In order to recommend appropriate departments and doctors to patients, this paper uses the ERNIE pre-trained language learning model to classify corresponding departments for patient’s disease descriptions, then calculates the similarity between patient’s disease descriptions and doctors’ expertise, so as to show patients the ranking of doctors in corresponding departments for patients to choose consultation. As is shown in the experimental results on Haodf.com datasets, the ERNIE model outperforms other models in terms of evaluation indexes, and the accuracy rate is improved by 3.54%-11.51%.
With the national Belt and Road strategy in place, meteorological operational systems will face domestic and international dangers, and it is very urgent and important to protect the safety of meteorological equipment and resources. The existing meteorological operational system uses a role-based access control model for permission management, whose permissions do not change during one authorized access, and cannot sense the threat of user behavior to the system in real time, much less adjust user permissions in time, thus failing to cope with the security needs of the weather business system. In this paper, we propose a Meteorological Operational System-oriented Usage Control (MS-UCON) model to address this problem. The model refines the UCON model according to industry characteristics and business needs, and adds a periodic trust assessment module and a collection of trust value assessment factors to monitor user behavior in real time, perform periodic trust assessment on user behavior, sense user risky behavior in time, dynamically update user trust value attributes, and dynamically adjust permissions. Finally, the security of the MS-UCON model is proved based on state machine theory. The results show that the MS-UCON model can solve the problem that user privileges cannot be dynamically adjusted in meteorological operational systems, and also meet the security requirements of meteorological operational systems.
LLVM is an open-source compiler framework that provides flexible, modular and reusable compiler components for building compilers, static analysis tools, code optimizers and more. It is based on a modern modular architecture and supports multiple programming languages. LLVM also provides a rich toolchain and libraries for building high-performance compilers and optimizers. Besides, LLVM has the advantages of scalability and portability, so it is widely used in the field of compilers. This paper will introduce the architecture, features, research hotspots and applications of LLVM.
The study of the state of health (SOH) of lithium-ion batteries can avert major mishaps while also grading retired batteries for secondary use. To overcome the flaws of the traditional back propagation (BP) neural network technology, which has low precision and a poor fitting effect, this study proposes a model for forecasting the SOH of lithium-ion batteries based on a novel health factor and an upgraded BP algorithm. To begin, the NASA lithium battery database is evaluated offline, and the metrics with the highest correlation in the charging process are chosen as new health variables. On the other hand, simulation experiments are carried out on various groupings of battery data, and the BP algorithm is optimized using a genetic algorithm. The experimental results show that the errors in the classic BP models are all less than 2.4%, whereas the errors in the optimized BP neural network model are all less than 0.38%, and that the fitting effect has better generalization than the traditional BP model, which improves the estimation accuracy of SOH of lithium batteries, and that the proposed health factor can be directly collected by the battery management system (BMS), which has some practical implications.
In recent years, deep neural networks have achieved tremendous success in image classification in both academic and industrial settings. However, the high hardware requirements imposed by their intensive and complex computations pose a challenge for deployment on low-storage devices. To address this challenge, lightweight networks provide a viable solution. This paper provides a detailed review of recent lightweight image classification algorithms, which can be categorized into low-redundancy network model design and neural network compression algorithms. The former reduces network computations by replacing traditional convolution with efficient lightweight convolution, while the latter reduces redundancy in the network by employing methods such as network pruning, knowledge distillation, and parameter quantization. We summarize the experimental results of some classical models and algorithms on ImageNet2012 and CIFAR-10 datasets, and analyze the characteristics, advantages and disadvantages of these models respectively. Finally, future research directions for lightweight algorithms in the field of image classification are identified.
Spiking neural networks (SNN) have good computational energy efficiency and biological plausibility, and are promising for applications in low-power computing and brain-like computing. To achieve hardware acceleration of convolutional SNN (i.e., SCNN), we propose a configurable and extensible hardware architecture based on clock-driven design. In this architecture we use a strategy of output-channel-wise parallelism and hierarchical pipeline structure to speed up the computation. Aiming at retaining the performance when reducing the quantization bits, the adaptive channel-wise logarithmic quantization (ACLQ) method for SCNNs is proposed to ensure performance while significantly reducing the overhead on hardware resources such as on-chip memory. We configured and evaluated two sizes of LeNet networks model based on the proposed architecture in FPGA. Experiments show that our architecture achieved recognition accuracy of up to 99.26% on the MNIST dataset, while achieving a recognition speed of up to 1605 frames per second (FPS) at a 100MHz clock speed and consuming only 0.65mJ of energy to process each image. Using the proposed logarithmic quantization method, the weights’ bit width can be reduced to 3 bits with almost none of accuracy loss, which significantly reduces the on-chip resources and power consumption for weights storage.
