Automatic Violence Detection and Classification (AVDC) with deep learning has garnered significant attention in computer vision research. This paper presents a novel approach to combining a custom Deep Convolutional Neural Network (DCNN) with a Gated Recurrent Unit (GRU) in developing a new AVDC model called BrutNet. Specifically, we develop a time-distributed DCNN (TD-DCNN) to generate a compact 2D representation with 512 spatial features per frame from a set of equally-spaced frames of dimension 160×90 in short video segments. Further to leverage the temporal information, a GRU layer is utilised, generating a condensed 1D vector that enables binary classification of violent or non-violent content through multiple dense layers. Overfitting is addressed by incorporating dropout layers with a rate of 0.5, while the hidden and output layers employ rectified linear unit (ReLU) and sigmoid activations, respectively. The model is trained on the NVIDIA Tesla K80 GPU through Google Colab, demonstrating superior performance compared to existing models across various video datasets, including hockey fights, movie fights, AVD, and RWF-2000. Notably, our model stands out by requiring only 3.416 million parameters and achieving impressive test accuracies of 97.62%, 100%, 97.22%, and 86.43% on the respective datasets. Thus, BrutNet exhibits the potential to emerge as a highly efficient and robust AVDC model in support of greater public safety, content moderation and censorship, computer-aided investigations, and law enforcement.

Secure and reliable electricity supply is a prerequisite for the development of smart cities, and the trustworthy and efficient transmission of electrical data is the foundation for the safe and stable operation of the power grid. This paper introduces a real-time data transmission blockchain technique based on parallel proof of work algorithm. The new block generation progress of proposed blockchain is divided into five subroutines: hash pointer computation, real-time data pudding, signature value iteration, interruption, block header assembly. The real-time data pudding and signature value iteration are parallel processed, which brings the effect of decreasing energy loss of blockchain system, and upgrades the speed of new block generation and the bandwidth of data storing on blockchain. Computer simulation shows the proposed strategy can be effectively applied in real-time electrical data transmission application, raising the data transmission reliability with no harm to real-time data transfer function. This strategy provides a solution to guarantee data transmission safety in the digital conversion of power grid.

Grease in the normal operation of the RV reducer has a role that can not be ignored, for the variable working conditions of the RV reducer, the performance of the lubricant changes directly affect its reliable operation. Therefore, the study of the rheological properties of the grease has become the focus of the study of RV reducer performance.In this paper, SK-1A grease is taken as the research object, and its rheological characteristics under wide temperature range working conditions (-20℃~40℃) are investigated through rheological experiments, to analyze the potential influence of SK-1A grease on the performance of RV reducer.In addition, to better study the rheological properties of grease under different working conditions, the Elman neural network model was used to predict the rheological properties of grease based on the rheological experiments of grease, and the results were compared with those of the BP neural network and the RBF neural network.The prediction accuracy of the Elman neural network model was assessed using Mean Absolute Error (MAE) and Mean Absolute Percentage Error (MAPE) in a cross-validation approach.The results show that the viscoelastic properties of SK-1A grease generally show a decreasing trend as the temperature rises in a wide temperature range, and the degree of entanglement of soap fibers decreases more obviously, but its fluidity is more stable.The results of the three neural network prediction models show that the Elman neural network model used for the prediction of grease rheological properties shows high prediction accuracy, and the model can provide a valuable theoretical reference for the accurate prediction of the rheological properties of grease affected by complex multifactor.

In this paper, Short-term predicting of load and spinning reserve is first performed using a combination of ANFIS and meta-heuristic algorithms including Differential Evolution (DE), Genetic Algorithm (GA) and Particle Swarm Optimization (PSO). The ANFIS-PSO combination is selected as the best ANFIS combination in load and spinning reserve prediction with a lower error criterion than other methods. As a DL method, LSTM network can provide good accuracy for load and spinning reserve forecasting. In the optimal ANFIS-PSO method, the average error value is low, but the error variance is high, on the contrary, in the LSTM method, the average error value is high, and the error variance is low. Therefore, we use the combination of ANFIS-PSO and LSTM to reduce the average error and error variance to an acceptable level. The weighted average method is as follows: the accuracy of each Method is obtained in the training step, then the predicted value for each data in the test step is calculated in each Method, then they are multiplied, and after that added together, finally will be divided to the total accuracy of two methods. The results obtained from the weighted average Method show the success of the proposed Method.

