Short term traffic flow prediction and timing optimization at signalized intersections based on SG-LSTM and particle swarm optimization

In addressing the challenge of spatio-temporal correlation in traffic flow at signalized intersections, this study introduces an innovative methodology that integrates a SG-LSTM neural network with a particle swarm optimization algorithm. The proposed methodology involves the pre-processing of original traffic data to enhance its predictability, followed by the development of a short-term traffic flow prediction model utilizing the LSTM neural network. Utilizing the outcomes of the traffic flow predictions, a dynamic timing model for signalized intersections is formulated through the application of the particle swarm optimization algorithm. Furthermore, a multi-objective optimization model is established to refine the timing scheme of the signalized intersections, incorporating constraints that consider the maximization of overall intersection capacity, minimization of average delay, and reduction of average queuing times. The Pareto compromise programming method is employed to perform dimensionless processing of the three performance indicators, while the fuzzy preference method is utilized to ascertain the weight relationships among the objective functions. The optimized signal timing scheme is derived through the implementation of the particle swarm optimization algorithm in Matlab. Experimental results indicate that the proposed methodology surpasses the traditional Webster model in terms of performance indicators, thereby affirming the effectiveness and reliability of the approach.