Multi-Timescale Electrical Load Probabilistic Forecasting Method Based on Dual Attention Mechanism and Quantile Regression

Electrical load exhibits strong volatility and randomness in its fluctuations. Accurate electrical load forecasting can provide decision-making support for power system operation and scheduling. The electrical load is influenced by multiple factors. To deeply explore the autocorrelation and cross-correlation characteristics and capture the stochastic uncertainty of electrical load, a probabilistic electrical load forecasting method based on dual attention mechanism and quantile regression is proposed. Firstly, the standardization modeling of multi-time scale load forecasting is conducted to define the input-output structure of the model and construct a general framework for multi-time scale load forecasting. Then, the dual attention mechanism is utilized to extract the autocorrelation characteristics of electricity load in the temporal dimension, as well as the cross-correlation characteristics between electrical load and its influencing factors in the feature dimension. The probabilistic load forecasting is implemented based on quantile regression. Finally, the effectiveness of the proposed forecasting method is validated with electrical load data from some regions in Zhejiang province of China.