Stochastic neural networks for transportation systems modeling
Domenico Rosaci · Applied Artificial Intelligence · 1998
This work intends to analyze the modeling problem of urban transportation networks as a learning process in a biological-like neural network. An urban transportation network is a stochastic dynamic system in which random behavior is due to the uncertain choices of the users in carrying out their trips. Several internal parameters, such as the road capacities of the running lines and the traffic lights' cycles at the crossroad, influence the evolution of the system. Thus, the problem is that of determining a stochastic model capable of simulating the random behavior of such a system. This model will consist ofa number ofparametrical functional elements, each representing a specific subpart of the transportation system. Then, the modeling problem is that ofevaluating the internal parameters of the functional elements in such a way to satisfy the experimental observations representing the real behavior of the urban net. In this work an urban transportation network is regarded as a stochastic recurrent neural network in which the nodes represent the crossroads and the edges represent the running lines. The parameters ofthis neural network are the parameters of the probabilistic distributions associated to each neuron output. These parameters are estimated using supervised learning algorithm based on experimental observations, and the result is a transportation network model that reflects the real behavior. An application of the proposed approach to a real urban net is shown by applying a learning algorithm trained by the Hopfield rule.