Abstract: Improving the accuracy of coal-rock fracture acoustic emission signal is crucial to prevent gas outburst hazards. In this paper, a coal rock fracture acoustic emission time series prediction model was built on the basis of Principal Component Analysis(PCA) and Echo State Network(ESN), namely PCA-ESN. The acoustic emission time series data of coal-rock instability was obtained using uniaxial compression acoustic emission tests. PCA was applied to reduce the dimensionality of acoustic emission data. The three principal components were extracted as comprehensive indicators of coal rock destabilization rupture degrees. Small-World network(SW) was used to optimize the topology of the Echo State Network reserve pool and to reduce coupling of neurons in the pool. Another acoustic emission time series prediction model, PCA-SWESN was established by using the three comprehensive indexes extracted by PCA as the input of the echo state network. Compared to the PCA-ESN model, the PCA-SWESN model reduced the ill-conditioned solution of the ESN, improved the prediction accuracy of the acoustic emission signal of coal-rock instability. The PCS-SWESN technique provides a theoretical basis for preventing gas disasters.