Given multiple weather forecasts, how can we achieve higher photovoltaic (PV) prediction accuracy while maintaining prediction stability? As the significance of renewable energy sources increases, accurate prediction of PV generation has become increasingly important. However, the inherent volatility of the weather sets a challenge into achieving accurate PV generation predictions. To address this challenge, this study proposes a Transformer-based Photovoltaic power prediction model Utilizing multiple weather forecasts (TPU), which is a novel framework that utilizes multiple weather forecasts to enhance prediction accuracy while maintaining performance stability. TPU employs attentive Long Short-Term Memory (LSTM)to create embeddings of weather forecasts with different time horizons and cross-attention to fuse the features of different forecasts. Experimental results conducted using a six-month dataset comprising generation data from nine PV power plants spread across the whole country demonstrate that our model outperforms all baseline models in terms of mean absolute error (MAE) and root mean squared error (RMSE), while concurrently achieving a reduced mean absolute deviation (MAD).