DeepComp: Deep reinforcement learning based renewable energy error compensable forecasting

A taxonomy of short‐term solar power forecasting: Classifications focused on climatic conditions and input data

“…In [70], an automated reinforcement learning algorithm based on prioritized experience replay was proposed to support a multi‐period single‐step forecasting model for improving RES prediction accuracy. A deep reinforcement learning approach was introduced in [71], in order to make the predictive errors of the forecast more compensable. The contributions of the works that propose DRL‐based SPF are presented in Table 2.…”

“…Ultra‐short‐term forecasting or nowcasting uses solar data to estimate photovoltaic power predictive outputs with a forecasting horizon typically ranging from a few seconds to 30 min [92]. Reviewed works [28, 33–40, 42, 46, 49, 71, 73, 76, 83, 84] propose different models and methodologies for ultra‐short‐term SPF.…”

“…The forecasting horizon typically ranges from 1 h to 1 day [93]. Reviewed works [30, 32–36, 38, 39, 41–45, 47, 48, 59–63, 71, 75, 81, 82, 84] focus on short‐term SPF by using different models and methodologies.…”

“…The reviewed works [18–84] focus on providing accurate short‐term SPF models. Those models take into account specific factors in order to achieve accurate forecasts.…”

“…To further provide information over the climatic conditions' effect on the SPF results, Figures 4 and 5 provide further information over the European continent. Temperate dry [27,83] Temperate humid [33,34,43,46,63,71,91] FIGURE 5 European photovoltaic power production map (2021) [96].…”