<p>Most multivariate time series long-term forecasting models primarily extract features of the time series mainly in the time domain for prediction. However, they do not utilize the frequency domain characteristics of the time series, even though frequency domain features are more sensitive to seasonal variations in the time series. Furthermore, time series often exhibit a mixture of long and short-term patterns in the time domain. The long-term patterns reflect the overall trends and patterns of the time series over an extended period, while the short-term patterns reflect the rapid, cyclic fluctuations within a shorter period. If we can effectively extract both the long and short-term patterns of the time series, it will significantly enhance the accuracy of time series forecasting. This paper proposes a multi-dimensional time series long-term forecasting model to improve the accuracy of multi-dimensional time series long-term forecasting. The model incorporates a time series decomposition mechanism, where the time series is decomposed into seasonal and trend components in the time domain for separate forecasting. For the seasonal component, Global-TCN and Local-TCN networks are designed extract the long-term and short-term patterns of the time series The frequency domain features of the seasonal component are then extracted by transforming the seasonal component of the time series to the frequency domain via Fast Fourier Transform (FFT). Finally, the extracted long-term and short-term patterns of the seasonal component, along with the frequency domain features obtained through fast Fourier transform, are separately mapped through the output module of the forecasting model to obtain the time-domain and frequency-domain predictions of the seasonal component. For the trend component, the forecasting output module of the forecasting model is directly used for mapping to obtain the trend predictions. Lastly, the forecasting results of the trend component, as well as the time-domain and frequency-domain predictions of the seasonal component, are fused to obtain the final prediction output for the time series. In the evaluation of real data, the proposed model shows improved prediction accuracy compared to other multi-dimensional time series long-term forecasting models.</p>