Air conditioning makes up a substantial portion of building energy consumption. Analyzing the correlation between air conditioning set temperature, load, as well as meteorological factors is critical for studying load scheduling strategies of air conditioning. Among them, the air conditioning set temperature is an essential factor that affects building energy consumption and user comfort. Predicting the air conditioning set temperature reasonably based on meteorological conditions and load is a challenging problem. This paper presents a deep learning approach integrating a convolutional neural network (CNN) with long short-term memory (LSTM) for modeling and forecasting air conditioning set temperatures. The hybrid CNN-LSTM approach introduced in this study utilizes the CNN to extract local features of meteorological factors and air conditioning load. Then, it performs a time series analysis using the LSTM model. The proposed approach is employed for analyzing air conditioning operational data collected in a specific region within Zhejiang Province, China, and the prediction results are compared in four other ways: recurrent neural network, backpropagation, LSTM, and bidirectional long short-term memory. The results validate the superior prediction accuracy of the proposed method compared to the four alternative deep learning models. Furthermore, the generalization capability of the proposed CNN-LSTM model is evaluated by conducting tests on a separate dataset of users belonging to the same category. It provides a reference for the implementation of non-intrusive air conditioning control.