Pavlyuk O. Short-range forecast processes consumption of electric energy on the basis of neural networks with not iterative training

The thesis for Ph. D. degree on the speciality 01.05.04 - the system analysis and the theory of optimum decisions is presented. National university “L’vivska polytechnica”, L’viv, 2005. This thesis is dedicated to the problem solution of prediction accuracy improvement of electrical energy consumption when prediction is done by artificial neural network in the case of incomplete data specification with partially contradictory information. The new effective method of incoming data repairing has been developed. It allows to repair 90-98% of lost information. This method may be used to improve prediction accuracy of ANN and statistical methods. The point neighbourhood method is used to improve accuracy of the ANN prediction. This method allows to double dimension of the incoming realizations space and it improves ANN prediction accuracy by 2-3%. The new “functional on the tabular functions set” ANN method is developed. It is based on the usage of incoming realizations space expanding by k - nearest neighbourhood method. The k - nearest neighbourhoods method allows to multiply incoming realizations space by N (NєR) and it allows to get enhancement of ANN prediction accuracy by 2-5%. Also this method allows to execute incoming data clusterization. The method for increasing entries count “functional on the tabular functions set” ANN is proposed. The incoming data space fazification is used in this method to increase realization’s space dimension. It allows to short learning sample and it allows system usage in real time. As result the accuracy of short and intermediate prediction is improved by 2-4%. The system analyses principles of IAS construction are adapted for tasks solution of EE consumption forecasting. These principles are based on application of neural network modeling and neural network prediction facilities. It allows accuracy improvement of short- and intermediate forecasting by 30-50% and 20-30% accordingly. Also it allows to execute prediction in real time. The IAS “Prognoz” software application is developed. And it is used in energy supporting company OSC “L’vivoblenergo”. This application provides quality improvement of dispatcher’s management efforts via abilities to get prediction of EE consumption in real time mode.