Forecasting Short-Term Electricity Load with Combinations of Singular Spectrum Analysis

Area of Science:

• Power Systems Engineering
• Data Science
• Applied Mathematics

Background:

• Accurate electricity demand forecasting is crucial for economic control and secure power system operations.
• Electricity data exhibit nonlinearity and multi-seasonal features, posing challenges for traditional forecasting models.
• Existing decomposition methods often struggle with feature separability and border effects.

Purpose of the Study:

• To develop an effective decomposition-based forecasting approach for electricity load using singular spectrum analysis (SSA).
• To demonstrate the importance of separable feature extraction for enhancing individual model performance.
• To propose an SSA-based period decomposition method that addresses border effects.

Main Methods:

• Singular Spectrum Analysis (SSA) for decomposing electricity load data into distinct feature subseries.
• Individual forecasting models applied to each extracted feature series.
• SSA-based period decomposition to achieve separable decomposition and mitigate border effects.

Main Results:

• The proposed SSA-based decomposition method successfully extracts separable features from electricity load data.
• Individual models utilizing these separable features demonstrate improved ability to capture distinct data characteristics.
• Empirical studies confirm the effectiveness of the proposed approach in achieving accurate electricity demand forecasting.

Conclusions:

• The novel SSA-based period decomposition method offers a reliable and promising tool for electricity load forecasting.
• Separable feature extraction is key to improving the performance of decomposition-based forecasting models.
• The approach effectively handles data nonlinearity and multi-seasonal features, overcoming limitations of previous methods.