Machine learning methods against false data injection in smart grid

Abstract

The false data injection in the power grid is a major risk for a good and safety functioning of the smart grid. The false data detection with conventional methods are incapable to detect some false measurements, to remedy this, we have opted to use machine learning which we used five classifiers to conceive an effective detection [k-nearest neighbour (KNN) algorithm, random trees, random forest decision trees, multilayer perceptron and support vector machine]. Our analysis is validated by experiments on a physical bus feeding system performed on PSS/in which we have developed a dataset for real measurement. Afterward we worked with MATLAB software to construct false measurements according to the Jacobean matrix of the state estimation. We tested the collected data with different classification algorithms, which gives good and satisfactory results.