The finite availability and unsustainable nature of fossil fuel sources have driven increasing interest in renewable energy. However, significant efforts are still needed to fully integrate energy from renewable sources into existing power distribution networks. While reliability is essential for enhancing energy sustainability, the reliance of solar power plants on weather conditions presents challenges in maintaining a consistent output without incurring high storage costs. As a result, accurate forecasting of photovoltaic power generation is critical for effective grid management and energy trading. Machine learning models have emerged as a promising solution due to their ability to process large datasets and capture complex patterns. This study explores the use of metaheuristic optimization techniques to improve lightweight gated recurrent units (GRU) models for forecasting power generation in photovoltaic plants. Additionally, a modified metaheuristic optimization method based on the reptile search algorithm (RSA) is proposed to meet the demanding requirements of hyperparameter optimization. Extensive simulations were performed using real-world data from a power plant in India, accompanied by a thorough comparative analysis with other advanced metaheuristic algorithms. This research aims to address a gap in the literature, as lightweight GRU models have not been extensively studied for this particular challenge. The best-performing models achieved mean squared error (MSE) scores of 0.007407, demonstrating the significant potential of the proposed approach for real-world applications.