Abstract:
Aiming at the problem that the existing single model has a large prediction error of PM
2.5 concentration, the combined model of Autoregressive Integrated Moving Average (ARIMA) and Support Vector Machine (SVM) is proposed. Firstly, the hybrid kernel function based on linear combination is used instead of the single kernel function to increase the learning ability and generalization ability of the SVM model. Then it is considered that the ordinary particle swarm optimization algorithm is easy to get trapped into local optimum solution and late-time oscillation. The improved particle swarm optimization algorithm is proposed by adjusting the inertia weight factor adaptively in different phases of the process according to the variation of the cosine function and adding the momentum term. Finally, the PM
2.5 concentration data of a site in Beijing is verified. The results show that the root mean square error of the improved combined model is reduced by 1.741 μg·m
–3 and 6.72 μg·m
–3, respectively, compared with the unmodified combination model and the single ARIMA model, which has better prediction accuracy.