A neural prediction of multi-sensor systems

In actual engineering applications a typical problem concerns the prediction (classification) of successive states of a real-world system. The state is often characterized by several measures related to a multi-sensor array. We propose in the paper a clustering approach to the automatic determination of significant zones in the multidimensional space where data can be represented and by which the information about the characteristic system state can be classified. Using this approach we will obtain multidimensional time series, which will be predicted by an MoG (Mixture of Gaussian) neural network. The proposed system will be validated by considering a particular application concerning the prediction of the vehicular traffic flow.