Highway traffic parameter prediction model based on the combined deep learning model

Abstract

Traffic flow time series prediction is developing very rapidly. The research methods of prediction models mainly focus on model parameter optimization and data feature analysis to improve prediction accuracy. Based on this, this article adopts the form of model optimization and uses ETC gantry data to conduct optimal combination and applicability analysis of time series prediction models.This paper builds two deep learning time series combination prediction models based on the Internet of Vehicles and Internet of Things environments and combines them with current research results. Model 1 (INFO-LSTM) is an LSTM prediction model based on the vector-weighted optimization algorithm. Model 2 (ICSSA-GRU-ATT) is a GRU model based on the improved Sparrow Search Algorithm (ICSSA) under the attention mechanism. Short-term prediction was carried out using the ETC data of Shanlin Expressway in Gansu Province as the experimental content. The results showed that the data training and prediction capabilities of INFO-LSTM and ICSSA-GRU-ATT were ideal in the test. Model 1 performs very well in the prediction results of the training set and has strong data memory capabilities in data analysis. Model 2 is also able to meet the stability requirements.