Research object - energy consumption processes. Research target is the synthesis of new neural network methods for forecasting electricity consumption processes that take into account the specifics of these processes and relationships between them. Research methods - the theory of artificial neural networks (ANN), which allowed the synthesis of new architectures of ANN, optimization theory and linear algebra, which allowed to improve methods of ANN training, the fuzzy logic theory, which allowed to handle the data specified in ordinal and nominal measurement scales, simulation, which confirmed the reliability of the theoretical results, mathematical statistics, which allowed to examine the results of experiments. Theoretical and practical results - new ANN architectures and training methods are developed that allow improvement of various aspects of the electric load forecasting problem solving and can be used for specific tasks both individually and collectively. Scientific novelty - a locally-recurrent neural network architecture for long-term electric load forecasting is proposed, distinguished by the presence of the first hidden layer, which implements the nonlinear autoregressive - moving average model of various orders, which enables automatic selection and tracking of the order of the process under consideration; neural network forecasting method for trend-seasonal electric load forecasting is proposed, which is characterized by parallel processing of harmonic components, which allows to predict process with a fixed, a priori known number of harmonic components; neural network forecasting method for polyharmonic electric load forecasting is proposed, which is characterized by sequential processing of harmonic components, which allows to predict process with a priori unknown and changing number of harmonic components; training method for the specialized neuro-fuzzy networks for short-term forecasting is improved, which is characterized by the presence of a regularizer, which allows to avoid "paralysis" of the network; training method for a counter-propagation neural network is improved, which features the strategy "the winner takes more" using bipolar neighborhood function, which improves the load forecasting accuracy in the nodes of a power system. Degree of implementation - the research results are used in Khartep, ltd., Kharkiv, which is confirmed by the act of 07.06.2010, and are used in the "load forecast" task in the Western power system of GP NEC "Ukrenergo", which is confirmed by the act of 30.06.2010. Scientific results, conclusions and recommendations contained in the thesis are used in the preparation of courses "Neural network methods of computational intelligence" and "Intelligent control systems and diagnostics", which are taught to students of the specialty "Intelligent Decision Support Systems" of Kharkiv National University of Radio Electronics, which is confirmed by the act of 03.06.2010, and in research projects of Kharkiv National University of Radio Electronics, which is confirmed by the act of 23.06.2010. The scope of use - in organizations dealing with similar problems of development of intelligent systems and methods for complex processes forecasting, in the areas of information technology, energy, transport, economy, and in the educational process in the preparation of specialists in the areas of intellectual information processing.