It has been a hot topic in the takeaway industry that how to reduce total delivery costs while considering users’ satisfaction. Considering the traditional takeaway delivery pattern, due to the influence of partial merchant locations, some orders are assigned to different riders for delivery, so that the routes delivered by different riders have the same repetitive sections, resulting in an increase in delivery costs. In order to solve this problem, with considering the satisfaction of merchants and customers, this paper proposes the idea of rider handover and order combination for delivery, then a mathematical model with riding distance cost and time penalty cost as the optimization objectives is established. By improving the genetic algorithm to solve the arithmetic example, the feasibility and effectiveness of the model and algorithm are verified. The results show that delivery with rider handover and order combination can effectively reduce delivery costs, which provides a feasible reference for takeaway delivery management and prepares the foundation for further research.
Inspired by the idea of rectangular optimized coding scheme for binary images, this paper proposes three optimized coding methods aimed at gray image compression using the rectangular non-symmetry and anti-packing model (RNAMC). The adaptive arithmetic encoding and the adjacent position correlation encoding based on column bisections are used to improve representation efficiency of gray images. The experimental results show the effectiveness of the three optimized coding methods proposed in this paper. These methods can improve the compression ratio in the RNAMC algorithm and have a good application prospect in some commercial applications related to image processing, such as reducing storage space, increasing processing speed, and improving pattern matching efficiency.
A deep learning classifier is proposed for grading hypoxic-ischemic encephalopathy (HIE) in neonates. Rather than using handcrafted features, this architecture can be fed with raw EEG. Fully convolutional layers were adopted both in the feature extraction and classification blocks, which makes this architecture simpler, and deeper, but with fewer parameters. Here, two large (335 h and 338 h, respectively) multi-center neonatal continuous EEG datasets were used for training and testing. The model was trained based on weak labels and channel independence. A majority vote method was used for the post-processing of the classifier results (across time and channels) to increase the robustness of the prediction. A dimension reduction tool, UMAP, was used to visualize the model classification effect. The proposed system achieved an accuracy of 86.09% (95% confidence interval: 82.41–89.78%), an MCC of 0.7691, and an AUC of 86.23% on the large unseen test set. Two convolutional neural network architectures which utilized time-frequency distribution features were selected as the baseline as they had been developed or tested on the same datasets. A relative improvement of 23.65% in test accuracy was obtained as compared with the best baseline. In addition, if only one channel was available, the test accuracy was only reduced by 2.63–5.91% compared with making decisions based on the eight channels.
Multiple-choice question answering(MCQA) is one of the most challenging tasks in machine reading comprehension. MCQA task requires selecting the most appropriate answer from several relevant options for a given question. In recent years, many works have concentrated on designing models from the perspective of using the information of the question and options at a large granularity level. However, few studies have explored how the model uses the information to find the correct answer at a fine granularity level or a multi-granularity level. This paper proposed a multi-granularity representation enhancement method to use information from different granularities. The method introduces large-grained candidate option information into the question to guide the selection of fine-grained critical information and facilitate the information interaction between the answer and the question which is in line with the human reasoning processes. Experimental results show that the method proposed in this paper can effectively improve the accuracy of MCQA tasks without introducing external knowledge.
With the specializing and clustering development of the manufacturing, outsourced manufacturing plays a significant role in enhancing production efficiency. Meanwhile, small and medium-sized enterprises (SMEs) usually suffer capital shortage, as an important supply chain financing tool, order financing is receiving increasing attention. This paper investigates a brand new bi-objective uniform parallel machine scheduling issue considering outsourced manufacturing and order financing, with the objectives to minimize makespan and maximize profit simultaneously. To solve the problem, a bi-objective mixed-integer linear programming model is constructed, and is transformed into a set of single objective problems with $\varepsilon$-constraint method. The effectiveness of the proposed model is verified by CPLEX with a practice-based case and a series of randomly generated instances.
