The Study On Caenorhabditis Elegans Behavior Based On Biological Neural Network

The Caenorhabditis elegans(C.elegans)neural network possesses simple structure and complete function,so it becomes an important model organism for studying bio-intelligence and machine intelligence.Though the complete connectivities of C.elegans neural network have been determined,the intensities of synaptic connections between neurons are still undetermined,meaning that the mechanisms of C.elegans behaviors are not clear.In this thesis,in order to simulate the light-touch response,foraging and avoidance behavior of C.elegans,a C.elegans neural network model is constructed.This work can provide some inspiration for studying the mechanism of C.elegans behavior.The main research contents are as follows:1.The traditional feedforward neural networks mostly adopt the static neuron models and they topologies are hierarchical.So they are not suitable for modeling C.elegans neural network which has dynamic neurons and annular topology.This thesis constructs a C.elegans neural network model based on discrete Hopfield dynamic neuron,which models the chemical synaptic connections in the C.elegans neural network as unidirectional connections,and the gap connections as bidirectional connections.The C.elegans neural network model is a feedback neural network whose topology is consistent with the C.elegans neural network and neurons outputs also change over time,thus can keep a high similarity with the C.elegans neural network.2.The topology of the C.elegans neural network model is also annular,so the gradient can not be calculated.Therefore,this thesis applies the genetic algorithm which doesn't require gradient information for parameter learning.In order to improve the performance of genetic algorithm,this thesis proposes an improved genetic algorithm,which crossover operator and mutation operator use eugenic strategy.The steps of the crossover operator are as follows: first,it uses human behavior algorithm to find all local minimum points between the two parents;then,local minimum points with higher fitness are selected as offspring.The steps of the mutation operator are as follows: first,it lets the individual waiting for mutation be the start point and generate a random search direction;then,it uses human behavior algorithm to find the global minimum point along the direction;last,it lets the global minimum point be the mutated individual.And the effectiveness of the improved genetic algorithm is verified by experiments.3.Based on the C.elegans neural network model,local neural network models of the lighttouch response,foraging and avoidance behavior are established.The behaviors of C.elegans are characterized by input modes and ideal output modes.Aiming at the foraging behavior and avoidance behavior,a two-dimensional motion model with stochastic characteristic is constructed which reflects the characteristic that the random occurrence of rotation during the movement.The experimental results show that C.elegans can exhibit light-touch response,find food,avoid harmful chemicals successfully,and the simulated trajectories are similar to the true trajectories of C.elegans,verifying the effectiveness of the C.elegans neural network model,behavior representation method and two-dimensional motion model.