Aiming at the energy consumption prediction problem of an air-conditioning system in a shopping mall in Wuhan, the data mining method is used for modeling processing, in which the Savitzky-Golay smoothing algorithm is introduced to perform noise reduction processing on the original data, and the long short-term memory neural network algorithm is used to predict and analyze the instantaneous energy consumption. The results show that compared with methods such as back-propagation neural network, recurrent neural network and gated recurrent unit, the long short-term memory neural network has highest prediction accuracy, in which the determination coefficient is 0.861. The denoising of raw data using the Savitzky-Golay smoothing algorithm significantly enhances prediction accuracy by reducing the influence of noise, achieving a coefficient of determination of 0.955 with a increasing of 10.9%. The feasibility of the proposed method for energy consumption prediction in chilled water systems of commercial buildings is thereby validated.