Simulating long-term energy consumption prediction in campus buildings through enhanced data augmentation and metaheuristic-optimized artificial intelligence

Abstract

Forecasting long-term energy consumption is essential to enhance resource utilization and promote sustainability in campus buildings. This study employs a comprehensive approach, integrating artificial intelligence models, data augmentation techniques, and metaheuristic optimization algorithms to predict energy consumption one year in advance. The predictive models undergo training and validation using historical energy consumption data and weather patterns. However, predicting energy consumption a year ahead presents challenges due to instability and limited training data. To tackle this, the dataset is enriched using diverse data augmentation techniques. A metaheuristic algorithm, Jellyfish Search Optimizer (JSO), is also applied to refine the best predictive models and augmentation techniques. Results demonstrate a notable enhancement in prediction accuracy by integrating data augmentation techniques, convolutional neural network models, and metaheuristic optimization algorithms. The top-performing model, JSO-MobileNet with random noise injection, achieves a Mean Absolute Percentage Error (MAPE) value of 11.95% in forecasting energy consumption for campus buildings one year ahead. This study contributes to affirming the effectiveness of the metaheuristic optimization algorithm in fine-tuning artificial intelligence models integrated with data augmentation techniques, offering a practical approach to forecasting long-term energy consumption in campus buildings. The research outlines a systematic process for identifying the optimal predictive model, the most suitable data augmentation method, and the best set of hyperparameters.