Electricity load forecasting using support vector regression with memetic algorithms.

Area of Science:

• Power Systems Engineering
• Computational Intelligence
• Machine Learning

Background:

• Electricity load forecasting is critical for power system operations and electricity market transactions.
• Support Vector Regression (SVR) is a popular but parameter-sensitive forecasting model.
• Optimizing SVR parameters is essential for achieving high forecasting accuracy.

Purpose of the Study:

• To propose a novel hybrid optimization algorithm, the Firefly Algorithm-Memetic Algorithm (FA-MA), for determining optimal SVR parameters.
• To enhance the performance of SVR in electricity load forecasting through improved parameter selection.
• To compare the proposed FA-MA-based SVR model against other evolutionary and traditional forecasting methods.

Main Methods:

• The study developed a hybrid Firefly Algorithm (FA) and Memetic Algorithm (MA) named FA-MA.
• FA is utilized for global exploration of the solution space.
• Pattern search is incorporated for local exploitation and individual learning within the FA-MA framework.
• The FA-MA algorithm optimizes the parameters of the Support Vector Regression (SVR) model for load forecasting.

Main Results:

• The FA-MA-based SVR model demonstrated superior forecasting accuracy compared to four other evolutionary algorithm-based SVR models.
• The proposed model outperformed three well-known forecasting models in accuracy.
• Experimental results confirmed the effectiveness of the FA-MA approach in enhancing SVR performance.

Conclusions:

• The proposed FA-MA optimization technique effectively determines SVR parameters for electricity load forecasting.
• The FA-MA-based SVR model offers a significant improvement in forecasting accuracy over existing methods.
• This hybrid approach provides a robust solution for accurate electricity load prediction in power systems.