Research On The Icing Thickness Model Of Transmission Line Based On Intelligent Identification

This subject is sub-project of the science and technology projects in Shanxi Province Electric Power Company, which the title of is "to reduce power transmission line analysis and application of ice measures," .the earliest recorded transmission line icing incidents appeared in the United States in 1932.In recent decades, China has also suffered a wide range of ice incidents in January 2008, southern China suffered heavy snow,follow-up Freeze snow and other extreme weather attacks, leading to coal shortage and 17 provinces power grid blackouts, it caused very heavy losses. Icing problems suddenly become a hot problem. This paper is based on Xinzhou in Shanxi province which is easy to produce icing disaster, through a back line Xinzhou Kambara 109 towers at the meteorological parameters and related historical data analysis, neural network intelligent recognition of the transmission network covered ice thickness model, follow-up to simulate the future identification predicted ice thickness in order to guide the production before they occur.In this paper, the generation mechanism of ice cover as well as existing models are reviewed, followed by in-depth study of the neural network model building issues, and gives a more appropriate modeling methods and principles to be followed. In the study of a large number of literature based on the follow-up ice by analyzing the impact of meteorological factors, built a four-layer BP neural networks, established Xinzhou 220kVtransmission line model of intelligent identification. Prediction accuracy of the results of the relative level of traditional methods has been greatly improved. However, it's difficult to determine based on BP network hidden layer nodes, easy to fall into local minimum point, This paper discusses the RBF (Radial Basis Function) network to establish the mechanism, RBF network model training speed and good convergence characteristics, the same time can greatly reduce the number of hidden layer neurons, the paper by matlab as a platform, according to transmission line The historical data were used to four BP network and RBF network of model identification, the identification results show that: RBF network is better than BP network forecast.SVM is based on statistical learning theory, a new machine learning techniques. As a result of the principle of structural risk minimization Empirical risk minimization principle of substitution, making it a better solution to the small sample learning problems. SVM theory is precisely because the theoretical basis for a more complete and better learning performance, making it the study of neural network after following the new hotspot. In the paper, we introduce the theoretical basis of support vector machines, by analyzing the Xinzhou City, Shanxi Province, 220kV power transmission lines, humidity, temperature, rainfall, covered ice thickness and other historical data, set up follow-up SVM Xinzhou transmission line identification model ice thickness. For the SVM method, through the different kernel function, parameter analysis, to establish the optimal kernel function and related parameters, using SVM regression algorithm for network training, and ultimately determine the structure of the network, and use test data for testing. Through the error analysis to predict the results, and with BP and RBF neural network prediction results were compared, based on support vector machine Xinzhou transmission line models to predict follow-up ice thickness accuracy and speed is superior to neural network method.