Multi-Step Hourly Power Consumption Forecasting in a Healthcare Building with Recurrent Neural Networks and Empirical...

Area of Science:

• Artificial Intelligence
• Energy Systems
• Data Science

Background:

• Short-term electric energy consumption forecasting is crucial due to price volatility.
• Neural Networks are preferred AI tools for accurate energy predictions.
• Hospitals require reliable energy demand forecasts for operational efficiency.

Purpose of the Study:

• To develop and evaluate hybrid AI models for 24-hour ahead electric energy consumption forecasting in a hospital setting.
• To compare the performance of Long Short-Term Memory (LSTM) and Gated Recurrent Units (GRU) based hybrid models.
• To assess the impact of using univariate versus multivariate datasets on forecasting accuracy.

Main Methods:

• Hybrid models integrating Empirical Mode Decomposition (EMD) with LSTM and GRU neural networks were developed.
• Forecasting was performed using both univariate (consumption only) and multivariate (consumption, reactive consumption, temperature, humidity) datasets.
• Model performance was evaluated using Mean Absolute Percentage Error (MAPE) and Root Mean Square Error (RMSE).

Main Results:

• Hybrid models significantly outperformed simple neural network models.
• Multivariate datasets yielded better forecasting performance than univariate datasets.
• The best results (MAPE: 3.51%, RMSE: 55.06) were achieved using the LSTM-based hybrid model with the multivariate dataset.

Conclusions:

• Hybrid AI models incorporating EMD preprocessing enhance short-term electric energy consumption forecasting accuracy.
• Multivariate data analysis improves prediction reliability for hospital energy demand.
• LSTM networks demonstrate strong potential for precise energy consumption forecasting in critical facilities.