Abstract: This paper proposes a structural dynamic response prediction based on Physics-Informed Extreme Learning Machine (PIELM) to address the limitations of high computational demands in traditional numerical methods and the necessity of high-quality training data for data-driven machine learning models. Based on the Extreme Learning Machine (ELM) framework, the proposed method employs a differential operator derived from commonality of higher-order derivatives, transforming the structural dynamics problem into a task of solving output weights. The physical information is embedded into the loss function in the form of differential operators. The method obviates the reliance on training data for predictions and ensures physical interpretation and generalizability. The numerical studies demonstrate that the proposed method exhibits superior time step adaptability, prediction accuracy and error control capability than traditional numerical methods, providing an efficient and stable prediction method for structural dynamic response prediction.