Short-Term Load Forecasting Based on Data Decomposition and Dynamic Correlation

Abstract

The load in the power grid is often affected by many factors, and the coupling relationship among them changes dynamically with time. This study proposed a short-term load forecasting technique based on time pattern attention and a long short-term memory network considering time-dependent intrinsic cross-correlation (TDICC) for smart grids with massive amounts of data. Based on source-load dynamic correlation analysis, The proposed TDICC to track and correct the multi-timescale dynamic correlation of two signal overruns or lags, which realizes the transformation of correlation description from two-dimensional space to three-dimensional space and expands its ability to describe multi-timescale dynamic correlation. Finally, the actual load data are used for example analysis, and the results show that the proposed method can tap the dynamic correlation between multiple influencing factors and load and has higher prediction accuracy compared with other models, which provides a more accurate database for power system dispatching.

Keywords

• Time-dependent intrinsic cross-correlation
• empirical mode decomposition
• short-term load forecasting
• temporal pattern attention mechanism