Research On The Construction And Application Of Neural Network Based On Memristor

The memristor is the fourth basic electronic devices in parallel with the resistance,capacitance and inductance,which is proposed by Professor Cai in 1971,HP labs successfully developed nanometer memristor physical devices in 2008,Since then,researchers at home and abroad have deepened the research of the memristor.Many classic memristor models have been put forward,the application of memristor in the fields of chaotic circuit,secure communication and artificial intelligence is also increasing.Neural network based on memristor is a hot topic in recent years,but most of them have only studied the characteristics of neural networks,which do not involve the application of neural networks.This paper presents an improved memristor model and builds the Hamming Neural Network with this improved memristor model,then Hamming Neural Network is applied to the field of traffic pattern recognition.Traffic pattern recognition is a new research field in recent years,accurately predicting people's travel mode can reasonably plan traffic construction,reduce traffic congestion and improve people's travel speed.Firstly,an improved memristor model is proposed by analyzing many classic memristor models,the improved memristor model consists of four parts,which are cache unit,integral amplifying unit,voltage bias unit and linear optocoupler unit.Compared with the traditional memristor models,this improved memristor model can read the memristor more accurately and save the state of the circuit before power failure.After a simple analysis of the circuit model,the voltage and current curves of the memristor model are observed at different frequencies.Then through the improved memristor model to simulate the synapses.The structure of the synapse and the memristor is similar,so that memristor is very suitable for the simulation of synapses.We observe whether the constructed artificial neural synapse model meets the STDP rules,furthermore,we build artificial neuron cell circuits combined with Neuron MOS tube and memristor.The artificial neuron cells based on the memristor and the Neuron MOS tube are more efficient than the artificial neuron cells constructed by traditional methods.The neuron is the most basic unit of the neural network,after the artificial neuron cell circuit is constructed,we construct the double layer structure of the Hamming Neural Network.Finally,the Hamming Neural Network combined with GPS is applied to traffic pattern recognition,the simulation result shows that if we use GPS and memristor Hamming Neural Network to predict people's travel mode,the accuracy rate is higher than traditional methods.These indicate that memristor neural network have advantages and can be applied in other fields.