Accurate photovoltaic power production forecasting is critical for advancing the integration of renewable energy into modern power grids. This work introduces a hybrid metaheuristic approach that incorporates novel modifications to particle swarm optimization (PSO)for hyperparameter tuning of Long Short-Term Memory (LSTM) models. Using a real-world dataset, the proposed methodology achieved a mean squared error (MSE) of .000631 in one-step-ahead forecasts, outperforming other algorithms in comparative analysis. The modifications enable the modified metaheursitic to avoid local optima and achieve global solutions more effectively. The proposed approach addresses challenges inherent to the unpredictable nature of renewable energy production and demonstrates significant potential for improving forecasting accuracy. Practical implementation of this methodology can foster a transition from fossil fuels to renewable energy, bridging the gap between theoretical advancements in computational intelligence and their real-world applications, with broad social and policy implications.