Deep-Learning-Based Adaptive Model for Solar Forecasting Using Clustering

Accurate short-term solar forecasting is challenging due to weather uncertainties associated with cloud movements. Typically, a station comprises single prediction model irrespective of time and condition, which often results in suboptimal performance. In the proposed model, different categories movement are discovered using K-medoid clustering. To ensure broader variation movements, neighboring stations were also used that selected dynamic warping (DTW)-based similarity score. Next, cluster-specific models constructed. At time, current condition first matched groups found through clustering, subsequently chosen. As result, multiple dynamically for particular day station, improves performance over site-specific model. The achieved 19.74% 59% less normalized root mean square error (NRMSE) rank compared benchmarks, respectively, was validated nine across two regions three climatic zones India.