Abstract Traditional air conditioner (AC) temperature control typically sets temperature goals or establishes a uniform temperature range without considering industrial characteristics and user comfort preferences. As a result, users cannot achieve differentiated and precise control. This paper proposes a data-driven approach to predict the range of AC set temperatures under the constraint of user comfort. Firstly, a random forest feature importance screening method is employed to extract key factors influencing AC set temperatures. Secondly, a convolutional neural network (CNN) model is constructed to estimate AC set temperatures for five typical user categories. Finally, based on the Predicted Mean Vote (PMV) model, user thermal comfort constraints are established to calculate the flexible adjustment range of AC for users. Through simulation analysis using real AC data from a certain region in Eastern China, the proposed standard for estimating user air conditioner set temperatures in this paper ensures a root mean square error not exceeding 0.5 and an average absolute percentage error not surpassing 1.8%. The obtained air conditioner adjustment ranges reflect the characteristics of users in different industries, meeting engineering requirements.