Highly automated assembly lines enable significant productivity gains in the manufacturing industry, particularly in mass production condition. Nonetheless, challenges persist in job scheduling for make-to-job and mass customization, necessitating further investigation to improve efficiency, reduce tardiness, promote safety and reliability. In this contribution, an advantage actor-critic based reinforcement learning method is proposed to address scheduling problems of distributed flexible assembly lines in a real-time manner. To enhance the performance, a more condensed environment representation approach is proposed, which is designed to work with the masks made by priority dispatching rules to generate fixed and advantageous action space. Moreover, a Monte-Carlo tree search based soft shielding component is developed to help address long-sequence dependent unsafe behaviors and monitor the risk of overdue scheduling. Finally, the proposed algorithm and its soft shielding component are validated in performance evaluation.
Sea clutter is the backscattered echo of radar irradiating the sea surface, and this signal data includes not only pure sea clutter, but also often accompanied by many interference signals and noise. Based on the effective sea clutter region in the radar echo image, certain characteristics of the sea clutter can be accurately calculated, which plays a crucial role in the subsequent study of sea clutter and radar performance. The traditional method of using manual detection of effective sea clutter region often requires a lot of time, and with the rapid development of artificial intelligence technology in recent years, it is necessary to establish an effective sea clutter region detection method based on deep learning in Range-Pulse (RP) images. In this paper, based on the sea clutter data collected by shore-based ultra high frequency (UHF) radar, an improved model (H-YOLOv4) based on YOLOv4 network architecture is proposed to solve this problem, which uses the H-swish activation function to improve the performance of the model for effective sea clutter region extraction. In our test set of sea clutter RP images, it is experimentally demonstrated that the AP50 of the model reaches 89.75% with a score threshold of 0.25, which over state-of-the-art method. Compared with the traditional manual detection methods, the model can quickly and accurately extract the effective sea clutter region and maintain a good detection speed.
The microprocessor suitable for the physical network is studied, and a processor model more suitable for the Internet of Things is proposed. This solution adopts the structure of five-stage pipeline and tailors the MIPS instruction set architecture CPU. The designed data path is more concise, the power consumption is lower, and it has a high degree of scalability, making it more suitable for the Internet of Things field. The scheme was tested on the EDA platform. The test took the temperature and humidity of the greenhouse as the test object. The test results showed that the scheme met the design requirements.
With the rapid development of social economy and the acceleration of urbanization, energy consumption has increased rapidly, and power resources have become a scarce resource. In order to actively respond to the national green lighting strategy, this paper uses wireless sensor network and Internet of Things technology to design a smart street light management system, and incorporate urban street lights into the new urban Internet of Things system. The system collects the data of the surrounding environment of the street lamp in real time through the street lamp controller with multiple types of sensors installed in the street lamp pole, receives the data of the street lamp node through the LoRa network, and then uploads the data to the cloud monitoring platform through the NB-IoT technology. The city street lamp information is effectively collected and stored; at the same time, combined with WEBGIS technology, the current, voltage, power and other street lamp status data are visually displayed in the electronic map. For abnormal data, the system analyzes, locates and alarms, so as to realize the information interaction between the street light and the cloud platform, the precise control of the street light, and the intelligent monitoring and management. The experimental results show that the system has good stability, can effectively save energy and reduce consumption, and save the costs of management and operation and maintenance.
The external power receiving capability of the overall AC/DC receiving-end system is mainly restricted by the HVDC bipolar blocking failure. This paper presents a method to evaluate the maximum AC external receiving capability of the receiving-end power grid. Firstly, the power flow transfer ratio of the AC transmission line is analyzed after the HVDC bipolar blocking occurs in the receiving-end power grid under the given operation mode. Then, according to the thermal stability limit constraint of the AC transmission line, the initial power flow is calculated when the power flow just satisfies the power constraint after the power flow transfer. The calculated initial power flow is used as the optimal power flow distribution under the maximum power receiving capability of the receiving-end power grid. Next, the power flow transfer capability indexes of AC transmission lines and transformers and the maximum power receiving capability margin are defined. The objective function is established to analyze and evaluate the maximum external receiving capability of the receiving-end power grid under different operation modes. Finally, the effectiveness and adaptability of the proposed method are verified by taking a provincial receiving-end power grid as an example.
