With the development of industry, urban air pollution has become increasingly serious. In the face of increasing air pollution, forecasts of air quality are particularly important. Air quality data is highly complex nonlinear time series, and traditional linear methods are difficult to obtain good prediction results. Support vector machine regression is based on statistical machine learning methods. It overcomes the problems of traditional methods and has good robustness and generalization ability. For SVM prediction problems, the choice of model parameters (penalty factor C, kernel function type, and kernel function parameters) has a crucial influence on the accuracy of the prediction results. How to select the model parameters to obtain the optimal prediction results is a hot issue in the current research field of support vector machines. At present, commonly used SVM parameter optimisation methods include genetic algorithm (GA), standard particle swarm algorithm (PSO), and ant colony algorithm (ACO). For SVR difficult to determine the optimal parameters, this paper proposes the LSA-SVR model, and then use the model to simulate the Wuhan Meteorological Station air quality index. Through simulation experiments on Wuhan historical weather data and error analysis and comparison with the traditional SVR and PSO-SVR models, it was found that the LSA-SVR method has a better predict effect.