Previsão de carga no periodo de demanda de ponta utilizando redes neurais artificiais

AUTOR(ES)
DATA DE PUBLICAÇÃO

1996

RESUMO

The ability to accurately predict the system load is vital to the efficient, economic, and secure operation and planning of a power system. This work investigates the use of artificial neural networks for short and very short-term load peak demand forecasting. Two forecasting algorithms are tested and evaluated based on their precision and computational load. The influence of weather conditions on load demand is investigated. We sought a robust and compact topology which considers annual load sazonality, in order to preserve the forecast precision. The algorithm (MWS) uses data from the last 10 typical days to forecast the load peak demand profile with static and dynamic methods. The second algorithm (AAS) uses historical data from the previous year s load and weather database to forecast current year using static and dynamical methods. The MWS algorithm with dynamic forecasting yields the best 1000peak demand forecasting results for 10 minutes (very short-term forecasting), 1 and 24 hours ahead. The maximum temperature is the most significant weather variable. Scarce consistent evening weather data prevent a more complete evaluation of the influence of weather conditions on load forecasting. -The results show good performance of neural networks with around 2% mean percent absolute error for forecasts on the three horizons evaluated

ASSUNTO(S)

redes neurais (computação) sistemas de energia eletrica

ACESSO AO ARTIGO

Documentos Relacionados