In view of the complex operating enviro nment of apple picking robot, a G-YOLO apple detecti on algorithm based on lightweight YOLOv4 was propos ed, which solved the problems of slow detection speed and large memory consumption of YOLOv4 model. The G-YOLO detection network model draws lessons fromt he idea of GhostNet. The traditional convolution operat ion is carried out in two steps, and the lightweight ope ration is used to enhance and extract features, so astor educe the calculation amount of model parameters, and the detection ability is improved by introducing attentio $n$ mechanism into the backbone network.The PANet net work is optimized by the BI-FPN of EfficientDet.Based on the SqueezeN et principle, the SPPN et structure was optimized by FireModule and SPPF, and the model par ameters were further reduced by deep separable convol ution. Finally, the size of the model was only33.36MB, and the memory occupancy was reduced by 86 % .The r esults showed that the mAP of G-YOLO was97.05 % , w hich was 0.5% higher than YOLOv4.G-yolo improves t he detection speed on GPU by about 24%, on CPU by 333%, and on Raspberry PI by 406.96%. Itis suitable f or small embedded devices with weak computing power such as Apple picking robots.
Production loss characterizes the difference between the expected horizon and the actual system performance. In this paper, an analytical method is proposed to model the production loss for assembly systems during the warming-period considering Bernoulli machines and finite buffers. First, the modeling of the assembly system is performed based on the transient analysis. Second, dynamic states of machines and buffers are formulated to characterize the production loss for assembly systems. At last, the impact of system layouts, machine reliabilities, and buffer capacities on production loss are numerically investigated in case studies. The results indicate that this method can evaluate the production loss for assembly systems during the warming-period, which exhibits the potential to design or manage more complex assembly systems.
Emotion affects human being's health to a great extent and it attracts lots of attention recently. Objective measurements are necessary for identifying various emotion states. As a powerful technique on high temporal resolution, Electroencephalogram (EEG) provides an effective way to quantify emotion impersonally, especially uncovering the dynamic characteristics. The implementation of time window length is one popular method for extracting dynamic interactions in the brain network. In this paper, the effect of time window length is revealed on dynamic brain network investigation. A quantitative pipeline is proposed based on brain network extraction and community mining in this paper. The EEG data of healthy subjects are recorded under sadness, happiness and neutral emotions that induced through video stimulation. Since beta (13-30Hz) band EEG signals are believed to play important role in cognitive processing, this paper focuses on the beta band EEG signals. Phase locking value (PLV) was adopted to calculate the functional interactions among brain areas. A nonoverlapping time window was taken in to produce dynamic connections. As a comparison, the dynamic brain networks were extracted by the length of 2s and 5S sliding time windows respectively. Later, the Louvain algorithm was used to detect the communities. As the member of community changes over time, the stationarity is quantified to measure the evolution of the varying communities in the brain network. The preliminary results illustrated that significant difference exists among communities from the beta band EEG networks that derived from 2s and 5s time windows separately. It is suggested that the choice of sliding time window will affect the stationarity quantification of the dynamic evolution of the brain network under various emotional conditions. It is necessary to pay attention to the choice of time length window when uncovering dynamic brain network changes in the future.
Miss Distance is an important parameter to characterize the missile strike performance. Using single-station radar measurement is a commonly used method for measuring the miss distance in the range. The error of miss distance calculated by directly using the original radar measurement data is large. This paper considers that the missile and target have a high speed and a short time at the intersection stage, and the two can be equivalent to a uniform linear motion. Under this premise, this paper proposes a method of least square linear fitting the original radar measurement data and calculating the miss distance based on the fitting data. The simulation results show that, when the method in this paper is used to solve the miss distance, the error of the miss distance calculated based on the measurement data of the typical phased array radar in the range can be controlled at the meter level, which basically satisfies the measurement accuracy requirements of the miss distance in the missile shooting test.
