Conditional Wasserstein Generative Adversarial Nets Based Cross-modality Photovoltaic Output Data Generation Approach

According to the “Analysis Report on Top 500 of Chinese Energy Groups in 2019”,the national energy and power system is a multi-source hybrid system architecture consisting of fossil energy and renewable energy.And,it is waiting for the intelligent innovation and industrial upgrade urgently.The intelligent innovation of power grid is an important part of national energy development strategy.Smart grid includes various components,such as autonomous situation awareness and regulation based on advanced sensing technology.Among them,situation awareness of smart grid is a fundamental and critical part of the intelligent innovation process.However,current research on the situational awareness of smart grid often limited by weak big data storage and analysis.Meanwhile,it doesn’t integrate advanced artificial intelligence technology into power grid effectively.Therefore,PV output data generation are not precise enough in fine time granularity.It is often unable to generate valid time-series scenario data effectively.In view of the above important issues of fine-granularity time-series PV output data generation,this paper proposes a cross-modality technology based conditional Wasserstein generative adversarial nets,named as CWGAN.Further functional improvements are made to CWGAN,and connect effectively different functional modules of CWGAN.This method can effectively mine the spatio-temporal correlation between observed nodes of power grid,and realize the cross-modality data generation.In the experimental section,our generation method is tested on the public PV dataset of NREL,which contains five data sets from different states of America.We compare CWGAN model with various models on the aforementioned dataset.It is found that the CWGAN model has comprehensive advantages in practical effect of data generation and running time.In addition,the stability and robustness of the CWGAN model are also tested.It is verified that CWGAN model can run stably in different data sets,which means its performances are less affected by changes of data sets.