Abstract:In response to the increasing demand for scientific exploration in proximity space, a temperature sensor for nearspace sounding instruments has been devised, predicated upon miniature bead-shaped thermistors. This methodology encompasses several pivotal stages: Primarily, the Computational Fluid Dynamics technique was enlisted to simulate and quantify the solar radiation error entailing the sensor probe. Subsequently, a backpropagation network and optimized through a genetic algorithm-based backpropagation neural network were employed to train on the accumulated dataset, thus comparing and facilitating the construction of a predictive model. Furthermore, a low pressure wind tunnel experimental platform was erected to emulate conditions reminiscent of those in the near-space milieu. This permitted the evaluation of solar radiation errors under diverse parameter configurations. The obtained test results were then compared with the data output from the predictive model to validate the precision of the sensor′s measurements. The experiments revealed that the average measurement error of the sensor probe was 0.007 3 K, with a root mean square error of 0.009 8 K.

Abstract:Aiming at the problems such as excessive impact of bus voltage and unsynchronized phase of multiple PV inverters during the blackstart process in the solarstorage farm, the structure of solar-storage microgrid and micro-source control mode are studied. An improved V/f control strategy with variable d-axis bus voltage reference value is proposed to reduce the bus voltage impact during black start process. Based on the grid-forming virtual oscillator, the self-synchronization control topology of multiple parallel inverters with three-phase circuit are developed. An independent microgrid model is established. The tests of the bus boost, the PV inverter phase synchronization and the microgrid power balance are designed under different working conditions to verify the effectiveness of the control strategies. The results show that based on the designed control strategy and the structure, the bus voltage surge during the black start process can be reduced to within 2% of the rated voltage from original 5%. The unsynchronization degree of parallel inverters is also reduced. The maximum phase difference of multiple inverters is decreased to less than 1%; and the energy storage unit can quickly respond to photovoltaic power and load changes within 0.03 s. The black start process can be smoothly realized.

Abstract:Aiming at the problems of poor robustness of fatigue driving detection system and too simple classification of fatigue degree, use MediaPipe face key point detection technology and a fuzzy inference system to fuse a variety of facial fatigue features to study the quantitative assessment method of driver fatigue degree in video sequences and realize realtime scoring of driver fatigue degree and fatigue warning. In this paper, firstly, the MediaPipe face detection model is used to locate the facial key points; secondly, the detected key points are used to extract the dynamic features of facial fatigue from the video frames, and four evaluation indexes are obtained: PERCLOS, yawn length, whether or not to nod off, and the approximate entropy of face oscillation; finally, the fuzzy inference system is designed to quantify the fatigue degree and realize the real-time evaluation of driver fatigue. The proposed method shows that the proposed method is scientifically sound. The study shows that the proposed method scientifically and effectively achieves the quantitative assessment of driver fatigue degree and further improves the robustness and reliability of fatigue driving detection based on facial features.

Abstract:In order to realize a fully coherent shore-air bistatic radar, time-frequency synchronization between transmitting and receiving stations must be realized. The time-frequency synchronization of the coherent shore-air bistatic radar contains two parts. One part is time synchronization, that is to ensure that the transmitted signal and the local vibration signal are generated at the same time and have the same sequence. The other part is the synchronous clock source, which ensures that the DDS generating the transmitted signal and the local vibration signal has a reference clock with the same frequency and phase. Article puts forward a clock synchronization scheme that taming local high stable constant temperature crystal with satellite synchronous pulse, this scheme uses the GPS clock chip can output two signals at the same time, one output time pulse to synchronize the time between double station, the other output MHz time pulse tame local high stable constant temperature crystals in order to realize the synchronous clock source. Time synchronization makes the timing error of the double-ended signal less than 1 μs required by the system index. The synchronous clock source to tame after the local clock has less than 22 ns synchronization precision, close to the timing precision of GPS chip 20 ns, compared with the phase detection in FPGA with PPS, improves the frequency of the phase detection, therefore has the shorter tame time, and does not depend on the FPGA chip, make the system modular while saving cost.

