Abstract:Aiming at a domestic 12-bit 4-channel digital to analog convert(DAC),a high-precision DAC test and evaluation system based on the FPGA is designed in this paper.It mainly includes the FPGA implementation of digital signal generator,desk-top discrete instrument and evaluation board of the chip under test.The parameters that is difficult to complete by ATE system,such as peak pulse interference,small signal amplitude-frequency response,wide-band noise,crosstalk.The result demonstrates that the performance of the system is stable,reliable and has a strong adaptability,thus it can fully meet the demands of engineering ap-plications.

Abstract:Aiming at the random error and non-line-of-sight error problems of ultra-wideband(UWB)positioning technolo-gy,this paper proposes the wild dog optimization algorithm(DO)for the first time to apply to UWB technology,and improves the wild dog optimization algorithm,and proposes the UWB positioning technology based on the target tracking wild dog optimization al-gorithm.First,the position selection of the label is reconstructed by the average trilateral positioning algorithm.Then,the DO algo-rithm is used to match the label position locally.Finally,the tag position is locked by introducing target tracking mechanism.The simulation results show that the DO algorithm improves the accuracy by 67%and 31%respectively compared with the trilateral posi-tioning algorithm and the average trilateral positioning algorithm,and its positioning stability reaches 87%.Compared with the DO algorithm,the TTDO algorithm improves the positioning accuracy by 22%and the algorithm operation time by 34%.Therefore,the UWB positioning algorithm based on target tracking wild dog optimization can effectively improve the accuracy of tag positioning in UWB technology,and has higher stability and reliability.

Abstract:As a common problem of small sample identification,tunnel crack detection based on deep learning will cause low classification accuracy due to too few crack samples.Based on the CycleGAN model,a crack sample augmentation method GDCycle-GAN based on the global feature fusion double-layer lightweight model is proposed.The GFF(Global Featue Fusion)module is in-troduced into the generator of CycleGAN,which integrates the semantic information of the upper and lower layers of the crack image to obtain multi-scale information,and uses the attention mechanism to strengthen the selection of the crack image feature informa-tion to weaken the crack background information.Aiming at the problem that the efficiency of sample generation is not high,the DL(Double Lightweight)module is introduced into the discriminator of CycleGAN,and group convolution and deep separable convolu-tion are used instead of the original convolutional layer to reduce the amount of convolution operation parameters of each layer and improve the quality of generated samples and the speed of network training.Experimental results show that after the small sample ex-pansion of the proposed algorithm,the speed of expanded sample training and the subsequent crack classification accuracy are im-proved.

Abstract:As one of the important components of overhead lines,insulators are used to maintain good insulation between the live body and the pole tower.However,insulators are exposed to the natural environment all year round and have a high probability of failure.In view of the problem of identifying insulator damage that needs to be solved urgently due to the damage of insulators af-fected by environmental external forces and thus affecting the safety and stability of transmission line operation,on the basis of the object detection algorithm YOLOv5s(You Only Look Once v5s),the convolution in the C3 module in the network is replaced with a multi-head self-attention layer(MHSA)to improve the algorithm's attention to the global information of the image.The attention mechanism CBAM(Convolutional Block Attention Module)is introduced to enhance the fusion of attention features of the target in the dual dimensions of channel and space.This paper replaces the loss function of the network with SIoU(Scylla Intersection over Union),introduces directionality into the cost of the loss function,and improves the training and inference performance of the algo-rithm.The experimental results show that compared with the original YOLOv5s,the improved model not only ensures the accuracy and detection speed,but also reduces the number of model parameters,increases the recall rate by 13.3%,increases the average accuracy by 7.5%,and significantly improves the recognition effect,which effectively solves the problem caused by insulator fault identification.

Abstract:Runtime verification is a technique to verify the correctness of program behavior during program operation,which has been applied to many fields.The monitor statute is used to describe properties in the runtime validation,and multiple properties can be defined in the monitor statute.However,existing runtime validation technologies have performance drawbacks.When multi-ple properties are defined in the monitor statute,existing technologies can only verify multiple properties in a serial manner,result-ing in poor runtime performance of the program and thus limiting the utility of runtime validation.Therefore,in order to improve the efficiency of runtime verification,a parallel runtime verification method based on path slices is proposed,and the existing tool MOVEC is extended to implement the above parallel runtime verification method.Moreover,experiments are also compared on the test set Mibench with MOVEC under the serial mechanism.The experimental results show that the parallel run-time verification method based on the path slice can realize the effective parallel verification of the C language program,and the performance can be improved by about 70%,achieving the purpose of making more use of computing resources and verifying the correctness of the pro-gram faster.

