A robust maximum correntropy forecasting model for time series with outliers.

Area of Science:

• Time series analysis
• Robust statistics
• Signal processing

Background:

• Traditional time series forecasting models struggle with outlier-corrupted data, reducing reliability and accuracy.
• Outliers significantly impact the performance of autoregressive models in real-world applications.
• Accurate power demand forecasting is crucial for grid management and stability.

Purpose of the Study:

• To propose a robust Maximum Correntropy Autoregressive (MCAR) forecasting model.
• To enhance the accuracy of time series forecasting in the presence of outliers.
• To evaluate the MCAR model's performance against deep learning methods using real power series data.

Main Methods:

• Developed a robust MCAR model incorporating maximum correntropy principles.
• Utilized Gaussian kernel width for correlation entropy to measure local data similarity.
• Employed semi-definite relaxation for parameter estimation in the MCAR model.
• Applied the model to actual power series data from Hanzhong City, China.

Main Results:

• The MCAR model demonstrated superior performance compared to deep learning methods.
• MCAR achieved a 1.63% lower average Mean Absolute Percentage Error (MAPE).
• Maximum correntropy was confirmed to be effective in mitigating outlier interference.

Conclusions:

• The proposed MCAR model offers a robust and accurate solution for time series forecasting with outliers.
• Maximum correntropy is a valuable tool for enhancing the reliability of autoregressive models.
• The MCAR model shows significant potential for application in power system forecasting.