Abstract:In order to obtain multimode and multichannel remote sensing images of remote sensing satellite cameras, this paper proposes the design and implementation method of spaceborne multi-mode and multi-channel image data acquisition and processing system, which can collect and transmit multi-channel image data of satellites in conventional mode and deorbit mode to the storage system, which solves the problem that the existing on-board storage unit cannot store multi-channel data in parallel. A multi-channel image acquisition module compatible with conventional mode and down-orbit mode is designed, and the collected image data is processed in multiple modes, multi-channel image normalization processing is realized through the polling scheduling algorithm, and finally the processed image data is transmitted to the storage system through the Aurora64/66B transmission protocol, completing the efficient acquisition and processing of data. After experimental testing, the system data acquisition transmission rate can reach 9.2 Gb/s, and the collected data is accurate and zero bit error. The system is stable and reliable, with good real-time performance, and meets the actual satellite use needs.

Abstract:In order to solve the problems of light reflection on the surface of metal cabinet and the limitation of detection site, this paper studies the 3D reconstruction of metal cabinet based on depth camera, and uses the depth camera to collect the RGBD images of metal cabinet to achieve high quality data acquisition on the surface of metal cabinet. After calculating the point cloud data, outlier elimination and downsampling are carried out to reduce the influence of noise and improve the computing speed. The three-dimensional model of metal cabinet is constructed by using the reconstruction method based on point cloud data and three-dimensional information. Through comparison experiments, it is proved that the point cloud data obtained by the method in this paper is obviously superior to the binocular vision method in terms of quantity, average error of precision and acquisition speed. Compared with the real values, the error percentage of the three metal cabinet models is about 3%.

Abstract:Aiming at the problems of early blind search, slow convergence and easy to fall into local optimum in the traditional ant colony algorithm for UAV 3D path planning, an improved ant colony algorithm is proposed in this paper. The algorithm uses spatial location to initialize the pheromone distribution and set a concentration threshold, which enhances the directionality of the early search of the ant colony and avoids the algorithm from falling into the local optimum. The heuristic function which takes into account both distance and direction factors is designed to improve the quality of path planning. The adaptive volatility factor is used to control the volatility of the pheromone, which improves the convergence speed of the algorithm. Compared with the traditional algorithm, two experiments show that the proposed algorithm reduces the average path length by 18.6%, the average iteration times by 63.3% and 78.7%, and the average corner times by 62.5% and 42.3%, respectively.

Abstract:This paper investigates the problem of anti-jamming communications with intelligent frequency hopping in complex electromagnetic environment. Essentially, this paper proposes a new mixed deep recurrent Q-learning network (MixDRQN) for reinforcement learning (RL) of the optimal antijamming strategy. The proposed deep RL algorithm effectively combines double deep Q-learning network(DoubleDQN) and dueling deep Q-learning network(DuelingDQN), and further introduces long short-term memory (LSTM) layer for preprocessing the time-sensitive inputs. With the use of DoubleDQN, the proposed RL algorithm solves the problem of Q-value over-estimation caused by ε-greedy algorithm. In the mean time, the use of DuelingDQN and the LSTM layer has been proved to be very efficient for learning the time-correlated feature of inputs. Extensive experimental results show that both the convergence speed and anti-jamming performance are significantly improved, and in particular, the convergence speed of the proposed RL algorithm is more than 8 times higher than that of the existing RL algorithms.

Abstract:In order to improve the safety of the point-to-point operation of unmanned aerial vehicles(UAV), this paper proposes a low risk path planning strategy. Low risk path planning strategy includes risk assessment and path planning. In the part of risk assessment, the risk value assessment model and risk grade assessment model are established by taking the risk of UAV to pedestrians on the ground as the assessment index, and the risk map is taken as the result to represent the safety degree of UAV operation. In the path planning part, combined with the unique background and characteristics of the risk model, an improved rapidly-exploring random tree* algorithm is proposed. This algorithm uses the strategies of improved collision detection and priority exploration to explore the low-risk path more effectively. Finally, the simulation results show that the low-risk path planning strategy can effectively plan the operation path to avoid the high-risk area. The risk value is reduced by 13.16% compared with the straight-line flight, and the planning time is reduced by 60.63% and 12.20% compared with the contrast algorithm. Therefore, this planning strategy has practical value and can improve the safety of UAV operations.

