WARNs: a time series model integrating wavelet convolution and channel attention mechanism

Time series forecasting has wide applications in various fields. The primary goal is to extract patterns from historical data and predict future trends. Convolutional neural networks are extensively used in time series forecasting due to their ability to identify local patterns. However, they encounter challenges when it comes to capturing long-term patterns. To solve this problem, we propose a model named WARNs, which integrates wavelet convolution layers and channel attention mechanism. This model is capable of detecting long-term trends, short-term fluctuations and periodic characteristics of time series data while dynamically adjusting feature weights. Experiments on public datasets show that WARNs outperforms existing models in time series forecasting, achieving lower error metrics such as MSE and MAE.