Author :
Dalvand, Mohammad Moradi ; Azami, Seyed Bahram Zahir ; Tarimoradi, Hadi
Abstract :
Development planning of electrical power systems starts with load forecasting. Accurate load planning can be useful in appropriate expansion and development of power plants and transmission and distribution devices. In developing countries, where power demand increases more dynamically, load forecasting is even more important. In this paper, artificial neural network (ANN) is used to predict electric loads. In long-term load forecasting, economical factors are more influential than weather conditions. Weather factors were omitted in our study, to prevent noise. In this work, inputs of neural network are selected for Iran as a fast developing country and according to its particular conditions. A contribution factors is defined to specify the influence of the selected factors on the peak load. The contribution factors show a heavy dependence of the Iranian economy, and hence the peak load, on oil. Finally, 12 economic factors that have more contribution factors are selected as the inputs of the ANN: (1) gross domestic product (GDP), (2) gross domestic product without accounting for oil, (3) gross national product (GNP), (4) Iranian oil price, (5) value-added of manufacturing and mining group, (6) oil income, (7) population, (8) consumer price index (CPI), (9) gas consumption, (10) electricity, water and gas supply, (11) exchange rate, (12) gold price. A three-layer feed-forward, which is trained with error back-propagation algorithm, is designed to predict peak load of Iran. 19 neurons are selected for the hidden layer as a trade-off between overfitting and learning capability. Finally, power demand in 2004, 2005 and 2006 are used to test and to examine the accuracy of the prediction and the test error is about 1%. Monte Carlo and fuzzy membership functions are used to estimate economic inputs for future. In previous works, a peak load is forecasted as an single value. In this paper, Gaussian probability density function (PDF) is presented for the peak loads of the fu
Keywords :
Gaussian processes; Monte Carlo methods; feedforward neural nets; fuzzy neural nets; load forecasting; power engineering computing; power grids; power system economics; power system planning; Gaussian probability density function; Iranian power grid; Monte Carlo functions; PDF; artificial neural network; distribution devices; economic factors; electrical power system planning; error back-propagation algorithm; fuzzy membership functions; long-term load forecasting; power plants; three-layer feed-forward; transmission devices; Artificial neural networks; Fuzzy neural networks; Impedance; Inductors; Load forecasting; Magnetic variables control; Power grids; Reactive power; Sliding mode control; Static VAr compensators; ANN; Monte Carlo; contribution factors; economical factors; fuzzy membership function; long-term load forecasting;