Abstract: To address the problem of lacking power forecast accuracy due to the general lack of new energy resource monitoring devices for wide-area distributed new energy sources, this paper proposes a wide-area distributed photovoltaic power forecast method based on meteorological resource interpolation and transfer learning. First, based on geographic information and coarse particle meteorological data, the meteorological resource data under a wide area are gridded and interpolated. Secondly, according to the interpolation results, self-organizing mapping (SOM) network clustering is carried out for photovoltaic power plants with the same meteorological characteristics, and the source domain and target domain of transfer learning are divided for photovoltaic power plants in each category, which ensures forecast accuracy. Then, combined with the long short term memory network (LSTM), an error correction link is introduced to establish a dual transfer model from the source domain to the target domain. Finally, the effectiveness of the method is verified by taking the distributed photovoltaic power station in Shaoxing City, Zhejiang Province as an example. Compared with the separate modeling of each photovoltaic power station, this method can improve the training speed of photovoltaic power stations in the target domain by more than ten times, and it also has a significant improvement in forecast accuracy, which has certain promotion and application value.