Multi-Scale Dilated Convolution Network for Long-Term Time Series Forecasting

Feifei Li,Suhan Guo,Feng Han,Jian Zhao,Furao Shen

2024-05-14

Abstract:Accurate forecasting of long-term time series has important applications for decision making and planning. However, it remains challenging to capture the long-term dependencies in time series data. To better extract long-term dependencies, We propose Multi Scale Dilated Convolution Network (MSDCN), a method that utilizes a shallow dilated convolution architecture to capture the period and trend characteristics of long time series. We design different convolution blocks with exponentially growing dilations and varying kernel sizes to sample time series data at different scales. Furthermore, we utilize traditional autoregressive model to capture the linear relationships within the data. To validate the effectiveness of the proposed approach, we conduct experiments on eight challenging long-term time series forecasting benchmark datasets. The experimental results show that our approach outperforms the prior state-of-the-art approaches and shows significant inference speed improvements compared to several strong baseline methods.

Machine Learning,Artificial Intelligence

What problem does this paper attempt to address?

The paper mainly discusses the problem of long-term time series prediction. Accurate prediction of long-term time series data is important in many decision-making and planning applications, such as power consumption prediction, economic forecasting, and weather forecasting. However, capturing long-term dependencies in time series is a challenge.

To address this problem, the paper proposes a method called Multi-Scale Dilated Convolution Network (MSDCN). This method uses a shallow architecture of dilated convolutions to capture the periodic and trend characteristics of time series. By using convolutional kernels of different sizes and exponentially increasing dilation rates, MSDCN can sample time series data at different scales, effectively extracting multi-scale information. In addition, the model combines traditional autoregressive modules to directly model the linear relationship of the data, simplifying the complex modeling process.

Experiments on eight challenging benchmark datasets for long-term time series prediction validate the effectiveness of MSDCN. The results show that this method outperforms previous state-of-the-art approaches and demonstrates faster speed in both training and inference phases. Compared to Transformer, MLP, and CNN-based methods, MSDCN has advantages in capturing short-term local dependencies, long-term global dependencies, and handling time and space complexity.

In summary, the paper aims to address the problem of accurately and efficiently predicting long-term time series data, especially in dealing with periodicity, trend changes, and multi-scale information. By proposing MSDCN, the researchers provide a new solution that can better capture key features in time series, improving the accuracy and speed of prediction.