CGAN-filter for model-free photovoltaic scenario generation with conditional maximum mean discrepancy indicator. (28th March 2021)

• Record Type: Journal Article
• Authors: Li, Jie; Gao, Min; Xiang, Yi; Fan, Xiaochao
• Is Part Of: International journal of system control and information processing. Volume 3:Number 2(2020)
• Page Start: 129
• Page End: 149
• Publication Date: 2021-03-28
• Subjects: CGAN -- conditional generative adversarial network -- PV -- photovoltaic -- C-MMD -- conditional maximum mean discrepancy -- scenario generation; System design -- Data processing -- Periodicals; Information technology -- Periodicals; 003.5
• Languages: English
• ISSNs: 1759-9334

Abstract

It is difficult to produce credible photovoltaic (PV) scenarios depending on required accuracy, climate, cloud, and other environmental conditions. To resolve it, this paper presents a model-free approach to generate efficient PV scenarios through learning the inherent distribution of historical data based on conditional generative adversarial network (CGAN). Unlike probabilistic models with specific system models, the proposed CGAN-Filter obtains solar energy production patterns by discovering distribution characters and uncertainties in temporal dimensions. In order to offset severe disturbances in PV scenarios caused by random clouds and climates, a filter layer is designed between the two neural networks in CGAN to reduce amplified environmental randomness. Additionally, to observe learning processes of the proposed approach, a conditional maximum mean discrepancy (C-MMD) is theoretically and practically demonstrated to be able to monitor the performance of PV scenario generation. Experimental results on real-world PV scenario datasets have validated the effectiveness of our proposed approach.