PM2.5 CONCENTRATION PREDICTION MODEL USING IMPROVED DBN AND SLTSA

In view of the low accuracy of the existing PM2.5 prediction model, a PM2.5 concentration prediction model based on improved deep belief networks (DBN) and supervised local tangent space arrangement algorithm (SLTSA) is proposed. Firstly, the original PM2.5 concentration sequence was decomposed by the complementary set empirical mode decomposition method. Then, SLTSA is used to map the high-dimensional data to the low-dimensional feature space and extract the concentration feature. Finally, the reduced dimension features are input into the improved DBN model, and the PM2.5 concentration prediction is obtained through deep learning training. Based on the sampling data of Zhoukou monitoring point, the simulation experiment of PM2.5 concentration prediction model is carried out. The results show that the changing trend of PM2.5 concentration prediction of the proposed model is consistent with the actual value in the prediction period, and compared with other models, the average relative error of the proposed model is the smallest and the convergence speed is the fastest, and the prediction effect of PM2.5 concentration is the best.