Teleoperation Robot Network Delay Prediction Research Based On Artificial Neural Network

Teleoperation system based on Internet is a kind of comprehensive control system, which put the Internet as communication medium and the operator can control remote robots to complete specified actions. It makes full use of the advantages of Internet such as universal, easy access, and cheap resource and so on, and will have broad prospects for development in the future. But, the characters of transmission delay of Internet such as time-varying, jitter problem have also become the bottleneck problems for teleoperation system.This paper first through the actual network round trip time delay (RTT) measurement, we get a lot of delay data. And we make a deep research of the autocorrelation and probability density distribution of Internet delay characteristics, concluded that the segmented time delay distribution is approximated to translation gamma distribution.Then, we come up with a time delay prediction model which based on RBF (Radial Basis Function) neural network. This model first uses support vector machine algorithm to classify the time delay data, and then build different neural network models which will be trained by classification sample sets, and at last use the multiple trained neural network to achieve the Internet time delay prediction by switch regulation. The differences between this several different RBF neural network models are the number of neurons and center vector in hidden layer, and the connection weights between hidden layer and output layer. This article use bisection K-means clustering algorithm to optimize cluster centers of hidden layer neurons, and the number of neurons in hidden layer is just the cluster centers’ number, and improved local particle swarm optimization algorithm is used to optimize the weight matrix between hidden layer and output layer. And the validity of the model is verified by experimental contrast.At last, we try to improve the existing TCP retransmission mechanism and congestion control mechanism, and the improved TCP timeout timer algorithm and improved TCP congestion window increases and decreases algorithm is proposed. The improved algorithm is applied to robotic arm simulation experiment as a whole. At last, by NS2 simulation experiments, the effectiveness of the new TCP in improving the network throughput and reducing network shock and some other aspects are verified.