Developing a Mixed Neural Network Approach to Forecast the Residential Electricity Consumption Based on Sensor...

Area of Science:

• Smart Grid Technology
• Artificial Intelligence in Energy Management
• Renewable Energy Integration

Background:

• Accurate electricity consumption forecasting is crucial for smart homes, especially those integrating renewable energy sources.
• Existing methods often rely on historical meteorological data, which may not be available or cost-effective to acquire.
• The need for granular, appliance-level consumption predictions is increasing for efficient energy management.

Purpose of the Study:

• To develop an accurate short-term electricity consumption forecasting method for smart homes utilizing renewable energy.
• To overcome the limitations of missing historical meteorological data and associated costs.
• To refine forecasts to the individual appliance level using sensor data.

Main Methods:

• A hybrid artificial neural network (ANN) approach combining non-linear autoregressive with exogenous input (NARX) and function fitting neural networks (FITNETs).
• Utilized a comprehensive dataset of sensor-recorded electricity consumption data from individual appliances.
• Incorporated timestamp datasets as exogenous variables for the NARX model.

Main Results:

• Successfully developed and validated a mixed ANN model for precise electricity consumption forecasting.
• The method accurately predicts energy usage at the appliance level without historical weather data.
• The forecasting approach is adaptable for cloud-based service delivery.

Conclusions:

• The proposed hybrid ANN method offers a viable solution for accurate, appliance-level electricity consumption forecasting in smart homes.
• This approach effectively addresses the challenge of unavailable historical meteorological data and reduces reliance on external forecasting services.
• The developed system can be deployed as a cloud-based service, providing valuable energy management insights to operators and consumers.