Research On Biological Neural Network Structure Identification Method Based On Nonlinear Granger Causality

Biological Neural Network(BNN)is a very complex dynamical nonlinear network system.It is an important field of brain science research to explore the functional connection map of biological neuron network and to deeply study the dominant relationship between the structure of biological neuron network and its network function.The idea of linear granger causality modeling and causality identification is extended to nonlinear granger causality modeling and network structure identification according to the essential nonlinear dynamic characteristics of neuron pulse ignition,using the Radial Basis Function(RBF)to fit the nonlinear multi-variable dynamic response characteristics of biological neural network,By detecting the synaptic neuron after pulse ignition signal prediction accuracy is improved,revealing the synaptic neuron functional connections from the former,which will be proven by research of biological neural networks to the flow of information distribution.To verify the effectiveness of the proposed method,the nonlinear granger causality network structure identification method is applied to the network structure discovery process of Spiking Neural Network(SNN)simulation model based on Integrate and Fire(IF)mechanism.The main work contents are as follows:(1)The biological neural network is modeled and simulated.The real firing behavior of neurons is simulated by artificial construction of biological reality pulse neural network SNN model.Firstly,the SNN model based on IF mechanism is established.Then,the parameters in the model are determined,and a neuron and multiple neural networks are simulated.The SNN model simulates the firing behavior of real biological neurons when receiving stimulation,and generates multi-channel pulse neuron sequence data through network simulation.(2)The nonlinear granger causality method is used to identify the network structure.The causal synaptic connections and strength in the network are identified reversely using the multi-channel pulse neuron sequence data generated by the simulation of the network.The classical granger causality is extended to the nonlinear space by using RBF to fit nonlinear model,and the causal effects between neurons are judged by comparing the causality of interactions between neurons,so as to identify the structure of biological neuron network.(3)The identification results are obtained by MATLAB simulation.The linear granger causality method and the nonlinear granger causality method based on the RBF are used to identify the same biological neural network structure,and the identification results show that for 2 nodes,3 nodes,4 nodes,5 nodes,6 nodes of the small-scale neural networks,medium-scale neural networks with 10 nodes,15 nodes,20 nodes,25 nodes,30 nodes and large-scale neural networks with 50 nodes,60 nodes,80 nodes,100 nodes,and the identification accuracy of nonlinear ganger causality method is significantly higher than the linear ganger causality method in the same scale networks.