A Type-2 Fuzzy Traffic Signal Control Method based on Short-Term Traffic Flow Prediction Using LSTM

Traffic demand at signalized intersections generally exhibits temporal continuity and short-term evolution patterns, allowing recent traffic observations to provide useful information for anticipating near-term demand. To improve the responsiveness of isolated-intersection signal control, this study proposes a prediction-assisted interval type-2 fuzzy traffic signal control method integrating rolling one-step-ahead LSTM forecasting with queue-based feedback. The LSTM uses the most recent 20 min of observed lane-level traffic-flow data to predict demand for the next sampling interval. Rather than treating traffic prediction as an independent forecasting task, the proposed method converts lane-level forecasts into phase-level anticipatory demand and embeds this information directly into the traffic state used for signal control. The predicted demand is integrated with the observed queue of the active phase, while the observed next-phase queue provides complementary traffic-state feedback. Interval type-2 fuzzy inference then maps the resulting observed–predicted state to adaptive green-time decisions while accommodating uncertainty in traffic-state representation. In this way, an interpretable prediction-to-control mechanism is established, enabling near-term demand information to influence signal timing while preserving real-time queue feedback. Across three traffic-demand conditions and 30 paired stochastic runs, LSTM-T2 consistently achieves the best delay and queue performance among the compared controllers. Under dynamically varying mild congestion, the prediction mechanism further reduces delay and queue length by 8.74% and 7.09%, respectively, relative to conventional type-2 fuzzy control, demonstrating that incorporating short-term traffic-flow prediction can improve the responsiveness and effectiveness of traffic signal control strategies.