. The results show that the proposed method significantly improves prediction performance. This work is supported by Natural Science Foundation of Zhejiang Province (No. LY20E070004).