TSP is one of the most famous problems in graph theory, as well as one of the typical NP-hard problems in combinatorial optimization. Its applications range from how to plan the most reasonable and efficient road traffic to how to better set up nodes in the Internet environment to facilitate information flow, among others. Reinforcement learning has been widely regarded as an effective tool for solving combinatorial optimization problems. This paper attempts to solve the TSP problem using different reinforcement learning algorithms and evaluated the performance of three RL algorithms (Q-learning, Sarsa, and Double Q-Learning) under different reward functions, ε-greedy decay strategies, and running times. The results show that the Double Q-Learning algorithm is the best algorithm, as it could produce results closest to the optimal solutions, and by analyzing the results, better reward strategies and epsilon-greedy decay strategies are obtained.

Today’s healthcare system relies on MRI (medical resonance imaging) for early diagnosis and treatment planning. For open MRI systems to achieve resolutions of about a hundred microns, a high voltage is required, as well as a specialized power supply. NP0 (Negative-Positive-Zero) ceramic is selected for the fabrication of adjustable capacitors. Specifically, it stands for which is a classification based on the temperature coefficient of capacitance (TCC) of the ceramic material used in the capacitor. NP0 capacitors have a TCC of 0 ±30 ppm/°C, which means that their capacitance value does not change significantly with temperature and frequency. They are known for their stability and low losses, making them ideal for applications that require high accuracy and reliability, such as timing circuits for RF applications. In this paper, MgTiO-CaTiO ceramic is used to make an adjustable capacitor with desired properties for MRI systems. To enhance the dielectric properties of MgTiO3 ceramics,CaTiO3 was added in varying concentrations. After pressing and sintering, the resulting samples were tested using a vector network analyzer in the frequency range of 10 MHz to 130 MHz.The adjustable capacitor fabricated using high co-fired NP0 ceramic may have been used for MRI applications such as tuning circuits and matching networks, where precise capacitance values and low loss are critical[1]. MRI systems with resonance frequencies of 128 MHz require trimmers with ceramic cores.

5G network slicing is a promising solution to prioritize time-critical protection communication in wireless networks. However, recent trends indicate that a 5G slice could encompass all smart grid applications lacking the necessary granularity. At the same time, while substation communication standards recommend prioritization of protection communication traffic to improve reliability, these recommendations are only for wired connections. Therefore, this paper investigates traffic shaping and uplink (UL) bitrate adaptation of video stream based on existing commercial solutions as methodologies for prioritizing the protection communication in a 5G slice. These methodologies are validated in an experimental setup combining controller-hardware-in-the-loop (CHIL) simulation with a quality of service (QoS) measurement system. The system under test consists of commercial 5G networks, commercial intelligent electronic devices (IEDs), and merging units to validate the methodologies on three smart grid applications: fault location, line differential, and intertrip protection. The results show improvement in protection communication when traffic shaping and UL bitrate adaptation are applied. Traffic shaping even improves prioritization with a wired connection.

Abstract: Path planning is a crucial component for ensuring the safety and efficiency of flight missions, especially for fighter aircraft. To enhance the combat effectiveness of fighter aircraft, it is important to consider how to avoid danger sources and terrain obstacles, reduce fuel consumption, and utilize the aircraft’s own performance to accomplish the mission objectives. In the modern battlefield environment, the shortest path is not the only criterion for planning, but also other factors such as the threat level to the aircraft, fuel consumption, mission completion time, and minimum turning radius. In this paper, we propose a multi-constraint path planning method for fighter aircraft that incorporates these factors into an improved particle swarm algorithm. We transform the constraints of three-dimensional terrain, threat source, fuel consumption, and mission time into an aggregated fitness function. We construct a limit curvature matrix to evaluate the feasibility of the generated path. We also introduce an adaptive adjustment strategy based on the activation function for the parameters in the particle swarm algorithm. The weights of each constraint are determined according to the actual demand. The experiment results show that our method can efficiently plan the optimal path that satisfies the requirements. Compared with other improved particle swarm algorithms, our method has higher optimal search efficiency and better convergence effect. We also provide optimal values for important parameters such as mission energy consumption, mission time, flight speed and others to support the overall mission planning. Our method has certain practical application value.

