Combination of Manifold Learning and Deep Learning Algorithms for Mid-Term Electrical Load Forecasting.

Area of Science:

• Electrical Engineering
• Data Science
• Artificial Intelligence

Background:

• Mid-term load forecasting (MTLF) is crucial for power system planning and operation.
• Existing MTLF methods struggle with complex load variations and high dimensionality, leading to low accuracy.
• The
• curse of dimensionality
• hinders accurate prediction due to numerous variables.

Purpose of the Study:

• To propose an advanced MTLF method that improves prediction accuracy and reduces computational load.
• To address the nonlinear characteristics and high dimensionality inherent in electrical load data.
• To leverage manifold learning for effective feature extraction in MTLF.

Main Methods:

• A novel MTLF approach combining manifold learning and Long Short-Term Memory (LSTM) neural networks.
• Manifold learning is employed for nonlinear dimensionality reduction and feature extraction from complex load data.
• LSTM networks are utilized to build forecasting models within the reduced low-dimensional space.

Main Results:

• The proposed method demonstrates superior prediction accuracy compared to established techniques on mid-term timescales.
• Successful application of the method for forecasting electrical load 24, 168, and 720 hours ahead.
• Validation on Independent System Operator (ISO) New England datasets confirms the method's effectiveness.

Conclusions:

• The manifold learning-based MTLF method significantly enhances prediction accuracy.
• The approach effectively reduces dimensionality and handles nonlinear load characteristics.
• This method offers a promising solution for more reliable power system planning and operation.