Temperature Time Series Prediction Using Convolutional Bidirectional LSTM Optimized by FPA

As global climate change intensifies, accurate temperature predictions have become increasingly important across various fields, including agriculture, meteorology, and environmental protection. This paper proposes a temperature time series prediction model that integrates a convolutional neural network (CNN) with a bidirectional long short-term memory network (BiLSTM), utilizing the flower pollination algorithm (FPA) for hyperparameter optimization. The model effectively captures bidirectional dependencies within sequence data and enhances prediction accuracy through the global optimization capabilities of the FPA algorithm. In a case study focused on temperature prediction in Beijing, the model exhibited superior prediction accuracy and faster convergence compared to traditional deep learning models. The findings offer a novel perspective on regional temperature prediction and highlight the model's potential for practical applications.