TCN-BiGRU: A Hybrid Deep Learning Architecture for Enhanced Temperature Time Series Forecasting
Tao Gu, Yajuan Zhang, Limin Wang · 2024
This study proposes a novel hybrid deep learning architecture that combines a temporal convolutional network (TCN) with a bidirectional gated recursive unit (BiGRU), aiming to improve the accuracy of temperature prediction. The model effectively addresses the challenges of complexity, non-stationarity, and high volatility in temperature time series prediction. TCN can capture long-term temporal dependencies through its scalable convolutional structure, thereby identifying complex patterns and periodic changes in data, BiGRU enhances adaptability to the dynamics and unpredictability of meteorological phenomena by fully considering historical and future influences. Through the synergistic effect of TCN and BiGRU, the proposed model exhibits stronger multi-scale feature extraction capability and complex temporal relationship modeling ability in temperature prediction and can better handle nonlinear dynamics and long time-dependence problems compared with the traditional methods. The experimental results show that the hybrid model can provide better prediction performance than the traditional methods in a wide range of geographic regions and climatic conditions, with strong noise immunity and adaptability to extreme weather events. This study provides an effective technical framework for deep learning applications in climate science, which is of great theoretical significance and application value in the fields of environmental monitoring, energy management and climate change response.