<p>Accurate prediction of renewable energy stock prices is crucial for investors and policymakers. To achieve this, a Deep Learning (DL) model called the Long Short-Term Memory (LSTM) network has been employed to predict renewable energy stock prices based on Investors’ Sentiment (IS) from environmental television newscasts. However, it may not capture the full spectrum of IS due to a lack of social media data, and inappropriate hyperparameter configurations can lead to significant errors in predictions. Hence, this manuscript introduces a new optimized DL model for predicting renewable energy stock prices based on the ISs from both environmental television newscast data and social media texts. The main aim of this model is to integrate data from multiple sources and reduce the prediction error. First, a dataset is created by collecting environmental television newscast data, social media texts related to renewable energy stocks, and stock prices of top-k companies. Thereafter, Natural Language Processing (NLP) is applied to extract IS scores. These scores, along with the stock price data, are passed to the Self-Adapted and Attention-based Bidirectional Gated Recurrent Unit (SA²BiGRU) network, with hyperparameters optimized by an Improved Secretary Bird Optimization (ISBO) algorithm to predict renewable energy stocks. Finally, the experimental results show that the SA²BiGRU model outperforms existing models, reducing prediction errors and enhancing the reliability of forecasts for renewable energy stock prices using additional data sources.</p>