Summary of the methods in current research.

Reduced forecast efficiency and accuracy are the result of traditional traffic flow prediction algorithms’ inability to adequately capture the spatiotemporal characteristics and dynamic changes of traffic flow. To address this problem, this study proposes a short-term traffic flow prediction method based on an improved convolutional neural network and a bidirectional long short-term memory algorithm. The method firstly identifies, repairs and decomposes the abnormal traffic flow data by smoothing the estimation threshold and adaptive noise integration empirical modal decomposition method to improve the data quality and stability. The suggested model is then supplemented with the enhanced Adam and Lookahead algorithms in an effort to increase the model’s prediction accuracy and rate of convergence. The outcomes indicated that the method showed faster convergence and lower loss values during both training and validation. The training loss decreased from 0.0250 to 0.0021, and the validation loss decreased from 0.0010 to 0.0008. Compared with the traditional convolutional neural network with bidirectional long short-term memory algorithm, the training loss decreased by 42.86% The suggested algorithm outperformed the current advanced algorithms in terms of prediction precision, with an average absolute percentage error of 0.233 and a root mean square error of 23.87. The findings display that the study’s suggested algorithm can effectively and precisely forecast the short-term traffic flow, which serves as a solid foundation for planning and traffic management decisions.