Abstract:Both fields of machine learning and antenna design are developing rapidly.In response to the shortcomings of tradi-tional antenna design software such as HFSS and CST,which have low design efficiency and over-reliance on the designer's experi-ence,a strategy based on the swarm intelligence algorithm is proposed to optimize the antenna design to accelerate the antenna de-sign process and find the desired parameters quickly.Tunicate swarm algorithm mainly simulates the behavior of a group of tunicate organisms in navigating and foraging,so as to quickly find the global optimal solution.The simulations are mainly performed using Matlab programming software,HFSS and the Matlab-HFSS-API toolbox.The dual U-slot antenna optimized by the algorithm has good performance and has a dual band covering both the 2.4 GHz～2.484 GHz and 5.47 GHz～5.825 GHz bands.The return loss S11 of the antenna is below-30 dB in both high and low frequency bands and can be used in the 2.45 GHz and 5 GHz bands of WLAN.

Abstract:Service migration technology brings great convenience to mobile users,it also faces many problems.In the process of service migration,illegal nodes will deploy fake base stations or gateways to eavesdrop on service data,which will lead to the leakage of users'private data and threaten user security.For the security problem of the service migration process,a transmission strategy combining fountain coding and artificial interference based on game theory will be adopted.Since idle servers that generate jamming noise will consume their own power and will not provide jamming for free,an incentive mechanism based on game theory is designed to urge jammers to help the communication link.Experiments show that the incentive mechanism can motivate the jammer to help the communication link.Finally,the security of the migration process is measured by the secrecy rate.

Abstract:Traditional text generation has shortcomings such as unstable sentiment polarity of generated short texts,one-sided semantics,and single comments.This method first uses TextRank to extract important sentences,then uses sliding windows to se-lect key sentences to form key sentence groups,then uses RoBERTa and Sentence-BERT to obtain emotional and semantic fea-tures,and finally uses GPT-2 to fuse these features to decode and generate high-quality short text.Experiments show that on the Nanfang Daily's news comment dataset,NLPCC-2017 dataset and CNewSum dataset,this paper has improved the three indicators of ROUGE.At the same time,the effectiveness of each component is verified by ablation experiments.

Abstract:Deep learning is one of the important fields of machine learning.Multi-task learning(MTL)based on convolutional neural networks(CNNs)has achieved great success in the field of computer vision.The key of multi-task learning is to learn the shared representation of multiple tasks when the structure of the model is unchanged,so that the model of multi-task learning has more generalization ability.This algorithm designs an efficient adaptive feature interaction layer.MobileNetV3 and feature interac-tion model is used to allow different task features to adaptively determine the sharing of features between tasks.Then,through the improved multi-task loss function,the loss of different tasks is weighted and balanced,so that the difference in the training efficien-cy of the multi-task training can achieve better results.The experimental results show that the precision of multi-task training is higher than that of single task training,which meets the experimental requirements.

Abstract:In this paper,an improved artificial bee colony algorithm is proposed for hybrid constrained flow shop(HCFSP)scheduling problem.Based on the fusion of tabu search algorithm,this algorithm uses NEH heuristic algorithm to generate high-quality initial solution.Because NEH does not necessarily give the shortest or optimal sequence,but it can ensure local opti-mality to a certain extent,and because of its excellent solution speed,it can be used as the initial solution set of other algorithms,that is,it can be used to generate the initial honey source space.The tabu search algorithm is integrated in the detection bee stage.By maintaining the tabu table and prohibiting local search,the problem of falling into the local optimal solution is effectively avoid-ed and the population diversity is increased.After integrating the advantages of the two algorithms,this paper verifies the improved artificial bee colony algorithm through four scale examples,and proves that the improved artificial bee colony algorithm has good convergence and robustness,In terms of convergence speed and accuracy,it has been greatly improved compared with the previous one.

