Use of Sensors and Analyzers Data for Load Forecasting: A Two Stage Approach.

Area of Science:

• Building Energy Management
• Smart Buildings
• Internet of Things (IoT)

Background:

• Increasing sensor deployment in buildings offers data for energy management.
• Load forecasting is crucial for optimizing building energy consumption.
• Effective sensor data selection is key to improving load forecasting accuracy.

Purpose of the Study:

• To propose a novel two-stage methodology for sensor data utilization in building load forecasting.
• To enhance the accuracy of load forecasting by optimizing sensor input selection and adaptive model retraining.
• To demonstrate the methodology's effectiveness using real-world data from an office building.

Main Methods:

• A two-stage approach for sensor data selection and load forecasting.
• Stage 1: Relevance assessment of sensor parameters for specific building contexts.
• Stage 2: Adaptive forecast updating based on forecast error, with selective retraining of Artificial Neural Network (ANN) or K-Nearest Neighbors (KNN) models.

Main Results:

• The proposed methodology effectively selects relevant sensor data for improved load forecasting.
• The adaptive updating mechanism reduces forecasting errors by retraining models with recent data when a threshold is met.
• Case study data from an office building over six months validated the methodology's advantages.

Conclusions:

• The developed methodology enhances building energy management through intelligent sensor data utilization.
• The two-stage approach offers a significant improvement in load forecasting accuracy for smart buildings.
• Agent-based sensor prototypes facilitate data collection and support the proposed energy management strategy.