Abstract:Accurately predicting changes in the fire environment helps to accurately grasp the development trend of the fire and ensure the safety of personnel. Due to the coexistence of multiple parameters of the fire scene environment, the complex coupling relationship, and the time series and nonlinearity, it is difficult to establish an accurate prediction model. Therefore, this paper proposes a long-term and short-term memory network model of self-attention mechanism based on the improved Harris Hawk algorithm, which realizes the accurate prediction of the fire scene environment data. Firstly, the logistic mapping strategy, cosine weighting factor, and Gaussian perturbation strategy are introduced into the Harris Hawk optimization algorithm to enrich its population diversity, balance its global exploration and local development capabilities, and improve its convergence accuracy. Then, the improved Harris Hawk optimization algorithm is used to optimize the hyperparameter in the self-attention mechanism short-term memory network model, and the fire environment is predicted based on the optimized parameters. The simulation results show that the self-attention mechanism based on the improved Harris Hawk optimization algorithm has better long-term memory network model fitting effect and higher prediction accuracy.

Abstract:Aiming at the problem of declining packet delivery rate (PDR) and sharp increase in energy consumption of large-scale LoRa networks, a LoRa spreading factor (SF) and coding rate (CR) allocation method based on weighted utility function is proposed. Firstly, the coupling influence relationship between SF and CR on PDR and energy consumption is established. Then, the utility functions of PDR and energy consumption are established based on the reward and punishment mechanism, and their weighted values are taken as the objective function, and the greedy algorithm is used to jointly allocate the SF and CR. Finally, the optimal weight coefficient is determined by traversing the search. The simulation results show that compared with the latest CA-ADR algorithm, the proposed algorithm obtains a 21.2% increasment in terms of the average PDR, and a 165% increasement of the average energy efficiency. The algorithm in this paper effectively improves the communication reliability and life cycle of terminal devices in LoRa networks, and has high application value in many scenarios such as smart cities and smart agriculture.

Abstract:Aiming at the defects of low accuracy, low reliability and large error of ultrasonic flowmeter in low velocity fluid measurement, a four-channel gas ultrasonic flow measurement technology based on DSP is proposed. Firstly, according to the measurement principle of ultrasonic flowmeter, the time difference method is selected as the flow measurement algorithm.Secondly, the acoustic channel layout scheme of the ultrasonic flowmeter is determined, and the weighting coefficient of the instantaneous flow is calculated by using the mathematical integration method. TI′s TMS320F28335 DSP chip is selected to complete the system hardware design. Finally, the received signal is filtered to simulate the flow in the pipeline under static and quantitative conditions.The experimental results show that the average measurement errors of each acoustic channel in the static flow measurement of pipeline are 0.142%, 0.217%, 0.273 % and 0.362%, and the error is less than ±0.4 %, which meets the standard requirements. In the quantitative measurement of low flow rate, the average error between the prototype and the standard meter is 0.945%, and the experimental data show that the prototype detection error decreases with the increase of gas flow rate, which also meets the industrial standards and meets the design requirements.

Abstract:UAV aided wireless communication has the advantages of flexible deployment and large coverage, but it still has the disadvantages of large path loss, low throughput and transmission capacity. Beamforming technology is introduced into UAV aided wireless communication, which can effectively compensate the communication path loss and mitigate intra cell or inter cell interference. Therefore, this paper constructs a UAV aided wireless communication system model, which is equipped with a uniform plane array on the ground base station and UAV to achieve beamforming, so as to maximize the system capacity. For the nonconvex problem with high dimensional and highly coupled optimization variables, an efficient iterative algorithm is proposed to jointly optimize the position of the relay UAV and beamforming vector. Firstly, the original problem is transformed into two subproblems of the position optimization of the relay UAV and beamforming vector optimization by the block coordinate descent method, and then the two subproblems are transformed into a convex optimization problem by the successive convex approximation algorithm. Experimental results show that the proposed optimization theory and algorithm have good convergence performance and can effectively improve system capacity.

