Deterioration of Electrical Load Forecasting Models in a Smart Grid Environment

Area of Science:

• Electrical Engineering
• Data Science
• Artificial Intelligence

Background:

• Smart Grids (S.G.) generate heterogeneous data streams in dynamic environments.
• Current machine learning methods are static and struggle with S.G. data variations and diverse generation modalities (D.G.M.).
• Existing models exhibit performance degradation (5-15% accuracy loss) when encountering parameter changes.

Purpose of the Study:

• To address the limitations of static machine learning models in smart grid environments.
• To develop an adaptive framework for more accurate electrical load forecasting.
• To enhance model adaptability to dynamic data, features, and generation modalities.

Main Methods:

• Evaluated the performance of ARIMA, Artificial Neural Network (ANN), and Long Short-Term Memory (LSTM) models.
• Observed model behavior using two open-source and one real-world dataset.
• Investigated model response to changes in input parameters.

Main Results:

• Static models (ARIMA, ANN, LSTM) failed to adapt to input parameter changes without manual intervention.
• Models showed significant performance degradation when input parameters varied.
• Accuracy deterioration ranged from 5% to 15% due to parameter shifts.

Conclusions:

• A novel adaptive framework is proposed to overcome the limitations of existing electrical load forecasting models.
• The framework aims to improve model accuracy and adapt to dynamic parametric variations in S.G. and D.G.M. environments.
• Adaptive models are crucial for reliable forecasting in evolving smart grid systems.