Research on traffic flow optimization and signal light configuration based on GIS multidimensional data-driven approach

Abstract

This study proposes an intelligent traffic management solution based on GIS multidimensional data-driven approach to address the problems of increasing traffic congestion and insufficient effectiveness of traditional management models in the process of urbanization. By integrating short-term traffic flow prediction and dynamic signal timing optimization technology, the solution achieves intelligent expansion of road resources. Taking Nanxun Ancient Town as an empirical object, integrating multi-source heterogeneous data such as road network structure, real-time traffic flow, spatio-temporal distribution, holiday features, etc., a three-dimensional feature matrix containing dynamic function perception features, multi-scale time series coding and spatio-temporal map diffusion features was constructed, and a hybrid prediction model (DTP model) of XGBoost and Multi-Layer Perceptron (MLP) was innovatively proposed. Through differential evolution algorithm to optimize weight distribution (XGBoost: 0.62, MLP: 0.38), the accuracy of traffic flow prediction reached 95%, which was 19.1% higher than the performance of a single model. On this basis, this article combines the SCATS signal light dynamic timing system to establish a five-step timing model that includes traffic data collection, dynamic sub zone division, public cycle calculation, green signal ratio allocation, and phase difference optimization. The green light duration allocation and regional coordination control strategy are optimized to significantly improve the efficiency of road network operation.