Abstract:Planning a flyable path for an unmanned aerial vehicle(UAV)operating in a cluttered environment is a very typi-cal and challenging problem in the field of robotics,and there are various explorations in this field.Aiming at the problem that the RRT*algorithm does not include flight kinematic constraints and the path is rough in UAV planning,an RRT*UAV path planning optimization algorithm based on improved heuristic search is proposed.The heuristic search method combined with flight kinematics constraints is used to achieve path optimization and refinement of RRT*algorithm,and a smooth flight path of UAV is generated.The numerical optimization method based on nonlinear optimization is used to achieve high performance under the premise of guaran-teeing convergence,which significantly reduces the time of path calculation.Simulation experiments are established in Matlab,and the simulation results show that the heuristic search based on the improved RRT*has faster convergence speed and smoother path.

Abstract:In view of the problems of"data saturation"and"system noise"in parameter identification of permanent magnet synchronous motor,the recursive least square method has slow convergence speed,large fluctuation range and biased identification results.In this paper,a recursive least square method based on adaptive forgetting factors and instrumental variables is proposed by analyzing the properties of forgetting factors and instrumental variables.In this method,the error between the predicted output value and the real value of the motor model is used to construct the dynamic adaptive forgetting factor adjustment function,so as to better balance the influence of the old and new data on the identification of motor parameters.At the same time,instrumental variables are added to solve the problem that the identification results are unbiased under the influence of colored noise.The simulation results show that the algorithm has good convergence ability and stability under the adaptive forgetting factor adjustment,and the instrumen-tal variables are used to ensure that the identification results are unbiased.

Abstract:Aiming at the problems of low efficiency,large randomness and unsmooth path of the rapid expansion random tree(RRT)algorithm in UAV path planning in different environments,a dual-strategy RRT algorithm is proposed.The weight coeffi-cient is added to the random function of the traditional RRT algorithm to constrain the growth direction of the random tree,and the expansion step size is changed from fixed to adaptive.For the problem that the planned path is too long and not smooth,an optimiza-tion method of deleting redundant nodes is proposed,which effectively reduces the length of the path,and this paper uses the cubic B-spline curve method to smooth the track path,optimize the inflection points between the path nodes,and improve the smoothness of the path.The simulation results show that the path planning method of UAV based on RRT algorithm proposed in this paper has the advantages of short planning path,fast search speed and strong flightability,and it is suitable for UAV path planning require-ments in different environments.

Abstract:The existing routing algorithms usually choose the transmission path with few hops and short distance.Due to the limited energy of wireless sensor,the energy of some nodes is rapidly consumed and the nodes die,thus reducing the life cycle of the network.To solve these problems,a WSN routing method based on ant colony optimization algorithm and energy prediction is proposed.By establishing a multidimensional pheromone model of node distance,hop number and energy,the route is dynamically adjusted according to the residual energy of nodes in the process of data transmission,so as to ensure the balanced decline of node energy and prolong the life cycle of WSN network.At the sink node,the residual energy of the sensor node is predicted by the ener-gy prediction method to ensure that the path node can quickly find the optimal solution of the repair path in case of multiple self-re-pair or repair failures.The simulation results show that the network life cycle is increased by 33%on the basis of ensuring high trans-mission efficiency.

Abstract:Aiming at the problem of insufficient positioning accuracy of sensor nodes in wireless sensor networks,a node loca-tion method based on improved multi-objective particle swarm optimization algorithm for wireless sensor networks(LIMOPSO)is proposed.The multi-objective location model is improved,and the location information of the unknown node near the anchor node is used as the target guidance to improve the location efficiency.An optimal individual selection strategy based on the number of domination is proposed to enable the algorithm to quickly locate the location range of nodes and improve the location speed.The Fi-bonacci mutation strategy is proposed to improve the multi-objective particle swarm optimization algorithm to estimate the node posi-tion and improve the accuracy and stability of location.The experimental results show that the average positioning error of the im-proved multi-objective positioning model is significantly improved compared with the average positioning error of the improved mod-el.At the same time,compared with the three comparison algorithms,the positioning method proposed in this paper has higher posi-tioning accuracy and stability,and can effectively reduce the impact of the proportion of anchor nodes and communication radius on positioning.

