Two-stage scheduling of integrated energy systems based on a two-step DCGAN-based scenario prediction approach

Integrated energy systems (IESs) are developing rapidly as a supporting technology for achieving carbon reduction targets. Accurate IES predictions can facilitate better scheduling strategies. Recently, newly developed unsupervised machine learning tool, known Generative Adversarial Networks (GAN), has been used to predict renewable outputs and various types of loads its advantage in that no prior assumptions about data distribution required. However, the structure traditional GAN leads problem uncontrollable generations, which be improved deep convolutional (DCGAN). We propose two-step prediction approach takes DCGAN achieve higher accuracy generation results uses K-means clustering algorithm scenario reduction. In terms strategies, common two-stage is generally day-ahead intraday stages, with rolling stage. To account impacts on results, Conditional Value at Risk (CVaR) added The intra-day process also ensure inputs each domain updated real-time. simulations typical show proposed describe load-side demands significantly reduced computational complexity strategy improve economy results.