Abstract: The intelligent working face has abnormal working conditions such as lost support, uneven straightness, and support skew after the automatic following of hydraulic support. Therefore, manual regulation and control are needed. At present, the research lacks the knowledge discovery of manual regulation and control working conditions after the hydraulic support automation in the production process of the intelligent working face. This is not conducive for workers to quickly judge the number of hydraulic support requiring manual regulation and control. In order to solve the above problems, based on the identification of the number of hydraulic support that is not up to the standard after the hydraulic support automation, a manual regulation and control decision model of middle hydraulic support cluster automation in the intelligent working face is put forward. Firstly, the historical data of the working face is analyzed. It is concluded that after the automatic following of the hydraulic support is finished, three characteristic values can be used as important characteristics for judging whether the hydraulic support carries out manual regulation and control after the automatic following of the hydraulic support. The characteristic values include the distance of the automatic following of the hydraulic support, the stroke variation of the pushing oil cylinder before and after the automatic following of the hydraulic support, and the absolute difference between the number of the hydraulic support at the position of the shearer and the number of the judged hydraulic support. According to the above conclusion, the structure of the manual control decision model after the hydraulic support cluster automation is proposed. The data acquisition module is used for providing the original data. The data preprocessing module prepares the original data by outlier processing, filtering, sorting and correlation analysis. The characteristic engineering module calculates and standardizes the above three characteristic values to provide a sample set for the classification model. After the classification model divides the sample set, the ID3 decision tree is used for classification. Finally, the number of hydraulic supports needing normal working conditions and the number of hydraulic supports nedeing manual control are output. The results of the model evaluation show that, compared with the traditional K-nearest neighbor (KNN), support vector machine (SVM), logical regression (LR) classification algorithms, the training set accuracy of the ID3 decision tree based classification model for the working conditions of hydraulic supports in the middle of the intelligent working face is 92.27%. The test set accuracy is 93.75%. The model can better distinguish the manual control hydraulic support number after automation.