In the Automatic Demand Response System (ADRS), the electrical load characteristics of the user's side will vary with the change of the response strategy due to the influence of holidays, weather, temperature and other factors, when a large amount of power load data is collected automatically in real time. The prediction accuracy will be reduced using the traditional load forecasting method, which will no longer meet the requirement of ADRS. In this paper, based on the response characteristic of flexible load and the rich load power data of ADRS, the Automatic Demand (AD) response signal is introduced into the system as an input variable when each link of "predictive- dispatch -response" is consided as a closed loop control system. The closed loop ultra-short-term forecasting model is established by using the "black box" idea. The validity of the model is verified by simulation results. The results show that the prediction accuracy of the model is higher than that of the traditional prediction model. This work is supported by National Natural Science Foundation of China (No. 61703144), Henan Open Laboratory Project (No. KG2016-7), Key Scientific Research Project of Colleges and Universities in Henan Province (No. 18A470014) and Doctoral Fund of Henan Polytechnic University (No. B2017-20).