Traditional simultaneous localization and mapping ( SLAM) mostly performs under the assumption of an ideal static environment, which is not suitable for dynamic environments in the real world. Dynamic real-time object-aware SLAM ( DRO-SLAM) is proposed in this paper, which is a visual SLAM that can realize simultaneous localizing and mapping and tracking of moving objects indoor and outdoor at the same time. It can use target recognition, oriented fast and rotated brief (ORB) feature points, and optical flow assistance to track multi-target dynamic objects and remove them during dense point cloud reconstruction while estimating their pose. By verifying the algorithm effect on the public dataset and comparing it with other methods, it can be obtained that the proposed algorithm has certain guarantees in real-time and accuracy, it also provides more functions. DRO-SLAM can provide the solution to automatic navigation which can realize lightweight deployment, provide more vehicles, pedestrians and other environmental information for navigation.

Short-term load forecasting (STLF) plays a crucial role in the smart grid. However, it is challenging to capture the long-time dependence and the nonlinear relationship due to the comprehensive fluctuations of the electrical load. In this paper, an STLF model based on gated recurrent unit and multi-head attention (GRU-MA) is proposed to address the aforementioned problems. The proposed model accommodates the time series and nonlinear relationship of load data through gated recurrent unit (GRU) and exploits multi-head attention (MA) to learn the decisive features and long-term dependencies. Additionally, the proposed model is compared with the supportvector regression (SVR) model, the recurrent neural network and multi-head attention (RNN-MA) model, the long short-term memory and multi-head attention (LSTM-MA ) model, the GRU model, and the temporal convolutional network (TCN) model using the public dataset of the Global Energy Forecasting Competition 2014 (GEFCOM2014). The results demonstrate that the GRU-MA model has the best prediction accuracy.

The difficulty of bumblebee data collecting and the laborious nature of bumblebee data annotation sometimes result in a lack of training data, which impairs the effectiveness of deep learning based counting methods. Given that it is challenging to produce the detailed background information in the generated bumblebee images using current data augmentation methods, in this paper, a joint multi-scale convolutional neural network and multi-channel attention based generative adversarial network (MMGAN) is proposed. MMGAN generates the bumblebee image in accordance with the corresponding density map marking the bumblebee positions. Specifically, the multi-scale convolutional neural network ( CNN) module utilizes multiple convolution kernels to completely extract features of different scales from the input bumblebee image and density map. To generate various targets in the generated image, the multi-channel attention module builds numerous intermediate generation layers and attention maps. These targets are then stacked to produce a bumblebee image with a specific number of bumblebees. The proposed model obtains the greatest performance in bumblebee image generating tasks, and such generated bumblebee images considerably improve the efficiency of deep learning based counting methods in bumblebee counting applications.

The analysis of dissolved gas in oil can provide an important basis for transformer fault diagnosis. In order to improve the accuracy of transformer fault diagnosis, a method based on the relational teacher-student network (R-TSN) is proposed by analyzing the relationship between the dissolved gas in the oil and the fault type. R-TSN replaces the original hard labels with soft labels, and uses it to measure the similarity between different samples in the space, to a certain extent, it can obtain the hidden feature information in the samples, and clarify the classification boundary. Through the identification experiment, the effect of R-TSN diagnosis model is analyzed, and the influence of the compound fault of discharge and thermal on the diagnosis model is studied. This paper compares R-TSN with support vector machines (SVMs), decision trees and multilayer perceptron models in transformer fault diagnosis. Experimental results show that R-TSN has better performance than the above methods. After adding compound faults in the sample set, the accuracy rate can still reach 86.0%.

A wireless powered communication network (WPCN) assisted by intelligent reflecting surface (IRS) is proposed in this paper, which can transfer information by non-orthogonal multiple access ( NOMA) technology. In the system, in order to ensure that the hybrid access point (H-AP) can correctly decode user information via successive interference cancellation ( SIC) technology, the information transmit power of user needs to satisfy a certain threshold, so as to meet the corresponding SIC constraints. Therefore, when the number of users who transfer information simultaneously increases, the system performance will be greatly restricted. To minimize the influence of SIC constraints on system performance, users are firstly clustered, and then each cluster collects energy from H-AP and finally, users transfer information based on NOMA with the assistance of IRS. Specifically, this paper aims to maximize the sum throughput of the system by jointly optimizing the beamforming of IRS and resource allocation of the system. The semi-definite relaxation (SDR) algorithm is employed to alternately optimize the beamforming of IRS in each time slot, and the joint optimization problem about user's transmit power and time is transformed into two optimal time allocation sub-problems. The numerical results show that the proposed optimization scheme can effectively improve the sum throughput of the system. In addition, the results in the paper further reveals the positive impact of IRS on improving the sum throughput of the system.

As a more trackable performance metric for secrecy capacity, the secure degrees of freedom (SDoF) are widely studied for most multiuser networks in the high signal to noise ratio (SNR) region. However, the SDoF for these networks under rank-deficiency and arbitrary antenna configurations have not yet been determined. In this paper, the SDoF of two-user general multiple-input multiple-output ( MIMO ) interference channel with confidential messages (ICCM) under rank-deficiency are derived. For the two-user rank-deficient MIMO ICCM, the model is generalized to fully asymmetric settings, where the transmitters and receivers are equipped with arbitrary antennas. The outer bound of SDoF is the union of three outer bounds that are based on the Fano's inequality and the secrecy constraints, the secrecy penalty lemma and the role of a helper lemma, and the transmitters cooperation, respectively. The SDoF region is subdivided into five regions according to the number of transceivers antennas, and each region has an achievability scheme with designed null space transmission and alignment techniques. Numerical results indicate that the SDoF increase at first and then decrease as the rank of the channel matrix decreases. The SDoF improve by increasing the transmitting antenna or reducing receiving antenna, but the effect of the transmitting antenna is greater.

Aiming at the factory with high-complex and multi-terminal in the industrial Internet of things ( IIoT), a hierarchical edge networking collaboration ( HENC ) framework based on the cloud-edge collaboration and computing first networking (CFN) is proposed to improve the capability of task processing with fixed computing resources on the edge effectively. To optimize the delay and energy consumption in HENC, a multi-objective optimization (MOO) problem is formulated. Furthermore, to improve the efficiency and reliability of the system, a resource prediction model based on ridge regression (RR) is proposed to forecast the task size of the next time slot, and an emergency-aware ( EA) computing resource allocation algorithm is proposed to reallocate tasks in edge CFN. Based on the simulation result, the EA algorithm is superior to the greedy resource allocation in time delay, energy consumption, quality of service (QoS) especially with limited computing resources.

In view of the fact that current data delivery methods are not enough to meet the security requirements of today's distributed crowd sensing, and the data delivery methods are not flexible enough, this paper proposes a crowd sensing data interaction method based on tangle directed acyclic graph (DAG) network. In this method, users and platforms are regarded as nodes of the network in the process of performing crowd sensing tasks. First, the heaviest chain is generated through the main chain strategy to ensure the stability of the network. Then, the hidden Markov model ( HMM ) prediction model is used to improve the correlation of the perceived data to improve the performance. Then, the confidential transaction and commitment algorithm is used to ensure the reliability of the transaction, overcome the security risks faced by the trusted third party, and simplify the group intelligence aware transaction mode. Finally, through simulation experiments, the security and feasibility of the group intelligence aware data delivery method based on tangle DAG network are verified.