An Lstm 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 solar station has a single prediction model irrespective of the time and the cloud condition, which often results in sub-optimal performance. In the proposed model, different categories of cloud movement are discovered using K-medoid clustering. To ensure a broader variation of the cloud movements, neighbouring stations are also used, selected using Dynamic Time Warping (DTW) based similarity score. Next, cluster-specific models are constructed. At the prediction time, firstly, the current weather condition is matched with the different weather groups found through clustering, and subsequently, a cluster-specific model is picked. As a result, different models will be used dynamically for a particular day and a solar station which improves the performance over a single site-specific model. The proposed model has achieved 19.74%, and 59% lower nRMSE, mean rank compared to the benchmarks, and it is validated for nine solar stations across two regions and three climatic zones of India