Application And Research On QIA-SVR For Friction And Wear With Electric Current In Pantograph-Catenary System

Pantograph and catenary is an important devices of the train power supply system, moving train uses the pantograph to obtain power from the contact wire, the sliding contact state of the train directly affects the stability of traction force and running speed, along with six times acceleration of the train, current-carrying instability problem in the pantograph and catenary system have become increasingly prominent. Therefore, carring out the analysis, evaluation of the friction and wear property and current-carrying stability characteristics in the pantograph and catenary system have important realistic significance.Firstly, this paper analyses the performance of support vector machine for regression and introduces its advantages and disadvantages from the VC dimension, slack variables, insensitive loss function, kernel function and other aspects, for the problem of the penalty parameter and kernel function parameter selected by virtue of experience can not guarantee the current prediction effect is optimal, this paper proposes the use of quantum immunity algorithm to improve it, through qubits encoding the antibodies, the dynamics quantum rotating gate achieves evoluion and the quantum catastrophe is joined, this algorithm can effectively solve the multi-peak, high dimensional and nonlinear optimization problems.Then introduced the experimental program of the sliding electrical contact testing machine, according to the experimental measurement data, skateboard wear rate and contact current stability coefficient regression model based on quantum immune optimized support vector machine for regression were established respectively, after regression models are verified by validation sample and compare with regression models of the tradition immune optimization algorithm, they have achieved higher prediction accuracy. In order to depth analysis of the skateboard wear resistance and current-carrying characteristics, the regression models are used to predict data other than the experimental program, multiobjective programming methodology is introduced to strike pareto solution-optimal load.