As a typical kind of the Internet of Things, smart grid has attracted a lot of attentions. The power energy management of smart grid is of great importance for energy distribution, system security and market economics. One of the most important issues is the accurate and stable prediction of wind speed for the optimal operation and management of wind power generations connected to smart grid. In this study, a novel two-layer nonlinear combination method termed as EEL-ELM is developed for short-term wind speed prediction problems such as ten-minute ahead and one-hour ahead. The first layer is based on extreme learning machine (ELM), Elman neural network (ENN) and long short term memory neural network (LSTM) to separately forecast wind speed by making use of their merits of calculation speed or strong ability in forecasting, and obtain three forecasting results. Then, we propose the second layer by making use of ELM-based nonlinear aggregated mechanism to alleviate the inherent weakness of single method and linear combination. Two real-world case studies, gathered from Inner Mongolia’s wind farm in China, are implemented to demonstrate the effectiveness of the proposed EEL-ELM method. By comparing with other eight wind speed prediction methods, the simulation results reveal that EEL-ELM can achieve better forecasting performance according to three evaluation metrics and three statistical tests.

With the rapid development of communication, control, and computer technology, traditional power systems have evolved into cyber-physical power system (CPPS). However, CPPS not only affords convenience but also introduces more cyber-attacks. Among many types of cyber-attacks, false data injection attacks (FDIAs) have drawn much attention in the CPPS security domain due to their stealthiness. Most FDIAs are based on the static model on a single snapshot and often ignore the reality of dynamically time-evolving CPPS. Accordingly, this article proposes a novel three-stage dynamic false data injection attack (DFDIA) model in CPPS by considering potential dynamic behaviors. To consider both attack location and attack amplitude, the designing DFDIA is formulated as two constrained single-objective optimization problems. Two versions of constrained differential evolution are presented as the solver to determine the attack location and optimize the attack vector for collaboratively altering the meter measurements. Then, an interval state forecasting-based countermeasure is proposed to detect the established DFDIA. In this detector, the variation bounds of state values are determined via ensemble deep learning-based state forecasting method. Finally, extensive simulation results on several IEEE bus systems demonstrate the feasibility of DFDIA and the effectiveness of the defense mechanism.

Autonomous Vehicles ( $AV$ s) take advantage of Machine Learning (ML) for yielding improved experiences of self-driving. However, large-scale collection of $AV$ s’ data for training will inevitably result in a privacy leakage problem. Federated Learning (FL) is proposed to solve privacy leakage problems, but it is exposed to security threats such as model inversion, membership inference. Therefore, the vulnerability of the FL should be brought to the forefront when applying to $AV$ s. We propose a novel Byzantine-Fault-Tolerant (BFT) decentralized FL method with privacy-preservation for $AV$ s called BDFL. In this paper, a Peer-to-Peer (P2P) FL with BFT is built by extending the HydRand protocol. In order to protect their model, each $AV$ uses the Publicly Verifiable Secret Sharing(PVSS) scheme, which allows anyone to verify the correctness of encrypted shares. The evaluation results on the MNIST dataset have shown that introducing decentralized FL into $ AV$ area is feasible, and the proposed BDFL is superior to other BFT-based FL method. Furthermore, the experimental results on KITTI dataset indicate the practicality of BDFL on improving performances of multi-object recognition in $ AV$ areas. Finally, the proposed PVSS-based data privacy preservation scheme is also justified its characteristic of no side-effect on models’ parameters by the experiments on the MNIST and KITTI datasets.

Accurate and stable short-term traffic flow prediction is an indispensable part in current intelligent transportation systems. In this paper, a novel short-term traffic flow forecasting model termed as EnLSTM-WPEO is proposed based on ensemble learning of long short term memory neural network (LSTM), no negative constraint theory (NNCT) weight integration and population extremal optimization (PEO) algorithm. In the first stage, a cluster of LSTMs is constructed to separately forecast with different time lag, which is a significant element to affect the prediction performance. In the second stage, the PEO-based NNCT weight integration strategy is introduced to determine the weight coefficients of the ensemble model. The simulation results for six different datasets from highways of Seattle have testified the superiority of the proposed EnLSTM-WPEO to other six popular traffic flow forecasting models in terms of two commonly used performance indices and three statistical tests.