A Hybrid Predictive Meta-Heuristic Framework for Dynamic Traffc Signal Optimization in Multi-Intersection Networks

Abstract

One of the most important issues facing contemporary city administration is urban traffc congestion, which results in major delays, higher fuel usage, and environmental damage. The predominant method in Algerian urban networks, fixed-time signal management plans, are essentially unable to handle the dynamic and unpredictable character of actual traffc demand. In this thesis, an actual nine-intersection road network in the Algerian city of Blida is used to test a hybrid predictive meta-heuristic framework for dynamic traffc signal optimization. A Long Short-Term Memory (LSTM) neural network that forecasts the overall network waiting time sixty seconds in advance, a population-based meta-heuristic optimizer that looks for the best signal plan at each control horizon, and the SUMO microscopic traffc simulator connected via the TraCI real-time interface are all integrated into the framework and function within a rolling horizon control loop. Three optimization methods are assessed: a reactive Genetic Algorithm (GA), which serves as a benchmark without prediction, and the Whale Optimization Algorithm (WOA) and Harris Hawks Optimization (HHO), both of which are guided by LSTM prediction. Predictive HHO reduces waiting times by 27.9%, 38.2%, and 34.6% in comparison to the fixedtime baseline, according to experiments carried out under three traffc scenarios: low (200 veh/hr), medium (500 veh/hr), and peak (900 veh/hr) and confirmed over five independent runs. Reductions of 18.8%, 27.0%, and 18.8% are obtained with predictive WOA. Predictive information is crucial for effcient signal optimization under high congestion, as demonstrated by the reactive GA’s 19.1% performance degradation under peak conditions. With the best average performance and the least run-to-run variability in every case, HHO proves to be the most dependable algorithm