Joint forecasting of regional wind and solar power based on attention neural network

Abstract

In the future high proportion of renewable energy grid-connected scenario, a large number of centrally developed wind farms and photovoltaic power stations will be integrated into the power system. There is a certain spatial and temporal correlation between wind and solar resources, which can provide additional information to help improve the power prediction accuracy. However, most of the existing wind and solar power prediction methods are focus on a single energy form, a single station or a single unit, which cannot fully and effectively reflect the space-time correlation between wind and solar resources. In addition, a single-field power prediction is no longer well enough to meet the needs of grid dispatching agencies. On the one hand, the power system as a whole, the dispatchers are more concerned about the total amount of uncertainty in the wind and solar power prediction; on the other hand, in order to make reasonable dispatch to the grid and avoid some off-grid events, the cluster power prediction is needed. Therefore, this paper proposes a joint forecasting method of regional wind and solar power. With NWP wind speed, irradiance and temperature data of several wind farms and photovoltaic power stations as input and measured power data as prediction target, based on the attention neural network algorithm, constructs a joint forecasting model which can reflect the spatial-temporal correlation of regional wind and solar resources. The proposed method is verified with the data of 8 wind farms and 7 photovoltaic power stations. The analysis results show that the proposed model can not only improve the power prediction accuracy, but also obtain the power prediction results of each target stations at the same time, which reduces the workload and has certain engineering application value.