Abstract: To ensure the normal interaction of demand response information to carry out demand response safely in various regions, a method to identify and detect the distribute denial of service attack by improved long- short-term memory (abbr. LSTM) network under the interaction of power demand response information, which was suitable for the detection and classification of distributed denial of service (abbr. DDoS) attack in demand response interaction traffic under the form of multiple categories and multiple characteristics, was designed. Firstly, a classification and selection mechanism of power demand response traffic characteristics supported by demand response information exchange specification was presented. Secondly, to recognize the bidirectional traffic within the demand response interaction system, the linear element of Gussian error was led in and based on the improved long- short-term memory network a model to detect distributed denial of service attack was established. Finally, by means of selecting the traffic data set under the demand response, a method of setting up different attack rates under different states of power grid was established to verify the established model, and it was proved that the proposed method possessed high recognition rate for multi-class of distributed denial of service attack in the demand response information interaction and the accurate classification of distributed denial of service attacks could be accurately classified.