By the end of 2021, the average penetration rate of gas(LNG:Liquefied Natural Gas) nationwide reached 85% in Korea. Being extensively utilized in various fields including households and industries, ensuring a consistent supply of gas is crucial. Additionally, detecting pressure anomalies is important as it allows for timely adjustments when required. This paper presents a model for anomaly detection in the secondary pressure of gas regulators located in the Gyeryong city Korea.In addition, weather data that may affect the gas pressure was collected, and the analysis was carried out considering temperature and wind speed. Given the interconnected nature of most gas regulators, this study models four adjacent gas regulators in Gyeryong city simultaneously to identify abnormal pressure values. The models used in this analysis are VAR, a traditional multivariate time series model, Long Short-Term Memory Auto Encoder (LSTM-AE), a machine learning technique for time series anomaly detection, and Multivariate Time Series Anomaly Detection via Graph Attention Network (MTAD-GAT), which directly incorporates variable correlations into the model.The study develops proper models that analyzes four gas regulators in Gyeryong City using three algorithms to identify abnormal pressure points. The performance evaluation of each model utilized an F1 score, and the MTAD-GAT model demonstrated the highest F1 score. Furthermore, the model not only identifies abnormal values at a specific time but also considers the value immediately preceding the anomaly. When implemented on-site, this model will be able to enhance the efficiency of anomaly detection.
The continuous development of modern mobile communication, radar and electronic countermeasure system has driven communication equipment towards miniaturization and multi-functionalization, so the communication system has higher requirements on the operating bandwidth characteristics, radiation characteristics, and volume size of antenna and array antenna. A 5G dual-polarized high gain small directional radiating Sub-6GHz antenna is proposed. The designed antenna consists of a coplanar waveguide feed, a dual element butterfly radiating patch, and a parasitic rectangular strip around the patch. The parasitic strip is used to increase the bandwidth of the antenna and also used to improve the directional characteristics of the antenna. It is verified by HFSS simulation that the antenna has a directional antenna with wide bandwidth and good directional radiation pattern, which is suitable for micro-deformation monitoring radar applications.
For 5G communication, the angle of arrival (AOA) positioning has low success rate and low accuracy caused by base station shortage, put forward a combination positioning solution for AOA positioning technology and redundant inertial navigation systems (INS). And adopt a fusion resettlement federal Kalman filtering, which ameliorates the problem of decreased accuracy of INS for a long-term positioning and solving accuracy due to low accuracy and random errors. Simulations show that redundant INS/AOA combined navigation effectively improves the problem of positioning loss caused by 5G signal interruption and the problem of filtering divergence caused by long-term solution of inertial navigation system. This combination of positioning accuracy and reliability improved over single 5G positioning or INS positioning greatly. The use of fusion reset Federal Kalman can effectively detect system faults and isolate them, improving filtering accuracy and system reliability.
For the workflow scheduling question with delivery time constraints in cloud computing environments, a time-aware scheduling-driven cloud computing scheduling algorithm is proposed. This algorithm considers the impact of the workflow task path length on the overall workflow delivery time and introduces a sub-deadline allocation strategy. Additionally, a time-aware scheduling strategy is presented to address the cost reduction challenge of workflow execution under the pricing policies of cloud services. The sub-deadline allocation strategy assigns different sub-deadlines to cloud users based on the length of the workflow task path, ensuring compliance with user delivery time constraints while increasing the flexibility in selecting appropriate resource servers for workflow tasks. The time-aware strategy schedules the pre-scheduled workflow tasks at suitable idle times, considering the current workflow task scheduling situation. Experimental results demonstrate that the proposed algorithm can satisfies the delivery time constraints for cloud users and achieves a significant reduction in the execution cost of workflows.
