Short-Term Load Forecasting Based On Artificial Neural Networks

Electric Power Systems Research 63 (2002) 185 Á/196 www.elsevier.com/locate/epsr

Short-term load forecasting based on artificial neural networks parallel implementation
K. Kalaitzakis *, G.S. Stavrakakis, E.M. Anagnostakis
Department of Electronics and Computer Engineering, Technical University of Crete, GR-73100, Chania, Greece Received 5 February 2002; received in revised form 23 April2002; accepted 6 May 2002

Abstract This paper presents the development and application of advanced neural networks to face successfully the problem of the shortterm electric load forecasting. Several approaches including Gaussian encoding backpropagation (BP), window random activation, radial basis function networks, real-time recurrent neural networks and their innovative variations are proposed,compared and discussed in this paper. The performance of each presented structure is evaluated by means of an extensive simulation study, using actual hourly load data from the power system of the island of Crete, in Greece. The forecasting error statistical results, corresponding to the minimum and maximum load time-series, indicate that the load forecasting models proposed here providesignificantly more accurate forecasts, compared to conventional autoregressive and BP forecasting models. Finally, a parallel processing approach for 24 h ahead forecasting is proposed and applied. According to this procedure, the requested load for each specific hour is forecasted, not only using the load time-series for this specific hour from the previous days, but also using the forecasted load data ofthe closer previous time steps for the same day. Thus, acceptable accuracy load predictions are obtained without the need of weather data that increase the system complexity, storage requirement and cost. # 2002 Elsevier Science B.V. All rights reserved.
Keywords: Short-term load forecasting; Moving window regression training; Gaussian encoding neural networks; Radial basis networks; Real timerecurrent neural networks

1. Introduction Load forecasting is a very crucial issue for the operational planning of electric power systems, especially for the isolated ones. Short-term load forecasting (STLF) aims at predicting electric loads for a period of minutes, hours, days, or weeks. STLF plays an important role in the real-time control and the security functions of an energy managementsystem. STLF applied to the system security assessment problem, especially in the case of increased renewable energy sources (RES) penetration in isolated power grids, can provide, in advance, valuable information on the detection of vulnerable situations. Long- and the mediumterm forecasts are used to determine the capacity of generation, transmission, or distribution system additions, along withthe type of facilities required in

* Corresponding author

transmission expansion planning, annual hydro and thermal maintenance scheduling, etc. Short-term forecasts are needed not only for power system control and dispatching, but also as inputs to load-flow study or contingency analysis. A short-term load forecast for a period of 1 Á/24 h ahead is important for the daily operations of apower utility. It is used for unit commitment, energy transfer scheduling and load dispatch. With the emergence of load management strategies, the short-term forecast plays a broader role in utility operations, especially in the case of isolated power grids with increased RES penetration, as in the case of Crete Island. Development of an accurate, fast and robust STLF methodology is crucial to both,the electric utility and its customers. Many techniques have been proposed during the last few decades regarding STLF [1]. Traditional techniques applied to STLF include Kalman filtering, the Box and Jenkins method, regression models, the autoregressive (AR) model and the spectral expansion techniques [1,2].

0378-7796/02/$ - see front matter # 2002 Elsevier Science B.V. All rights reserved....