FPGA Implementation of ANN Based MPPT Algorithm with A Hardware Friendly Custom Activation Function

Abstract

Despite numerous studies and advances in renewable energy harvesting technology, it is still not mature and robust enough to replace conventional fossil fuels completely. This study explores avenues for improving the performance of one of the most promising sources of renewable energy, solar energy. Calculating the real-time maximum power point (MPP) of a solar system by utilizing an artificial neural network (ANN) implementation in a field programmable gate array (FPGA) device shows considerable improvement in terms of speed and accuracy. In order to accomplish this, a hardware friendly custom activation function was defined. The efficiency of the system improved from 97.4% to 99.21% after the implementation of this custom activation function. And the response time of the system was around 5ms under varying operating conditions. Moreover, data from a laboratory setting for solar development is collected and utilized to create a dataset for training an ANN model. The neural network design is subsequently created for an FPGA implementation and described in Verilog hardware description language (HDL). For validation and verification, hardware-in-the-loop simulation was used throughout the entire process. This study shows that an ANN-based algorithm on an FPGA offers a superior solution for the MPP calculation problem.