Collapsing coal-rock identification based on wavelet packet entropy andmanifold learning

In order to recognize the collapsing coal and rock, a feature extraction method of vibration signals based on Wavelet packet entropy and manifold learning is proposed.The vibration signals are caused by the impact of collapsing coal-rock and hydraulic support tail beam.Firstly, the vibration signals are decomposed by Wavelet packet and the signals at different frequency bands are reconstructed, then the Wavelet packet energy entropy is calculated to determine the complexity of the signal energy distribution, the sample entropy at each frequency band is calculated to determine the complexity of the Wavelet packet coefficient at each frequency band.Wavelet packet energy entropy and sample entropy of each frequency band are used as feature vectors and the input of BP neural network to identify the collapsing coal and rock.Then, the low-dimensional manifold of feature vector is extracted by local linear embedding (LLE).The low-dimensional manifold is input into neural network to compare the recognition effect with the feature vector as the input.And an unknown sample low-dimensional estimation method is proposed to get its low-dimensional embedding.The results show that the feature vector based on Wavelet packet entropy and LLE is both accurate and simple, and the neural network identification rate reaches 92.5% using it as the input.Low-dimensional embedding of unknown sample based on low-dimensional estimation method is also accurate.