Freeway Ramp Metering Control Based on Neural Dynamic Optimization

In this paper, we consider the optimal metering control problem for a local freeway ramp. The traffic model is formulated as one with stochastic and nonlinear properties. The control objective is to maintain the freeway operated at a desired traffic density and to diminish the queue length on the ramp as possible. Such a nonlinear stochastic optimal control problem is known hard to solve exactly, then we concentrate on providing approximate solution. Our approach consists of two steps: the first is to solve a finite-horizon optimal control problem based on neural dynamic optimization; the second is to adapt the finite-horizon optimal controller to work for the infinite-horizon case by using the receding-horizon control mechanism. We choose the multilayer feedforward neural network as the approximating structure, and the stochastic steepest descent method for the training. Extensive simulation studies demonstrate the efficiency of our approach.