This paper presents a novel hybrid model that integrates predictive and optimization techniques to enhance the scheduling and management of electricity generation in large-scale power systems, with a focus on the variability of photovoltaic (PV) energy. By combining a long short-term memory (LSTM) neural network with an optimization framework, the model forecasts PV power generation over a one-month horizon using historical data, validated against actual production. The optimization component, built on a refined large-scale power system model, incorporates these predictions using a block representation approach to simulate diverse generation technologies, including natural gas, fossil fuel-based thermal units, hydroelectric, PV, nuclear, and wind power plants. This integrated approach addresses the stochastic nature of renewable sources, distinguishing it from prior studies that focus solely on prediction or optimization. The Argentine Interconnection System (SADI) serves as the case study, leveraging over a decade of time-series data to evaluate the model’s performance. Results demonstrate reliable prediction and scheduling capabilities, achieving a low prediction error of approximately 0.01% for key PV sources. Implemented in Python within the Spyder environment, with TensorFlow and Keras for LSTM predictions and PYOMO for optimization, the model offers a practical and effective solution for system operators to optimize resource allocation in renewable-heavy power systems.