Detection of False Data Injection Attacks for VRB Energy Storage Systems Based on Long-& Short-term Memory and Generative Adversarial Networks
As the information technology improves by leapes and bounds,the DC microgrid energy storage system has become a deeply integrated information physical system,and the accurate SOC estimation is critical to the real-time monitoring and safe and stable operation of this system.Aiming at the problem of the abnormal data detection in the SOC estimation,we propose a long short-term memory and generative adversarial network(LSTM-GAN)detection that enables us to test the false data injection attacks in the SOC estimation of the vanadium redox flow battery(VRB)energy storage system.First,the VRB equivalent circuit model and the false data injection attack model are established respectively;Then,by training the cyclic network composed of the LSTM network and the GAN network,the LSTM neural network is embedded into the GAN framework as a generator and discriminator to analyze the battery timing data,and the abnormal loss is obtained through the discrimination loss error in the discrimination network and the reconstruction residual error in the generation network for comprehensive judgment;Finally,the accuracy and feasibility of the detection method are verified by the construction of the false data injection attacks with different strengths for experiments of the CEC-VRB-5kW battery.The results show that by comparing the results of the proposed method with those of the recurrent neural networks,the random forest,the Auto Encoder and the LTSM encoder,the proposed method has higher detection accuracy and is able to more effectively identify the FDIAs in the estimation of the SOC in the VRB energy storage system.
long short term memorygenerative adversarial networksenergy storage systemSOC estimationFDIAs
万方数据
维普
