Year: 2025
Type: article
Abstract: The increasing annual temperatures have led to various extreme weather events such as droughts and heatwaves, posing significant threats to human production and daily life. Therefore, accurate temperature prediction is crucial for effective disaster prevention. With the continuous development of computer parallel technology and high-performance computing devices, artificial intelligence has increasingly become an effective solution for addressing such challenges. Leveraging these technological advancements and historical temperature data, this study proposes a multivariate time series prediction approach based on Recurrent Neural Networks (RNN) to address the limitations of Convolutional Neural Network-Long Short-Term Memory models in handling long-term temporal dependencies for temperature forecasting. This paper first develop an RNN-LSTM model for univariate time series prediction, then extend it to multivariate scenarios. The inherent capability of RNN to retain state information from previous time steps enables effective capture of long-range dependencies. Experimental results demonstrate that our model achieves an RMSE of 0.15, MAE of 0.10, and coefficient of determination (R²) of 0.92 on standardized data, with a test set prediction accuracy of 92%. Compared with baseline models, our approach reduces parameter quantity by 34% while decreasing prediction error by 21%, validating its effectiveness in processing multivariate temporal features and long-term dependencies. This research provides an interpretable deep learning framework for meteorological prediction, where the dual-layer LSTM architecture combined with Dropout regularization strategy offers universal reference value for time series forecasting tasks.
Author: <p>Zerui Yu</p>
Institution: Guilin University of Electronic Technology
Language: English
Subfield: Environmental Engineering
Field: Environmental Science
Domain: Physical Sciences
SDG: Climate action
Open Access status: diamond
OpenAlex ID: W4413065037