Long-term electricity demand forecasting of a university campus based on advanced deep learning artificial neural network algorithms

• Advanced hybrid approach combining CNN with BiLSTM and BiGRU for accurate long-term electricity demand forecasting on university campuses. • Leveraging real electricity consumption and weather conditions enables more accurate long-term demand forecasting. • Optimizing energy use on university campuses, promoting sustainability through data-driven decision-making. The increasing global energy demand needs effective energy management strategies, especially in large environments with diverse infrastructure like university campuses that consume significant energy. This study focuses on electricity demand forecasting for university campuses based on various advanced Artificial Neural Network (ANN) models, aiming to find the best model for long-term demand forecasting. The research examines fourteen models, including Feedforward Neural Networks (FNNs), Recurrent Neural Networks (RNNs), Conventional Neural Networks (CNNs), and Hybrid Convolutional Neural Networks-Recurrent Neural Networks (CNN-RNN), using an hourly dataset collected over seven years (1/1/2017 – 12/31/2023) from the University of Missouri’s Combined Colling and Heating Power Plant. In addition to the campus energy consumption the dataset includes various weather conditions to determine the dynamic relationships between weather and energy usage. The results show that while all models performed well, the hybrid CNN-RNN models, particularly CNN-BiLSTM and CNN-BiGRU models, showed higher performance, achieving 0.98 and 0.97 accuracy during training and 0.92 and 0.93 during testing, with a validation loss error of 0.073 and 0.062, respectively. Subsequently, the models were used to predict energy demand for the entire year 2024. The findings aligned closely with historical demand trends, showing strong reliability in handling long-term data. Finally, during the deployment phase, we compared the models' predictions with actual energy consumption in 2024 gathered from the power plant. This comparison confirmed that the hybrid models performed exceptionally well under real-world conditions. The results show the ability of hybrid models to optimize energy management strategies and supporting sustainability efforts for university campuses.