Traffic flow prediction based on nonlinear weight decreasing PSO-SVR univariate time series prediction algorithm

The aim of this paper is to investigate the application of car flow prediction in the field of transport in order to solve the problem of urban traffic congestion. For this purpose, we adopt the nonlinear weight decreasing PSO-SVR univariate time series prediction algorithm to predict the car flow, and divide the data set into training set and test set according to the ratio of 7:3. By analysing the scatterplot and line graph between the predicted and actual values of the training and test sets, we find that the prediction effect is better, but there is a certain deviation. Specifically, the scatter plot shows a Y=X distribution, but the actual values have a wide range of variation relative to the predicted values. Meanwhile, the line graph shows that the actual traffic flow and the predicted traffic flow have the same trend, but the actual traffic flow value changes more, while the predicted traffic flow value changes less. This may be due to the long time span resulting in too many cycles, and shortening the time span can reduce the number of cycles and thus improve the prediction accuracy. Further analysis of the MAE values for the training and test sets shows that the MAE values for both the training and test sets are relatively small, 1.8437 and 2.6408, respectively, where the MAE for the test set is on the large side, but the overall prediction results of the model are better. Therefore, the nonlinear weight decreasing PSO-SVR univariate time-series prediction algorithm used in this paper can provide powerful decision support for traffic management departments and help them to better formulate traffic planning and management strategies.