Abstract:In this study, the traffic cone used to construct the temporary road is taken as the research objective, and the three-dimensional point cloud data of the temporary road collected by multi-line LiDAR is taken as the input. A graph neural network model based on graph theory is proposed, which can realize the segmentation of point cloud data and improve the learning effect of the model on the disordered point cloud data. Take the driverless formula car as the experimental platform, train and test for traffic cone, the experimental results show that the segmentation accuracy of the graph neural network model reaches 88.6%, which is about 10% higher than that of the PointNet model. In addition, the model also has a certain generalization ability under sparse LiDAR point cloud data, and has good applicability.

Abstract:Aiming at the problem of low fault diagnosis accuracy of charging piles, this paper first proposes to use multi-dimensional scale analysis method to process sample data, map the original data to a lower dimensional space, and reduce the cost of model calculation. Secondly, Sin chaos mapping and dynamic adaptive weighting are integrated into the sparrow algorithm to improve its global search ability and optimization accuracy, and then the improved sparrow algorithm is used to optimize the parameters of the support vector machine model, and the optimal diagnosis model is established. Finally, the obtained model is used to diagnose the fault diagnosis of the charging pile and output the diagnosis results. The final experimental results show that the diagnostic accuracy of the fault-diagnosis model of the charging pile proposed in this paper is as high as 95.135 1%, which is significantly higher than that of some existing commonly used models. At the same time, the support vector machine model selected in this paper has better effect and higher efficiency than other classification models.

Abstract:Phase denoising is a key technology for digital speckle pattern interferometry, but the existing denoising methods represented by sinecosine mean filtering and window Fourier transform filtering cannot fully meet the requirements in terms of phase fidelity, adaptive noise reduction, and ease of operation. In this article, a new adaptive denoising method is proposed. The method estimates the noise variance of a raw phase map at the first, and then performs the sinecosine transformation of the phase map to obtain two phase maps. The two phase maps are then smoothed respectively by using several layers of wavelet threshold denoising and non-local mean filtering. The two phase maps are subjected to arctangent operation and their noise variances are estimated again. The above-mentioned denoising operations are iteratively performed according to the criterion of image noise variance to realize adaptive noise reduction of the phase maps. Experimental results show that compared with the traditional sin-cosine mean filtering, The noise variance of the proposed method is reduced by 0.38, the sum of L operators is reduced by 0.2, and the SSIM is increased by 0.16. Meanwhile, the difference of image information entropy is only 0.1. This method can effectively suppress the coherent noise in the phase maps, preserve the phase edge information, and avoid phase distortion or noise residue caused by inappropriate filtering cycles.

Abstract:The randomness of sample selection during the construction of traditional random forest model leads to a large number of decision tree classifiers with low classification accuracy and similar classification performance in random forest, which affects the accuracy and efficiency of the overall random forest model classification. In order to improve the accuracy and efficiency of random forest model in point cloud classification, a random forest algorithm based on feature importance weighted voting was proposed. Firstly, decision trees with low classification accuracy and similar classification performance are eliminated from the aspects of classification accuracy and inconsistency measurement of decision trees. Secondly, the voting weight of each decision tree is calculated based on the similarity between random forest and decision tree feature importance. In this paper, three sets of densely matched point clouds are taken as examples to compare the improved stochastic forest classification model with the traditional stochastic forest, support vector machine classifier (SVM), neural network and decision tree. The experiments show that the improved random forest classification algorithm is 0.20%, 15.159%, 5.893%, 6.316% and 28.935% higher than the traditional random forest, support vector machine, decision tree, neural network and point-based feature classification method, respectively. In terms of classification efficiency, the improved random forest classification algorithm takes about 75% less time than the traditional random forest.

Abstract:When the oil immersed transformer fails, it will cause abnormal vibration of the iron core and winding, excessive temperature rises and transmission to the box. In order to simultaneously monitor the vibration and temperature status of transformers during operation, an RFID sensor based on dual antennas is designed, which integrates temperature and acceleration sensor chips. In addition, an "L" shaped metal reflector is added to enhance the radiation efficiency. The accuracy of the RFID sensor was tested in the laboratory, and the results showed that the test errors for vibration and temperature were only 0.7% and 1.6%. Finally, OFPSZ-150000/220 transformer is taken as the experimental object for field monitoring. The results show that the vibration signals of the transformer box monitored by RFID sensors are mainly concentrated in the range of 100~500 Hz with 100 Hz as the fundamental frequency; The temperature of the box is higher when it is closer to the middle and lower when it is closer to the edge. The experimental results show that the measured data is very close to the field data, the measurement errors of acceleration and temperature are only 6.25% and 1.9% which proves the reliability of the RFID sensor.

