Abstract: The congestion of coal flow in the fully mechanized mining face may cause the scraper conveyor to operate under overload or even be crushed, which significantly affects the production efficiency of the fully mechanized mining face. Aiming at the problem of coal flow congestion in scraper conveyors caused by insufficient coordination of mining and transportation operations in the fully mechanized mining face, this paper takes the coal shearer as the control object and proposes a method for regulating the traction speed of the coal shearer based on ensemble decision-making to prevent coal flow congestion. This method constructs a bidirectional temporal attention coal flow speed regulation model TC-BGRU was constructed by combining the self-attention mechanism. The coal flow regulation accuracy of the deep learning model was improved by mining the dependency information between the features of the coal flow time series. To enhance the adaptability of TC-BGRU under different working conditions, taking TC-BGRU as the base decision-maker, a double-layer ensemble decision-making method for regulating the traction speed of coal mining machines, dubbed DI-TC-BGRU, is proposed. This method can dynamically select base decision-makers with different characteristics under different coal flow congestion working conditions. To verify the effectiveness of the proposed method, the performance of the proposed method was verified in combination with the actual data of the 52604 fully-mechanized mining face of a certain coal mine in Shendong Mining area. The experimental results show that compared with the TC-BGRU algorithm, the evaluation indicators of the DI-TC-BGRU algorithm, namely the mean absolute error, the mean square error, and the logarithmic error of the root mean square, have decreased by 36.82%, 22.58%, and 25.69% respectively. Compared with the Transformer algorithm, the evaluation indicators of the DI-TC-BGRU algorithm, namely the mean absolute error, mean square error, and root mean square logarithmic error, have been reduced by 42.05%, 32.87%, and 44.27% respectively. Our method can meet the requirements of coal flow congestion control and treatment in fully mechanized mining faces, providing an effective solution for coal flow congestion prevention control and treatment in fully mechanized mining faces.