MAlA NEURAL NETWORK: AN APPLICATION TO THE RAILWAY ANTI-SKATING SYSTEM! G. Pappalardo, M.N. Postorino, D. Rosaci, G.M.L. di Reggio Calabria - Reggio Calabria,ltalia tel. 39 965 875204 - fax 39 965 875220 - E-mail: sarne@ns.ing.unirc.it I.Introduction The Neural Network (NN) model proposed in this paper has got a very strong biological analogy; in fact, the basic mechanisms of the signal transmission, particularly the frequency, among biological neurones are considered to model this kind of NN. The information on the NN synapses are transformed in a sinusoidal time-varying signal while in the traditional NN they are constant; furthermore, as in the biological case each neurone can have a proper working frequency. The complete theory of the Modulated Asynchronous Information Arrangement NN will not be presented here [14], but only the basic concepts as well as the techniques used to resolve both linear and non linear NN though the MAlA algorithms in supervised learning modality are presented. The MAlA NN can be largely used and it has been tested to resolve different kinds of problems [6], [9], [10], [11]. In this paper it has been applied to the railway field, particularly in the refining of the performances of a railway anti-skating system. by comparing the results obtained with those of a supervised back-propagation NN. 2. The mono-neural linear MAlA NN The following general problem can be defined: "given two sets I and Y of atomic elements (respectively of cardinality equal to N and M), the unknown relationship F between the elements of I and the elements of Y has to be reached". The aim of the MAlA NN is to search of the unknown relationship F [9], [10]. In order to obtain the general MAlA NN, it is convenient to define the mono-neurone NN, in order to characterize the basic elements of a NN: neurone and synapsis. Then, the following linear stationary system (LSS) described by the following equations and represented in fig. 1 has been considered: x = Ax+ Bi; y = Cx+ Di (1) where iE RN is the vector of inputs, yE RM is the vector of outputs, and XE R S is the vector of the system states. All these vectors can be considered as functions of the time t; A, B, 1 M.N. Postorino dealt with the transport aspects, while G. Pappalardo, D. Rosaci, G.M.L. Same dealt with the neural network aspects. M. Marinaro et al. (eds.), Neural Nets WIRN VIETRI-97 © Springer-Verlag London Limited 1998