The smart grid is now increasingly dependent on smart devices to operate, which leaves space for cyber attacks. Especially, the intentionally designed false data injection attack (FDIA) can successfully bypass the traditional measurement residual-based bad data detection scheme. Considering that the smart grid data naturally contain linear and nonlinear components, inspired by parallel ensemble learning, especially by the stacking method, this article presents an effective two-level learner-based FDIA detection scheme using the Kalman filter and recurrent neural network (KFRNN). The first level includes two base learners, in which the Kalman filter is used for state prediction to fit linear data, and the recurrent neural network is used to fit the nonlinear data feature. The second-level learner uses the fully connected layer and backpropagation (BP) module to adaptively combine the results of two base learners. Then, through fitting Weibull distribution of the sum of square errors (SSEs) between the observed measurements and the predicted measurements, the dynamic threshold is obtained to judge whether FDIA occurs or not. Comprehensive simulation results show that our scheme has better performance than other neural network-based and ensemble learning-based FDIA detection schemes.

We propose the PSDRNN and TriPSDRNN schemes that employ the explicit feature extraction before the Deep Recurrent Neural Network (DRNN) model. Specifically, considering that Power Spectral Density (PSD) feature can capture the frequency characteristics and meanwhile retain the successive time characteristics of data gathered from smartphone accelerometer, PSD feature vectors are respectively extracted from linear accelerations and tri-axle accelerations, and explicitly used as the input data to the following DRNN classification model. Thorough experiments based on a real dataset demonstrate that PSDRNN can achieve the comparable effectiveness as the xyz-DRNN (the best accurate DRNN based HAR scheme using only acceleration data), and the average recognition and training time were reduced by 56% and 80% respectively. Moreover, TriPSDRNN advantages over the xyz-DRNN in terms of recognition accuracy, and the running time is lower than xyz-DRNN. Besides, PSDRNN scheme can achieve superiority in the recognition of complex transition activities.

WiFi fingerprint-based indoor localization is regarded as one of the most promising techniques for Location Based Services (LBS). However, it faces serious problem of privacy disclosure of both clients’ location data and provider’s fingerprint database. To address this issue, this paper proposes a Differential Privacy (DP)-based privacy-preserving indoor localization scheme, called DP3, which is composed of four phases: AP (Access Point) Fuzzification and Location Retrieval in client side, and DP-based Finger Clustering and Finger Permutation in server side. Specifically, in AP Fuzzification, instead of providing the measured full finger (including AP sequence and the corresponding Received Signal Strength (RSS)), a To-Be-Localized (TBL) client only uploads the AP sequence to the server. Then, the localization server utilizes the DP-enabled clustering to build the fingerprints related to the AP sequence into k clusters, permutes these reference points in each cluster with exponential mechanism to mask the real positions of these fingerprints, and sends the modified dataset to TBL client. At client side, Location Retrieval phase estimates the location of the client. Theoretical and experimental results show that DP3 can simultaneously protect the location privacy of the TBL client and the data privacy of the localization server.