Abstract: In order to solve the problem of generator characteristic difference and meteorological data deviation caused by geographical location offset in distributed photovoltaic power short-term forecasting,further improving the prediction performance,a dynamic time warping（DTW）-multi-head attention neural network（MANN）-factorization machine（FM）prediction model combined with user profile is proposed. Firstly,we analyze the profile data and historical power generation data of generators to count the user profile. Then we use the optimization algorithm based on DTW standard,to correct offset of meteorological characteristics,forming a reasonable and perfect“user+meteorological”feature combination. Finally we utilize the weighted data samples to train the model. In the simulation stage,the real photovoltaic data of Jiangsu Province are used to compare the proposed model with several advanced photovoltaic prediction models in the industry. The results show that the proposed model has higher accuracy and robustness,indicating better predictive performance.