Detection And Repair Of Grid False Data Injection Attacks Based On Machine Learning

Smart grid is a new type of power grid formed by the high integration of modern advanced sensing and measurement technology,communication technology,information technology,computer technology and control technology with the physical power grid based on the physical power grid.As smart grids continue to evolve,there are many potential risks,and False Data Injection Attacks(FDIAs)are currently the most common way to receive human attacks on smart grids.Through the analysis of the attack mechanism of false data injection,this paper uses the advantages of machine learning for data processing and analysis,and starts from three aspects: data sample construction,feature extraction and model establishment,and designs a machine learning-based model to perform a classification detection on the measurement data attacked by false data injection,and reconstructs the separated and screened data to complete the defense repair of nodes.Key studies include:1.The power system model based on DC state is studied,the power system state estimation module and the bad data detection mechanism based on residual estimation are introduced,and the shortcomings in the mechanism are analyzed.This shows that the reason why fake data detection attacks can bypass the detection mechanism is that they can not change the size of the residual.By introducing actual load data,simulation and experiments on IEEE14 nodes can verify the attack possibility and provide a basis for subsequent data simulation under the actual power system.2.An improved random forest classification model based on supervised learning algorithm is proposed to inject attack classification detection method into false data,and the model parameters are optimized by swarm intelligence optimization algorithm.Combined with the high-dimensional and complex nature of power grid measurement data,the data features are extracted by dimensionality reduction,and the detection performance is verified under the IEEE14 node system in combination with the simulation and experiments in Chapter 2.3.Design a combined defense repair model that first separates abnormal data and then repairs missing data.Considering the characteristics of small attack samples and small scale of power grids in practice,an anomalous data separation model based on the isolated forest algorithm in the unsupervised algorithm is adopted,and an attention mechanism is added to improve the accuracy of the model separation.After completing the separation and elimination of abnormal data in the measurement data,the neural network based on deep learning is used to make short-term predictions on the data,and make up for the bad data after separation and elimination,so as to improve the self-healing ability of the power grid after being attacked.Finally,the simulation is completed in the IEEE14 node system and the validity of the model is verified.