Accurate short-term forecasting of bus passenger demand is vital for efficient smart transit operations. In this study, we propose a novel ensemble Deep Learning approach that integrates real-time and forecasted weather data with historical ridership patterns to enhance prediction accuracy. Our ensemble combines Recurrent Neural Networks (RNN), Long Short Term Memory (LSTM), and Bidirectional Long Short Term Memory (BiLSTM) architectures, each optimized using Bayesian hyperparameter tuning, to capture both short-term fluctuations and long-term dependencies. We incorporate a comprehensive set of meteorological variables, such as temperature, humidity, wind speed, and visibility, into the prediction pipeline. Experiments using large-scale datasets demonstrate that weather-aware models significantly improve 1 h forecasts, with the ensemble achieving R \(^{2} > 0.97\) across all time horizons. These results highlight the potential of weather-informed, ensemble models to support dynamic scheduling and resource allocation in smart bus systems.