ResearchBib Share Your Research, Maximize Your Social Impacts
Sign for Notice Everyday Sign up >> Login

Application of neural networks for the calculation of technical losses of electric energy in air power lines 6-35 kV

Journal: Reporter of the Priazovskyi State Technical University. Section: Technical sciences (Vol.30, No. 2)

Publication Date:

Authors : ;

Page : 152-160

Keywords : neural networks; the loss of electricity; overhead power lines; model;

Source : Downloadexternal Find it from : Google Scholarexternal

Abstract

A model for the calculation of technical losses of electricity in the air lines with voltage of 6-35 kV based on neural networks with due regard to meteorological factors has been worked out; the main components of the model have been considered and researched; the best ones being selected, that is: a set of input variables, volume of excerpts (training, control and testing), architecture and net-work activation function, network learning algorithm was proposed. Simulation was conducted in OS STATISTICA Neural Networks. Input variables are: transmission line (TL) active load, transmission line rated voltage, transmission line cross section and length of wire, average air temperature, wind speed, rainfall availability; output variable ? that is technical losses in electric transmission line. To select the optimal input vector model the data selection methods were used: variables testing using trial and error method, variables stepped inclusion and exclusion algorithm. It has been proved that the most important variables are TL active load and average air temperature. all input variables under review should be included in the created artificial neural network (ANN). It was determined that the optimal volume for ANN training set given parameters made 250 observations, control and test excerpts volume were respectively 250 and 332 observations. It has been proved that the best type of architecture is multilayer perceptron ANN that being compared to radial basis functions and generalized regression network is characterized by minimal errors and complexity of the network. ANN of the following architecture: multilayer perceptron, 7 neurons in the input layer, 5 neurons in the hidden layer, 1 output neuron, logistics as activation function ? has been taken optimal

Last modified: 2015-11-26 22:54:04