Abstract: In order to solve the problem of identifying the different degrees of pipeline blockage during the use of the thin feeding system, an improved recurrent neural network-Long and short-term memory（LSTM） algorithm based on complete ensemble empirical mode decomposition（CEEMD） was used to detect the blockage status of feeding pipes, that is, CEEMD was used to decompose the collected sound feedback signals in feeding pipes to obtain the intrinsic mode function（IMF） component. The energy proportion and approximate entropy were extracted from IMF components as feature vectors for constructing feature set M₁. According to the characteristics of Pearson correlation coefficient and energy proportion, the IMF components with strong correlation and more information were selected to reconstruct the feature set M₂. Then back-propagation network and RNN-LSTM algorithm model were used to classify and identify five working conditions of three-way pipeline, non-blocked pipeline, mildly blocked pipeline, moderately blocked pipeline and severely blocked pipeline. The results showed that the recognition accuracy of single feature was lower than that of multiple features, and the recognition accuracy after feature screening was higher than that without screening. Under the same conditions, the recognition accuracy of RNN-LSTM algorithm was significantly higher than that of back-propagation network in identifying the blockage status of feeding pipeline. The results indicated that RNN-LSTM algorithm model could effectively identify the different degrees of blockage in feeding pipeline, and could realize the prediction of pipeline blockage, which had certain practical application value.