Deep learning and multi-objective optimization for real-time occupancy-based energy control in smart buildings.

Area of Science:

• Building Science
• Artificial Intelligence
• Energy Management

Background:

• Indoor environmental conditions can predict room occupancy, impacting energy efficiency and occupant comfort.
• Traditional energy management systems lack adaptability and struggle with non-linear relationships.

Purpose of the Study:

• To investigate the predictive power of indoor environmental parameters (CO2, illumination, humidity, temperature) for room occupancy.
• To introduce a novel deep learning-augmented predictive energy modeling (DL-PEM) framework with multi-objective particle swarm optimization (MOPSO) for intelligent buildings.
• To optimize real-time energy management by minimizing energy consumption and CO2 levels while maximizing thermal comfort.

Main Methods:

• Utilized a deep feedforward neural network (DNN) within DL-PEM to model non-linear relationships between environmental variables and occupancy.
• Integrated MOPSO for Pareto-optimal management of trade-offs between energy consumption, CO2 reduction, and thermal comfort.
• Adjusted adaptive HVAC and lighting controls in real-time based on occupancy predictions.

Main Results:

• Achieved 99.8% accuracy in prediction and up to 85% optimization efficiency.
• Outperformed baseline models including KNN, DT, AO-ANN, and LSTM for prediction and control tasks.
• Demonstrated significant improvements over traditional models in empirical evaluations with real building data.

Conclusions:

• The DL-PEM-MOPSO framework offers a scalable, adaptable, and future-ready solution for smart building energy management.
• Enhanced decision-making transparency and improved occupancy data analysis.
• Successfully optimized thermal comfort and enabled accurate power demand forecasting, leading to better overall energy utilization.