A Feature Set for Structural Characterization of Sphere Gaps and the Breakdown Voltage Prediction by PSO-Optimized Support Vector Classifier

Abstract

Air insulation strength relates closely to the electrostatic field distribution of the gap configuration. To achieve insulation prediction on the basis of electric field (EF) simulations, the spatial structure is characterized by a feature set including 38 parameters defined on a straight line between sphere electrodes. A support vector classifier (SVC) with particle swarm optimization (PSO) is used to establish a prediction model, whose input variables are those features. The EF nonuniform coefficient f of each sample gap is calculated and used for training sample selection according to the ranges of f values. Trained by only 11-sample data, the PSO-optimized SVC model is employed to predict the power frequency breakdown voltages of 260-sphere gaps with a wide range of structure sizes. The predicted values coincide with the standard data given in IEC 60052 very well, with the same trend and minor relative errors. The MAPEs of the five predictions with different training sets are within 2.0%. The model is also effective to predict the breakdown voltages of 9.75-cm sphere-6.5-cm sphere gaps, whose MAPEs are within 2.6%. The results demonstrate the effectiveness of the EF feature set and the generalization ability of the SVC model under the case of limited samples. This paper lays the foundation for estimating the dielectric strength of other air gaps with similar structures.INDEX TERMS Breakdown voltage prediction, sphere gap, electric field features, shortest interelectrode path, support vector classifier (SVC), particle swarm optimization (PSO).

A New Combined Prediction Model for Ultra-Short-Term Wind Power Based on Variational Mode Decomposition and Gradient Boosting Regression Tree

“…By adjusting and experimenting with its parameters iteratively, PSO can partially avoid the issue of being trapped in local optima and instead find the global optimum. The PSO algorithm has gained attention and been widely applied in various fields [26,27] due to its simplicity, ease of implementation, robust performance, fast convergence speed, and reduced tendency to become stuck in local optima.…”

Sphere Gap Breakdown Voltage Prediction Based on ISSA Optimized BP Neural Network and Effective Electric Field Feature Set

“…To further validate the feasibility of this prediction method, it was extended to nonstandard sphere gaps with different sphere diameters, whose power frequency breakdown voltages were recorded in Ref. [14]. The diameters of the high‐voltage sphere and the grounded sphere are 9.75 and 6.25 cm.…”

“…An idea to characterize air gap structure by electrostatic field distribution features was proposed in Ref. [12][13][14], and a data association model based on support vector machine (SVM) was constructed to learn the relationship between the electric field features and air gap breakdown voltages. This method offers an alternative way to realize breakdown voltage calculation by data association analysis rather than physical models.…”