基于改进相关向量机绝缘子污闪电压预测研究

Research on the Prediction of Insulator Pollution Flashover Voltage Based on Improved Relevance Vector Machine

污闪电压是评估绝缘子电气性能最重要的指标,提出一种改进的相关向量机(RVM)污闪电压预测模型,选择组合核函数作为最终核函数,通过差分进化算法优化RVM的核函数宽度和组合函数的权值,以绝缘子表面盐密和灰密为输入样本,污闪电压为输出样本。仿真结果表明,所建立的污闪电压预测模型较之BP神经网络模型、支持向量机(SVM)模型和未改进的相关向量机模型,精度更高,泛化能力更强,能有效地克服传统方法的局限性,适应于绝缘子污闪电压的实时预测,有一定的应用价值。

Flashover voltage is the most important indicator to assess the performance of electrical insulators. This paper proposes a flashover voltage model based on relevance vector machine. In this model, the combination kernel function is selected as the final kernel function; the kernel width and combination function weight of RVM are optimized using differential evolution algorithm; and the salt and non-soluble deposit density on insulator surface and the pollution flashover voltage are selected as the input and output. Simulation results show that the improved RVM model has higher accuracy and generalization ability than BP neural network, SVM model and non-improved RVM model. It overcomes the shortcomings of traditional methods and is suitable for real-time prediction of pollution flashover voltage. The proposed model has certain application value.