FPGA-implementation of the Decision Block for Numerical Prediction System Based on Artificial Neural Networks

This paper considers the method of cascading neural network modules to build the operating device of a specialized computing system. Each module implements some model of neural network. The cascade of modules is formed strictly in accordance with the logical links, found after the analysis of training samples. The structure of the input vector is fragmented according to the logical dependencies of the parameter set. Individual nodes of the cascade process certain fragments of the input vector, taking into account the features of neural network models used in them. The library containing the basic elements was prepared in the Quartus II CAD. The prototype with the ensemble of the main modules is implemented on the Cyclone III FPGA. Comparative assessments of the hardware- and software- implementation of the system for predicting numerical results in sports disciplines and multisport competitions are presented.