Prediction Of Twisted Wire Crosstalk Based On BAS-BP Neural Network Algorithm

With the development of highly integrated power electronic technology,the cable harnesses as the main body of the power electronic equipment network are becoming more and more dense.As the number of signal transmission lines increases,it will also cause an increasingly harsh electromagnetic compatibility environment.As for the electromagnetic compatibility of cables,it can be divided into conduction and radiation.Cable crosstalk is the most important component of conducted interference,and crosstalk prediction is the primary goal of cable electromagnetic compatibility problems.At present,most researches on cable crosstalk focus on uniform multi-conductor transmission lines,while the cables in practical problems are mainly non-uniform multi-conductor transmission lines.As one of the representatives of non-uniform transmission lines,twisted wire has extremely high anti-interference ability and is suitable for high-frequency and high-loss occasions.Therefore,the research on the prediction of twisted wire crosstalk has high theoretical guiding significance and practical engineering value.Starting from the method of crosstalk prediction,the traditional electromagnetic method has complicated calculations,poor repeatability,large memory usage,and lack of self-learning ability.Based on the above considerations,based on the theory of multi-conductor transmission lines,this thesis carries out research on the prediction of stranded wire crosstalk,solves the problem of extracting parasitic parameters of stranded wire,and analyzes its crosstalk law with finite difference time domain algorithm(FDTD).It mainly consists of the following research contents:Firstly,based on the theory of multi-conductor transmission line,modeling and crosstalk analysis of multi-conductor transmission line are carried out.Constructing an equivalent circuit model of multi-conductor transmission line.Extracting the parasitic parameters of the multi-conductor transmission line based on the model analysis and the principle of mirroring.Deriving the multi-conductor transmission line equation from the "way method".Introducing the FDTD algorithm to analyze the multiconductor transmission line equation.Then,based on the BAS-BP neural network algorithm,the twisted wire crosstalk is predicted.Constructing a stranded wire crosstalk model based on the solid geometry method.Constructing a stranded wire parasitic parameter extraction network through model analysis,matrix transformation and the introduction of beetle antennae search(BAS)-back propagation(BP)neural network algorithm.On this basis,combine the FDTD algorithm to solve the noise voltage and noise current to realize the stranded wire crosstalk prediction.Then,carry out the research on the crosstalk modeling and crosstalk prediction of twisted pair when the pitch of the twisted wires is not uniform.Constructing a nonuniform pitch twisted pair crosstalk model from a process point of view.Based on the Monte Carlo algorithm,set up multiple groups of non-uniform pitch twisted pair data networks.On this basis,it combined with the simulated annealing(SA)BAS-BP neural network algorithm to build the parasitic parameter average extraction network and combined with the FDTD algorithm to analyze the crosstalk average law to achieve non-uniform pitch Crosstalk prediction for twisted pair cables.Finally,relying on the Key Engineering Laboratory of Electrical Equipment and Electromagnetic Compatibility of Jiangsu Province,this thesis carried out a verification experiment for the prediction of stranded wire crosstalk.In this thesis,a semi-anechoic chamber and a vector network analyzer are used to extract the S-parameters of the stranded wire by means of frequency sweeping,and the near-end and far-end crosstalk of the stranded wire are obtained through parameter conversion.In this thesis,a threecore twisted wire is used as an example to verify the validity of the prediction of crosstalk and the applicability of the proposed method through comparison with actual measurement data.