In some hospitals, it is often seen that because there are more patients and fewer doctors, the attending doctor often rushes between the patient's wards to query the patient's physiological data, or the nurse is responsible for recording the patient's physiological data and reporting it to the doctor. The work is very labor-intensive, and the data seen by the doctor is not real-time data, so the judgment of the patient's condition is not real-time and effective. To this end, this paper develops a medical big data statistical management system, which collects, counts, and stores patients' medical information in the system's management database for centralized management and facilitates query. For the storage of medical big data(BD), this paper uses a distributed cache framework to store medical data information, and implements distributed data processing through the C4.5 algorithm of the decision tree(DT) algorithm. The experimental results of large-scale(LC) medical data processing based on the C4.5 algorithm are also The data processing efficiency of the algorithm is illustrated, which provides great convenience for medical data management.
SGD (Stochastic gradient descent) is widely used in deep learning, however SGD cannot get linear convergence and is not effective in large amounts of data. This paper use SSAG to improve the efficiency. SSAG contains two optimization strategies, one is stratified sampling strategy and the other is historical gradient averaging strategy. It has the advantages of fast convergence of variance, flexible application to big data, and easy work in deep network. This paper studies the efficiency of SSAG gradient optimization algorithm based on RNN framework. The proposed RNN framework comprises a feature extraction layer, a stacked RNN layer, and a transcription layer. The experimental results confirm that the accuracy of SSAG is better than the SGD and the Momentum. Both stratified sampling and historical averaging strategies have the effect of improving task accuracy. Experimental results verified that SSAG has better effect in image classification task.
The efficiency of production is greatly determined by the assembly line's balance. It entails an even distribution of work among all workstations in order to ensure a smooth flow of work with minimal bottlenecks. The Kilbridge and Wester Column (KWC) is a heuristic approach that is often used in line balancing. The balancing of assembly line of an automotive industry that manufactures the air condition duct ASSY of a car is focused in this study. The Problem reported by management is an unequal allocation of job load across employees resulting in a high ideal time. The assembly line is optimized by using the Kilbridge and Wester Column approach. To test and verify the results, the present and proposed models are simulated using arena software.
In order to overcome the shortcomings of traditional networks, Network Function Virtualization (NFV) is widely used. For meeting the diverse service needs of users, the deployment of service function chain based on NFV has become a current research focus. In the actual scenario that the network service may be interrupted due to the failure of the physical server, this paper proposes a noval reliable deployment method of service function chain based on multi-agent reinforcement learning. Firstly, an integer linear programming (ILP) model is established for the reliable deployment problem of service function chain. Secondly, an efficient MADDPG algorithm is designed to solve it. The experimental results show that the proposed algorithm reduces the cost of physical node computing resources and link bandwidth resources, and improves the acceptance rate of network service requests under satisfying reliability constraints with compared algorithm.
In view of the serious occlusion phenomenon of pedestrian fall detection, the difficulty of extracting small target details, and the slow detection speed, this paper proposes a high-precision lightweight detection network AC-YOLOv5s. First, the convolution module in the backbone is replaced by ACBConv, and the C3 module is replaced by ACBC3 to improve the detailed feature extraction capability. Secondly, a small target detection layer is added to the feature fusion network (FPN) to improve the detection accuracy of small targets. Finally, use Alpha IoU loss replaces CloU loss to improve the loss and regression accuracy of the High IoU target. Finally, compared with the original YOLOv5s, the network in this paper improves the mAP by 2.33%, and the FPS reaches 21 during detection. The experimental results show that our network achieves better results than other networks.
Aiming at the problem of insufficient exposure of images obtained by photography and photography in realistic low-light, backlight and other scenarios, this paper proposes a low-light image enhancement deep learning network model that improved the Learning-to-See-In-the-Dark (LSID) algorithm. This method implements the amplification factor estimation weight network by design. The defects of the previous artificially designed parameter amplification factors were compared, and through the training and testing of experimental data, the experimental results of different methods were compared in terms of Peak Signal-to-Noise Ratio (PSNR) and Structural SIMilarity (SSIM) parameter indicators, and image visualization was compared at the same time. The experimental results show that the proposed method in this paper can improve the PSNR and SSIM indicators by 0.81 and 0.025, respectively, and it can obtain better image dark light enhancement effect.
