Abstract: To solve the problem of complex calculation of polling system performance parameters, a performance prediction method of polling system based on Long Short-Term Memory (LSTM) network was proposed. By modeling and analyzing the existing experimental data, a LSTM model was constructed to predict the polling system performance under different information packets arrival rates. First, the experimental data were processed, and the average queue length under different information packets arrival rates was formed into a sequence. Then the LSTM network including input layer, hidden layer, fully connected layer and output layer was established to perform the prediction. Finally, the output of the network was re-input to predict the average queue length under unknown arrival rate. Through experiments, the predicted value and the real value curve show the same trend, which shows that this method can effectively predict the performance of the polling systems. Compared with the traditional mathematical analysis method, this method is more efficient.