Solar Irradiance Nowcasting for Virtual Power Plants Using Multimodal Long Short-Term Memory Networks

The rapid penetration of photovoltaic generation reduces power grid inertia and increases the need for intelligent energy resources that can cope in real time with imbalance between consumption. Virtual plants are a technology coordinating such monetizing them, example on electricity markets real-time pricing or frequency reserves markets. Accurate short-term forecasts essential virtual plants. Although significant research has been done medium- long-term forecasting, forecasting problem requires special attention to sudden fluctuations due high variability cloud cover related weather events. Solar irradiance nowcasting aims resolve this by providing reliable expected capacity. Sky images captured proximity panels used determine behavior solar intensity. This is computationally challenging task conventional computer vision techniques only handful Artificial Intelligence (AI) methods have proposed. In paper, novel multimodal approach proposed based two Long Short-Term Memory Networks (LSTM) receives temporal image modality stream sky images, numerical time-series past readings as inputs nowcasting. pipeline consists preprocessing module an augmentation implements detection, Sun localization mask generation. complete was empirically evaluated real-world case study across four seasons northern hemisphere, resulting mean improvement 39% multimodality.