With the rapid expansion of power system scale, demand response business is promoted to develop. More and more demand response terminals are connected to the smart grid, smart grid is an intelligent system that allows the grid to effectively perform its functions. Its data can be used in intelligent decision-making during grid operation, which may be attacked by hackers in practical applications, causing security problems of demand response terminals of the power network. The security feedback trust model establishes trust relationship through trust mechanism, which can effectively ensure the security of interaction between nodes and demand response terminals of the smart grid. Therefore, a security feedback trust model of power network demand response terminal triggered by hacker attacks is proposed. Analyze the role of smart grid in power grid, and use convolutional neural network in artificial intelligence technology to enhance the flexibility of smart grid. Aiming at the security problem of the demand response terminal of the power network being attacked by hackers, based on the trust theory, the security feedback trust model of the demand response terminal of the power network is designed through the main services provided by the security feedback trust model, the trust information storage of the power network nodes and the summary of the main work. Establish the identity trust relationship, adopt the distributed verifiable signature scheme, update the power grid node certificate, update the identity trust relationship, and revoke the identity trust relationship based on the trust evaluation and threshold value to prevent hackers from attacking the power grid demand response terminal. Based on information theory, trust is established and measured. Entropy is used to represent the trust value. Behavior trust evaluation and composition mechanism are introduced into the security feedback trust model of power network demand response terminals to achieve the credibility of identity and behavior among power network nodes. The experimental results show that the proposed method can judge the hacker attacks, reduce the impact of hacker attacks on the trust of power grid nodes, and improve the interaction security between power grid demand response terminals and power grid nodes.