Probability prediction method of transmission line icing fault based on adaptive relevance vector machine

Abstract

To precisely forecast the operation status of transmission line during an ice storm and achieve early warning, a method based on adaptive relevance vector machine (ARVM) is proposed for fault probability prediction of transmission line icing. According to the basic theory of RVM, this paper establishes a forecasting model, which consists of selection and preprocessing of data, initial parameter optimization, icing prediction with adaptive optimization and fault probability prediction of transmission line. The quantum particle swarm algorithm, together with K-fold Cross-validation is applied to optimize model parameters. The weight vector of the icing prediction model is corrected by repeating training to get the precise prediction result of ice thickness with its probability distribution. The case study with practical data from Zhejiang province shows that the proposed method can effectively improve the accuracy of icing prediction and further realize fault probability prediction of transmission line, which can provide early warning for the anti-icing and mitigation work of the electric power department.