Abstract: High precision wind-solar joint prediction is an important premise to give full play to the wind-solar complementary characteristics. Based on the coupling relationship between wind and PV, a distributed wind and PV joint prediction method is established in this paper. Firstly, considering the influence of geographical location, meteorological and other factors on the wind and solar power output, the coupling of wind power and photovoltaic output is analyzed. Secondly, the optimal intra-group variance algorithm is established to identify and process the abnormal data. Then, a combined prediction model based on improved attention mechanism and LSTM is proposed, which can effectively extract the coupling relationship between wind-solar and make accurate prediction. Finally, simulation results show that the proposed method can effectively improve the accuracy of landscape prediction.