Research on short-term power load forecasting based on VMD and GRU

Area of Science:

• Electrical Engineering
• Data Science
• Artificial Intelligence

Background:

• Traditional power load forecasting methods struggle with accuracy due to external factors like weather and seasonality.
• Extracting meaningful physical insights from power data is a significant challenge for existing prediction models.

Purpose of the Study:

• To develop a novel and accurate combined power load forecasting approach.
• To overcome the limitations of traditional methods by integrating advanced signal processing and deep learning techniques.

Main Methods:

• Variable Mode Decomposition (VMD) for extracting time-frequency features from power load data.
• Convolutional Neural Network (CNN) to process decomposed features.
• Attention mechanism to prioritize critical information from CNN outputs.
• Gated Recurrent Unit (GRU) for time series modeling and final forecasting.

Main Results:

• The proposed VCAG model demonstrated high accuracy and stability in power load forecasting across two public datasets.
• Experimental results confirmed the model's superiority over traditional forecasting techniques.
• The integration of VMD, CNN, attention, and GRU effectively addressed the limitations of existing methods.

Conclusions:

• The VCAG model offers a promising solution for accurate and stable power load forecasting.
• This approach has significant potential for widespread application in the energy sector.
• The study highlights the effectiveness of combining signal decomposition with deep learning for complex time series prediction.