Sparse Bayesian learning mechanism for research of audible noise in UHV transmission project

Audible noise produced by corona discharges is one of the more important considerations in the design of UHV AC transmission lines, which will greatly affect the electromagnetic environment and the technical economical index of transmission lines, etc. So it will be of very important practical significance that making scientific researches on AN prediction from UHV AC transmission lines. Based on the basic philosophy of sound propagation and attenuation, quantitative relationship of the model with sound pressure level and sound power level is deduced, which it will provide the theory basis for AN prediction. To overcome the limitation of existing prediction formulas, a novel machine learning technique, i.e. relevance vector machine (RVM) for AN prediction is presented in this paper. The RVM has a sparse Bayesian learning framework and has good generalization capability, as a result it can yield higher prediction accuracy and more universal application arrange. Based on the RVM regression prediction model, the AN from 1000kV AC UHV single-circuit lines and double-circuit lines on the same tower in China are calculated, and it is shown that the line configurations are rational and satisfied with the request of environment noise standard.