Efficiently mining the rich information from time-series data collected by intelligent sensing devices and reliably analyzing and predicting these time-series data hold significant practical importance for IoT applications. This paper proposes a Discrete Wavelet Transform Temporal Convolutional Network (DWTTCN) to deeply extract frequency domain information from the data. Firstly, Discrete Wavelet Transform is used for multi-resolution decomposition in frequency domain. Then, Frequency Convolutional Block (FCB) is used to extract the frequency domain features of each level after decomposition. Finally, the frequency domain feature and time domain feature are fused for prediction. By experiments on seven benchmark datasets, the results show DWTTCN outperforms state-of-the-art methods on five datasets and performs comparably on others. As a unique convolutional structure, DWTTCN not only delivers superior performance but also operates with greater efficiency. On the largest benchmark dataset traffic dataset, the number of parameters decrease 95% and the MACs decrease 90% than the most advanced time convolution networks. This study fully leverages frequency domain features to predict time-series data, achieving highly reliable prediction performance while minimizing the number of parameters and computational resources required.