Synthesis of multi-year PV production data using generative adversarial networks

Multi-year forecasts of PV production are important for economic assessment of behind-the-meter PV+BES (photo-voltaic plus battery energy storage) systems. Historical solar resource data is available for many locations in the United States, but these data are limited and must be converted from solar resource to PV production data before they can be used in BES control simulations. We propose both rule-based and generative adversarial network methods for synthesizing multi-year PV production forecasts. These methods use reference PV production and latitude-longitude inputs to generate hundreds of PV production scenarios which enable detailed simulation of behind-the-meter demand charge management.