Abstract: In order to recognize the collapsing coal and rock, the vibration signals caused by the impact of collapsing coal-rock and hydraulic support tail beam are collected at the scene of a fully mechanized caving face.Then a new feature extraction method based on wavelet packet energy flow and LTSA is proposed, which is achieved by three main steps: firstly, the wavelet packet transform is conducted to decompose the vibration signals into a set of different timefrequency subspaces; then the wavelet packet energy flow is formed through time-frequency subspaces from low layer to high layer, the energy is calculated in each subspace of each layer; finally, low-dimensional manifold features carrying class information are extracted from the wavelet packet energy flow by using the LTSA algorithm.To verify the effectiveness of the low-dimensional manifold feature, it is used as the input of BP neural network to identify the collapsing coal and rock.The experimental results show that the proposed feature based on wavelet packet energy flow and LTSA is both accurate and concise, and the neural network identification rate reaches 100%.