A Deep Learning Model for Forecasting Photovoltaic Energy with Uncertainties

Accurate forecasting of photovoltaic (PV) power generation is crucial for integrating this renewable energy into existing energy systems. Predicting PV output is a challenging task because of the uncertain and stochastic behavior of solar energy systems (e.g., unstable climatic conditions, errors in irradiance measurements). Conventional PV production forecasting approaches, such as non-deep neural networks, solely rely on high-precision and ideal data, and they may fail in predicting uncertain PV profiles. In this study, an uncertainty-aware deep neural network is proposed for predicting stochastic PV generation profiles. We combined convolutional neural networks (CNN) with long short-term memory (LSTM) to improve the robustness of our prediction model against uncertainties in the data. In the first step, CNN is utilized to capture local features in PV datasets. In the second stage, CNN outputs are injected into the LSTM neural network. LSTM is responsible for learning the temporal relationships in the data. To evaluate the performance of our proposed approach, it is applied to a real-world dataset. The effectiveness of our proposed hybrid model is compared to that of a baseline approach (a single LSTM neural network). In contrast to the existing PV forecasting approaches presented in the literature, our proposed framework provides highly accurate predictions of PV generation with flexible prediction horizons, and it is robust against noises in measurement devices and missing profiles.