基于STFT时频注意力机制的短期电力负荷预测OA

Accurate and efficient electrical load forecasting is a key for optimal power system dispatch.Aiming at the multi-periodic and non-stationary characteristics of load time series,this paper proposes a short-term load forecasting model based on short-time Fourier transform(STFT)and time-frequency attention mechanism.First,the load series is decomposed to separate seasonal and trend components,thereby mitigating the impact of non-stationarity.Then,the STFT time-frequency enhancement module is employed to mine the potential time-frequency characteristics of the seasonal component,followed by the STFT time-frequency attention module to extract key features along both temporal and frequency dimensions,enabling the modeling of local fluctuations and global periodic patterns for seasonal prediction.Meanwhile,reversible instance normalization and a multi-layer perceptron are adopted to forecast the trend component.Finally,the predicted results of each component are fused to obtain the final load forecasting.Experiments based on four public datasets from the United States and Australia demonstrate that the proposed model achieves determination coefficients of 0.893 and 0.539 for 8-hour and 48-hour ahead forecasting,respectively,fully verifying its forecasting performance and generalization ability.