Demand-side load forecasting in smart grids using machine learning techniques.

Area of Science:

• Electrical Engineering
• Data Science
• Artificial Intelligence

Background:

• Electrical load forecasting faces challenges due to dynamic environmental factors like weather.
• Traditional methods struggle with Big Data management from IoT devices and smart meters.
• Existing two-tier architectures often suffer from low precision and overfitting.

Purpose of the Study:

• To develop an advanced, robust electrical load forecasting system.
• To leverage Big Data and IoT for improved energy consumption prediction.
• To enhance forecasting accuracy by incorporating underutilized weather features.

Main Methods:

• A two-level forecasting approach using Daily Consumption Electrical Networks (DCEN) and Intra Load Forecasting Networks (ILFN).
• Implementation of a three-tier architecture: cloud, fog, and edge layers.
• Utilizing conventional neural networks to mitigate overfitting and employing Support Vector Regression (SVR).

Main Results:

• Support Vector Regression outperformed other methods in experimental evaluations.
• Achieved a Mean Absolute Percentage Error (MAPE) of 5.055.
• Obtained a Root Mean Square Error (RMSE) of 0.69 and R2 score of 0.86.

Conclusions:

• The proposed three-tier architecture and enhanced feature set significantly improve electrical load forecasting.
• The study validates the effectiveness of Support Vector Regression for this task.
• The findings demonstrate a more accurate and reliable approach to energy consumption prediction.