Abstract
Dual-ramp merging control requires coordination of safety risk, traffic operations, and command smoothness. We propose EA-QD-CVaR, which uses traffic-state entropy, risk entropy, and event proportions to update control objectives and local search parameters. Conditional value-at-risk (CVaR) evaluates the upper tail of time-to-collision (TTC) loss. In a SUMO dual-ramp merging scenario with demand scaled to real traffic conditions, each stochastic method undergoes 25 independent search trials. The main joint-risk measure is based on the union of events with TTC below 7 s or with net headway below 3 s. Compared with Uniform, SA, Greedy-global, and MPC-CVaR, EA-QD-CVaR reduces mean TTC risk by 6.22–24.95% and mean joint risk by 1.54–8.65%, while increasing mean downstream arrivals by 0.82–2.91%. Relative to the single observations for the deterministic references Greedy-local, No-Control, and ALINEA, mean TTC risk decreases by 13.23%, 28.56%, and 31.79%, respectively. The corresponding reductions in mean joint risk are 5.28%, 10.15%, and 7.89%, and mean downstream arrivals increase by 0.94%, 1.69%, and 1.69%. Relative to fixed-weight QD-CVaR, mean joint risk decreases by 4.71% at the 3 s net-headway threshold. At the 1.5 s threshold used for control feedback, the decrease is only 0.15%. Control variation increases. The method achieves lower mean TTC and joint risk and slightly higher mean downstream arrivals. It also supports the design and simulation-based evaluation of coordinated dual-ramp metering and variable speed limits under a downstream freeway bottleneck.