Research On Optimization Of Multi-dimensional Feature Noise Addition Algorithm In Smart Grid Privacy Protection

In the real-time monitoring system of the smart grid,with the widespread use of smart devices,smart meters and smart terminals,fine-grained measurement makes the problem of user privacy leakage more serious.In particular,the user's behavioral pattern can be inferred from the real-time aggregated dynamic sequence collected by the meter.Although fine-grained measurements cannot be directly accessed,detailed measurement data can be obtained by differential attack.In order to effectively protect user privacy,a differential privacy protection method that adds noise to the original query is generally adopted.The added noise has a certain degree of uncertainty,but compared with other high computational overhead protection methods,the differential privacy factor Its easy operation and low overhead can achieve flexible trade-off between data utility and privacy protection.How to achieve high data privacy protection and high data performance will be the research goal of this paper.The classic baseline Laplacian noise algorithm(BLN)directly adds Laplacian noise based on the global sensitivity of the data set.Higher global sensitivity leads to an increase in noise variance and a reduction in data utility.Uniform Laplacian noise algorithm(ULN)has an equal probability of noise distribution,that is,the noise amplitude added by each dimension of the data is consistent,but its efficiency is still not optimal.After in-depth study of the above two kinds of noise-adding algorithms,this paper proposes an optimized multidimensional feature plus noise algorithm(MFPN,Multidimensional feature plus noise algorithm).The algorithm is mainly divided into two steps.The first step is to solve the global sensitivity.The maximum eigenvalue of the data set is taken as the objective function.The objective function is solved by the combination of the average consensus and the iterative power method.The global sensitivity is calculated according to the maximum eigenvalue of each data set.The second step is to solve the total noise of each dimension independent noise index weighting.The global sensitivity,threshold value and privacy protection budget are taken as input parameters,the original data set is decomposed into multi-dimensional data,and the privacy sensitivity of each dimension is calculated according to the input parameters.And use the Laplace noise distribution to calculate the independent noise values to be added in each dimension,and the total random noise values are weighted by the independent noise values,through differential privacy analysis,performanceanalysis,optimal boundary selection,and noise mechanism.The six parts of discrete analysis,error analysis and complexity analysis optimize the algorithm.The Python is used to compare the proposed algorithm MFPN with the baseline Laplacian noise extraction algorithm BLN and the uniform Laplacian noise addition algorithm ULN.The simulation mainly compares the privacy protection strength and performance of the algorithm.The simulation results show that compared with BLN algorithm and ULN algorithm,MFPN algorithm has higher privacy protection intensity and higher data efficiency,which effectively realizes data integrity and strong security.