Security And Privacy Issues Of Electricity Usage Data In Smart Grid

Smart Grid is the combination of power gird and information technologies,which applies the advanced computing,communication and control technologies in all parts of grid,including the generation,transmission,distribution and consumption parts.Thanks to the two-way communica-tions of electricity and information,the conventional grid's problem of information asymmetry is alleviated,which helps the demand side responds to the dynamics of the power flow and the grid system's operations.However,with the enhancing openness of smart grid,the electricity usage da-ta generated in demand side is faced with more severe security issues and privacy leakage threats.Therefore,the security and privacy issues of smart grid have drawn considerable attentions of the governments and researchers all over the world.The researches of interests mainly concentrate on designing various mechanisms so as to im-prove the security of system or protect the sensitive individual information from being stolen by all means.Nevertheless,the existed works have not attached enough importance at the electricity usage data generated at the consumption part of smart grid.Specifically,once the confidential-ly,integrity and availability of those electricity usage data is compromised,not only is harmed the interests of users in smart grid,but also the performance of the entire grid system is affect-ed.Moreover,owing to the high complexity of grid system and strong interconnection among the generation,transmission,distribution and consumption parts,it is of great significance while chal-lenging to investigate the impact on the smart grid system performance resulted from the security or privacy issues of electricity usage data.Therefore,motivated by the confidentially,integrity and availability of information,this thesis focuses on the inherent quantitative relation between the grid system performance and the security or privacy issues of electricity usage data,and studies how much the system performance is affected by security operations or privacy protection.The main works and contributions are summarized as follows:1.A brief review on the security and privacy issues if electricity usage in smart grid,as well as the related literatures are provided.2.Research on performance degradation of smart grid communication network caused by the jamming attack.We propose LearJam,a novel two-phase energy-efficient learning-based jamming attack strategy against smart grid communication network,where the attacker esti-mates the probability distribution of transmission period in the learning phase,and schedules its jamming attacks in the attacking phase based on this estimated distribution.Then we jointly optimize the learning duration and the attacking duration under the energy constraint in order to degrade the network throughput to the maximal degree.We further discuss how the state-of-the-art mechanisms can defend against LearJam by re-scheduling transmission pattern,which will aid the researchers to improve the security of smart grid communication network.3.Research on the tradeoff between grid generation cost and the strength of privacy protec-tion.We leverage the notion of Differential Privacy(DP)to measure the privacy-protection strength,under the framework of Optimal Power Flow(OPF).And we are the first quan-titatively investigating DP preserving OPF problem.Starting with re-modeling the noise-injected OPF problem,we rigorously prove OPF solution's sensitivity with respect to the uncertainty of demand.Also,aiming at OPF-based pricing mechanism,Locational Marginal Pricing(LMP),we present explicit expression of the respective privacy-protection's con-tribution on LMPs.Furthermore,by combining the grid topology and privacy-protection strength,we propose a novel billing system to fairly charge the extra payment to subsidize the privacy-insensitive customers.4.Research on privacy-preserving demand forecasting based on a fast,asynchronous distribut-ed framework.We design a novel framework for neural-network-based short-term demand forecasting in the consumption part.Concretely,each smart meter builds up a homogeneous neural-network model,while energy management system acts as the parameter server,gath-ering and exchanging the parameters at each local training.Besides,in order to guarantee the training time of modelling and the precision of demand forecasting,we incorporate a gradient-descent-based algorithm which combines the stochastic optimization and the exist-ed algorithm,Adadelta,as well as the asynchronous data parallelism mechanism,into that distributed framework.Lastly,based on the practical dataset of electricity usage,the outper-formance of our framework is illustrated.In the end,the thesis is concluded and some future research works are discussed.