Hyperparameter Optimization of Bayesian Neural Network Using Bayesian Optimization and Intelligent Feature...

Area of Science:

• Electrical Engineering
• Artificial Intelligence
• Data Science

Background:

• Accurate short-term load forecasting (STLF) is crucial for power system operation and stability.
• Existing STLF models face challenges in feature selection, dimensionality reduction, and parameter optimization, impacting accuracy and convergence.

Purpose of the Study:

• To propose a novel hybrid framework, FE-BNN-BO, for enhanced STLF.
• To improve forecasting accuracy, stability, and convergence rate.
• To address feature redundancy and local optima issues in STLF.

Main Methods:

• Developed a hybrid feature selection using Random Forest (RaF) and Relief-F (ReF) with grey correlation analysis (GCA).
• Integrated Kernel Principal Component Analysis (KPCA) for feature extraction and dimensionality reduction.
• Employed Bayesian Optimization (BO) to fine-tune Bayesian Neural Network (BNN) parameters, avoiding local optima.

Main Results:

• The FE-BNN-BO model demonstrated significantly improved accuracy on PJM electricity market data.
• Achieved a faster convergence rate compared to benchmark models like LSTM, ANN-AFC, and ANN-MI.
• Successfully reduced the Mean Absolute Percent Error (MAPE), indicating superior forecasting performance.

Conclusions:

• The proposed FE-BNN-BO framework offers a robust and accurate solution for STLF.
• The hybrid approach effectively handles feature engineering and parameter optimization challenges.
• This model provides a valuable advancement for power grid management and energy market operations.