Abstract:The application of Beidou three-generation satellite technology in short-message communication and positioning technology in power is becoming more and more mature, but the related research on using Beidou short-messages for power emergency communication image transmission is not perfect, especially after the images are transmitted through Beidou short-messages. The resolution of the distortion problem is not ideal. This paper proposes a progressive transmission image recovery algorithm for Beidou short messages. First, taking the JPEG2000 image compression algorithm as an example, the characteristics of the progressive image transmission technology and Beidou short messages are analyzed, and then the JPEGRE image recovery algorithm is proposed. The peak signal-to-noise ratio and structural similarity parameters are used to reflect the performance of the algorithm. Finally, the experimental method proves that the JPEGRE image restoration algorithm has better restoration ability than other popular algorithms. The purpose of this research is to improve the degree of image recovery after lossy compression, and to make a better emergency plan when assisting power emergencies.

Abstract:A defect detection method for aluminum-plastic blister capsule packaging based on GoogLeNet network model is proposed in order to address the poor detection effect brought on by the color, size, and picture noise of aluminum-plastic blister capsules. To locate the Plate batch number region to be detected, the normalized product correlation gray scale matching method is first used. Next, the drug plate′s capsule blister region is divided by the improved gray value projection method, and the dataset for aluminum-plastic blister capsules is created. Finally, the improved GoogLeNet network model is trained and tested to realize the defect recognition of the missing grain, concave cap, and other defects. The experimental findings demonstrate that the improved horizontal-vertical projection method achieves 100% segmentation accuracy for capsule blister area, and the recall rate of network for defect recognition is over 98.64%. The improved gray value projection technique offers excellent segmentation capabilities and strong robustness. The improved network, which can be used for the quality inspection of aluminum-plastic blister tablet packaging, has considerably increased the accuracy of fault identification of aluminum-plastic blister capsule pharmaceutical board packaging when compared to previous methods.

Abstract:In view of the problem that the traditional insulation state detection methods of transformers and medium-voltage switchgear rely on manual labor, this paper based on the audible sound recognition method, by mixing the discharge fault sound of power equipment with the sound of normal conditions and environmental noise to make a sample set, in order to simulate the real operating environment of power equipment. After the fault sound is preprocessed, the spectrogram is used to extract the short-time frequency and energy distribution features of the sound, and the spectrogram data set is constructed. Combined with the improved convolutional neural network, the discharge fault detection is realized. By adding the attention mechanism, adjusting the exponential decay learning rate, the number of data set samples, the audio sampling rate and other ways to further improve the accuracy of the network model, the final design of the network model identification accuracy up to 99.2%, compared with other detection methods have obvious advantages, can realize the online detection of discharge faults.

Abstract:The reciprocating compressor valves are prone to fail during operation. From the perspective of the valve time frequency images analysis, it is proposed that fault diagnosis method based on GLCM-HOG and WOA-ELM for reciprocating compressor valves. First, the vibration signal of each operating valve is processed by wavelet generation time frequency images. The GLCM and the HOG were used to extract the time frequency image features of the valve, and fused to form GLCM-HOG features. Then, the WOA is used to optimize the ELM model for input layer node weight and hidden layer node and the valve fault diagnosis model is constructed. Finally, the GLCM features and GLCM-HOG features are fed into the WOA-ELM model to demonstrate the effectiveness and superiority of the proposed method for the diagnosis of reciprocating compressor valve fault. The experimental results show that compared with the GLCM features, the constructed GLCM-HOG features can accurately and comprehensively reflect the valve time frequency image features. The WOA-ELM model diagnoses valve failure with higher accuracy.

