Unmanned construction technology is being developed to aid in disaster recovery by remotely operating construction machinery, thereby reducing the possibility of secondary disasters. However, the work efficiency of unmanned operation is only about half that of the normal operation. Thus, various support methods based on operation states have been developed to improve the work efficiency. It is important to automatically determine the operation state of the construction machinery to apply these support methods. In this paper, we propose an extended operation state identification model that can identify various operation states. The proposed model consists of two parts: operation state classification system using the classification model of Long Short-Term Memory (LSTM), and novel state search and registration system using the regression model of LSTM. Simulation of unmanned construction was conducted to verify the potential of the proposed model.