Electricity Consumption Forecasting Scheme via Improved LSSVM with Maximum Correntropy Criterion.

Area of Science:

• Energy Economics
• Machine Learning
• Time Series Analysis

Background:

• China's electricity market reforms necessitate accurate electricity consumption forecasting (FoEC).
• Existing FoEC methods face challenges in accuracy and scientific evaluation, especially with small sample sizes.
• Key factors influencing electricity consumption include GDP and temperature.

Purpose of the Study:

• To propose a novel prediction scheme for electricity consumption (EC) forecasting.
• To enhance the accuracy and reliability of EC forecasting models.
• To scientifically evaluate the performance of the proposed forecasting scheme.

Main Methods:

• Utilizing a least-square support vector machine (LSSVM) model for prediction.
• Incorporating the maximum correntropy criterion (MCC) for LSSVM parameter optimization.
• Employing K-fold cross-validation and grid searching to improve model learning ability.

Main Results:

• The proposed LSSVM model with MCC demonstrates superior performance compared to traditional LSSVM.
• The scheme effectively forecasts electricity consumption using relevant influencing factors.
• Experimental results validate the efficacy of the novel prediction scheme on Shaanxi Province's EC data.

Conclusions:

• The novel LSSVM-MCC prediction scheme offers an effective approach for electricity consumption forecasting.
• This method provides a scientifically rigorous evaluation for EC prediction results.
• The findings contribute to more accurate energy management in deregulated electricity markets.