Abstract:In order to improve the accuracy of abnormal sound recognition of highway tunnel accident and to solve the problem that convolutional neural networks only pay attention to local information. An integrated voice recognition model based on CNN-RNN is proposed. The model used the Stacking integration strategy to combine the strong feature expression ability of CNN and the strong memory ability of RNN.The gated cyclic memory unit was used to reduce the computational complexity of RNN. SIREN sinusoidal periodic function was used as the implicit activation function of RNN to enhance the fitting ability of the model to sound data. The precision of multi-channel convolution refinement feature extraction was designed to achieve global feature extraction. The performance of the proposed sound recognition model was evaluated on the abnormal sound data set. Experimental results show that the proposed sound model has higher recognition performance than other models and is more robust, which can effectively identify the abnormal sound of highway tunnel accidents.

Abstract:In order to solve the problem of weak fault feature extraction in one-dimensional vibration signal of rolling bearing; In order to solve the problem that the deepening of deep learning model layer is easy to lead to the disappearance of gradient or the deterioration of gradient explosion, which leads to the low accuracy and poor robustness of fault diagnosis, this paper proposes a rolling bearing fault diagnosis method based on EMD-GAF and improved SERE-DenseNet. One-dimensional vibration signals of rolling bearings were decomposed and reconstructed by EMD after rolling sampling, and the reconstructed one-dimensional signals were converted into two-dimensional images by GAF as model input. DenseNet121 was selected as the main task in terms of model, and SERE module was introduced. The Dense Layer with 2 convolution layers is improved into 3 sparse modules with base number of 8. Feature extraction and fault classification are carried out by using 2D image as input. The bearing data set of Case Western Reserve University was used for simulation experiments. The experimental results show that the proposed method can accurately diagnose rolling bearings, with the maximum accuracy of 100% and the average accuracy of 99.91% in 10 experiments. Compared with the common deep learning model, the proposed method has great advantages. The fault diagnosis accuracy is 96.48% when the signal to noise ratio is 10 dB, and the proposed method has strong robustness.

Abstract:In recent years, with the development of artificial intelligence technology, deep neural network has been widely used in intelligent manufacturing. This paper combines deep neural network with aircraft deformation prediction, proposes a prediction method of aircraft segment deformation based on graph convolution and multi-mode. In the deformation analysis of aircraft segment, the model extracts the features of aircraft structure mode and working condition mode respectively, and fusion at the decision-making level. When extract features from aircraft segment structure data, the aircraft structure data is in point cloud format and has the characteristics of non-Euclidean data, this paper introduce the graph convolution. Based on ModelNet40 and real aircraft segment working condition data, construct aircraft segment deformation dataset deformation dataset including four aircraft segments, and experiments are conducted on this dataset. The experimental results show that the prediction mean square error of this method is 0.188, and get the best prediction in the nose segment of the aircraft, which can effectively predict the deformation of aircraft segments.

Abstract:For the presence of small targets and a large number of long bar-shaped objects in urban streetscape datasets, segmentation is difficult, and although current networks with coding and decoding structures can refine segmentation results, most of them do not make full use of spatial and contextual information, so this paper proposes a semantic segmentation algorithm based on pixel attention feature fusion. Firstly, using ResNet50 as the backbone network, the initial feature fusion is carried out using the null space convolutional pooling pyramid and strip pooling to obtain multi-scale features while circumventing useless information; then the pixel fusion attention module is used to aggregate contextual information and recover spatial information, and finally the attention feature refinement module is used to eliminate redundant information. The algorithm was experimented on the CamVid dataset and the results showed that the algorithm was able to achieve 75.22% mIoU on the validation set and 67.21% on the test set. This is an improvement of 2.51% and 2.86% respectively compared to the DeepLabv3+ network.

Abstract:At present, significant breakthrough has been made in the detection of significant objects of transmission lines, but there are still limitations in predicting the "integrity" of significant areas, and it is difficult to fully identify and locate the defects of insulator strings on transmission lines. In this paper, integrity awareness network is used to detect insulator strings on transmission lines. First, feature aggregation module is used to extract features at different levels. Second, integrity enhancement module is used to highlight significant target channels and suppress other interference channels. Finally, part whole inspection module is used to determine whether there is a strong consistency between parts and the whole of target features, which can improve the recognition accuracy of defective insulator strings. Through subjective and objective comparison between the algorithm in this paper and the three popular algorithms currently disclosed, it is found that the algorithm in this paper has more advantages in the significance detection when the insulator string and background are highly integrated.