Dynamic energy cost prediction for industrial and commercial buildings using advanced machine learning and multi-pricing models

Accurate electricity energy cost prediction is essential for demand-side management, enabling dynamic pricing strategies and reducing market volatility. This study aims to develop a robust predictive framework for forecasting energy costs in industrial and commercial buildings, addressing challenges such as volatility, non-linearity, and seasonal variations. The framework integrates advanced Machine Learning (ML) models, including Temporal Fusion Transformer (TFT), Convolutional Bidirectional LSTM (Conv-BiLSTM), and Convolutional Bidirectional Gated Recurrent Units (Conv-BiGRU), with imputation methods Variational Autoencoder (VAE), adaptive K-Nearest Neighbor (KNN), and Generative Adversarial Imputation Networks (GAIN) and Hyperparameter Optimization (HPO). These hybrid models are designed to capture complex spatial and temporal dependencies in energy data. Validation using a real-world dataset from South Korean commercial and industrial buildings demonstrates that the Conv-BiLSTM model outperforms other models, achieving the lowest Mean Squared Error (MSE), Mean Absolute Error (MAE), and Root Mean Squared Error (RMSE). Integrating VAE with HPO enhances predictive accuracy, addressing missing data and improving model robustness. Adaptive KNN also proves effective, particularly for industrial datasets, while GAIN shows moderate performance. Notably, the Time of use (ToU) pricing policy gives more accurate predictions than the uniform pricing policy. This study introduces a novel hybrid framework combining Conv-BiLSTM, VAE, and HPO, offering actionable insights for optimizing resource allocation and decision-making. The findings contribute significantly to sustainable energy management and informed decision-making in electricity markets. • Developed a method for imputing missing energy data. • Proposed hybrid deep learning models to capture spatial and temporal patterns. • Bayesian optimization applied for optimal parameter learning. • Exploratory data analysis for key trends identification.