Wide-Area Composite Load Parameter Identification Based on Multi-Residual Deep Neural Network.

Area of Science:

• Electrical Engineering
• Power Systems Analysis
• Artificial Intelligence in Energy

Background:

• Accurate load modeling is crucial for power system stability, control, and protection.
• Wide-Area Measurement Systems (WAMS) enable real-time monitoring of load consumption patterns.
• Existing methods struggle with the complex, time-varying nature of composite load models.

Purpose of the Study:

• To develop a robust and accurate WAMS-based load modeling method using a deep learning framework.
• To effectively capture both spatial and temporal features of load behavior in large-scale power systems.
• To improve the accuracy and robustness of load modeling, especially under noisy conditions.

Main Methods:

• Constructed a composite load model combining impedance-current-power and induction motor (IM) characteristics.
• Developed a multi-residual deep learning structure integrating Residual Convolutional Neural Network (ResCNN) for spatial features and Gated Recurrent Unit (GRU) for temporal features.
• Implemented a parallel structure with a weighted fusion method for parameter estimation and an error-based loss function for improved training and robustness.

Main Results:

• The proposed method accurately models the complex, time-varying behavior of composite load models.
• Achieved high accuracy with errors below 0.055% even in noisy conditions on IEEE 68-bus and Iranian 95-bus systems.
• Demonstrated at least 50% improvement over several state-of-the-art load modeling techniques.

Conclusions:

• The WAMS-based deep learning load modeling approach is effective and robust for power system studies.
• The method provides a significant advancement in load modeling accuracy and reliability.
• The proposed technique offers superior performance compared to existing methods, enhancing power system operational security.