Coal and gangue identification method based on EMD feature extraction and random forest

Identification based on vibration signals is an effective method to realize coal and gangue identification in fully mechanized mining. The existing method needs to be further studied in terms of identification accuracy and effectiveness. A coal and gangue identification method based on empirical mode decomposition(EMD) feature extraction and random forest(RF) is proposed in this study. The acceleration sensor and data acquisition instrument are used to collect the vibration signals generated by the impact of coal and gangue on the tail beam of the hydraulic support in a fully mechanized working face. Then the two signals are processed by EMD respectively so as to obtain a series of intrinsic mode functions(IMF). The effective IMFs are selected according to the EMD results, and the IMF energy, kurtosis, matrix singular values and corresponding entropy are extracted as feature vectors. Each feature vector is used to train the RF model independently. The feature vectors are filtered according to the identification results of each RF model on the test samples, and the feature data set is established. The feature data set is used to train the RF model, and the trained RF model is applied to realize the coal and gangue identification. The test results show that the identification accuracy of the method reaches 96.5% for 200 sets of coal and gangue test samples, and the highest identification accuracy is achieved when the number of decision trees in the RF model is set to 100 or 150. Furthermore, the time consumed for feature extraction and identification of test samples is less than 0.2 s, which meets the requirements of accuracy and real time of coal and gangue identification in fully mechanized working face.