The air-conditioning demand response has gradually become an effective approach to reduce peak load of the national grid. However, with the increasing penetration of renewable energy generation sources, static performance evaluation of demand response that only fits one certain scenario cannot meet the demandsofthe times. To address this issue, a novel dynamic demand response quantification and control framework was built based on the secondary development of building energy simulation software and multilayer perceptron algorithm. In this framework, the demand response evaluation method can be used for different global temperature adjustment strategies under different external disturbances and internal loads, and the dynamic demand response control method could be further adopted for load reduction prediction and setpoint adjustment recommendation. The performance prediction model based on artificial neural network can predict the maximum load shedding during demand response period with an average absolute percentage error of 15 %, and the prediction error of load shedding intensity is ± 1 W/m2 for over 80 % of samples. Besides, the setpoint adjustment prediction model achieved an average absolute percentage error of 3 % while its prediction error is ± 0.25 °C for over 99 % of samples. The results prove that this research can provide reasonable suggestions for customers and load aggregators to achieve accurate control of setpoint adjustment in dynamic demand response conditions.