Wind Power Generation Forecast Based on Multi-Step Informer Network

Abstract

Accurate forecast results of medium and long-term wind power quantity can provide an important basis for power distribution plans, energy storage allocation plans and medium and long-term power generation plans after wind power integration. However, there are still some problems such as low forecast accuracy and a low degree of integration for wind power physical processes. In this study, the Multi-step Informer network is proposed to add meteorological parameters to wind power generation forecast and make network interpretable. The Multi-step Informer network uses Informer to obtain the initial training model according to the historical data of wind power generation, introduces the Informer model of wind speed and air pressure training involved in the dynamic pressure model, and compares the historical data of wind power generation to obtain model modification, so as to further improve the forecast accuracy of Multi-step Informer network. The backpropagation process of the pre-trained Informer should be truncated to avoid being influenced by the pre-trained Informer during training of the Multi-step Informer network, which also guarantees the interpretability of the running results of the network. The Multi-step Informer network has the advantage of error correction of wind power generation, which improves the forecast accuracy. From the calculation results of the root mean square error, Multi-step Informer network improves forecast accuracy by 29% compared to Informer network.