FPGA-implementation of a Prediction Module Based on a Generalized Regression Neural Network

In many areas of human activity it is required to solve forecasting problems. Depending on the application area, they can cover a different time frame. Ultra-short-term parametric forecasting requires a rapid response from the system. Software systems that implement forecasts, due to a number of features may not provide sufficient performance. The article describes the possibility of implementing the prediction module on FPGAs. A generalized regression neural network was chosen as a mathematical forecasting apparatus, which showed quite good efficiency in solving such problems. The functional scheme of the module is given, the calculation of the required number of gates for the implementation of the neural network on the FPGA is made. On the basis of the test experiments, a comparative analysis of the module performance by software (MATLAB) and hardware implementation was carried out.