Analysis of False Data Injection Impact on AI based Solar Photovoltaic Power Generation Forecasting
This work addresses the reliability of solar power forecasting for grid integration, but it is incremental as it applies existing methods to a new security challenge.
The study evaluated machine learning models for solar PV power generation forecasting and tested their robustness against false data injection attacks, finding that Multi-layer Perceptron Regression performed best with lower error metrics like RMSE and MAE.
The use of solar photovoltaics (PV) energy provides additional resources to the electric power grid. The downside of this integration is that the solar power supply is unreliable and highly dependent on the weather condition. The predictability and stability of forecasting are critical for the full utilization of solar power. This study reviews and evaluates various machine learning-based models for solar PV power generation forecasting using a public dataset. Furthermore, The root mean squared error (RMSE), mean squared error (MSE), and mean average error (MAE) metrics are used to evaluate the results. Linear Regression, Gaussian Process Regression, K-Nearest Neighbor, Decision Trees, Gradient Boosting Regression Trees, Multi-layer Perceptron, and Support Vector Regression algorithms are assessed. Their responses against false data injection attacks are also investigated. The Multi-layer Perceptron Regression method shows robust prediction on both regular and noise injected datasets over other methods.