The cement industry has consistently consumed large amounts of coal and electricity resources. Optimizing energy scheduling and production process control can typically save energy and improve production efficiency. Therefore, the prediction of energy consumption holds great significance in the cement industry and other energy-intensive sectors. However, predicting energy costs is challenging due to multiple production factors, variable coupling, and time lags. In this research, we proposed the use of a dual-channel temporal convolution neural(TCN) network to forecast coal and electricity consumption in the cement calcination process for the upcoming production hour. Additionally, we employ the Spearman correlation coefficient method to select variables for the calcination system, aiming to reduce feature data dimensions and improve model training efficiency. To address parameter redundancy and mitigate the risk of overfitting, we devise a dual-channel structure. For comparison, we utilized various models including Recurrent Neural Network (RNN), Gate Recurrent Unit(GRU), Long Short-Term Memory(LSTM), Convolutional Neural Network(CNN), and Back-Propagation(BP) in prediction experiments using actual cement calcination process energy consumption data. The results indicated that with a kernel size of 13, dilation rates of [21, 22,…26 ], and a filter size of 36, the TCN model achieves an accuracy of 97.65%. Relative to other models, the TCN model achieved a reduction of at least 40% and 24% in the Mean Squared Error (MSE) for coal and electricity consumption forecasts, respectively, meeting the expected requirements.