Scalable multi‐site photovoltaic power forecasting based on stream computing

Photovoltaic (PV) is essential for global carbon neutrality, it imperative to forecast PV generation accurately power operation. With the rapid growth of distributed sites, a scalable cloud service tends play vital role in forecasting address increasing cost computing resources and data subscriptions. Such scheme creates possibility further enhance performance by re-using model, data, all cloud. In order achieve this goal, work proposes multi-site system design with message queue (MQ) stream engine, where hybrid neural network model trained continuously updated using real-time data. A benchmark up 60 sites served simultaneously was performed verify scalability based approach. Moreover, after incremental updating decrease normalized root mean square error absolute were observed, demonstrating that better short-term accuracy achieved.