Aiming at the problems of low prediction accuracy and long prediction time in the load forecasting method of air conditioning system participating in demand response, an air conditioning load forecasting model based on extreme learning machine (ELM) optimized by principal component analysis (PCA) and seagull optimization algorithm (SOA) is proposed. The main characteristics affecting the load data of air conditioning system are extracted through PCA, the ELM load forecasting model of air conditioning system is established, and the model parameters are iteratively optimized by SOA. In order to verify the effectiveness of the algorithm, taking the air conditioning load data of an office building in Xi'an as an example, the experimental results show that six principal components containing 98.00% of the original information are obtained after PCA feature extraction. The prediction results of SOA-ELM model are basically consistent with the actual values, with root mean square error of 0.013 7, average absolute percentage error of 0.839 2%, determination coefficient of 0.991 0 and training time of 3.482 s. Compared with the other three comparison models, the performance of the model is better. It is proved that the model has strong generalization performance and high prediction accuracy, and can effectively predict the load change in the demand response period of the air conditioning system.