Abstract: With the development of new power system and the participation of renewable energy, excavating the resource of large-scale demand side interactive response with high flexibility is in demand. Aiming at solving the problem of demand response (DR) potential analysis with a lack of historical data of load aggregators, a demand response potential assessment method based on deep subdomain adaptation was proposed. Firstly, key factors affecting users' participation in DR were proposed, and a quadratic regression parameter database was established on typical power consumption modes in buildings. Then, a DR potential evaluation method based on similarity evaluation was performed to obtain an estimate value of power users' participation value in DR. Key influencing factors and the load reduction rate of users when participating DR were applied to train the depth subdomain adaptive neural network. Besides, feature parameters extracted from the full connection layer of the network was aligned to attain the training of the potential evaluation neural network with a lack of historical data. The validity of the proposed method was verified by case studies.