In order to clarify the extent to which private express delivery enterprises fulfill their social responsibilities, help the government supervise their social responsibilities and promote the high-quality development of Chinese express delivery enterprises, this paper obtains annual reports and corporate social responsibility reports issued by Chinese private express delivery enterprises over the years from relevant channels such as Juchao Information Network, Orient Fortune Network, and the company's official website, and applies the grounded theory analysis method. QSR Nvivo 12.0 software is used to carry out open coding, spindle coding and selective coding of the original data, and summarizes and extracts six core categories of social responsibility of private express enterprises, namely, employee responsibility, management responsibility, economic responsibility, customer responsibility, environmental responsibility and charity responsibility. And further construct the structure dimension and core content system of private express corporate social responsibility. This study theoretically clarifies the structural dimension of the social responsibility of private express delivery enterprises, which not only provides a reference for private express delivery enterprises to fulfill their social responsibilities, but also makes a contribution to the government's supervision of enterprises to fulfill their social responsibilities.
The training and reasoning of convolutional neural networks (CNN) require a large amount of computation, among which the most core operation is convolutional operation. In order to accelerate convolutional operation, many methods for accelerating convolution have emerged in recent years, including sliding window method, Winograd algorithm, etc. Although the sliding window method and Winograd algorithm can improve the computational efficiency of convolution to a certain extent, they also have some drawbacks. The sliding window method requires a large amount of data reuse and transmission, thus requiring high bandwidth and low latency memory and data paths, which is also the main challenge of its hardware implementation; The Winograd algorithm requires preprocessing of convolutional kernels and input data, which increases the computational and storage burden in the early stage. At the same time, for convolutional kernels of different sizes, matrix transformations need to be regenerated, which increases the difficulty of algorithm implementation. Based on this, this article adopts a pulsating array convolution calculation method based on fixed weights.
Low Density Parity Check Code (LDPC) can meet the decoding requirements of a variety of code lengths and rates in different communication scenarios, but the LDPC code check matrix is irregular, which has problems with difficult storage and reading of the check matrix and high coding complexity. For the decoding requirements of multi-code length, this paper proposes a Quasi Crystal Low Density Parity Check Codes (QCLDPC) code based on Fibonacci Lucas sequence, which can obtain the codewords with different code length and code rate by changing the number of rows and columns of the exponential matrix. In terms of decoding algorithms, the Layered Non-Maximum Suppression (LNMS) decoding algorithm, which combines the minimum sum decoding algorithm with the layered algorithm, is proposed to improve the convergence speed of decoding iteration and decoder throughput. Through simulation and comparative experiments, it is proved that this design can support a variety of bit rates, block lengths and sub matrix sizes, and meet the requirements of high throughput and low complexity decoder in modern communication systems.
The study of Li-ion battery charging system control strategy can improve Li-ion battery charging efficiency and avoid safety accidents. To address the shortcomings of the traditional particle swarm optimization (PSO) algorithm, which is slow to converge and easily falls into local optimal solutions, the particle swarm algorithm is improved to iteratively find the optimal proportion integration differentiation (PID) parameters of the controller. Initially, a mathematical model is developed for the charging system, followed by the design of an objective function for the improved particle swarm algorithm that focuses on controlling the charging current and voltage. Secondly, the particle swarm algorithm optimization using linearly updated inertia factor and learning factor is used to rectify the PID parameters and to build a system controller model for validation using simulation experiments. According to the findings, the control strategy analyzed in this study notable control performance and adaptability, which surpasses that of traditional PID and particle swarm optimized (PSO-PID) control strategies, and can accurately output the target charging current and voltage to achieve precise control. Therefore, the control strategy of lithium battery charging system based on improved particle swarm algorithm PID control has good application value and promotion value.
The classical aspect-based sentiment classification combined with Recurrent Neural Network (RNN) and attention mechanism is weak and cannot fully capture long-distance dependencies. These methods only focus on the relative position and ignore sentence order features during sentiment classification. This paper proposed a Gated Convolutional Neural Network denoted as GTRU gating model PosATT-GTRU with improved position attention to achieve aspect-based sentiment classification. First, combine the Laplacian kernel function to define the location information matrix to obtain location features and word order features. Then, filters of different sizes are then used to semantically encode the text position and content. Finally, PosATT-GTRU is used to control the information flow of semantic features, location features, and aspect words while capturing the long-distance dependence of the text. The experimental results on the SemEval2014 data set show that model accuracy has been significantly improved compared with previous models.
