Anomaly Detection in California Electricity Price Forecasting

Dr. Joseph Nyangon has co-authored a new research publication titled “Anomaly Detection in California Electricity Price Forecasting: Enhancing Accuracy and Reliability Using Principal Component Analysis,” now available on arXiv. The study makes a significant contribution to electricity market analytics by addressing persistent challenges in price forecasting driven by complex generation dynamics and heteroskedastic market behavior.

Accurate and reliable electricity price forecasts are essential for grid management, renewable energy integration, power system planning, and managing price volatility. Focusing on the California Independent System Operator (CAISO) market, the study analyzes hourly electricity prices and demand data from 2016 to 2021 to enhance day-ahead forecasting performance.

The research applies a two-stage anomaly detection framework. It begins with traditional interquartile range outlier analysis and advances to robust principal component analysis (RPCA) to more effectively identify and eliminate anomalies. This process improves data symmetry and reduces skewness, resulting in cleaner and more informative datasets.

Multiple linear regression models are then developed using both raw and PCA-transformed features. Results show that models built on transformed features consistently outperform those using raw data. Notably, the application of the SAS Sparse Matrix outlier removal method delivers the most substantial gains in forecasting accuracy.

The findings highlight PCA-based anomaly detection as a powerful tool for improving electricity price forecasting, with important implications for supporting renewable integration and operational decision-making in day-ahead electricity markets.