In the aluminium smelting process, the accurate prediction of furnace temperature can ensure the quality of aluminium products. However, in a complex furnace environment, it is sometimes difficult to measure the temperature accurately. To address this issue, a soft sensor furnace temperature prediction model based on Multi-Scale ModernTCN is proposed. First, the raw data are decomposed into trend and seasonal components using the time series decomposition technique. Then, the trend component is modeled using a linear layer to capture its long-term trend, while the seasonal component is modeled using a multi-scale ModernTCN pyramid network to extract the multi-scale cyclical features. Finally, the processed trend and seasonal component features are combined to get furnace temperature prediction results. The results of the experiments show that the proposed model meets the accuracy requirements of the aluminium smelting process.