A novel short-term wind power scenario generation method combining multiple algorithms for data-missing wind farm Considering spatial-temporal correlativity

For newly-built or expanded wind farms with missing, insufficient or invalid wind power data, the existing methods often have limitations in describing their wind power characteristics and generating wind power scenarios. To this end, a novel effective short-term wind power scenario generation method is put forward in this paper, where similar data domain matching, transfer learning, conditional deep convolutions generative adversarial network (C-DCGAN) and parameter optimization are improved and combined in a unified framework with full consideration of the spatial–temporal correlativity among multiple adjacent wind farms. Specifically, a similar data domain matching process is firstly presented to quickly filter and purify the sufficient wind power data of adjacent wind farms, so as to extract their useful similar wind power characteristics. On this basis, an accurate wind power scenario generation model of data-missing wind farm can be constructed through transfer learning and C-DCGAN training. Then a constrained optimization model is proposed to control the noise parameter in order to obtain the short-term wind power scenarios for a specific day. After expounding the general principle and mathematical formulations of the proposed method, simulation studies and comparative analysis are conducted based on the WIND public dataset to verify the accuracy, effectiveness and superiority of the proposed method.