Intelligent recognition of coal and rock based on wavelet packet multi-scale fuzzy entropy and weighted KL divergence

Intelligent recognition of collapsed coal and rock is a prerequisite for intelligent coal caving. Real-time and precise recognition of collapsed coal and rock can avoid the problem of "under caving" or "over caving" of top coal caused by manual coal caving. Most existing coal and rock recognition methods obtain collapsed coal and rock feature vectors through data dimensionality reduction processing, and construct recognition models for coal and rock recognition. However, data dimensionality reduction, model establishment, and training all require a long time. To some extent, these factors affect the efficiency of continuous fully mechanized caving mining. In order to solve the above problems, an intelligent coal and rock recognition method based on wavelet packet multi-scale fuzzy entropy and weighted KL divergence is proposed. Wavelet packet decomposition is performed on the vibration signals of the tail beam after the hydraulic support is impacted by collapsed coal and rock under different working conditions (top coal collapse, rock collapse, and large top coal collapse) to obtain a series of frequency bands. The sequences of each frequency band are coarse-grained. The method calculates the fuzzy entropy under multiple scales of coarse-grained sequences in each frequency band, that is, wavelet packet multi-scale fuzzy entropy. The method uses it as a feature vector. The method uses the ratio of the energy of each frequency band after wavelet packet decomposition to the total energy of the vibration signal as the weight of the weighted KL divergence. The weighted KL divergence of the unknown samples to be tested and the sample feature vectors under different working conditions are compared. The real-time and precise recognition of collapsed coal and rock is achieved. The experimental results show that the method based on wavelet packet multi-scale fuzzy entropy and weighted KL divergence can effectively recognize the category of collapsed coal and rock. The method based on multi-scale fuzzy entropy and KL divergence and the method based on wavelet packet fuzzy entropy and KL divergence have poor recognition performance. When wavelet packet multi-scale fuzzy entropy is used as the feature vector, the recognition accuracy of the BP neural network reaches 95%. It further verifies that wavelet packet multi-scale fuzzy entropy can be used as the feature vector to characterize collapsed coal and rock. The entire coal and rock identification process takes 1.063 9 seconds, which basically meets the real-time requirements of intelligent recognition of collapsed coal and rock. At the same time, it greatly reduces the impact on the efficiency of continuous fully mechanized caving mining. Its comprehensive performance is superior to similar coal and rock recognition methods.