Distributed PV installations in the PV installations in the proportion of increasing, to carry out distributed PV power prediction is very important practical significance. Aiming at the problem that it is difficult to obtain accurate meteorological characteristic data for new distributed PV power prediction today, a distributed PV ultra-short-term power prediction method is proposed. First, an inverted transformer (iTransformer) model is introduced into the distributed PV power prediction problem, and the transpose operation of the iTransformer model on the input enables the model to capture the correlation between a small amount of meteorological feature data to fully exploit the data potential. In addition, Particle Swarm Optimization (PSO) algorithm is used to optimize the key hyperparameters in the model instead of manual parameter tuning. The simulation results show that the PSO-iTransformer model exhibits higher prediction accuracy than the traditional Long Short Term Memory (LSTM) neural network model and Decomposition-Linear (DLinear) model.