Design And Implementation Of Feedforward Neural Network And Particle Swarm Optimization Based On FPGA

Artificial Neural Networks is a branch of machine learning,a mathematical model of the biological neural network structure that mimics the characteristics of biological behavior.Deep learning is a new learning algorithm developed on the basis of neural networks.Its basic structure is deep neural network.Particle Swarm Optimization(PSO)is a swarm intelligence algorithm that optimizes targets.Aiming at some limitations(power consumption,time-consuming and cost)of feedforward neural networks and particle swarm optimization algorithms implemented by software,this paper presents a feedforward neural network and particle swarm optimization algorithm based on FPGA.The method,which realizes parallel,pipeline design and modular design,makes the neural network and PSO algorithm based on FPGA lower power consumption,shorter time and higher computational efficiency.The main contributions of the feedforward neural network and particle swarm optimization algorithm designed and implemented by FPGA in this paper include the following points.(1)The logic structure of BP neural network is designed and implemented by FPGA.Modular programming is adopted,and the circuit structure is designed according to the principle of the algorithm,which makes the circuit more portable and reconfigurable.In the system design,the input layer module to the hidden layer module and the hidden layer module to the output layer module adopt a pipeline structure,which can complete multiple operations in one clock cycle.The data calculation of the hidden layer module,the output layer module and the weight and threshold update module is implemented in parallel,so that the BP neural network designed by FPGA has lower power consumption,shorter time consumption and higher calculation efficiency.(2)The logic structure of designing and implementing convolutional neural networks with FPGA is studied.The convolutional neural network implemented by FPGA also adopts the modular idea,which can easily change the network structure,that is,increase or decrease the number of layers.In the design,the convolutional layer module to the pooling layer module,the pooling layer module to the full connection layer module realizes the pipeline operation,and the convolutional layer module,the pooling layer module and the fully connected layer module layer realize parallelism,and accelerate the convolutional neural network.Calculation.(3)The logic structure of FPGA design and implementation of PSO algorithm is studied.The FPGA design and implementation of PSO algorithm also uses modular design.According to the particle size,particle data of this scale is generated in parallel,and the fitness of particles is calculated in parallel.This greatly shortens the calculation time,making the PSO algorithm more advantageous in hardware implementation.In this paper,a design scheme of feedforward neural network and PSO algorithm based on FPGA is presented and verified in hardware.The implemented feedforward neural network and PSO algorithm are portable,parallel,and pipelined,and can be easily transplanted to other networks.Compared with the feedforward neural network and PSO algorithm implemented by software(MATLAB),FPGAs achieve less resources,lower power consumption,shorter time,higher computational efficiency and more real-time performance.