Abstract:To address the problem of chemical gas leakage occurring in industrial production,a semantic segmentation net-work based on infrared image of leaking gas is proposed for gas leakage detection.The method is based on the semantic segmentation network SegNet,which adds a hopping connection containing an attention mechanism to fuse shallow information in the process of upsampling,and adds a pyramid pooling module at the end of downsampling to enhance the feature extraction ability,and adds a Dropout module between the encoder and decoder to increase the generalization ability of the network.The network is trained by a synthetic dataset with pixel-level annotation,and experiments are conducted on the synthetic and real datasets separately,showing that the network can accurately mark the leaking gas in the infrared image and the speed can meet the requirements of real-time de-tection.

Abstract:In order to meet the needs of intelligent recommendation of college students,a method based on user explicit and implicit preferences is designed.By establishing user portrait,the basic information of college students is summarized,and the por-trait word cloud map is generated,so as to achieve a comprehensive coverage of user explicit and implicit preferences.Neighbor-hood analysis is conducted in the word cloud map to determine the similarity of information,this paper establishes the interactive mining relationship between users and recommended information,and improves the user portrait.Then,combined with the multi-source preference and the explicit implicit preference characteristics in the portrait,the sequence is processed in the form of ontology compatibility to be recommended,and the recommendation range is expanded through neighborhood search,and the rec-ommendation results are output to realize the design of intelligent employment information recommendation model.This paper uses the bilateral matching method,adds a one-to-one and one-to-many recommendation structure,adjusts the output results,and completes the recommendation of the final employment information.The experimental results show that the hit rate of the recommen-dation method designed in this study can reach over 60%,which is superior to other recommendation methods with significant fluctu-ations.

Abstract:In recent years,with the development of cities and transportation,alleviating traffic congestion has become a hot topic.At the same time,more and more researches on intelligent traffic lights have emerged.This paper proposes a new intelligent traffic light control algorithm TaskLight,which combines the idea of multi-task learning with the original DQN framework.In addi-tion,noise mechanism is introduced into the framework of multi-task learning to balance the importance of each task in each time period.The experimental evaluation results on real data-sets and synthetic data-sets show that compared with some of the most ad-vanced algorithms,TaskLight can significantly shorten the average travel time of vehicles and increase the throughput at intersec-tions.At the same time,TaskLight can make the main network converge to a better state by learning different tasks.In addition,the design of the algorithm framework is not limited to the field of intelligent traffic lights,and theoretically it is also transferable to oth-er areas of reinforcement learning.

Abstract:Aiming at the current situation of psychological problems existing in college students,multi-source data driving is used to predict the mental health of college students.Firstly,part of college students'life data is mined,multi-source data set of col-lege students'mental health prediction is designed,and relevant features are extracted according to the data form.At the same time,aiming at the problem of too many features brought by multi-source data sets,the Relief algorithm is optimized and integrated for feature selection,and finally lightGBM classifier is used to complete efficient and accurate mental health prediction.This method is used to analyze and predict the data of students.The experimental results show that the method can realize the effective identifica-tion of students'psychological problems,so as to provide the planning and decision-making basis for the mental health education of students in universities.

Abstract:In response to the poor performance of current methods such as manual inspection and physical equipment-assisted nondestructive testing in rail sleeper crack detection,this paper proposes an improved YOLOv3 model.This paper proposes an im-proved backbone feature extraction network fused with coordinate attention mechanism,which enables it to capture the long-range dependence of spatial information and thus better locate the crack location.The RFN module proposed in this paper is able to en-hance the deep semantic information and fuse the global information of the overall network before performing feature fusion.The co-ordinate regression loss using DIoU instead of YOLOv3 makes the network converge earlier.Experiments show that the improved YO-LOv3 has better detection performance,and the improved target detection algorithm proposed in this paper is tested on the same rail crack dataset for both,in which the average accuracy is improved by about 5.2%on average,the precision is improved by about 2.5%,and the recall is improved by about 7.2%.The ablation experiments designed in this paper show that the use of SPP module as well as RFN module improves the detection accuracy of the model,and the use of DIoU loss function not only accelerates the con-vergence of the network,but also provides higher recall and detection accuracy.Overall,the model is able to perform the task of rail sleeper crack detection with high speed and high accuracy.