Compressed sensing (CS) techniques have enabled efficient acquisition and recovery of sparse high-dimensional data via succinct low-dimensional projections, which usually consist of an encoder and a decoder. Unlike conventional CS techniques with the encoding-decoding architecture, the uncertainty autoencoder (UAE) can sample from the learned input data distribution without an explicit likelihood function, hence avoids potential uninformative latent representations. However, existing works on UAE mainly focus on the encoders and maximize the lower bound of the mutual information between input and measurements, rather than the decoders, which brings the shortcoming that the two may not cope well. In this letter, we propose a novel training scheme for UAE that blurs the measurements to learn the encoder and decoder simultaneously. Experimental results show that the proposed method improves the reconstruction performances when applied to UAE.

As self-driving cars perform more tasks, new challenges arise. One of these challenging tasks is autonomous driving decision-making due to the uncertainty of the vehicle’s complex environment. This paper provides an overview of decision-making technology and trajectory control for autonomous vehicles. The main common goal in decision-making is to consider uncertainties, unpredictable situations, and driving tasks to propose a global and robust solution adapted to each situation. The main concern is safety. Decision-making falls into three categories. The first is the traditional approach, which often consists of building a rule system and deriving optimal operations. The advantages of such an approach are well known for being easy to understand and applicable to small problems. The second category of decision-making is based on a probabilistic process and, due to its efficiency, has several applications in this area. The third category is learning-based approaches. Once a decision has been made, manipulate the steering angle or accelerator/brake pedals to perform the appropriate action. Two approaches are existing to designing autonomous driving controllers. Either based on imitating human drivers that includes approaches based on the use of driver models such as AI, or the use of approach-based models

Based on the data of indexed papers from Inspec database, scientometrics and statistics-related analysis methods are used to study the overall development trend of international cooperation in advanced manufacturing, the characteristics and evolution of major countries in four dimensions of cooperation scale, cooperation intensity, cooperation network and cooperation network from 2011 to 2020, and this article focuses on the changing trend of China and the differences existing with other countries. The study shows that cooperation in advanced manufacturing has an upward trend, with China and the US having a clear lead in cooperation scale and intensity. China leads the most in the aspect of international cooperation in advanced manufacturing. And the U.S. is at the heart of a network of cooperation across the field.

The continued quest for finding a low-power and high-performance hardware algorithm for signed number multiplication led to designing a simple and novel radix-8 signed number multiplier with 3-bit grouping and partial product reduction performed using magnitudes of the multiplicand and the multiplier. The pre-computation stage constitutes magnitude calculation and non-trivial computations required to generate partial products. A new partial product reduction strategy is deployed in the design to improve the speed with low cost. 8 X 8, 16 X 16, 32 X 32, and 64 X 64 designs are presented for the proposed architectures. Performance results include area, power, delay, and power-delay-product of synthesized and post-layout designs using 32 nm CMOS technology with 1.05 V supply voltage.

The IOT management platform is used to handle and transmit data from many types of power system terminal devices. The current IOT management platform has a low data processing efficiency and a high mistake rate when it comes to finding anomalous data. Furthermore, the effective selection and optimum decision of the convolutional neural network’s structural parameters has a significant impact on prediction performance. Based on this, the paper proposes a decision algorithm for locating anomalous data in an IOT integrated management platform using a convolutional neural network (CNN) and a global optimization decision of key structural parameters of a convolutional neural network using an improved particle swarm optimization (APSO) algorithm. First, an index model is created to determine if the data retrieved from the IOT management platform is anomalous or not. Second, the structure of the convolutional neural network-based decision method for finding anomalous data is examined. Following that, an enhanced particle swarm optimization technique is developed to optimize the structural parameters of the convolutional neural network, and an APSO-CNN with improved performance for anomalous data localization is generated. Finally, the established algorithm’s correctness, feasibility, and efficacy were evaluated using the Adam optimizer. The results reveal that the established APSO-CNN-based decision algorithm for anomaly data localization offers considerable benefits in terms of accuracy and running time, with extremely interesting application potential.

