This chapter examines the application of two advanced learning methodologies in addressing energy-related challenges: (i) a meta-learning methodology designed to dynamically integrate the core predictions of multiple deep learning models, and (ii) transfer learning approaches to enhance photovoltaic (PV) production forecasts. First, it is demonstrated that learning the conditions under which each model in a meta-learning scheme performs optimally enhances overall predictive accuracy. Specifically, four baseline Long Short-Term Memory (LSTM) models with distinct architectures are developed to predict short-term energy production in PV systems. The accuracy of the meta-learning model is evaluated using data from three PV systems located in Lisbon, Portugal. The results indicate that different baseline models excel with various PV systems, underscoring the potential of meta-learning to significantly improve accuracy, particularly during peak PV production periods. Conversely, the chapter illustrates that the use of transfer learning enhances PV forecasts. Specifically, the LSTM model is employed with three transfer learning strategies to improve accuracy. Transfer learning is applied for both initializing the LSTM model’s weights and for feature extraction, utilizing distinct approaches for each case. These strategies are compared against a conventional learning model that does not incorporate transfer learning, as well as the widely accepted smart persistence model. The methodology is assessed in the context of forecasting hourly production for six PV systems. Results reveal that transfer learning models significantly outperform conventional models, achieving a 12.6% reduction in mean squared error and a 16.3% improvement in prediction accuracy utilizing one year of training data.