Feature extraction of ship radiated noise (SRN) plays an important role for target detection and recognition. Aiming at extracting prominent and reliable features of SRN, auditory based multi-scale amplitude-aware permutation entropy (AMAAPE) is proposed in this paper. The proposed method consists of two parts that are signal decomposition and multi-scale entropy quantification. Firstly, a set of filters that can effectively simulate the mask effect and frequency response of human ears is designed, through which the SRN is decomposed into a series of band-limited signals. Then, multi-scale amplitude-aware permutation entropy (MAAPE) is employed to measure the complexity of each band-limited signal. Experimental results show that the proposed scheme achieves higher classification accuracy compared with multi-scale permutation entropy(MPE) and variational mode decomposition (VMD).
In recent years, multi-agent reinforcement learning has been applied in many fields, such as urban traffic control, autonomous UAV operations, etc. Although the Multi-Agent Deep Deterministic Policy Gradient (MADDPG) algorithm has been used in various simulation environments as a classic reinforcement algorithm, its training efficiency is low and the convergence speed is slow due to its original experience playback mechanism and network structure. The random experience replay mechanism adopted by the algorithm breaks the time series correlation between data samples. However, the experience replay mechanism does not take advantage of important samples. Therefore, the paper proposes a Multi-Agent Deep Deterministic Policy Gradient method based on classification experience replay, which modifies the traditional random experience replay into classification experience replay. Classified storage can make full use of important samples. At the same time, the Critic network and the Actor network are updated asynchronously, and the learned better Critic network is used to guide the Actor network update. Finally, to verify the effectiveness of the proposed algorithm, the improved algorithm is compared with the traditional MADDPG method in a simulation environment.
As one of the important equipment to ensure the normal operation of substation, the performance of grounding grid has been highly valued. With the in-depth application of the Internet of things technology in the power field, it is possible to monitor the status of the grounding grid online. The corrosion detection system for grounding grid based on the non-contact measurement technology is proposed in this paper. Through the application of the internet of things technology, the accuracy, flexibility and application scope of grounding grid detection are further improved while saving costs. The detection results in the actual substation verify the effectiveness of the proposed system.
Aiming at the problem of insufficient ability to collect and analyze exceptions in the whole process of application performance monitoring methods in cloud platforms, a cloud platform-based method is proposed. Application Anomaly Detection and Bottleneck Identification System for Service Components, which provides customizable metrics for applications on multi-tier cloud platforms Value monitoring and analysis capabilities. The system first collects cloud platform service call data at the front-end application service layer and associates it with abnormal events;A customized anomaly detection method is configured to achieve the optimal detection effect; finally, performance anomalies caused by non-workload changes are identified and bottlenecks are identified. The experimental results show that the monitoring system can quickly and accurately detect different types of abnormal events and identify performance bottlenecks, which can meet the performance requirements of applications under the cloud platform. Ability to monitor demand.
The intelligent question and answer platform can improve customer satisfaction, strengthen the self-service function of customers, and obtain business information conveniently and quickly. It is an effective means to reduce the company's overall operating costs and an important measure to promote the capacity building of remote service channels. The purpose of this paper is to study the optimization algorithm of the power marketing AI response system based on the era of intelligent technology. Combined with knowledge graph visualization and other technical means, the basic knowledge graph of the power industry is constructed, and an intelligent dialogue system is designed in combination with the constructed knowledge graph to help the power industry's customer service capabilities to improve, knowledge storage, knowledge management and other work to be carried out well. From the perspective of model evaluation indicators, this paper chooses the accuracy of the model as an important indicator to measure the efficiency of the model. In the relational classification model of information extraction, the comprehensive indicators of precision, recall, and F1 value of the BERT-BiLSTM-CRF knowledge extraction model are better than word embedding + BiLSTM + CRF.
