Quantitative prediction model and prewarning system of water yield capacity (WYC) from coal seam roof based on deep learning and joint advanced detection

As one of the most abundant fossil fuels in the world, the safe and efficient mining of coal resources has greatly affects all fields of human production and life. Accurate prediction of the water yield capacity (WYC) of coal seam roofs and advanced detection of water abundance is of great significance to preventing roof water disasters in coal mines. In the process of coal mining, complex roof water inrush mechanism, and changeable factors are faced. Moreover, the uncertainty of the interconnection between various factors increases the difficulty of roof water disaster prevention. To improve the accuracy and reliability of WYC prediction, this study analyzed the main control factors of coal seam roof water inrush and built a WYC quantitative prediction model based on the SAA-CNN algorithm. 47 samples (40 training sets and 7 test sets) were selected to train and learn the model, and the WYC of the 3rd coal mining roof in Jiangzhuang Coal Mine (JZCM) was predicted. To verify the reliability of the model, BP, RF, SVM, CNN, and SAA-CNN models were used for training and error comparison. Finally, the 1001 working face of JZCM was selected to use a variety of geophysical methods and drilling combined for advanced detection and construct the dynamic monitoring and prewarning system of coal seam roof water inrush risk. The results show that the accuracy of the WYC prediction model is SAA-CNN>RF>CNN>SVM>BP. The average absolute error of the SAA-CNN model is 0.87 m3/h, and the average relative error is 3.56 %. In this study, the quantitative and accurate prediction of WYC of coal seam roofs is realized. Combined with the advanced detection method of "electric-magnetic-drilling", the prewarning system of water inrush risk of mining is constructed, which provides a new method for the prevention of water disasters in coal seam roofs.