The tire lateral force control is crucial to vehicle lateral stability. Vehicle side slip and out of control can be prevented effectively by observing accurately the lateral force. Thus, a novel hyperbolic tangent sliding mode observation algorithm (NTSMO) is proposed. The algorithm adopts the longitudinal tire force error as feedback considering vehicle parameter uncertainties and without a complex tire model. First, the on-line verification of the algorithm was carried out by dSPACE to using the experimental data of the real vehicle linear acceleration and deceleration conditions, and comparison of experimental output with different observation algorithms. Furtherly, the simulation under emergency obstacle avoidance conditions and the double-line shifting conditions were conducted to verify the accuracy of the algorithm respectively. Simulation results show that the percentage errors between the tire lateral forces from the proposed NTSMO and the actual data are less than 5.35%, and the prediction accuracy of the NTSMO by 38.78% is higher than that of the sliding mode observation(SMO), which indicates that the NTSMO is superior to the SMO.

In this study, an optimisation model is developed for two-stage energy management of a residential building to minimise energy cost under monthly power-based tariffs for peak demand and time variable electricity prices. The expected peak demand is determined in the first stage, and then the energy management system minimises the energy cost in the second stage. The optimisation problem of the second stage is solved in a rolling time window for the real-time operation of the flexible energy sources in the building. The optimal charging and discharging of the battery energy system, the charging of the electric vehicle battery, the operation of the heating system and the optimal start times of washing machines and dishwashers are determined close to real-time. The proposed approach allows the user to determine the expected peak during the month ahead and try to keep the peak demand in daily operation below that value using a close to real-time energy management system. The performance of the two-stage approach for demand-side management of a residential building has been validated by a realistic case study.

Increasing Internet of Things (IoT) deployments present a growing surface over which villainous actors can carry out attacks. This disturbing revelation is amplified by the fact that a majority of IoT devices use weak or no encryption at all. Specific Emitter Identification (SEI) is an approach intended to address this IoT security weakness. This work provides the first Deep Learning (DL) driven SEI approach that upsamples the signals after collection to improve performance while simultaneously reducing the hardware requirements of the IoT devices that collect them. DL-driven upsampling results in superior SEI performance versus two traditional upsampling approaches and a convolutional neural network only approach.

The assembly process of flareless pipe joints is very important for the sealing performance of hydraulic pipeline system. Based on the theory of contact mechanics, a theoretical model of the assembly process of pipe joints is established to simulate the extrusion molding process of flareless pipe joints. It is found that tightening torque is an important factor affecting the sealing performance of pipe joints. By comparing the changes of the contact stress between the sleeve and the pipe joint under different tightening tortures, combined with the mechanical transfer and deformation results of the contact surface, the results show that the fitting situation of the sleeve and the pipe is good when the expansion pressure is 180Mpa, and the sealing performance of the pipe joint is good when the tightening tortures are between 15N·m and 18N·m.

In order to locate the mobile robots in three-dimensional indoor environment, mostly global navigation satellite system-denied space, a monocular visual space positioning algorithm based on deep neural network is proposed. First, we employ the lightweight YOLOv5 algorithm for target detection, and the LibTorch deep learning framework is used for model deployment to improve the inference speed. Moreover, a multi-layer perceptron (MLP) neural network with four inputs and two outputs is constructed, which regress the coordinates of the robot in the field coordinate system to complete the target localization, and this method is compared with the mathematical model solving algorithm to reflect the accuracy and superiority of positioning algorithm based on deep neural network. The proposed positioning and tracking system has been successfully applied to ICRA robot competition, and results show that the positioning error estimated by our method is within 10cm whilst having good real-time performance.