As the energy crisis environmental concerns rise, harnessing renewable energy sources like photovoltaics (PV) is critical for sustainable development. However, the seasonal variability and random intermittency of solar power pose significant forecasting challenges, threatening grid stability. Therefore, this paper proposes a novel renewable energy hybrid forecasting method, NCPO-ELM, to adequately capture spatial and temporal dependencies within meteorological data crucial for accurate predictions. To effectively optimize the performance of the Extreme Learning Machine (ELM), the Normal Cloud Parrot Optimization (NCPO) algorithm is developed, inspired by Pyrrhura Molinae parrots' flock behavior and cloud model theory. NCPO integrates five unique search strategies and utilizes a random structure to explore and exploit. By introducing the normal cloud model to generate random samples with specific distributions, the algorithm enhances solution space coverage. NCPO is subsequently employed to optimize the Single-Layer Feedforward Network (SLFN) hidden layer hyperparameters, yielding optimal weights and biases for the output layer, thereby reducing benchmark ELM's sensitivity to noise and instability from random initialization. The actual forecasting results of PV stations across different regions demonstrate that the proposed NCPO-ELM shows superior prediction accuracy and performance compared to existing hybrid forecasting approaches, particularly for time series with diverse data characteristics and seasonal variations.