Short term peak load forecast using detrended partitioned data training ofa neuro-fuzzy regression machine

Load forecasting using neural networks can suffer from poor quality of data , non-stationary load patterns and poor forecasting accuracy. To address so me of these problems, a neuro-fuzzy based forecasting model trained with de trended data is proposed. Feature extraction methods to provide better data partitioning, capture important correlations, and detrend non-stationary d ata are developed. As a result, forecasting accuracy and robustness are enh anced.